Associative Learning and Adverse Childhood Experiences: A Multi-System Protocol

Summary

This protocol describes an integrated 60-min battery — pavlovian-to-operant fear conditioning with avoidance, fear acquisition–extinction–reacquisition with social stimuli, and probabilistic reward learning — that records skin conductance, behavior, and phase-by-phase ratings concurrently to derive per-participant indices of threat and reward learning in adults.

Abstract

Adverse childhood experiences (ACEs) are established transdiagnostic risk factors for adult anxiety and depression, yet the neurocognitive mechanisms linking early adversity to symptoms remain unclear within a unified framework. This protocol presents an integrated, multi-system experimental battery that captures threat discrimination, safety learning, and reward sensitivity within a single laboratory session. Thirty-eight healthy adults completed three behavioral paradigms — Pavlovian-to-operant fear conditioning with avoidance (Experiment 1), fear acquisition-extinction-reacquisition with social stimuli (Experiment 2), and probabilistic reward learning (Experiment 3) — alongside validated self-reports of childhood adversity (CTQ-SF), benevolent childhood experiences (BCE), depression (PHQ-9), and anxiety (GAD-7, STAI). Multi-channel data were collected concurrently: skin conductance response (SCR) as a physiological index of autonomic arousal, button-press avoidance and probabilistic choices as behavioral indices, and phase-by-phase subjective ratings of expectancy, threat, and relief. Each paradigm produced its canonical effect: Pavlovian discrimination on SCR (p = 0.009, ηp² = 0.12), expectancy (p < 0.001, ηp² = 0.58), and threat ratings (p < 0.001, ηp² = 0.35); operant relief discriminating avoidable from unavoidable threat (p < 0.001, d = 0.86); differential autonomic responding present at acquisition but not extinction in Experiment 2 (acquisition p = 0.008; extinction p = 0.168) coexisting with persistent declarative knowledge of the contingency (p < 0.001, d = 3.08); and above-chance probabilistic reward learning in Experiment 3 (M = 73%, p < 0.001, d = 1.97). The multi-system design revealed dissociations that a single channel would have missed, most notably attenuated autonomic responding alongside persistent subjective threat. Per-participant indices derived from the protocol were sensitive to individual differences in childhood experience, with illustrative moderation analyses reported in the supplementary material. The protocol provides a portable, scalable framework for indexing how early adversity recalibrates threat and reward processing in adulthood.

Introduction

Adverse Childhood Experiences (ACEs), encompassing abuse, neglect, and household dysfunction, are highly prevalent worldwide, with rates exceeding 60% in many populations¹. Research syntheses indicate that most children are exposed to at least one adverse event during development². ACEs are well-established transdiagnostic risk factors for adult psychopathology³, contributing to an estimated 30% of anxiety disorders and 40% of depression cases¹. Despite this strong association, commonly used clinical assessments rely on categorical diagnoses and retrospective self-reports, which are vulnerable to recall bias and do not capture real-time neurocognitive processes⁴.

To overcome these limitations, there is an increasing emphasis on transdiagnostic approaches that target underlying mechanisms, particularly associative learning. Associative learning, encompassing both Pavlovian and operant conditioning, provides a framework for understanding how individuals encode relationships between environmental cues, behaviors, and outcomes to guide decision-making^4,5. Within this framework, the latent vulnerability theory posits that early adversity induces neurobiological adaptations, such as heightened vigilance, that are initially adaptive in unpredictable environments but may become maladaptive in later, safer contexts^6,7.

Evidence suggests that ACEs influence both threat- and reward-related learning processes. Individuals with a history of adversity often show reduced discrimination between conditioned danger (CS+) and safety (CS−) cues, indicative of altered threat learning⁸. In parallel, early adversity has been associated with diminished reward sensitivity, reflected in lower accuracy and slower acquisition of reward contingencies⁸. Under conditions of uncertainty, these individuals may also exhibit increased choice variability and a bias toward assuming random reward delivery, consistent with early exposure to volatile environments^5,9.

These learning patterns are further shaped by protective and state-dependent factors. Benevolent Childhood Experiences (BCEs) may buffer the effects of adversity on learning systems¹⁰, while current affective states, such as anxiety, can modulate conditioning processes¹¹. Physiological measures of autonomic arousal, including Skin Conductance Response (SCR), provide objective indices of these processes and are widely used in conditioning paradigms¹².

A major limitation in the current literature is the use of isolated tasks that assess threat or reward learning independently, limiting the ability to capture integrated neurocognitive functioning¹. The present protocol addresses this gap by providing a standardized framework that combines behavioral and physiological measures across multiple learning domains. Specifically, the protocol includes three experimental components: (1) a paradigm assessing transitions from Pavlovian to operant conditioning, (2) a social fear extinction task, and (3) a probabilistic reward learning task⁸.

The rationale for this particular combination of paradigms is not that any one of these three tasks is, in isolation, more sensitive to early adversity than the many other established learning paradigms available; it is a rationale of measurement. In their systematic review of 81 studies (38 threat, 43 reward), Ruge et al.¹ showed that the literature linking ACEs to associative learning remains heterogeneous and, in key respects, inconclusive precisely because studies typically deploy a single task, record a single response channel, and operationalize adversity in non-harmonized ways. Illustratively, half of the behavioral reward-learning studies reported blunted learning while the other half reported null results; the small extinction literature is dominated by null findings and by reports that omit the preceding acquisition phase; and no outcome measure has emerged as universally more reliable, so different response channels are not interchangeable proxies for one another. Their explicit recommendation is a move toward within-subject, multi-domain, and multi-channel protocols yielding individual-level indices. The present triad follows that recommendation because its three tasks span three computationally dissociable demands that no single paradigm covers, and which the theory of latent vulnerability⁷ predicts that adversity should shape in different ways: (i) threat–safety discrimination combined with instrumental controllability, with relief as the putative reinforcer of avoidance¹³; (ii) inhibitory (safety) learning and return of fear, probed with social face stimuli¹¹, the ecologically relevant stimulus domain for interpersonal maltreatment; and (iii) reinforcement learning under probabilistic uncertainty, which yields per-participant computational parameters (learning rate alpha, inverse temperature beta)⁸. Because all three are administered within the same participant, the design is constructed to ask whether ACE-related alterations are domain-general — a global recalibration of associative learning, as Hanson et al.⁸ propose and as a neurocomputational latent-vulnerability account would predict⁶ — or valence-specific, confined to threat. A single-task study cannot adjudicate between these competing hypotheses by construction.

The battery is designed to test a set of explicit, directional predictions, which the present sample is not powered to adjudicate and which are stated here as targets for adequately powered studies. Consistent with the blunted-learning pattern described by Ruge et al.¹, higher ACE load is predicted to be associated with (i) reduced CS+/CS− discrimination in Experiment 1, driven by attenuated responding to the CS+ and expected to be more pronounced in the autonomic channel (SCR) than in declarative expectancy, which may remain intact; (ii) greater avoidance responding and higher subjective relief to the avoidable CS+, consistent with relief-reinforced avoidance¹³; (iii) slower extinction and a faster, larger return of fear at reacquisition in Experiment 2, an effect predicted to be amplified by the social (face) stimuli that carry ecological relevance for interpersonal adversity¹¹; and (iv) lower choice accuracy, a lower learning rate (alpha), and greater choice variability (lower beta) in Experiment 3⁸. Crucially, the three tasks are jointly informative about a fifth, discriminating prediction: if ACE-related alteration is domain-general, attenuation should appear across all three tasks; if it is valence-specific, it should be confined to the two threat paradigms and absent in probabilistic reward learning. Because response channels are recorded concurrently, a further prediction is testable — that ACE effects will be channel-dependent rather than uniform, so that conclusions drawn from any single channel would not generalize to the others.

Experiment 1 adapts the Pavlovian-to-operant fear conditioning paradigm with avoidance and trial-by-trial relief ratings developed by San Martín et al.¹³, in which two CS+ cues (one avoidable, one unavoidable) and one CS− are used to dissociate Pavlovian discrimination from instrumental control over the aversive outcome. In the original paradigm, US-expectancy ratings during the Pavlovian acquisition phase were assessed retrospectively at the end of the phase by the experimenter rather than on a trial-by-trial basis. In the present adaptation, subjective ratings of expectancy, threat, and relief were collected concurrently throughout the task on a phase-by-phase basis using visual analog scales, allowing continuous tracking of learning dynamics across phases.

Experiment 2 adapts the fear acquisition–extinction protocol with social (face) stimuli from Dibbets and Evers⁹, extending it with a brief reacquisition test to evaluate the rapid re-emergence of conditioned responding (savings effect).

