Childhood trauma scores showed no meaningful link to amygdala-prefrontal coupling in 187 adults

A widely discussed idea holds that childhood maltreatment weakens the coupling between the amygdala, which responds to threat, and the ventromedial prefrontal cortex, which helps regulate that response. We tested one resting-state form of the idea with the analysis fixed before any questionnaire value or image was seen. The data are the Transdiagnostic Connectome Project on OpenNeuro (CC0): 241 adults with and without psychiatric diagnoses, preprocessed in full, of whom 187 met the motion and questionnaire rules. Childhood trauma scores were unrelated to amygdala-vmPFC resting connectivity in 187 adults: partial r -0.006, 95% interval -0.152 to 0.140.c1 The 90% interval, -0.129 to 0.117, lies inside -0.20 to 0.20, so effects of 0.20 or larger in size are excluded at the preregistered bound.c2 A secondary region gave r -0.025. Shuffling the trauma scores 10,000 times placed the observed value at the 47th percentile. Fifteen exploratory sensitivity analyses moved the primary r between -0.081 and 0.081, and none reached an association.c3 The emotional-neglect and physical-abuse subscale analyses, r 0.081 and -0.081, are inconclusive because their 90% intervals touch 0.20.c4 Power to detect a correlation of 0.20 was 0.77, so true effects between 0.10 and 0.20 remain open.c5

The amygdalaDefinition: amygdala neuroanatomy A pair of almond-shaped structures deep in the temporal lobes, one per hemisphere, involved in detecting threat and in emotional learning. Excludes its separate subnuclei, which the anatomical masks here do not distinguish. full definition and the ventromedial prefrontal cortex (vmPFC)Definition: ventromedial prefrontal cortex (vmPFC) neuroanatomy The lower middle part of the frontal lobe, linked to the regulation of emotional responses. Here defined by the Harvard-Oxford label Frontal Medial Cortex. Excludes functional definitions from a task, and the dorsal and lateral prefrontal cortex. full definition form a circuit often linked to threat and its regulation. A common hypothesis is that maltreatment in childhood changes this circuit, and that the change shows up as weaker resting-state functional connectivityDefinition: resting-state functional connectivity neuroimaging The correlation over time between the blood-oxygen signals of two brain regions while a person lies in the scanner with no task. Excludes anatomical connections, activity evoked by a task, and the direction of any influence between regions. full definition between the two.

Birn and colleagues [2] studied 27 male combat veterans, 17 of them with PTSD, and modelled amygdala and hippocampus connectivity from childhood maltreatment, combat exposure and PTSD symptoms together. Their abstract reports that scores on the Childhood Trauma Questionnaire (CTQ)Definition: Childhood Trauma Questionnaire (CTQ) clinical psychology A 28-item adult self-report questionnaire about childhood experiences, scored as five subscales (emotional abuse, physical abuse, sexual abuse, emotional neglect, physical neglect) of 5 to 25 each and a total of 25 to 125, plus minimisation items that flag denial. Excludes records of maltreatment from the time it happened, and clinical diagnosis. full definition predicted lower amygdala connectivity to the vmPFC.

Goltermann and colleagues [3] tested the same family of claims in 1,325 adults (624 with major depression, 701 healthy controls), with the CTQ and regions chosen from earlier reports, among them the amygdala. Maltreatment was not associated with the connectivity of any region of interest in the full sample. Within the depressed group, maltreatment went with greater, not weaker, connectivity between the amygdala and dorsolateral frontal areas. The authors conclude that most earlier connectivity correlates of maltreatment did not replicate.

Our study adds a smaller, independent test on open data with one feature the larger study did not state in its abstract: a preregistered smallest effect of interestDefinition: smallest effect of interest statistics (equivalence testing) The size of effect, fixed before the data are seen, below which an effect is treated as too small to matter for the decision at hand. Here a correlation of 0.20 in size. Excludes a claim that smaller effects are zero. full definition and an equivalence test (two one-sided tests)Definition: equivalence test (two one-sided tests) statistics A test that supports the absence of a meaningful effect when the 90 percent interval of the estimate lies wholly inside the band set by the smallest effect of interest, which is the same as rejecting both edges of the band at the 5 percent level. Excludes a test of whether the effect is exactly zero. full definition in the form Lakens [4] describes for correlations. An equivalence test can support "no effect of a size that matters", which a non-significant result alone cannot.

The preregistrationDefinition: preregistration research method (metascience) A dated written plan for a study, fixed and hash-locked before any result is seen: the question, the data, the statistic, the controls, the forecasts and the kill conditions. Excludes a plan revised after looking at the outcome, and a protocol written only for the record after the run. full definition was frozen on 9 October 2026, before any phenotype value or image from the dataset was downloaded or viewed. Only file names, the dataset description and the CTQ definitions file had been read. Three choices were made at that point: an anatomical vmPFC, the full sample with no diagnostic filter, and a smallest effect of interest of 0.20 in size, about 4 percent shared variance.

