Authors: Inge Timmers, Emma E. Biggs, Lauren C. Heathcote, Mats Fredrikson, Daniel S. Pine, Johan W. S. Vlaeyen, David Borsook, Laura E. Simons
Categories: Article, Human behaviour, Learning and memory
Source: Communications Psychology
Authors: Inge Timmers, Emma E. Biggs, Lauren C. Heathcote, Mats Fredrikson, Daniel S. Pine, Johan W. S. Vlaeyen, David Borsook, Laura E. Simons
Exposure therapy for the treatment of pain-related disability relies on extinction learning, forming new safety memories inhibiting fear expression. However, fear often returns. The behavioral memory updating hypothesis posits that a fear memory can be ‘updated’ to a safe memory while in a malleable state, preventing return-of-fear. To test this hypothesis, 78 adolescents with and without chronic pain (age: Mean=15 y, range=10-24 y) were recruited for a two-day neuroimaging study. Due to incomplete data/excess motion, 55 participants (pain=38; pain-free=17) were included in MRI data analysis. Participants underwent a fear conditioning protocol with a within-subjects ‘updating’ one CS+ (CS+Reminded [CS+R]) was reactivated to achieve a malleable state before extinction, while a second (CS+Not Reminded [CS+NR]) was not. We observed significantly less functional connectivity between the ventromedial prefrontal cortex and the amygdala for the CS+R and CS- compared to the CS+NR, consistent with the purported change in neural circuitry underlying the ‘updating’ effect, however observed no credible difference in fear ratings between the CS+R and CS+NR. This discrepancy may be crucial to understanding the mixed findings in the field and indicates that while some form of ‘updating’ may occur, it may be insufficient to reduce reported fear.
Decreasing maladaptive avoidance by reducing pain-related fear represents one core feature of exposure-based treatments for chronic pain. While effective in the short-term, pain-related fear and the subsequent avoidance behaviors often return^1,2^, suggesting that the extinction learning that took place is not persistent, and highlighting a need to obtain more long-lasting and robust effects. Basic science research suggests that this may be achieved by interfering with memory reconsolidation^3–5^ and updating an active threat-related memory into a safety memory^6–8^, as opposed to the competition between a threat-related and safety memory that is learned during extinction. Experimentally, this has been achieved using classical conditioning procedures, where a cue (conditioned stimulus, CS) is paired with an aversive outcome (unconditioned stimulus, US), establishing a CS-US memory, and evoking a conditioned response (CR). Following consolidation, the CS-US memory is reactivated by presenting the CS without the US, purportedly eliciting a prediction error and destabilizing the memory^9,10^. Subsequent extinction training repeatedly presents the CS without the US, ‘updating’ the CS-US memory with a CS-noUS memory. Consequently, the CS no longer evokes the conditioned fear response at return-of-fear tests. The crucial addition of reactivation differentiates this procedure from traditional extinction learning, where the absence of a labile state for the CS-US memory results in the formation of a CS-noUS memory, competing with the original CS-US memory at return-of-fear tests^11^.
On a neural level, the competition between the CS-US and CS-noUS memory trace reduces conditioned fear responding through inhibition of the amygdala by the ventromedial prefrontal cortex (vmPFC)^12–14^. However, preliminary evidence suggests that this inhibitory connection is unnecessary in post-retrieval amnesia which relies on the amygdala^15^. While there is evidence for the behavioral memory updating hypothesis in non-human animal research^6^, findings in humans remain inconsistent^16–19^, and the boundary conditions for post-retrieval amnesia remain unclear^19–24^. Understanding when and how behavioral memory updating may enhance exposure therapy is especially valuable in pediatric chronic pain. Adolescents with chronic pain demonstrate resistance to extinction learning, an effect that increases with higher pain catastrophizing^25^. Given that extinction learning is the key mechanism in exposure therapy, enhancing extinction learning processes through behavioral memory updating could improve treatments of pain-related disability among adolescents with chronic pain.
Presently, we examine post-retrieval amnestic effects among adolescents with and without chronic pain using a transdiagnostic aversive conditioning approach using a developmentally-appropriate Threat-Safety discrimination paradigm^26,27^. Based on the behavioral memory updating hypothesis, we anticipate observing post-retrieval amnesia for a reactivated CS+ (CS+R), operationalized as less return-of-fear following reinstatement for the CS+R compared to a non-reactivated CS+ (CS+NR) (Hypothesis 1a). Furthermore, we expect a decrease in fear for the CS+NR and no change in fear for the CS+R during subsequent re-extinction (Hypothesis 1b). Based on previous findings, we anticipate pain catastrophizing to be related to the degree of extinction, and individuals with chronic pain to show more persistent differential responding to the CSs+ versus the CS- after extinction, compared to their pain-free peers (Hypothesis 2), and we will examine whether pain catastrophizing is related to post-retrieval amnestic effects. We hypothesize that less connectivity will be observed between the vmPFC and amygdala for the CS+R compared to the CS+NR during re-extinction (Hypothesis 3). Finally, we will also explore changes in functional connectivity between secondary regions-of-interest (hippocampus, nucleus accumbens, and anterior insula), that have previously shown altered connectivity related to pain catastrophizing during threat/safety learning^25^.