Experiment 3 adapts the probabilistic reward learning task introduced by Hanson et al.⁸ for the study of early-adversity-related learning differences, with two reward contingencies (80/20 and 70/30) that allow estimation of per-participant reinforcement-learning parameters (Q-learning Alpha and Beta). By integrating these three paradigms within a single laboratory session, the present protocol enables direct within-participant comparison of threat and reward learning processes that are typically studied in isolation. Beyond per-trial accuracy and Q-learning parameters, the analysis pipeline also produces individual learning-trajectory slopes from a linear mixed model fit, available for use in larger samples.

Two distinct goals must be kept apart here, because they are not equivalent. The first is to establish that a battery of this kind can be run as a single, integrated 60-min session and that each of its components elicits the canonical effect it was designed to elicit; reproducing canonical effects is sufficient evidence for this first goal. The second is to establish that the resulting indices are sensitive and specific to the consequences of childhood adversity; reproducing canonical effects is not sufficient evidence for that, and the present sample is not powered to provide it. The primary objective of this report is therefore the technical and methodological feasibility of the integrated battery: whether threat acquisition, instrumental avoidance with relief, extinction and return of fear with social stimuli, and probabilistic reward learning can be acquired concurrently across autonomic, behavioral, and self-report channels within a single session, and whether the pipeline yields stable, reusable per-participant indices (differential SCR contrasts, the operant relief contrast, Q-learning alpha and beta parameters, and individual learning-trajectory slopes). Assessment of ACE-related alterations in learning constitutes a secondary, exploratory, and illustrative objective: the individual-difference analyses reported here are intended to demonstrate the measurement capacity of the protocol and to make its candidate indices available for future work, not to specifically test trauma–learning pathways. The protocol thus provides candidate indices; whether they possess the sensitivity and specificity required to characterize the neurocognitive consequences of childhood adversity remains to be established in larger, ideally extreme-group-enriched and pre-registered samples^1,14.

Protocol

The present protocol describes the procedures and instrumentation used to measure Skin Conductance Response (SCR) in healthy participants aged 18 or older, in compliance with the ethics committee's requirements. Before participating in the study, all participants signed an informed consent form that also included the possibility of being contacted again after six months for a follow-up assessment. All procedures described in this document were approved by the Scientific Ethics Committee of Universidad de los Andes (ID: CEC2025023).

NOTE: This protocol provides a detailed description of the complete experimental workflow (Figure 1), including participant scheduling, laboratory preparation, physiological signal acquisition, data organization, and post-experimental data handling. In addition, it outlines the staff training procedure implemented to ensure that SCR recordings are collected consistently, standardized, and reliable across all experimental sessions.

Figure 1: Overall schematic of the study design. Each participant completed a single laboratory session covering three sequential behavioral paradigms: Pavlovian-to-operant fear conditioning with avoidance (Experiment 1), fear acquisition–extinction–reacquisition with social stimuli (Experiment 2), and probabilistic reward learning (Experiment 3). Self-reports of childhood adversity (CTQ-SF), benevolent childhood experiences (BCE), and current symptoms (PHQ-9, GAD-7, STAI) were also collected. Multi-channel data — skin conductance response, behavior (avoidance, probabilistic choices), and subjective ratings (expectancy, threat, relief) were recorded concurrently. Per-participant indices were entered into individual-difference analyses (see Supplementary File 2). Please click here to view a larger version of this figure.

1. Staff training

1. Train every team member in three progressive stages to ensure standardized and accurate data measurement.
   1. Run internal trials in which each staff member takes samples and executes all three experiments according to their respective protocols, so that the trainee becomes familiar with the equipment and with participant handling.
   2. Next, have each trainee take samples from real participants under the supervision of at least one of the principal researchers, in order to verify their ability to run the protocol and to correct any errors.
   3. Authorize the assistant to take samples autonomously and to schedule new participants only after the complete protocol has been applied correctly under supervision.

2. Scheduling

1. Recruit and schedule participants through social media, providing the general objectives of the research and the participation criteria.
   1. Create a digital poster and distribute it through the recruitment channels to reach as many potential participants as possible.
   2. Instruct interested candidates to contact the research team by email and schedule a session on a day convenient for the participant.
   3. Send the participant a link to the online questionnaire, entitled "The Invisible Marks: The Role of Associative Learning in the Relationship between Adverse Childhood Experiences and Mental Health in Adulthood."
   4. Begin the questionnaire with an informed consent form, and require the participant to read and accept it before any further items are presented.
   5. Include in the questionnaire a socio-demographic section and the following validated psychometric scales: Patient Health Questionnaire-9 (PHQ-9)¹⁵, Generalized Anxiety Disorder Screener (GAD-7)¹⁶, State-Trait Anxiety Inventory (STAI)¹⁷, Childhood Trauma Questionnaire—Short Form (CTQ-SF)¹⁸, and Benevolent Childhood Experiences (BCEs)¹⁰.

3. Preparations

1. Ensure that at least one staff member is present in the laboratory before the participant arrives. Prepare the experimental environment, verify the correct functioning of the equipment, and confirm that all protocol requirements are met before the session begins:
   1. Turn on the laboratory computer and verify that the operating system, experimental software, and all peripherals are functioning correctly.
   2. Confirm that the experimental files and data acquisition system are ready to use.
   3. Open and review the daily running file before the participant enters the laboratory.
   4. Identify the corresponding identification code (ID) assigned to the incoming participant.
   5. Verify the execution order of the experimental conditions according to the predefined counterbalancing.
   6. Ensure that the participant is assigned to the correct experimental order before any data acquisition begins.
      NOTE: The daily running file contains all necessary information to assign the participant's ID and determine the execution order of the experiment while maintaining the required counterbalancing across all experimental conditions. For example, a participant may be assigned the ID "015".
   7. Organize data storage on the laboratory computer in the standardized directory structure required by the automated processing pipeline: one top-level folder per experiment, one subfolder per counterbalancing condition, and one folder per participant ID, plus two global folders that receive the processed data and the consolidated databases (see Supplementary File 1).
   8. Create the participant folder corresponding to the ID and counterbalancing condition listed in the daily running file before beginning data acquisition, and follow the directory and file-naming conventions specified in Supplementary File 1 throughout the session, because the pre-processing and database scripts locate files by path and by name.
      NOTE: The directory tree, the counterbalancing subfolder labels, the file-naming conventions, and a worked example are given in Supplementary File 1. Deviating from these conventions prevents the automated pipeline from locating the participant's files.
   9. Verify that the experimental desk is completely free of external or distracting objects. Allow only the data acquisition system and the computer peripherals required for the experiment on the desk.
   10. Invite the participant into the room in a calm and professional manner.
   11. Ask the participant to leave all personal belongings aside in the designated area.
   12. Instruct the participant to place their mobile device in silent mode to avoid interruptions or distractions during the experimental session.
   13. Ask the participant to remove all accessories from their non-dominant hand, particularly rings, bracelets, or any object located in the middle and ring fingers, as they may interfere with the measurements.
   14. Guide the participant to sit comfortably in the chair positioned in front of the laboratory computer.
   15. Provide the participant with a printed physical copy of the informed consent document.
   16. Allow sufficient time for the participant to read the document and ask questions if necessary.
   17. Confirm that the informed consent form is correctly signed before starting any experimental procedure.
       NOTE: The informed consent procedure is repeated in physical format following the requirements established by the ethics committee and in order to maintain a physical backup of the participant's consent documentation.

4. Experimental setup

NOTE: The experimental setup must follow a strict and standardized protocol to ensure that all physiological measurements are acquired under identical controlled conditions across participants.