The decision rule had three outcomes, fixed in advance:

The preregistration also stated that effects between 0.10 and 0.20 could be missed at the expected sample size, and that any write-up must say so.

Six deviations are logged in the preregistration, each dated with the point at which it was decided:

The data are the Transdiagnostic Connectome Project, OpenNeuro dataset ds005237, snapshot 1.1.3 [1], released under CC0, a public-domain dedication. It is a transdiagnostic adult sample with up to four resting-state runs per participant. This paper reports group statistics only; per-participant tables were not published and no individual record appears here.

All 241 participants were preprocessed with fMRIPrepDefinition: fMRIPrep neuroimaging software An open preprocessing pipeline for functional MRI that corrects head motion and distortion, aligns images to a standard brain space and writes nuisance signals for later cleaning. Excludes the denoising and statistics that follow, which were run by our own code. full definition 25.2.6, with no processing errors. Of these, 230 had four runs, 6 had three and 5 had two. Distortion correction used acquired field maps for 198 participants and the field-map-free method for 43.

The analysed sample of 187 had a mean age of 33.4 years (SD 13.3). It included 112 women, 73 men and 2 people who reported another sex; 122 were scanned at one site and 65 at the other; 114 were in the patient group and 73 in the general-population group. The mean CTQ total was 43.0 (SD 17.3) on a scale of 25 to 125.

Regions came from the Harvard-Oxford atlasDefinition: Harvard-Oxford atlas neuroimaging A probabilistic atlas of cortical and subcortical regions in standard brain space. At a 50 percent threshold a voxel belongs to a region when at least half of the reference brains placed it there. Excludes regions drawn from each participant's own anatomy or from functional data. full definition at the 50 percent threshold in standard space: left and right amygdala combined; "Frontal Medial Cortex" for the primary vmPFC target; and "Subcallosal Cortex", the label closest to Brodmann area 25, as the pre-declared secondary target.

Each run was cleaned by regression of 24 motion parameters, aCompCorDefinition: aCompCor neuroimaging (denoising) Anatomical component correction: the main components of the signal in white matter and cerebrospinal fluid, where little neural signal is expected, are regressed out as noise. Excludes global signal regression, which removes the mean of the whole brain. full definition (up to five white-matter and five cerebrospinal-fluid components) and cosine drift terms, with no global signal regressionDefinition: global signal regression neuroimaging (denoising) Removing the average signal of the whole brain from every region before connectivity is computed. Excludes aCompCor and motion regression, which remove other nuisance signals. full definition in the primary analysis. Framewise displacementDefinition: framewise displacement neuroimaging (motion quality control) A summary of how far the head moved between two consecutive brain volumes, in millimetres, from the six rigid-body motion parameters. Excludes motion within a single volume. full definition was computed from low-pass filtered motion parameters, and censoringDefinition: censoring neuroimaging (motion quality control) Dropping single brain volumes whose head motion exceeds a threshold before connectivity is computed. Excludes dropping whole runs or whole participants, which follow separate rules. full definition removed volumes above 0.1 mm together with uncensored segments shorter than five volumes. A run was unusable with more than 20 percent of volumes censored, and a participant needed at least two usable runs.

For each usable run $j$, $r_j$ is the Pearson correlation of the mean denoised time series of the two regions. It is converted with the Fisher zDefinition: Fisher z statistics The transform z = artanh(r) of a correlation, which makes its sampling spread close to normal so that correlations can be averaged and given intervals. Excludes any change to the sign or ranking of correlations. full definition transform,

and averaged over the participant's $R$ usable runs:

The exposure is the CTQ total, the sum of the five subscales. The statistic is the partial correlationDefinition: partial correlation statistics The correlation between two quantities after the linear effect of a set of covariates has been removed from both. Excludes adjustment for covariates that were not measured, and any statement about cause. full definition $r_p$ between CTQ total and $Z$, adjusting for age, sex, site, mean filtered framewise displacement and the number of retained volumes, which the result file records as $k=6$ model terms. Intervals use the Fisher transform of $r_p$ with standard error

and run from $z_p-c·\mathrm{SE}$ to $z_p+c·\mathrm{SE}$, transformed back to the correlation scale, with $c$ 1.96 for the 95 percent interval and 1.645 for the 90 percent interval. With $n=187$ this gives the intervals in Table 1. Statistical powerDefinition: statistical power statistics The chance that a test detects an effect of a stated size, given the sample size and the threshold for significance. Excludes the chance that a detected effect is real. full definition uses the same normal approximation.