Youth with chronic pain (pain duration >3 months) were recruited from a pediatric pain management clinic at Stanford Children’s Health when they presented for multidisciplinary evaluation and treatment for their pain. Participants between the ages of 10 and 24 were included, in line with some definitions of adolescence^28^. We excluded participants taking opioid or antipsychotic medications, with significant cognitive impairment, or significant psychiatric conditions (e.g., active suicidality, eating disorder). Participants were not excluded for comorbid anxiety or depression, nor for taking selective serotonin reuptake inhibitors. Pain-free peers (i.e., no current or history (>3 months) of chronic pain) were recruited through advertisements, with the same inclusion and exclusion criteria as for youth with chronic pain.
A total of 78 participants were recruited (n = 54 chronic pain, n = 24 pain-free peers). Participants were excluded from the final MRI data analysis due to incomplete MRI data collection (n = 7 chronic pain, n = 2 pain-free peers), technical errors during data collection (n = 3 chronic pain, n = 2 pain-free peers), excessive motion (>6 mm/degrees framewise displacement) during MRI data collection (n = 5 chronic pain, n = 3 pain-free peers), or incidental MRI findings (i.e., abnormalities of potential clinical significance, n = 1 chronic pain). Participants who were excluded from the MRI analyses were significantly younger than the included participants, t(76) = –3.09, p = 0.003, Meanexcl = 13 years, Meanincl = 15 years, and the ratio of female to male participants was equal in the excluded sample (50% female, 50% male), whereas the included sample was predominantly female (76% female, 24% male), Χ^2^ = 5.09, p = 0.024. There were no credible differences in race, Χ^2^ = 6.15, p = 0.188, or ethnicity, Χ^2^ = 2.94, p = 0.230, between excluded and included participants. Levels of pain catastrophizing did not differ between the excluded and included samples, t(71) = –0.64, p = 0.524, however the individuals with chronic pain who were excluded had significantly shorter pain durations than those included, t(47) = –2.88, p = 0.006, Meanexcl = 13 months, Meanincl = 57 months. The final sample for MRI data analysis consisted of 38 youth with chronic pain and 17 pain-free peers. Demographics for the full sample can be found in Table 1, and demographics for the MRI sample can be found in Table S1 and Fig. S1.Table 1Sample characteristicsChronic pain n = 54Pain-free n = 24Chronic pain population (n = *2713)Age (years), M ± SD14.35 ± 2.3315.91 ± 4.6713.91 *± 3.00Sex, n (%) Female36 (67%)17 (71%)1953 (72%) Male17 (31%)7 (29%)759 (28%) Other0 (0%)0 (0%)2 ( < 1%) Unknown1 (2%)0 (0%)1 ( < *1%)*Pubertal development, M ± SD2.99 ± 1.053.04 ± 1.05Ethnicity, n (%) Hispanic or Latinx8 (15%)2 (8%)508 (19%) Not Hispanic or Latinx28 (52%)12 (50%)1688 (62%) Decline to answer18 (33%)10 (42%)167 (6%) Unknown0 (0%)0 (0%)341 (13%)Race, n (%) American Indian/Alaska Native1 (2%)0 (0%)12 ( < 1%) Asian5 (9%)7 (29%)240 (9%) Black or African American1 (2%)0 (0%)64 (2%) Native Hawaiian or Pacific Islander0 (0%)0 (0%)12 ( <1%) White28 (52%)6 (25%)1423 (53%) Multiracial3 (6%)3 (13%)0 (0%) Other451 (17%) Unknown14 (26%)7 (29%)329 (12%) Decline to answer2 (4%)1 (4%)*173 (6%)*Pain catastrophizing, M ± SD23.38 ± 11.7612.96 ± 8.00Pain duration (months), M ± SD45.56 ± 51.39Pain type, n (%) Neuropathic6 (11%) Musculoskeletal37 (69%) Visceral9 (4%) Unknown2 (4%)In order to evaluate whether the samples tested are representative of the population that presents for pain care in a tertiary pain setting, demographics for these study samples are presented alongside data obtained from the initial evaluations conducted at the Pediatric Pain Management Clinic (PPMC) at Lucile Packard Children’s Hospital Stanford between September 2019 and June 2022 (population pool).M mean, SD standard deviation.
The Screaming Lady paradigm is an age-appropriate Threat-Safety discrimination paradigm^26,27^ with neutral female faces serving as conditioned (threat-safety) stimuli (CSs) and a scared face paired with a loud (95 decibels) aversive scream serving as the unconditioned (threat) stimulus (US; see Fig. S2). The screaming lady paradigm provides a more generalizable and developmentally sensitive framework for examining fear learning mechanisms that underlie avoidance behaviors, including those relevant to pain. This paradigm allowed us to explore how youth differentiate between threatening and safe stimuli, which parallels the core cognitive and emotional processes involved in pain-related fear without requiring the direct induction of pain. By using non-pain-related stimuli, we ensured the experimental procedure was less distressing for participants and aligns with ethical standards for research with youth. Face and CS type were counterbalanced across participants.