1. Turn on the acquisition system and verify that the device is correctly connected to the laboratory computer and recognized by the acquisition system software (see Figure 2A–E).
2. Confirm again that all required modules and peripherals are functioning properly before proceeding with the participant preparation.
3. Take two electrodermal activity (EDA) electrodes (see Figure 2F).
4. Apply a small quantity of conductive gel to the contact surface of each electrode (see Figure 2G,H).
5. Ensure that the conductive gel is evenly distributed across the electrode surface to improve electrical contact with the participant's skin and optimize signal quality.
   NOTE: If conductive gel is not applied to the electrodes, the recorded signal may contain excessive noise, and the skin conductance response (SCR) measurement quality may be severely compromised.
6. Place the electrodes on the participant's non-dominant hand.
7. Attach one electrode to the first phalanx of the middle finger (third finger) and the second electrode to the first phalanx of the ring finger (fourth finger) (see Figure 2I).
8. Verify that both electrodes are firmly attached and maintain stable contact with the skin without causing discomfort to the participant.
9. Place the wrist strap on the participant's wrist.
10. Leave the EDA electrodes disconnected from the wrist strap during the calibration stage (see Figure 2J).
11. Turn on the wrist strap in order to establish wireless communication with the physiological acquisition system.
12. Confirm that the device is successfully detected.
    NOTE: Special attention must be paid to the wrist strap throughout the entire experimental session. The staff members must continuously verify that the device remains powered on and connected before starting the experiment, since any interruption in communication may compromise the physiological recording.
13. Start the acquisition process in the acquisition software.
14. Wait for the calibration window, which appears automatically on the screen once acquisition begins.
15. Perform the calibration procedure while the electrodes remain disconnected from the wrist strap.
16. Follow the calibration instructions provided by the software until the procedure is completed.
17. After successful calibration, connect the EDA electrodes to the wrist strap (see Figure 2K).
18. Verify in the software that the skin conductance response (SCR) signal is being correctly acquired.
19. Instruct the participant to keep their non-dominant hand completely still throughout the entire experiment.
20. Explain that unnecessary movements, finger contractions, or changes in posture may introduce noise to the SCR signal.
21. Ask the participant to maintain a comfortable and relaxed posture during the session.
22. Provide the participant with headphones before starting the experimental task.
23. Verify that the headphones are functioning correctly and that the participant can clearly perceive the auditory stimuli presented during the experiment.
24. Launch the experimental task in PsychoPy¹⁹.
25. When the PsychoPy interface opens, enter the participant name in the additional window that requests it.
26. Enter the participant's assigned ID exactly as specified in the daily running file before initiating the experiment.
    NOTE: Following the previous example, the participant ID entered in PsychoPy should be "015".

Figure 2: Acquisition system, materials, and electrode placement for skin conductance recording. (A) Main unit Biopac MP200. (B) Physiological signal acquisition module (PPG/EDA). (C) Sensors connector module. (D) Analog input module. (E) TTL module for communication between PsychoPy and AcqKnowledge. (F) Disposable electrodes (EL507A). (G) Isotonic gel (GEL101A) to improve skin contact. (H) Application of the gel on the electrode. (I) Electrodes placed in the first phalange of the third and fourth fingers of the non-dominant hand. (J) Wrist strap placement on the non-dominant hand for SCR measurement. (K) Electrodes connected after the calibration process. Please click here to view a larger version of this figure.

5. Beginning of the experiment

1. Before starting the experiment task, explain to the participant in detail how to respond to the questions presented during the experiment.
2. Ensure that the participant clearly understands: (i) the response method to be used during the task, (ii) the meaning of the instructions displayed on the screen, and (iii) the importance of remaining attentive and minimizing unnecessary movements throughout the experiment.
3. Answer any remaining questions from the participant before beginning the recording session.
4. Perform a final verification of the complete experimental setup to ensure that all systems are functioning correctly according to the established protocol, including: physiological acquisition system communication, wrist strap connection, SCR signal quality, PsychoPy execution, audio presentation through the headphones, and correct participant ID registration.
5. Once all systems have been verified and the participant confirms readiness to begin, start the experimental task.
6. Leave the experimental room once the task has started, in order to minimize external influences and distractions during data acquisition.
7. Remain nearby and available in case technical assistance or participant support is required, and wait until the participant completes the session.

6. Between experiments

1. Re-enter the experimental room once the participant has finished the task, and stop the physiological recording in the acquisition software.
2. Save the physiological recording in the participant's folder as an acquisition file, and then export the same recording to MATLAB format with events, so that the exported file carries the stimulus markers generated by PsychoPy through the TTL module; the export additionally produces a separate marker file.
   NOTE: The event and marker information is required by the pre-processing pipeline (Section 8) to segment the signal by trial and by stimulus type.
3. Save the PsychoPy output spreadsheet generated at the end of the task in the same participant folder. Use the standardized file-naming convention detailed in Supplementary File 1 for all four files.
   NOTE: At the end of each experiment, the participant's folder must contain the acquisition file, the exported MATLAB file, the marker file, and the PsychoPy spreadsheet. Exact names, paths, and a worked example are given in Supplementary File 1.
4. Carefully disconnect the EDA electrodes from the participant, ensuring no excessive movement or discomfort occurs during removal.
5. Open the next PsychoPy experimental task and restart the acquisition software in preparation for the following experiment.
6. Repeat the complete preparation and acquisition procedure starting from step 4.1.
   NOTE: The complete experimental session consists of three independent experiments. Therefore, once the second experiment has finished, the same procedure described above must be repeated for the third experiment.
7. At the end of the complete experimental session, ensure the participant has three subfolders for each experimental task, each containing all the previously mentioned files.
   NOTE: The approximate duration of each experimental task was previously estimated in order to allow staff members to monitor the participant and verify the correct functioning of the acquisition system through the session. The estimated duration of each experiment is as follows:
   Experiment_1: approximately 24 min.
   Experiment_2: approximately 17 min.
   Experiment_3: approximately 8 min.
   Considering the preparation, calibration, transition periods between experiments, and finalization procedure, the complete experimental session has an approximate duration of 60 min.

7. End of the experiments

1. Close the physiological recording session once all three experimental tasks have been completed.
2. Stop the acquisition process in the acquisition software if it has not already been stopped.
3. Disconnect the EDA electrodes from the wrist strap and remove the wrist strap from the participant's wrist.
4. Allow the participant to remove the EDA electrodes themselves to minimize discomfort and ensure safe removal from the skin.
5. Verify that all experimental files and physiological recordings have been correctly saved before concluding the session.
6. Thank the participant for their collaboration and offer a light snack as a token of appreciation after the experimental procedure is complete.
   NOTE: If any issue occurs during the physiological recording due to technical problems, including, but not limited to: loss of communication between the wrist strap and the physiological acquisition system module, unexpected shutdown of the wrist strap, loose or partially detached electrodes, or temporary signal interruptions or excessive noise caused by connection problems, the experiment must continue as planned and the participant must not repeat the experimental session.

8. Pre-processing of the data

NOTE: Once the experimental session has been completed and all physiological recordings have been properly stored, the SCR data must undergo a standardized pre-processing procedure prior to analysis.