The controlDefinition: control statistics and experimental design A version of the data with the structure under test removed (for example shuffled or pooled), scored the same way as the real data, so the real score has something to beat. Excludes a comparison with a different dataset or with a published number. full definition shuffled CTQ totals across participants 10,000 times (seed 20261029) and recomputed the partial correlation each time, giving a null distributionDefinition: null distribution statistics The spread of scores a control gives over many random draws. Its 95th percentile is the bar a real score must clear. Excludes a theoretical distribution assumed rather than drawn from the data. full definition for both the primary and the secondary test. It shows where the observed value sits among values that carry no link by construction.

There is no positive controlDefinition: positive control experimental design A test of a known effect run through the same pipeline, to show that the pipeline can find an effect of the expected kind. Excludes shuffled controls, which show what no effect looks like. full definition. No known effect was planted in the data, and no planted checkDefinition: planted check research method; called a positive control in experimental science Data built so that the right answer is known, run through the same analysis before the real data, to show that the analysis can see the effect when it is there. Excludes any check made on the real data, which cannot show what the analysis misses. full definition shows that this pipeline would find a connectivity effect of size 0.20 when one is present. The equivalence call rests on the interval width, which depends on the sample size and not on the pipeline's sensitivity. A pipeline that destroyed real signal would also produce a narrow interval around zero. This is the main weakness of the design.

Of 241 participants, 17 were excluded for having fewer than two usable runs after motion censoring, 16 had no CTQ record and 21 were missing at least one subscale, leaving 187.

Childhood trauma scores were unrelated to amygdala-vmPFC resting connectivity in 187 adults: partial r -0.006, 95% interval -0.152 to 0.140.c1 The 90% interval, -0.129 to 0.117, lies inside -0.20 to 0.20, so effects of 0.20 or larger in size are excluded at the preregistered bound.c2 The decision is "no meaningful association". The secondary region gave the same decision (Table 1).

The unadjusted p values were 0.94 and 0.74. Under the shuffle control the primary value sat at the 47th percentile of 10,000 shuffles (two-sided permutation p 0.93) and the secondary at the 37th (p 0.73), as Figure 1 shows. The shuffled primary values ran from -0.119 to 0.124 between the 5th and 95th percentiles.

Power to detect a correlation of 0.20 was 0.77, so true effects between 0.10 and 0.20 remain open.c5 At the Holm-adjusted threshold the power was 0.68. The smallest correlation detectable with 80 percent power was 0.207. The chance that the equivalence test succeeds when the true correlation is zero was 0.71.

Fifteen exploratory sensitivity analyses moved the primary r between -0.081 and 0.081, and none reached an association.c3 Figure 2 draws all of them with the two registered tests; Table 2 in Appendix A lists them. Thirteen of the fifteen met the equivalence rule on the primary region. The emotional-neglect and physical-abuse subscale analyses, r 0.081 and -0.081, are inconclusive because their 90% intervals touch 0.20.c4 Their 90% intervals run from -0.041 to 0.201 and from -0.202 to 0.042. Each of these single subscales neither shows an association nor excludes one of 0.20.

The same sensitivity analyses were also run on the secondary region. Two of them were inconclusive there: the band-stop motion rule (n 131, r -0.110, 90% interval -0.253 to 0.039) and the sexual-abuse subscale (r -0.098, 90% interval -0.218 to 0.025). None reached an association.

The registered result is an honest nullDefinition: honest null research method (our usage) A working test that found no effect beyond a stated bound, a forecast that failed, or an error of our own found and fixed, reported with the same weight as a positive result. Excludes a void run, and a result left unpublished because it was negative. full definition with a stated bound. Within this sample and this measure, the correlation between CTQ total and amygdala-vmPFC resting connectivity is smaller than 0.20 in size. That is consistent with Goltermann and colleagues [3], whose abstract reports no maltreatment association with the connectivity of its regions of interest in 1,325 adults.

It does not contradict Birn and colleagues [2] directly. Their sample was 27 male veterans, most with PTSD, and their model held combat exposure and PTSD symptoms fixed alongside maltreatment; ours is a mixed transdiagnostic sample of women and men (112 and 73), with demographic and motion covariates. An effect confined to a population like theirs could exist and be diluted here. The difference in design is the main reason the two results can both stand.

The inconclusive subscale results deserve plain reporting. Emotional neglect and physical abuse gave correlations of opposite sign, each with a 90 percent interval that crosses 0.20 by a small margin. They come from fifteen exploratory analyses with no correction, so one or two such edges are expected by chance. They do not support an association, and they also do not let us extend the equivalence claim to each subscale.

The deeper limit is the missing positive control. A null from a pipeline never shown to detect a known effect is weaker than a null from one that has. Adding one, for example a well-replicated connectivity difference with age or between homologous regions, is the next step before this pipeline is used for further trauma questions.