Pre-acquisition (PRE). During the pre-acquisition phase participants were presented with three CSs (4 times each), neither of which was paired with the US.
Acquisition (ACQ). Immediately following PRE, the three CSs were presented again (10 times each) with two of the CSs co-terminating with the US on 80% of the trials (CS+R & CS+NR) and the third CS (CS-) never paired with the US.
Reactivation (REACT). After a 1-h break, participants were presented twice with the CS+R, in the absence of the US.
Extinction (EXT). The subsequent extinction phase began 10 min after reactivation and included 16 presentations of the CS+R, and 18 presentations of the CS+NR, and CS-, with no US presentations.
Reinstatement (REIN). Return-of-fear tests took place during visit two, which included MRI data acquisition and began with a test of fear reinstatement. Participants were presented with the US four times, un-cued and without any CS presentation. For the test of reinstatement, each CS was presented once without reinforcement.
Re-extinction (RE-EXT). Immediately following the test of reinstatement, participants repeated EXT training, consisting of 15 presentations of each of CS+R, CS+NR, and CS-, in the absence of the US (across two runs; 7 presentations in early RE-EXT and 8 in late RE-EXT).
The study was approved by Stanford University Institutional Review Board (#38432). The study was divided into two visits. In visit one participants and legal guardians provided written assent/consent, completed surveys, and participants underwent the acquisition, reactivation, and extinction phases of the Screaming Lady paradigm. Session 2 took place one week later (mean=7 days, median=7 days, range 1–27 days) later in the MRI scanner and consisted of the reinstatement and re-extinction phases of the Screaming Lady paradigm. The paradigm was administered using E-prime 2.0 and 3.0. There was no preregistration of this study.
Demographics. Participants self-reported age and sex, while pain duration (months) and pain type were derived from medical records.
Pain Catastrophizing Scale for Children (PCS-C). The PCS-C is a 13-item questionnaire for children older than 9 years, that assesses catastrophic thinking about pain, including rumination, magnification, and helplessness, with higher scores indicating greater catastrophizing^29^. The PCS-C showed high internal consistency in the current sample (Cronbach’s α = 0.92).
After each phase of the Screaming Lady paradigm, participants were asked to provide ratings via a 11-point numerical rating scale (0–10) for fear of the CSs, participants were asked “How scared are you of this woman?”, with scale anchors of “not scared” to “extremely scared”.
MR data acquisition. Data were acquired on a 3T GE Premier MR 750 system (GE Healthcare, Richard M. Lucas Center for Imaging) using a 32-channel Nova head coil. For the functional images, a T2* simultaneous multi-slice (SMS) EPI sequence was used to acquire 45 axial slices (3 mm isotropic) covering the entire cortical volume, using the following repetition time (TR) = 1.11 s, echo time (TE) = 30 ms, flip angle (FA) = 70°, SMS factor = 3, field of view = 228 × 228 mm. In total, 123 functional volumes were collected for the first single CS presentation (test of reinstatement), 489 volumes for block 1 of the RE-EXT phase (early RE-EXT), and 552 volumes for block 2 RE-EXT phase (late RE-EXT). Prior to functional images, an ASSET calibration scan and a higher-order shimming protocol were used to measure coil sensitivity profiles and field inhomogeneities, and correct gradients accordingly. Structural T1-weighted anatomical images were acquired using a standard GE 3D BRAVO sequence, an IR-prep, fast 3D spoiled gradient-recalled (SPGR) sequence, using the following TR = 8.6 ms, TE = 3.4 ms, FA = 15°, FOV = 256 × 256 mm, 176 sagittal slices, voxel size 1 mm isotropic.
MR data preprocessing. Results included in this manuscript come from preprocessing performed using fMRIPrep 20.2.0^30,31^ (RRID:SCR_016216), which is based on Nipype 1.5.1^32,33^ (RRID:SCR_002502). The T1-weighted (T1w) image was corrected for intensity non-uniformity (INU) with N4BiasFieldCorrection [13], distributed with ANTs 2.3.3 [1] (RRID:SCR_004757), and used as T1w-reference throughout the workflow^34,35^. The T1w-reference was then skull-stripped with a Nipype implementation of the antsBrainExtraction.sh workflow (from ANTs), using OASIS30ANTs as target template. Brain tissue segmentation of cerebrospinal fluid (CSF), white-matter (WM) and gray-matter (GM) was performed on the brain-extracted T1w using fast (FSL 5.0.9, RRID:SCR_002823)^36^. Volume-based spatial normalization to one standard space (MNI152NLin2009cAsym) was performed through nonlinear registration with antsRegistration (ANTs 2.3.3), using brain-extracted versions of both T1w reference and the T1w template. The following template was selected for spatial ICBM 152 Nonlinear Asymmetrical template version 2009c (RRID:SCR_008796; TemplateFlow ID: MNI152NLin2009cAsym)^37^.