1. Run the Ledalab_Format MATLAB script to reorganize the data structure of the physiological recording exported from the acquisition software, so that it is compatible with Ledalab, a MATLAB-based toolbox designed for the analysis of EDA signals.
   NOTE: In the Ledalab_Format.mat code, define the path of the {Identification_Code}.mat file as an input variable. This allows the exported physiological recording file to be converted into a format that can be correctly read and processed by Ledalab. Once the conversion process has been completed, a new MATLAB file must be generated and saved inside the corresponding participant folder using the following naming convention: Ledalab_{Identification_Code}.mat. Following the previous example, if the data from Experiment_1 are converted for participant "015", the resulting file must be named: Ledalab_015.mat.
2. Open Ledalab, in the menu bar located in the upper-left corner, select File > Import Data > Matlab File (*.mat). Then, load the newly generated data file corresponding to the participant's experimental condition.
3. Enter the sampling parameters in the configuration window that Ledalab displays once the file is loaded.
4. The original acquisition frequency of the SCR signal is 2000 Hz. In order to reduce computational load, downsample the signal to 10 Hz using the following parameters: Target sampling frequency: 10 Hz, Downsampling factor: 200, Resampling method: Steps.
5. After completing the downsampling procedure, perform a continuous decomposition analysis (CDA)²⁰ within Ledalab to decompose the EDA signal into its tonic and phasic components.
6. In the menu bar, select Analysis > Continuous Decomposition Analysis (Extraction of Continuous Phasic/Tonic Activity).
7. Wait for the configuration window to appear on the screen.
8. In this window, select the Optimize option to automatically adjust the analysis parameters.
9. Once the optimization process is completed, click Apply to start the CDA analysis.
10. Once the CDA analysis has been completed, select Results > Event-related Analysis, and a configuration window will appear on the screen.
11. Set the minimum amplitude threshold to the Ledalab default of 0.01 µS and leave the z-scale option unselected. The event markers exported from the acquisition software tag the onset of the conditioned stimulus and the onset of the unconditioned stimulus separately, so the event-related analysis returns a response for each. Set the response window as follows:
    1. For Experiment 1, use 0 s to 9 s after the event marker.
    2. For Experiment 2, use 0 s to 6 s after the event marker.
12. Export the result as a spreadsheet file. Ledalab generates a spreadsheet containing the relevant physiological and event-related information extracted from the SCR signal.
    NOTE: The response window corresponds to the full duration of the conditioned stimulus in each task (Supplementary Table 1: 9 s in Experiment 1 and 6 s in Experiment 2), so that the CS-evoked response covers the entire CS-US interval and is not contaminated by the unconditioned response, which is scored separately on its own event marker. Responses falling below the 0.01 µS threshold are written as zero by Ledalab and are retained in the analysis as zeros; do not delete them, because their removal would inflate the mean response of low-responding participants.
13. Repeat the complete pre-processing, downsampling, and CDA procedure for Experiment 2 (see the NOTE below for Experiment 3, which follows a different procedure and does not require the CDA output).
    ​NOTE: At the end of the pre-processing phase, each experimental folder must contain an additional spreadsheet file corresponding to the CDA analysis results generated by Ledalab. This file must subsequently be converted to .xlsx format for storage, organization, and further analysis. Experiment 3 follows a different pre-processing procedure. For this experiment, only the MATLAB script must be used, and it does not require the CDA output, only the alpha and beta values. When the script is executed for "Experiment_3", it automatically extracts the alpha and beta values, merges them with the behavioral information, and saves the resulting spreadsheet in Global_Data/Exp_3.
14. Perform signal quality control, non-responders, and artifact handling.
    1. Before any statistical analysis, display the full downsampled skin conductance trace of each participant in Ledalab and inspect it visually from the first to the last event marker. Perform this inspection for every participant and for every task separately, because signal quality is channel- and task-specific.
    2. Reject a whole trace, and treat the SCR channel of that task as missing for that participant, if any of the following is observed: (i) a flat or absent signal over the task, indicating loss of wireless communication with the wrist strap, a wrist strap powered off, or a detached electrode; (ii) a tonic skin conductance level below 0.05 µS sustained across the task, which indicates inadequate electrode-skin contact rather than low arousal; (iii) uninterpretable drift or a sudden, sustained inflation of amplitude that is not time-locked to any event; or (iv) saturation of the amplifier range.
       ​NOTE: Do not repeat the session and do not exclude the participant from the other tasks or from the behavioral and self-report channels.
    3. Define a participant as a non-responder for a given task if no phasic response reaches or exceeds the 0.01 µS threshold on any event of that task, including the unconditioned stimulus events of the reinforced trials — that is, if the participant produces no measurable electrodermal response even to the aversive outcome itself.
       1. Exclude a non-responder from every channel of that task — skin conductance, behavior, and subjective ratings alike — because the absence of any electrodermal response even to the unconditioned stimulus indicates that the electrodermal system was not measurable in that session, and calls into question whether the participant engaged with the contingency at all.
       2. Retain that participant in the two remaining tasks. Distinguish this case from technical signal loss (a detached electrode, or a wrist strap that lost wireless communication), in which the participant did respond, but the signal was not recorded: there, treat the SCR channel of that task as missing and retain the behavioral and subjective data of that task, which are unaffected by electrodermal signal quality.
       3. Report the non-responder rate and the technical-loss rate separately for every task.
    4. Identify movement artifacts on the trace as (i) abrupt rising transients that are not time-locked to any event marker, or (ii) step-like offsets in the tonic level produced by finger contraction, by a shift in posture, or by rubbing the electrodes against the chair or the desk.
    5. Mark every trial containing such an artifact within its scoring window and exclude it at the trial level; do not exclude the participant.
    6. Exclude a participant from every channel of a task, and not only from its SCR analysis, if more than 25% of the trials of that task are marked as artifact-contaminated: the remaining trials no longer support a stable per-participant estimate, and the recording conditions of that session are not adequate for any channel. Retain that participant in the two remaining tasks.
    7. Transform the trial-level amplitudes with log(SCR + 1) before averaging, in order to normalize the positively skewed distribution of electrodermal amplitudes; clamp any negative CDA value, a rare consequence of baseline drift, to zero before the transformation.
    8. Aggregate the transformed trials to obtain one value per participant, per phase, and per stimulus.
    9. Report the analytic n separately for each channel (SCR, behavior, ratings) and for each task, because channel-specific loss makes these n diverge.
    10. Record every decision taken in steps 8.14.1–8.14.9 in a per-participant quality log containing: participant ID, task, counterbalancing condition, trace accepted or rejected (with reason), tonic level range, number and identity of artifact-rejected trials, percentage of artifact-rejected trials, non-responder status, and the final analytic n contribution for each channel.
    11. Store the quality log with the consolidated databases so that channel-specific data loss is auditable and reproducible.
        NOTE: In the present sample, no participant met the non-responder criterion of step 8.14.3, and no participant exceeded the 25% artifact threshold of step 8.14.6; consequently, no participant was excluded from all channels of a task. The divergences in the analytic n are of two independent kinds: technical loss of the electrodermal channel, which leaves the behavioral and subjective records of that task intact, and incomplete behavioral or self-report records (a task not completed, or missing visual-analog responses), which are independent of electrodermal signal quality. The exact analytic n for every channel and every task is reported in the analytic sample paragraph of the Representative Results. Experiment 3 does not include an SCR channel, so its n reflects behavioral rather than electrodermal data loss. Reporting these criteria explicitly addresses the lack of harmonization in the operationalization and reporting of electrodermal outcomes highlighted by Ruge et al.¹.

9. Databases

1. Run the Python script Data_Bases.py to merge, for each participant, the physiological measurements obtained from the SCR recordings with the responses collected during the experimental task.
   NOTE: The script integrates the information obtained from the physiological recordings processed through Ledalab, the CDA analysis outputs, and the PsychoPy response spreadsheets generated during the experiment.
2. After the script has run successfully, ensure a consolidated spreadsheet containing both physiological and behavioral information for a single participant is generated and stored in the Global_Data folder.
   NOTE: Following the previous example, if Experiment_1 is selected for analysis, the Python script will automatically traverse the folders of all participants associated with that experiment, merge the CDA outputs with the corresponding behavioral data for each participant, and generate an individual spreadsheet file for every participant inside the Global_Data/Exp_1 folder.
3. Once the individual participant databases have been generated, use the Python script Completed_Bases_Per_Experiment.py to merge the data from all participants for each experimental task.
   NOTE: The script combines the information from all participant folders into a unified database for each experiment, facilitating subsequent statistical analysis and data processing. If the procedure is completed successfully, three final spreadsheets will be generated, each corresponding to one of the three experiments:
   One spreadsheet for Experiment_1
   One spreadsheet for Experiment_2
   One spreadsheet for Experiment_3
4. Store the resulting complete databases in the Complete_Data_Bases folder.

Results

Analytic sample

A total of 38 healthy adults completed the full experimental session and the self-report battery (CTQ-SF, BCE, PHQ-9, GAD-7, STAI). Per-analysis sample sizes varied slightly because of channel-specific data loss, not participant withdrawal. Sample sizes by analysis: Experiment 1 — SCR and rating ANOVAs (Pavlovian and operant phases), n = 37; operant button-press ANOVA, n = 38. Experiment 2 — rating analyses (acquisition, extinction, reacquisition) and acquisition SCR, n = 36; extinction SCR and the SCR Phase × Stimulus ART-ANOVA, n = 35. Experiment 3 — all analyses, n = 37. Correlations involving only self-report measures (CTQ, BCE, symptoms) were used for the full sample (N = 38).

Experiment 1: Pavlovian and operant conditioning with avoidance

Pavlovian phase: Repeated-measures ANOVAs across the three CSs (avoidable CS+, unavoidable CS+, CS−) confirmed conditioned differentiation on both explicit measures: expectancy ratings, F(2, 72) = 50.45, p < 0.001, ηp²= 0.584, and threat ratings, F(2, 72) = 19.54, p < 0.001, ηp² = 0.352. Both CS+ types were rated as more predictive of the US (avoidable: M = 72.84; unavoidable: M = 70.97) and more threatening (52.45 and 50.75) than the CS− (16.14 and 18.80; all Bonferroni-corrected ps < 0.001, d_zs = 0.67–1.49), without differing from each other (ps = 1.00). The autonomic measure showed the same direction but a weaker effect: SCR varied across stimuli, F(2, 72) = 5.09, p = 0.009, ηp² = 0.124 (Greenhouse-Geisser p = 0.018), driven by the unavoidable CS+ (M = 0.014, SD = 0.012) relative to the CS− (M = 0.010, SD = 0.008), although this pairwise contrast did not survive Bonferroni correction (p = 0.052, d_z = 0.41). Collapsing the two CS+ into a single index recovered a significant CS+ versus CS− differentiation on SCR, t(36) =2.29, p = 0.028, d_z = 0.38. This dissociation between robust declarative differentiation and weaker, less precise autonomic differentiation is consistent with previous reports that explicit threat appraisals and skin conductance reflect partially separable conditioning channels²¹ (Figure 3, Table 1).

Figure 3: Pavlovian phase of Experiment 1: skin conductance response, expectancy, and threat ratings. (A) Log-transformed skin conductance response (log[microsiemens + 1]). (B) Expectancy rating on a 0–100 visual analog scale. (C) Threat rating on a 0–100 visual analog scale. Boxplots show the median (center line), interquartile range (box), whiskers extending to 1.5 × IQR, and individual outliers. Brackets indicate significant Bonferroni-corrected pairwise paired-samples t-tests (*p < 0.05, **p < 0.01, ***p < 0.001); non-significant pairs are not shown. n = 37. Please click here to view a larger version of this figure.

Measure	Avoidable CS+	Unavoidable CS+	CS-
SCR (log[microsiemens + 1])	0.0109 (0.0086)	0.0140 (0.0123)	0.0100 (0.0075)
Expectancy (VAS 0–100)	72.84 (23.92)	70.97 (26.25)	16.14 (27.94)
Threat (VAS 0–100)	52.45 (27.83)	50.75 (31.67)	18.80 (27.32)

Table 1: Descriptive statistics for the Pavlovian phase of Experiment 1. Cells display M (SD). SCR = skin conductance response, log-transformed as log(SCR + 1) prior to analysis; VAS = visual analog scale ranging from 0 to 100. n = 37.

In the operant phase, once the avoidance contingency was introduced, the three measures converged. Participants pressed the response button more often during both CS+ than during the CS−, F(2, 74) = 9.51, p < 0.001, ηp² = 0.204 (avoidable: M = 8.53; unavoidable: M = 9.04; CS−: M = 7.26; both Bonferroni ps ≤ 0.027). Expectancy, threat, and relief ratings clearly separated the two CS+ in accordance with controllability: expectancy, F(2, 72) = 136.34, p < 0.001, ηp² = 0.791 (unavoidable > avoidable, d_z = 1.56); threat, F(2, 72) = 61.09, p < 0.001, ηp² = 0.629 (unavoidable > avoidable, d_z = 1.02); and relief, F(2, 72) = 14.94, p < 0.001, ηp² = 0.293 (avoidable > unavoidable, d_z = 0.86; all Bonferroni ps < 0.001), reflecting that participants understood that the button reduced shock probability only for the avoidable CS+. Skin conductance did not differ between stimuli during this phase, F(2, 72) = 0.78, p = 0.462, ηp² = 0.021, suggesting that active engagement of the avoidance response attenuated the differential autonomic signal observed during the Pavlovian phase²²(Figure 4, Table 2).

Figure 4: Operant phase of Experiment 1: skin conductance response, avoidance behavior, and relief ratings. (A) Log-transformed skin conductance response (log[microsiemens + 1]). (B) Number of button presses (avoidance behavior); pressing the button cancels the unconditioned stimulus only during the avoidable CS+, not during the unavoidable CS+. (C) Relief rating on a 0-100 visual analog scale. Boxplots show the median, interquartile range, whiskers extending to 1.5 × IQR, and individual outliers. Brackets indicate significant Bonferroni-corrected pairwise paired-samples t-tests (*p < 0.05, **p < 0.01, ***p < 0.001); non-significant pairs are not shown. n = 37 to 38. Please click here to view a larger version of this figure.

Measure	Avoidable CS+	Unavoidable CS+	CS-
SCR (log[microsiemens + 1])	0.0197 (0.0150)	0.0196 (0.0151)	0.0212 (0.0169)
Button presses (count per CS)	8.53 (3.58)	9.04 (4.01)	7.26 (4.27)
Expectancy (VAS 0–100)	33.78 (26.97)	83.92 (19.80)	7.54 (14.58)
Threat (VAS 0–100)	37.12 (25.41)	70.60 (27.33)	11.60 (20.12)
Relief (VAS 0–100)	62.98 (26.36)	29.97 (27.93)	58.75 (36.29)

Table 2: Descriptive statistics for the operant phase of Experiment 1. Cells display M (SD). SCR = skin conductance response, log-transformed as log(SCR + 1); VAS = visual analog scale ranging from 0 to 100. Button presses are the average count per CS presentation. The CS− column reports button presses, expectancy, threat, and relief on trials on which the aversive outcome could not occur. n = 37 for SCR and for the ratings, and n = 38 for button presses; the analytic n per channel is reported in the analytic sample paragraph of the representative results.

Experiment 2: Acquisition, extinction, and reacquisition

A 2 (Phase: acquisition vs. extinction) × 2 (Stimulus: CS+ vs. CS−) within-subject ANOVA was used to formally test extinction; the Phase × Stimulus interaction is the critical test of a selective decrease in the CS+ response. Given the non-Gaussian distribution of SCR, its 2 × 2 ANOVA was computed with the aligned rank transform²³. Differential SCR was present at acquisition (Wilcoxon p = 0.008) and disappeared at extinction (p = 0.168); the ART-ANOVA indicated a non-significant trend toward an overall decrease of autonomic responding across phases (F(1, 34) = 3.24, p = 0.081) but no Phase × Stimulus interaction (F(1, 34) = 0.87, p = 0.358). Expectancy and threat ratings, by contrast, showed large Phase × Stimulus interactions in the parametric 2 × 2 ANOVA (expectancy: F(1, 35) = 67.71, p < 0.001, ηp² = 0.66; threat: F(1, 35) = 5.85, p = 0.021, ηp² = 0.14), confirming selective extinction at the declarative level. The reacquisition phase, by experimental design, presented only the CS+ at the autonomic level; subjective ratings to the CS+ showed a significant rebound, with expectancy rising from M = 25.90 at extinction to M = 55.94 at reacquisition (t(35) = 7.95, p < 0.001, d_z = 1.32) and threat rising from M = 31.26 to M = 42.04 (t(35) = 3.61, p < 0.001, d_z = 0.60), whereas the autonomic response showed no such recovery, with SCR to the CS+ at M = 0.012 at extinction and M = 0.010 at reacquisition (p = 0.568). This pattern, autonomic extinction without subsequent recovery, alongside persistent and rapidly reactivated declarative knowledge, illustrates the value of the multi-system protocol²¹ (Figure 5, Table 3).

Figure 5: Experiment 2: dependent measures across the three phases. (A) Log-transformed skin conductance response (log[microsiemens + 1]). (B) Expectancy rating (visual analog scale 0–100). (C) Threat rating (visual analog scale 0–100). The acquisition and extinction phases included both CS+ and CS- trials; the reacquisition phase, by experimental design, presented only the CS+ at the autonomic level to test the rapid re-emergence of conditioned responding (savings effect). Subjective ratings were collected for both stimuli at every phase. Boxplots show the median, interquartile range, whiskers extending to 1.5 × IQR, and individual outliers. Brackets in panel A indicate Wilcoxon signed-rank tests of CS+ vs. CS- per phase; brackets in panels B and C indicate paired-samples t-tests of CS+ vs. CS- per phase (*p < 0.05, **p < 0.01, ***p < 0.001); non-significant pairs are not shown. The Reacquisition phase has no SCR bracket because, by experimental design, only the CS+ was presented at the autonomic level. Although the median CS+ - CS- difference appears similar between acquisition and extinction in panel A, the proportion of participants showing CS+ > CS- differed (72% at acquisition vs. 60% at extinction), which the Wilcoxon test detects as a difference in within-subject consistency. n = 35 to 36. Please click here to view a larger version of this figure.

Measure	Acquisition CS+	Acquisition CS-	Extinction CS+	Extinction CS-	Reacquisition CS+	Reacquisition CS-
SCR (log[microsiemens + 1])	0.0121 (0.0124)	0.0109 (0.0125)	0.0113 (0.0126)	0.0098 (0.0102)	0.0103 (0.0126)	- (by design)
Threat (VAS 0-100)	47.64 (27.47)	16.83 (20.71)	31.26 (26.32)	12.99 (17.62)	42.04 (26.71)	12.17 (18.90)
Expectancy (VAS 0-100)	73.34 (21.31)	11.42 (18.21)	25.90 (24.07)	7.18 (11.65)	55.94 (23.04)	6.53 (10.38)

Table 3: Descriptive statistics for Experiment 2 (acquisition, extinction, reacquisition). Cells display M (SD). SCR = skin conductance response, log-transformed as log(SCR + 1); VAS = visual analog scale ranging from 0 to 100. The reacquisition phase, by experimental design, presented only the CS+ at the autonomic level in order to test the rapid re-emergence of conditioned responding after extinction (savings effect); the CS− cell for SCR is therefore left empty and is not missing data. Subjective ratings were collected for both the CS+ and the CS− in all three phases. n = 36 for the ratings and for acquisition SCR; n = 35 for extinction SCR.

Experiment 3: Probabilistic reward learning

Participants reliably learned the probabilistic contingencies. The mean proportion of high-probability choices was M = 0.734 (SD = 0.119), significantly above chance, t(36) = 11.97, p < 0.001, d = 1.97. The 80/20 pair was learned more accurately than the 70/30 pair, F(1,36) = 29.56, p < 0.001. Reinforcement-learning parameters fitted per participant produced moderate Q-learning rates (Alpha: M = 0.42, SD = 0.23) and inverse temperatures consistent with moderate exploitation (Beta: M = 4.68, SD = 2.33).