For the functional data, the following preprocessing was performed. First, a reference volume and its skull-stripped version were generated using a custom methodology of fMRIPrep. Susceptibility distortion correction was omitted. The BOLD reference was then co-registered to the T1w reference using flirt (FSL 5.0.9) with the boundary-based registration cost-function^38,39^. Co-registration was configured with nine degrees of freedom to account for distortions remaining in the BOLD reference. Head-motion parameters with respect to the BOLD reference (transformation matrices, and six corresponding rotation and translation parameters) are estimated before any spatiotemporal filtering using mcflirt (FSL 5.0.9)^40^. BOLD runs were slice-time corrected using 3dTshift from AFNI 20160207^41^ (RRID:SCR_005927). The BOLD time-series (including slice-timing correction when applied) were resampled onto their original, native space by applying the transforms to correct for head-motion. These resampled BOLD time-series will be referred to as preprocessed BOLD in original space, or just preprocessed BOLD. The BOLD time-series were resampled into standard space, generating a preprocessed BOLD run in MNI152NLin2009cAsym space. First, a reference volume and its skull-stripped version were generated using a custom methodology of fMRIPrep. Confounding time-series (framewise displacement (FD) and global CSF signal) were calculated based on the preprocessed BOLD. FD was computed using two formulations following Power (absolute sum of relative motions) and Jenkinson (relative root mean square displacement between affines)^40,42^. FD is calculated for each functional run, using the implementation in Nipype (following the definitions by Power et al. 2014). The CSF global signal is extracted within the CSF mask. Gridded (volumetric) resamplings were performed using antsApplyTransforms (ANTs), configured with Lanczos interpolation to minimize the smoothing effects of other kernels^43^. Prior to effective connectivity analysis, the first 6 volumes were discarded from the functional data to allow for stabilization effects, and data were denoised using simultaneous regression of the twelve realignment parameters and their first order derivatives and high pass filtering (0.008 Hz).
Self-reported fear. To examine differences in self-reported fear for the PRE, ACQ, and EXT phases, we used 3 (CS+R/CS+NR/CS-) X 2 (Chronic pain/Pain-free) repeated measures analysis of variance (ANOVA), as implemented in JASP (version 0.14.1)^44^. For the critical test of memory reconsolidation interference (Hypothesis 1a), we used a 2 (CS+R/CS+NR) X 2 (EXT/REIN) X 2 (Chronic pain/Pain-free) repeated measures ANOVA, with a significant interaction between CS type and Phase indicating successful interference. For a secondary test of interference (Hypothesis 1b), we conducted a similar test comparing the REIN and RE-EXT phases, with a significant interaction between CS type and Phase indicating differential re-extinction. To investigate the impact of individual differences on the observed findings, these tests were repeated including PCS as a covariate (Hypothesis 2). Violations of sphericity were assessed using Mauchly’s test and corrected using Greenhouse-Geisser adjustments (noted as FGG). Inference of significance was based on an alpha criterion of α < 0.05. In case of significant effects of CS type, planned comparisons were carried out with Holm-correction (noted as pholm). To aid the further evaluation of the size of observed (null) effects, effect sizes (partial eta-squared, ηp^2^) and inclusion Bayes Factors (BFincl) were calculated. If the inclusion Bayes Factor is greater than one, this indicates that the inclusion of that particular effect improves model fit, relative to the null model^45^. The specified priors were Cauchy distribution with scale parameter r = 1/√2 ≈ 0.707 for all fixed effects. The Inclusion probability was 0.5 for each effect (equal prior probability for inclusion vs. exclusion). The analyses used Bayesian Model Averaging (BMA) for repeated measures ANOVA with a random effects structure. All models included random intercepts for subjects and random slopes for all repeated measures factors (Maximal Random Effects specification). The techniques used BMA which rather than testing single models, evidence was averaged across all possible model specifications weighted by their posterior probabilities. The Inclusion Bayes Factors (BFincl) was calculated as the ratio of posterior odds to prior odds for including each effect, representing evidence across all models that include vs. exclude that effect. For post-hoc comparisons we used uncorrected Bayes factors (BF₁₀,U) with Cauchy priors, with posterior odds corrected for multiple testing using the Westfall, Johnson, and Utts (1997) method. All models included random slopes for all repeated measures factors.
Functional connectivity. Task-modulated changes in functional connectivity were tested using a generalized psychophysiological interaction (gPPI) analysis, as implemented in CONN (version 21.a^46^). The basolateral and central amygdala (blA & ceA from the CIT168 human brain template)^47^, and the vmPFC (Brodmann area 25 & 32, from the Talairach atlas)^48^, were included as the primary regions-of-interest. Additionally, the anterior and posterior hippocampus (aHipp & pHipp, from the Brainnetome atlas)^49^, nucleus accumbens (NAc, from the Freesurfer atlas)^25,50^, and anterior insula (aINS, from a recent meta-analysis of fear conditioning)^25,51^, were included as secondary regions-of-interest (for further details see^25^). All regions were separated into seeds for the left and right hemispheres and masked by individual grey matter masks. Time courses were extracted from each mask (unsmoothed functional data) and combined with CS predictors to create PPI predictors, such that there were three types of the psychological predictors (i.e., the task or CS regressors), the physiological predictor (i.e., the seed time course regressor, to control for non-task-related connectivity), and the interaction term (i.e., the product of the CS regressors and time course regressor). We investigated if, on average, connectivity between amygdala and vmPFC differed between the three CS types (Hypothesis 3), by inspecting differences across the three interaction terms. We included PCS as a covariate to investigate if this measure was related to any potential differences in connectivity. We conducted all gPPI analyses for early (run 1) and late (run 2) RE-EXT separately and controlled for age. Significance was assessed at FDR-adjusted p < 0.05. To visualize the effects, the regression coefficients associated with the interaction term were extracted per CS.
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
As expected, we observed no credible differences in self-reported fear for the three CSs at PRE (F(2,144) = 0.59, p = 0.556, BFincl = 0.08), no credible differences related to Group (F(1,72) = 0.05, p = 0.816, BFincl = 0.17), and no interaction between CS type and Group (F(2,144) = 0.52, p = 0.594, BFincl = 0.01). Furthermore, there was successful fear acquisition (main effect of CS type at ACQ: FGG(1.66,115.98) = 11.10, p < 0.001, BFincl = 9210.51), which did not differ between groups (main effect of Group: F(1,70) = 0.02, p = 0.888, BFincl = 0.18; CS type X Group FGG(1.66,115.98) = 0.33, p = 0.678, BFincl = 0.10). Post-hoc tests showed that the degree of fear acquisition did not differ between CS+R and CS+NR (t = –0.67, pholm = 0.503, BF₁₀,U = 0.192), and participants reported significantly more fear for both CSs+ compared to the CS- (CS+R vs. CS-: t = 4.32, pholm < 0.001, BF₁₀,U = 145.477; CS+NR vs. CS-: t = 5.00, pholm < 0.001, BF₁₀,U = 511.168). We did not observe full extinction, with a main effect of CS type remaining at EXT (FGG(1.52,108.06) = 7.27, p = 0.003, BFincl = 51.00). In addition, there were no credible differences in self-reported fear between the CS+R and CS+NR (t = –0.42, pholm = 0.675, BF₁₀,U = 0.182) at the end of EXT, nor were there any group-related effects (F(1,71) = 0.01, p = 0.921, BFincl = 0.24).
For the critical test of memory reconsolidation interference, we did not observe the expected interaction between CS type and Phase (F(1,66) = 1.53, p = 0.221, BFincl = 0.07), indicating that there was no credible difference among the CSs in the return-of-fear (i.e., from EXT to REIN). Furthermore, there was no interaction among CS type, Phase, and Group (F(1,66) = 0.02, p = 0.885, BFincl < 0.01), indicating that this lack of interference was present for both groups. For the secondary test of memory reconsolidation interference, we assessed whether the CSs differed in the amount of re-extinction that took place following reinstatement. Results showed no evidence for interference for this test either (CS type X Phase F(1,66) = 1.90, p = 0.172, BFincl = 0.14), nor an interaction between CS type, Phase, and Group (F(1,66) < 0.01, p = 0.967, BFincl = 0.07). However, we did observe a small Phase by Group interaction (F(1,66) = 7.05, p = 0.010, BFincl = 1.06), with only pain-free participants showing a significant decrease in self-reported fear after re-extinction (t = 2.99, pholm = 0.023, BF₁₀,U = 2.696). Results across all phases of the experiment are shown in Fig. 1.Fig. 1Mean self-reported fear across all phases of the conditioning paradigm for adolescents with chronic pain (colors and solid lines; n = 54) and pain-free peers (grey tones and dashed line; n = 24). Error bars indicate standard error.PRE pre-acquisition phase, ACQ acquisition phase, REACT reactivation phase, EXT extinction phase, REIN reinstatement phase, RE-EXT re-extinction phase, CS+R CS paired with US during acquisition and reactivated prior to extinction, CS+NR CS paired with US during acquisition and not reactivated prior to extinction, CS CS never paired with the US.