Sensitivity to individual differences in childhood experience

The analyses reported in this subsection are illustrative of the measurement capacity of the protocol and are not confirmatory tests. The study was powered to detect within-subject effects — the canonical paradigm effects reported above — and not between-subject interactions; with N = 38, the individual-difference models are exploratory and are presented only to show that the indices derived from the battery carry interpretable between-participant variance. Four CTQ-by-learning-index moderation models met a pre-specified dual-significance criterion (a significant bivariate correlation between the CTQ subscale and the symptom score, together with a significant interaction term). These four models, the conceptual diagram of the moderation model²⁴, and the featured interaction plot are reported in full in Supplementary File 2 (includes Supplementary Figure 1 and Supplementary Figure 2), which is provided with this submission and is also deposited, together with the analysis code, in the public repository (https://doi.org/10.5281/zenodo.21365750). They should be read as a demonstration that protocol-derived indices are sensitive to individual differences, and not as evidence for any specific neurocognitive signature of childhood adversity; establishing such signatures requires a confirmatory design in a substantially larger sample.

Benevolent childhood experiences (BCE) and CTQ subscales were strongly and inversely correlated, with the largest negative coefficients between CTQ-EN (emotional neglect) and BCE-CPF/DPF, rs = -0.68 and -0.67, both ps < 0.001, N = 38. In addition, BCE scores were associated with higher reinforcement-learning accuracy and lower contingent avoidance, a pattern consistent with a less defensive and more exploratory behavioral signature. These bivariate associations are descriptive: they document that the battery yields indices with sufficient between-participant variance to covary with early-experience measures, and they do not test any mediational or causal pathway. The full bivariate matrix is in Supplementary Table 2 and the accompanying text in Supplementary File 3.

Conclusions of the representative results

Across the three experiments, the protocol produced the canonical effects expected from each paradigm: Pavlovian discrimination (robust at the declarative level and weaker, less precise at the autonomic level), operant avoidance with clear controllability-dependent expectancy, threat, and relief ratings, selective extinction of declarative responding alongside a non-selective decrease of autonomic responding, and probabilistic reward learning above chance. Rather than uniform convergence, the multi-channel design yielded a mixture of convergence (e.g., operant differentiation in both button presses and ratings) and theoretically informative dissociation between physiological, behavioral, and subjective measures (e.g., Pavlovian SCR weaker than declarative differentiation; the autonomic response failing to rebound at reacquisition despite a strong declarative rebound), in line with the view that explicit threat appraisal and skin conductance index partially separable conditioning channels²¹.

It is important to delimit precisely what these results do and do not establish. What is demonstrated is that the battery is technically feasible within a single 60-min session, that each paradigm elicits its canonical effect with the expected direction and magnitude, and that the pipeline yields a set of interpretable per-participant indices (differential SCR contrasts, the operant relief contrast, contingent-avoidance counts, and Q-learning Alpha and Beta parameters) that can be redeployed in larger samples without modification of the protocol. What is not demonstrated is the sensitivity or the specificity of these indices for detecting adversity-related alterations. The sample was dimensioned for within-subject effects, not for individual-difference effects; accordingly, the CTQ and BCE analyses reported above are illustrative of measurement capacity and do not constitute a confirmatory validation of neurocognitive signatures of adverse childhood experiences. Determining which of these indices, if any, discriminate exposure groups or predict clinical outcome is the task of the adequately powered confirmatory studies outlined in the discussion.

The analysis pipeline (R and Python scripts), figure-generation code, and aggregate-level summary tables that support the findings of this study are publicly available in the Zenodo repository, https://doi.org/10.5281/zenodo.21365750 (DOI: 10.5281/zenodo.21365750). De-identified individual-level data (skin conductance traces and behavioral responses) are available from the corresponding author upon reasonable request, subject to a signed data use agreement and approval by the Scientific Ethics Committee of Universidad de los Andes (ID: CEC2025023). Code and tables relevant to the moderation models reported in Supplementary File 2 are included as supplementary files. The analysis script (paper_analyses_R.R; Supplementary Coding File 1) reproduces every statistic reported in the Representative Results, and the figure-generation script (paper_figures_R.R; Supplementary Coding File 2) regenerates Figure 3, Figure 4, and Figure 5 directly from the raw data files. The experimental designs for Experiment 1, Experiment 2, and Experiment 3 are provided in Supplementary Table 3, Supplementary Table 4, and Supplementary Table 5, respectively.

Measure	M	SD
Proportion of high-probability choices (overall)	0.73	0.12
80/20 contingency pair	0.79	0.12
70/30 contingency pair	0.68	0.14
Q-learning rate (Alpha)	0.42	0.23
Inverse temperature (Beta)	4.68	2.33

Table 4: Descriptive statistics for Experiment 3 (probabilistic reward learning). Cells display M (SD). The proportion of high-probability choices indexes accuracy across the experimental choice blocks (chance level = 0.50; see Supplementary Table 5 for the trial structure): overall = 0.73 (0.12); 80/20 contingency pair = 0.79 (0.12); 70/30 contingency pair = 0.68 (0.14). The 80/20 pair provides more deterministic feedback, whereas the 70/30 pair is more probabilistic and has a smaller signal-to-noise ratio. One-sample test of overall accuracy against chance: t(36) = 11.97, p < 0.001, d = 1.97. Within-subject contrast 80/20 vs. 70/30: F(1, 36) = 29.56, p < 0.001. Reinforcement-learning parameters were fitted per participant: Q-learning rate Alpha = 0.42 (0.23), bounded in [0, 1], which governs the size of the value update (moderate updating); inverse temperature Beta = 4.68 (2.33), which governs the exploration–exploitation balance (moderate exploitation). n = 37.

Supplementary Figure 1: Conceptual diagram of the moderation model. The predictor X (a CTQ subscale indexing childhood adversity) is hypothesized to influence the outcome Y (anxiety or depressive symptoms) directly through coefficient b1, while the moderator W (a per-participant learning index derived from the protocol — e.g., operant relief contrast, residual SCR during extinction, or differential threat rating) modulates the strength of the X Y relationship via the interaction term b3. A significant b3 indicates that the effect of childhood adversity on adult symptoms varies as a function of the learning profile, providing the conceptual framework for the four moderation models reported in Supplementary File 2. In the present manuscript four specific W variables are reported (each corresponding to one of the four moderation models that met the dual-significance criterion): (1) the operant relief contrast = subjective relief to the avoidable CS+ minus relief to the CS− (Experiment 1, operant phase); (2) the residual differential SCR during extinction = log-SCR to CS+ minus log-SCR to CS− at the end of the extinction phase (Experiment 2); (3) the operant SCR to the avoidable CS+ = log-SCR to the avoidable CS+ minus log-SCR to the CS− (Experiment 1, operant phase); and (4) the differential threat rating to the unavoidable CS+ = threat rating to the unavoidable CS+ minus threat rating to the CS− (Experiment 1, operant phase, 0–100 visual analog scale).Please click here to download this file.

Supplementary Figure 2: Interaction plot for the featured moderation. The effect of emotional neglect (CTQ-EN) on depressive symptoms (PHQ-9) is plotted at three levels of the moderator W (operant relief contrast: subjective relief to the avoidable CS+ vs. the CS-): low (-1 SD), mean, and high (+1 SD). Diverging regression lines visualize the interaction term b3 from Supplementary Figure 1: at low values of operant relief, the slope of CTQ-EN PHQ-9 is positive (greater neglect, more depressive symptoms), whereas at high values the slope flattens, illustrating a buffering effect of subjective agency over aversive outcomes (B = -2.05, p = 0.041; full model in Supplementary File 2).Please click here to download this file.

Supplementary Table 1: Stimulus parameters in Experiments 1, 2, and 3. This table summarizes the physical and procedural parameters of the stimuli used in the three experiments. For each experiment, it reports the nature and duration of the conditioned stimulus (CS) and the unconditioned stimulus (US), the CS+ US intensity and contingency, the response required from participants, the inter-trial interval (ITI), the dependent measures recorded, and the counterbalancing scheme. Experiment 1 (Pavlovian/operant avoidance) used colored lamps as CSs and an aversive 95-dB sound — white noise or a fork sound, counterbalanced — as the US. Experiment 2 (acquisition/extinction/reacquisition) used neutral human faces as CSs and a 95-dB male scream paired with an angry face as the US. Experiment 3 (probabilistic choice) used pairs of abstract images with probabilistic visual feedback. CS = conditioned stimulus; US = unconditioned stimulus; SCR = skin conductance response; VAS = visual analog scale.Please click here to download this file.

Supplementary Table 2: Pearson correlation matrix among CTQ subscales. Negative correlations between the BCE and the symptom scales reflect the expected protective role of positive childhood experiences. Values are Pearson correlation coefficients. The top section reports correlations among CTQ subscales and BCE scales; the bottom section reports correlations of CTQ subscales and BCE scales with current symptom scores. Significance markers: *p < 0.05, **p < 0.01, ***p < 0.001, +p < 0.10. N = 38 (self-report measures only; the full enrolled sample). Descriptive text accompanying this matrix is provided in Supplementary File 3. EA = emotional abuse, PA = physical abuse, SA = sexual abuse, EN = emotional neglect, PN = physical neglect, BCE scales: CPF = caretaking and protection from family, DPF = positive family dynamics, BCE-Total = overall benevolent childhood experiences score) and symptom scores (PHQ-9 = depression; GAD-7 = generalized anxiety; SAI = state anxiety [STAI-S]; TAI = trait anxiety [STAI-T]).Please click here to download this file.

Supplementary Table 3: Experimental design of Experiment 1 (within-subject). This table presents the within-subject design of Experiment 1, a Pavlovian/operant avoidance paradigm with three conditioned stimuli: an avoidable CS+ (A), an unavoidable CS+ (B), and a CS− (C). Rows correspond to the stimuli and columns to the four phases of the experiment (Pavlovian conditioning, Pavlovian test, operant training, operant test); cells give the number of trials and the contingency in standard associative-learning notation^25,26, where "+" denotes a CS followed by the US, "−"a CS presented without the US, and "R" the button-press response. During the operant phase, R cancels the US for the avoidable CS+ but is ineffective for the unavoidable CS+. The final row indicates the VAS ratings (threat, expectancy, relief) collected after the relevant phases. Each participant completed 52 trials.Please click here to download this file.

Supplementary Table 4: Experimental design of Experiment 2 (within-subject). This table presents the within-subject design of Experiment 2, a standard acquisition–extinction–reacquisition paradigm with two conditioned stimuli: a CS+ (A) and a CS− (B). Rows correspond to the stimuli and columns to the six phases (acquisition, acquisition test, extinction, extinction test, reacquisition, reacquisition test); cells give the number of trials and the contingency, where "+" denotes a trial followed by the US and "−"a trial without the US. During acquisition the CS+ was reinforced on 80% of trials (8 of 10). The reacquisition phase presents only the CS+ by design — this is not missing data. The final row indicates the VAS ratings (expectancy and threat) collected after the relevant phases. Each participant completed 52 trials.Please click here to download this file.

Supplementary Table 5: Experimental design of Experiment 3 (within-subject). This table presents the within-subject design of Experiment 3, a probabilistic learning task with two pairs of abstract stimuli (A–B and C–D). After a practice phase, participants completed two consecutive choice blocks; each block presented only one of the two pairs, and the second block always tested the pair not seen in the first. Rows correspond to the two between-subjects counterbalancing orders of the choice blocks (A–B first vs C–D first) and columns to the phases of the experiment. Reinforcement contingencies were p(A) = 0.80, p(B) = 0.20, p(C) = 0.70, and p(D) = 0.30; the C1/C2 counterbalancing additionally reverses which member of each pair carries the higher probability. Choice was coded as 1 (high-probability stimulus), 0 (no choice/omission), or −1 (low-probability stimulus). Each participant completed 200 trials.Please click here to download this file.

Supplementary File 1: Data management, directory structure, and file-naming conventions. Full description of the directory tree, the counterbalancing subfolders, the participant folders, the file-naming conventions, and the database scripts, with a worked example. Please click here to download this file.

Supplementary File 2: Individual-difference moderation analyses. Rationale and analysis plan for the individual-difference models, the four CTQ-by-learning-index moderations that met the dual-significance criterion, the conceptual diagram of the moderation model (Supplementary Figure 1), and the featured interaction plot (Supplementary Figure 2). These analyses are illustrative of the measurement capacity of the protocol and are not confirmatory tests.Please click here to download this file.

Supplementary File 3: Benevolent childhood experiences and behavioral signatures. Correlations among CTQ subscales, BCE scales, and symptom scores (Supplementary Table 2), and the behavioral phenotypes associated with benevolent childhood experiences.Please click here to download this file.

Supplementary File 4: Procedural recommendations and troubleshooting. The six steps that most strongly determine whether a session yields analyzable data, the troubleshooting rules for a degraded physiological recording, and the modifications available to laboratories with different equipment or recruitment constraints. Please click here to download this file.

Supplementary Coding File 1: paper_analyses_R.RPlease click here to download this file.

Supplementary Coding File 2: paper_figures_R.RPlease click here to download this file.

Discussion

The present results establish the technical and methodological viability of an integrated, single-session battery covering threat acquisition with instrumental control, safety learning, and return of fear, and probabilistic reward learning. In a sample of 35 to 38 healthy adults (analytic sample sizes differed slightly across experiments because of channel-specific signal loss, not participant withdrawal), every paradigm reproduced its canonical effect: differential subjective and — more weakly — autonomic conditioning during the Pavlovian phase, controllability-dependent avoidance and relief during the operant phase, selective extinction of declarative threat alongside a non-selective decrease of autonomic responding in Experiment 2, and reliable above-chance probabilistic learning in Experiment 3. It is important to state precisely what this demonstrates and what it does not. What is demonstrated is that the three paradigms can be run concurrently in one 60-min session without mutual interference, that all three canonical effects survive the compression of trial numbers that such a session imposes, and that the pipeline yields a set of per-participant indices that can be redeployed unchanged in other samples. What is not demonstrated — and what remains to be established — is that these indices are sensitive and specific to the consequences of childhood adversity. Reproducing canonical within-subject effects is a necessary but not a sufficient condition for that claim, and the present sample is underpowered for confirmatory individual-difference tests²⁷. The individual-difference results are accordingly illustrative of measurement capacity, not confirmatory evidence of trauma-learning pathways.

The methodological contribution lies less in any single paradigm than in their joint deployment. As set out in the introduction, the three tasks span three computationally dissociable demands, and measuring them within the same participant permits a question that no single-task study can answer by construction: whether ACE-related alterations reflect a domain-general recalibration of associative learning or a valence-specific alteration confined to threat. A battery combining two threat paradigms, or two reward paradigms, could not falsify either hypothesis against the other, while the present triad can. Three further features distinguish it from single-task protocols: threat and reward learning are measured in the same session, which removes between-session state variability; the multi-channel design yields a reusable toolbox of per-participant indices; and administering the BCE scale alongside the CTQ-SF models the adversity–benevolence continuum rather than adversity alone, in line with Narayan et al.¹⁰ The moderate inverse correlation between the two scales (|r| ≈ 0.50–0.68; Supplementary File 3) confirms that they are distinct yet partially overlapping, justifying their joint inclusion.

The weak and, in places, absent physiological effects deserve extended consideration, because they are not a footnote to the protocol's performance but arguably its most informative outcome. Three patterns stand out. In the Pavlovian phase of Experiment 1, declarative discrimination was robust while the autonomic effect was weak and did not survive correction for multiple comparisons. More strikingly, skin conductance ceased to discriminate between the stimuli altogether once the avoidance response was introduced in the operant phase, at the very moment when expectancy, threat, and relief ratings separated them precisely according to controllability. And in Experiment 2, autonomic responding extinguished and then failed to rebound at reacquisition, while declarative responding rebounded strongly. The corresponding statistics are reported in the representative results section. Theoretically, the operant result is not a measurement failure but an expected consequence of successful instrumental control: when an available action reliably cancels the aversive outcome, defensive arousal is down-regulated even though the organism continues to represent the cue as dangerous. This is precisely the reconceptualization of avoidance advanced by LeDoux et al.²² and the classical controllability effect described in the learned-helplessness literature²⁸: the instrumental response is itself a regulator of arousal, and its introduction should therefore remove the very autonomic differential that preceded it. The Experiment 2 pattern, in turn, is a textbook case of the partial coupling between response systems documented by Lonsdorf et al.²¹: skin conductance and explicit appraisal index are partially separable processes, and Dibbets and Evers¹¹ themselves reported an individual-difference effect on ratings with no corresponding effect on SCR in the very paradigm adapted here. This matters specifically for research on childhood adversity. If early adversity preferentially recalibrates the autonomic channel while leaving declarative appraisal intact — or the reverse — then a single-channel protocol would yield opposite conclusions depending on which channel the investigator happened to record, and this is one plausible source of the heterogeneity that Ruge et al.¹ document across a literature in which no outcome measure has been shown to be universally more reliable. Multi-channel recording is therefore not redundant but a precondition for interpretable individual-difference work. Clinically, the persistence of subjective threat after autonomic extinction is the laboratory analogue of the residual threat appraisal that survives exposure therapy and predicts relapse, and it is a plausible treatment target in its own right. We nevertheless emphasize the honest alternative explanation: attenuated SCR effects are also consistent with low power and low reliability of the autonomic channel under the trial-number constraints of a 60-min battery — few trials per cell, habituation across three consecutive tasks, and the well-known presence of SCR non-responders. This account cannot be ruled out with the present data, and the two explanations are not mutually exclusive. Studies seeking to arbitrate between them should apply the signal-quality criteria specified in step 8.14 of the Protocol, pre-register the exclusion of non-responders, and expect that a longer protocol with more trials per cell is required before an absent autonomic effect can be interpreted as evidence of dissociation rather than of insufficient measurement precision.