Given that a difference between the CSs+ and CS- remained following extinction, we conducted a post-hoc repeated measures ANOVA comparing fear responses to the CSs at the end of ACQ versus EXT, which showed no significant overall effect of Phase (F(1,70) = 3.42, p = 0.069, BFincl = 2.69), although the Bayes Factor indicated that a small effect may be present. However, as there was also no interaction between CS type and Phase (FGG(1.70,119.27) = 1.63, p = 0.203, BFincl = 0.09), the data largely indicated unsuccessful fear extinction. We therefore repeated the critical test of memory reconsolidation interference in a subgroup of participants who showed at least some fear extinction (i.e., at least one-point decrease in fear for both CS+R and CS+NR; npain = 18, npain-free = 10) and still did not observe a significant interaction between Phase and CS type (F(1,23) = 0.51, p = 0.482, BFincl = 0.28). However, a significant return-of-fear was now observed (for both CSs+), shown by a significant main effect of Phase (F(1,23) = 16.79, p < 0.001, BFincl = 28.28). There was still no main effect of Group (F(1,23) = 3.34, p = 0.081, BFincl = 0.85), nor interaction between Group and Phase (F(1,23) = 1.32, p = 0.263, BFincl = 0.81), Group and CS type (F(1,23) = 0.094, p = 0.762, BFincl = 0.21), or Group, Phase, and CS type (F(1,23) = 1.14, p = 0.297, BFincl = 0.09). For the secondary test of re-extinction we also did not observe an interaction between Phase and CS type (F(1,23) = 0.70, p = 0.412, BFincl = 0.20).
To examine the impact of individual differences in fear learning, PCS was included as a covariate in analyses of fear acquisition, extinction, return-of-fear, and re-extinction. While there was no interaction between PCS and CS type (FGG(1.66,114.18) = 0.69, p = 0.477, BFincl = 0.18), there was a significant main effect of PCS at ACQ (F(1,69) = 9.54, p = 0.003, BFincl = 4.76). During EXT, there was also a significant effect of PCS (F(1,70) = 16.86, p < 0.001, BFincl = 65.40) and the previously significant effect of CS type was no longer significant (FGG(1.52,106.52) = 0.52, p = 0.545, BFincl = 35.75), indicating successful fear extinction when controlling for PCS. The Bayes Factor did still suggest substantial evidence for CS type effects (BFincl=35.75), reflecting the difference between classical single-model testing and BMA approaches. When examining the change in fear from EXT to REIN, despite a significant effect of PCS (F(1,65) = 18.18, p < 0.001, BFincl = 26.44), we still did not observe the expected interaction between CS type and Phase (F(1,65) = 0.94, p = 0.335, BFincl = 0.08), indicating that there was no credible difference among the CSs in the return-of-fear from EXT to REIN, even when controlling for PCS. This pattern was also present when examining the return-of-fear from EXT to RE-EXT, with a main effect of PCS (F(1,66) = 17.21, p < 0.001, BFincl = 54.70), but no interaction between CS type and Phase (F(1,66) = 0.66, p = 0.421, BFincl = 0.11). Overall, these results indicate that individuals who are higher in pain catastrophizing reported higher fear across all phases of the conditioning paradigm, were (more) resistant to fear extinction, but did not differ in the (lack of) reconsolidation interference.
There were no credible differences across CS types in the connectivity between vmPFC and amygdala during early RE-EXT, however at late RE-EXT there was a significant difference across CS types in connectivity between left vmPFC (BA25) and right (basolateral) amygdala (F(2,106) = 4.70, puncorr = 0.01). The pattern (Fig. 2, connection a) showed more positive connectivity with right basolateral amygdala related to the CS+R and CS-, and more negative connectivity related to the CS+NR. When we included the secondary regions-of-interest, we did not observe any other significant effects of CS type.Fig. 2Overview of functional connectivity findings between regions-of-interest at late RE-EXT.Top: Regions-of-interest overlaid on a standard brain, plus arrows to indicate the connections (a, b, c) that are plotted below. Solid lines refer to connections showing the CS effect; dashed lines refer to connections showing an interaction with PCS. Bottom: Extracted parameter estimates (betas) from the gPPI analysis plotted per CS predictor for the connections (a, b, c). Note that to enhance readability and since there were no group-related differences, both groups are taken together (n = 55 participants). For illustrative purposes, a median split of PCS is used to visualize the interaction with PCS in connections b & c. L/R left/right, BA32/25 vmPFC ROIs (Brodmann area 32/25), ceAm central amygdala, blAm basolateral amygdala, CS+R CS paired with US during acquisition and reactivated prior to extinction, CS+NR CS paired with US during acquisition and not reactivated prior to extinction, CS CS never paired with the US, PCS pain catastrophizing, gPPI β parameter estimate for the generalized psychophysiological interaction predictor.
There were no credible differences in connectivity related to Group, nor a Group X CS type interaction, for either early or late RE-EXT. However, there was a significant relationship between PCS and the difference in connectivity between left basolateral and centromedial amygdala (CS type x PCS F(2,104) = 5.099, puncorr = 0.008), and between left and right vmPFC (BA32), related to CS type (CS type x PCS F(2,104) = 3.109, puncorr = 0.049). Findings are visualized using a median split of PCS in Fig. 2 (connections b & c). Those with lower PCS showed more differential modulation of connectivity within the left amygdala (central and basolateral nucleus connectivity) compared to those with higher PCS. Specifically, as PCS decreased, stronger connectivity related to the CS+NR emerged (consistent with more inhibition of the amygdala by the vmPFC, and potentially less fear expression, for those with lower PCS). However, connectivity between left and right vmPFC (BA32) showed a different pattern, with lower PCS being related to larger differences in connectivity between CS- and the CSs+, and higher PCS being related to relatively smaller differences in connectivity between the CSs.