Beyond reproducing canonical paradigm effects, the protocol-derived indices were sensitive enough to detect theoretically meaningful associations with childhood adversity in a modest sample. Four CTQ-by-learning-index moderation models met a dual-significance criterion (a significant bivariate correlation and a significant interaction term), and the CTQ and BCE scales showed the expected moderate inverse correlation. Both findings are reported as illustrative of the protocol's measurement capacity rather than as confirmatory tests of trauma-learning pathways; full statistical detail and theoretical interpretation are provided in Supplementary File 2.

Three factors determine the quality of the electrodermal record and warrant explicit attention during data collection: an even application of conductive gel across both electrodes, with firm but comfortable contact on the first phalanx of the middle and ring fingers of the non-dominant hand; a continuous wireless connection of the wrist strap, verified before each task and again at each inter-task transition, because brief losses of communication are silent at the participant's end but produce flat segments in the trace; and immobility of the non-dominant hand, because finger contractions and postural shifts generate transients that automated phasic decomposition cannot distinguish from genuine responses. If communication is lost mid-task, continue the experiment and do not repeat the session: treat the affected channel as missing data for that task only, while the participant remains usable for the other paradigms and for the behavioral and subjective channels of the same task. This decision rule explains why the analytic n varied between 35 and 38 across experiments. The remaining operational recommendations, the counterbalancing and calibration checks, the Ledalab parameters, and the full troubleshooting guide are given in Supplementary File 4.

Several limitations should be considered when interpreting these results, and they jointly define the status of this work as a translational methodological proposal rather than a confirmatory study. First, and most importantly, the sample (n = 35–38, varying across experiments) is adequate for the within-subject paradigm effects reported here but is underpowered for confirmatory individual-difference analyses. Fritz and MacKinnon²⁴ estimate that N ≥ 71 is required to detect a medium-mediated effect, and McClelland and Judd²⁷ show that detecting an interaction in a modest community sample requires several times the sample needed for a main effect. The moderation models are therefore reported as illustrative of the measurement sensitivity of the protocol, not as tests of trauma–learning pathways, and no claim of a neurocognitive signature of adversity is made. The direct consequence is that the clinical value of the protocol remains to be established: the indices derived here must next be submitted to adequately powered, pre-registered validation in larger and, ideally, extreme-group-enriched or clinical samples before any of them can be considered a candidate marker. Second, skin conductance is intrinsically vulnerable to movement artifacts and to intermittent wireless connection loss, and its yield varied across the three experiments in the present sample; the behavioral and subjective channels were more robust, which is precisely why the protocol records them concurrently rather than relying on the autonomic channel alone. Laboratories adopting the battery should expect channel-specific data loss and should report the analytic n per channel and per task (see step 8.14.9). Third, the protocol is bounded by the duration of a single laboratory visit (approximately 60 min, including setup), which constrains the number of trials per paradigm and therefore places a floor on the within-subject reliability of the derived per-participant indices; longer or multi-session protocols would yield more stable individual estimates at the cost of fatigue or within-session control. Finally, the battery has so far been administered only to healthy adults aged 18 or older, and its applicability to minors, older adults, and clinical populations remains to be established. What the present manuscript contributes is the standardized, reproducible measurement infrastructure — paradigms, acquisition parameters, quality criteria, and index derivation — on which those confirmatory clinical validations can now be built.

Three future applications follow naturally from the protocol. First, larger samples (n ≥ 71 to detect medium mediated effects)²⁴ would support formal mediation tests of the early-adversity -> learning -> symptoms pathway, ideally with pre-registered hypotheses and pre-specified moderator-by-predictor cells. Second, clinical validation comparing healthy controls to clinical groups (post-traumatic stress disorder, major depressive disorder) would identify which protocol-derived indices best discriminate diagnosis or predict treatment response. Third, the 6-month re-test wave already covered by the informed consent will allow evaluation of within-person stability and treatment sensitivity of the indices, which is a prerequisite for their use as candidate biomarkers. We recommend that studies on the long-term psychological impact of childhood adversity adopt integrated, multi-system protocols of this type rather than single-task or self-report-only designs.

Materials

List of materials used in this article

Name	Company	Catalog Number	Comments
AcqKnowledge software	BIOPAC Systems Inc.	ACK100W (Windows) / ACK100M (Mac)	Data acquisition and analysis software for physiological signals. RRID:SCR_014279 (verify with current SciCrunch entry).
afex (R package)	Singmann, Bolker, Westfall, Aust & Ben-Shachar	Version 1.5-1	R package for factorial experiments. Used for all repeated-measures ANOVAs (Experiments 1-3) via aov_ez(). RRID:SCR_022761 (verify with current SciCrunch entry).
ARTool (R package)	Wobbrock, Findlater, Gergle & Higgins	Version 0.11.2	Aligned Rank Transform for nonparametric factorial ANOVA. Used for the Phase x Stimulus ANOVA on skin conductance in Experiment 2, whose distribution is non-Gaussian.
Benevolent Childhood Experiences (BCE) scale	Narayan, Rivera, Bernstein, Harris & Lieberman (2018)	Public-domain instrument (10 items)	Spanish adaptation used. No RRID assigned to this instrument; cite original validation paper.
BioNomadix Wrist Strap (Transmitter)	BIOPAC Systems Inc.	BN-PPGED-T	Wearable transmitter for EDA and PPG signals. Used in pair with the BioNomadix receiver module.
Childhood Trauma Questionnaire - Short Form (CTQ-SF)	Bernstein et al. (2003)	Validated 28-item retrospective self-report	Five subscales (EA, PA, SA, EN, PN). No RRID; cite original validation paper. Used Spanish adaptation.
Conductor gel for EDA electrodes	BIOPAC Systems Inc.	GEL101A	Isotonic gel for skin conductance measurements.
Disposable Ag/AgCl EDA electrodes	BIOPAC Systems Inc.	EL507A	Disposable electrodes for skin conductance recording.
Generalized Anxiety Disorder Scale (GAD-7)	Spitzer, Kroenke, Williams & Löwe (2006)	7-item self-report	Validated Spanish version used. No instrument RRID; cite original.
ggplot2 (R package)	Wickham	Version 4.0.3	Grammar-of-graphics plotting system. RRID:SCR_014601 (verify with current SciCrunch entry).
Headphones (over-ear)	HyperX (HP Inc.)	Cloud III	Wired closed-back gaming headset; used to deliver the auditory unconditioned stimulus during fear conditioning.
Ledalab	Benedek & Kaernbach (2010)	Open-source MATLAB toolbox v3.2.5	Continuous Decomposition Analysis (CDA) of skin conductance signals. RRID not assigned (cite original paper instead).
lme4 (R package)	Bates, Maechler, Bolker & Walker	Version 1.1-37	Linear mixed-effects models. Used to estimate the per-participant learning-trajectory slopes in Experiment 3. RRID:SCR_015654 (verify with current SciCrunch entry).
MATLAB	MathWorks	R2025a	Used to run Ledalab and the SCR pre-processing pipeline. RRID:SCR_001622.
Patient Health Questionnaire-9 (PHQ-9)	Kroenke, Spitzer & Williams (2001)	9-item self-report	Validated Spanish version used. No instrument RRID; cite original.
Physiological data acquisition unit (PsychoPy)	BIOPAC Systems Inc.	MP200	Main physiological signal acquisition unit. Combined with BioNomadix wireless EDA module.
Python	Python Software Foundation	Version 3.12	Used for the pre-processing pipeline that merges the Ledalab output with the PsychoPy behavioural records (Data_Bases.py, Completed_Bases_Per_Experiment.py). RRID:SCR_008394 (verify with current SciCrunch entry).
R (statistical computing environment)	R Foundation for Statistical Computing	Version 4.5.0	Environment in which all statistical analyses were performed. Moderation models were estimated by ordinary least squares regression with the lm() function of the base stats package, with predictors z-standardized and the interaction term computed as their product; no specialized moderation package was used. RRID:SCR_001905 (verify with current SciCrunch entry).
State-Trait Anxiety Inventory (STAI)	Spielberger et al. (1983)	STAI Form Y (40 items, SAI + TAI)	Validated Spanish version used. No instrument RRID; cite original.
Stimulus presentation monitor	Samsung	LS22D300GALXZS (S22D300, 22-inch LED)	Display used for visual CS+ / CS- stimuli and rating scales.