To support interpretation of null results, we conducted sensitivity analysis for key null comparisons (CS+R vs CS+NR) across three Cauchy scale narrow (r = 0.35), default (r = 1/√2 ≈ 0.707), and wide (r = 1.414). For the ACQ phase, Bayes factors were BF₁₀ = 0.348 (narrow), 0.192 (default), and 0.099 (wide), providing anecdotal to moderate evidence for no difference. For the EXT phase, Bayes factors were BF₁₀ = 0.333 (narrow), 0.182 (default), and 0.094 (wide), providing anecdotal to moderate evidence for no difference. All tested priors yielded consistent evidence favoring the null hypothesis (BF₁₀ ≤ 1/3), confirming that our interpretation of no difference between CS+R and CS+NR was robust to reasonable variations in prior choice.
To explore whether the participants’ depressive symptom scores (CDI) may explain the findings, we re-ran the analyses using the CDI as a covariate. Controlling for depressive symptoms did not alter the pattern of findings for the memory reconsolidation interference effect (Hypothesis 1; see Tables S2 & S3 for details), nor for the impact of individual differences (Hypothesis 2; Tables S4 & S5). In addition, the significant difference across CS types in connectivity between left vmPFC (BA25) and right (basolateral) amygdala (connection a) also remained (Hypothesis 3; Table S6). Lastly, the interactions between PCS and CS type also remained when correcting for depression scores (Table S7).
Behavioral memory updating has the potential to enhance the effects of extinction learning by disrupting memory reconsolidation processes^3,6^. With pain-related fear memories implicated in the development and maintenance of chronic pain, targeting this process could enhance the effectiveness of current exposure-based treatments. In the present study, we investigated if interfering with memory reconsolidation via a reactivation-extinction procedure would reduce return-of-fear among adolescents with chronic pain and pain-free peers. We observed neural circuit evidence of disruption of memory reconsolidation, however, we did not find evidence to support this effect in self-reported fear. We did observe a significant difference between adolescents with chronic pain and their pain-free peers after re-extinction, with only the pain-free participants showing a reduction in fear. This resistance to extinction among adolescents with chronic pain was also present in a previous study^25^. In both the present study and the previous study of Heathcote et al.^25^ the results indicated that increasing levels of pain catastrophizing are related to increasing resistance to extinction, an effect that has also been demonstrated in adults with chronic pain^52^. This suggests that conditioning effects are more enduring in patients with chronic pain and may reflect one mechanism through which pain is sustained.
While there was no self-reported evidence of post-retrieval amnesia, the fMRI data did indicate neural circuit differences relating to the reactivated (CS+R) and non-reactivated (CS+NR) CSs. Connectivity between the left vmPFC (BA25) and right amygdala (basolateral nucleus) was more similar between the reactivated CS+R and safety cue CS-, consistent with the idea that the CS+R had been ‘updated’ to a safety cue. For the non-reactivated CS+NR the connectivity between these regions was negative, consistent with the inhibitory role of the vmPFC: increased deactivation of vmPFC results in more suppression of amygdala activation^12^.
The patterns of altered connectivity for BA25 and BA32 highlight the different roles these regions play in fear extinction and expression. Research on both human and non-human animals have demonstrated that BA25 (homologous to rat infralimbic cortex, IL)^48^ has a high density of bidirectional connections to the amygdala^48^, terminating primarily in specific neuronal sub-populations in the basolateral nuclei which are responsible for extinction memory formation and expression^53^. Evidence in both non-human and human animals has shown that the BA25/IL region is crucial for the consolidation and retrieval of extinction learning^53,54^. On the other hand, BA32 (homologous to rat prelimbic cortex, PL^48^) has been shown to be related to fear expression during extinction, but not extinction learning^54^. Through connections with the amygdala, BA32/PL typically shows an opposite response pattern to BA25/IL and excites fear expression^53^.
While the vmPFC connections with the amygdala are largely with basolateral nuclei, centromedial nuclei are also necessary for fear memory consolidation, and current consensus is that intra-amygdala circuits are needed for formation, consolidation, and retrieval of extinction memories, and receive cortical input to also further modulate inhibitory behavior (e.g., hippocampus for context effects)^53,55^. Given the importance of intra-amygdala connectivity for extinction memory retrieval^56^, the overall more blunted intra-amygdala connectivity for adolescents with high pain catastrophizing could be related to the lack of extinction we observed among these individuals.
While our results show extensive alterations in vmPFC-amygdala circuitry between recall of a reactivated CS+ versus a non-reactivated CS+, we did not observe any differences in self-reported fear. Previous research in humans has also repeatedly failed to show behavioral effects^16,18,21^ and our findings support the conclusion that post-retrieval amnesia may not occur under all conditions, which makes it unsuitable for clinical application. Due to the discrepancy between the neural and behavioral results obtained in this study, and the discrepancy between findings with humans and non-human animals, we posit that behavioral memory updating may be occurring on the memory traces formed by Hebbian associative learning, but the compensatory effects of other forms of associative learning (e.g., propositional^57,58^) modulate the exhibited behaviors. This added complexity, and uniquely human aspect of learning may explain the inconsistent translation from non-human animal research. Moreover, it is important to consider whether the observed neural signal may not be strong enough to impact human behavior or whether individual differences in memory reconsolidation or greater/less susceptibility to retrieval effects could also be playing a role in explaining these divergent findings.
In the reactivation-extinction procedure used in the present study, many processes critical for post-retrieval amnesia may modulate the the stability of the original memory trace, the efficacy of the destabilization procedure, the reintegration of new learning, the long-term storage of the memory, and the recall of the memory at the later memory tests. There are several considerations for each of these stages which may have implications for the conclusions that can be drawn. These are discussed in more detail below.
In the present study, acquisition training and reactivation took place on the same day, meaning that sleep-dependent consolidation did not take place. As a result, the degree to which the acquisition memories were consolidated prior to destabilization is likely to be less than in previous studies. Consolidation is the process of stabilizing a newly acquired memory trace, beginning in the minutes following learning (synaptic consolidation) and continuing for weeks (systems consolidation), with sleep and/or restful waking being crucial for several consolidation mechanisms^59,60^. Reconsolidation is a separate process that integrates new information into a memory trace and restabilizes it, relying on similar but distinct protein-synthesis processes^23,61–63^. Given the differences in underlying biological mechanisms, the applicability of our findings using the same-day acquisition/reactivation design to the clinic is limited. Exposure therapy is usually targeting pain-related fear memories acquired months or years previously, meaning they are likely to be fully consolidated and highly stable, and potential interference with the memory would require destabilization and reintegration through reconsolidation. Thus, exposure therapy centers on targeting and reducing avoidant behaviors rather than fear itself as behavior can change even in the presence of persistent fear. These findings further underscore the importance of this approach.
Furthermore, the destabilization of a consolidated memory trace is a critical component of behavioral memory updating. To destabilize the memory, the right amount of prediction error must be elicited by the reactivation trial(s). If acquisition takes place under partial reinforcement, then a single trial may not be sufficient to elicit a prediction error, however, too many trials may elicit new extinction memory formation, purportedly due to sustained phosphorylation in hippocampal cells^64^. Previous research has demonstrated that, for studies with ~80% reinforcement, two reactivation trials are optimal^9^. However, a limitation in both the present study and in previous research is that prediction error is not measured and therefore cannot be confirmed. While methods such as the third interval of skin conductance or startle reflexes have been developed, their relative novelty means they have not yet been implemented in current research^65,66^.
Assuming the memory has been destabilized, an effective interference must then take place during the reconsolidation window. As in previous studies, the present design used a 10-min gap between reactivation and extinction, thought to be optimal for reconsolidation^21^. However, confirming whether this reconsolidation window has been achieved is not possible, and there are factors which may influence this timing. For example, emotional memories tend to be (re)consolidated faster due to their importance for survival^67^, and therefore a better understanding of when reconsolidation is taking place is likely to be important as research in this topic expands to different populations and different types of stimuli. In addition to the timing of the extinction training relative to reactivation, the efficacy of the extinction training will determine what the original memory is updated to. In the present study, we observed resistance to extinction, an effect that was larger for individuals with high levels of pain catastrophizing. Therefore, the lack of post-retrieval amnesia we observed could be a result of the lack of extinction learning. However, even when we post hoc examined only participants who showed extinction, we still did not observe any evidence of post-retrieval amnesia.
Following extinction training, the extinction memory (CS+NR) and ‘updated’ safety memory (CS+NR) then need to be stored until the return-of-fear tests, and subsequently retrieved. The storage period for this study (on average one week) was considerably longer than in previous studies that did demonstrate post-retrieval amnesia (one day)^15,68^. This longer period was both a practical decision to facilitate participation by the adolescent sample and reflects the clinical reality that any amendment to exposure therapy would require that the effects be long-lasting. The post-retrieval amnesia observed due to reconsolidation interference is not always persistent, and in some cases can also be reversed^3^. Further investigating the persistence of the ‘updated’ memory is critical to evaluating its utility as a clinical tool.
In the present study, we tested the hypothesis that behavioral memory updating could be a method to reduce the return of fear following fear extinction. Our findings suggest that while the behaviors for the ‘updated’ versus ‘extinguished’ memory are comparable, the underlying neural mechanisms driving these responses are different. Further investigating these nuances in learning mechanisms could shed light on the conflicting findings in the field so far and will be necessary to establish whether there are circumstances in which approaches such as behavioral memory updating could have clinical utility.
Transparent Peer Review file Supplemental Material Reporting summary