Neural bias to social feedback moderates link between childhood abuse and subsequent harsh parenting

Abstract

Childhood abuse is often transmitted intergenerationally, leaving many families trapped in cycles of abuse. Because child abuse is transmitted through social interactions, developing effective intervention targets to disrupt these cycles may require understanding the mechanisms that support processing social feedback. This study examines whether neural biases in processing positive or negative social feedback moderate the link between history of childhood abuse and current expression of harsh parenting. Using fMRI and well-matched social and monetary feedback tasks, we studied female caregivers (N = 56; 46.16 ± 7.49 years) of adolescents. In the social feedback task, activation to negative relative to positive feedback in the anterior hippocampus, putamen, amygdala, dorsal anterior cingulate cortex (dACC), posterior cingulate cortex (PCC), and medial prefrontal cortex (mPFC) moderated the link between childhood abuse and harsh parenting. Specifically, abuse severity was most strongly associated with harsh parenting among individuals with greater activation to negative relative to positive feedback. No comparable effects emerged from the monetary task. These findings suggest sensitivity to social feedback, and negative feedback in particular—may promote risk for intergenerational abuse transmission.

Introduction

Childhood maltreatment by a parent or caregiver impacts over a third of all children (Stoltenborgh et al. 2015). Maltreatment contributes to 25%–50% of psychiatric diagnoses (McLaughlin et al. 2012, Li et al. 2016), increased risk of drug use, risky sexual behavior, and suicide attempts (Norman et al. 2012), resulting in an annual population lifetime burden of nearly half a trillion dollars in the United States alone (Peterson et al. 2018). Parents with a history of childhood maltreatment are 2–3 times more likely to maltreat their own children (Assink et al. 2018, Madigan et al. 2019). This intergenerational transmission of maltreatment often manifests as harsh parenting, including corporal punishment or verbal hostility (Vaillancourt et al. 2017, van IJzendoorn et al. 2020, Lotto et al. 2023). Identifying factors that potentiate or buffer the association between childhood abuse and harsh parenting could inform interventions to disrupt intergenerational risk for transmission of maltreatment.

One factor contributing to the challenge of understanding neurocognitive mechanisms implicated in intergenerational transmission of abuse is that childhood maltreatment includes two distinct types: abuse, characterized by harm or threat of harm to the child, and neglect, characterized by the absence of expected environmental inputs during development (McLaughlin et al. 2014, 2021, Sheridan and McLaughlin 2014). Abuse and neglect result in distinct neural and cognitive sequelae. Abuse results in overgeneralized threat learning, driven by alterations in the hippocampus, amygdala, medial prefrontal cortex (mPFC), and anterior insula, which may enhance rapid threat detection in abusive environments (McLaughlin et al. 2014, 2019, Lambert et al. 2019, Weissman et al. 2022, Peverill et al. 2023). In contrast, neglect is characterized by impaired cognitive control and executive function, linked to disruptions in the lateral prefrontal and parietal cortices (McLaughlin et al. 2014, 2019, Machlin et al. 2019, Rosen et al. 2020, Johnson et al. 2021). Abuse also has greater intergenerational persistence compared to neglect (Berlin et al. 2011, Madigan et al. 2019, Buisman et al. 2020). For these reasons, the present analyses focus on abuse. Because abuse is inherently interpersonal, social processing may be a key pathway through which risk is transmitted. Indeed, childhood abuse is associated with biases toward negative social information (Ayoub et al. 2006), including a greater tendency to interpret neutral facial expressions as negative (Pfaltz et al. 2019), and recalling more negative vs. positive self-referent words (Vrijsen et al. 2017). From a social learning perspective, these biases may reflect expectations that shape how social feedback is interpreted and responded to (McLaughlin and Sheridan 2016), and when paired with overgeneralized threat learning, may lead to negative distortions of social feedback.

A negativity bias in social processing has direct implications for parenting. Parents with greater adversity exposure demonstrate stronger hostile attribution biases toward their children’s behavior, which are, in turn associated with harsher parenting (Milner and Crouch 1997, Griffith et al. 2022, Milner et al. 2024). Childhood maltreatment has also been linked to deficits in regulating negative affect during interpersonal stress, which may impair responses to challenging social interactions (Kim and Cicchetti 2010), contribute to attributional biases (Dadds et al. 2003, Berlin et al. 2011, Chen et al. 2012, Schofield et al. 2013, Martoccio et al. 2022), and thereby promote intergenerational maltreatment. However, resilience following childhood abuse is not uncommon (Mcgloin and Widom 2001). Just as biased processing of negative feedback may potentiate abuse risk, heightened engagement with positive social feedback may be protective. For example, individuals who are phenotypically predisposed to be more attuned to positive social cues, despite abuse exposure, may experience greater positive emotions during social interactions and be less likely to engage in harsh parenting. Indeed, high quality social relationships can buffer intergenerational abuse transmission (see Schofield et al. 2013 for a meta-analysis of moderators). Thus, greater sensitivity to negative social feedback may index vulnerability, whereas greater sensitivity to positive social feedback may index resilience. Isolating phenotypic differences in how the brain responds to both positive and negative social feedback may shed light on mechanisms implicated in intergenerational abuse transmission.

To test these relations, it is crucial to isolate neural responses to social feedback from more general sensitivity to positively or negatively valenced feedback. In this study, female primary caregivers with varying levels of early childhood abuse exposure and current harsh parenting completed social and monetary feedback tasks during fMRI that were equivalent in structure and timing (Quarmley et al. 2019, Jarcho et al. 2022, Wyngaarden et al. 2024). We assessed the extent to which individual differences in neural processing of social or monetary feedback moderate the association between childhood abuse and risk for harsh parenting. We focused on a set of overlapping a priori regions of interest (ROIs) implicated in social feedback processing (Vijayakumar et al. 2017, Feng et al. 2021, Martins et al. 2021, Mwilambwe-Tshilobo and Spreng 2021) and the neural sequela of childhood abuse (Hart and Rubia 2012, Heany et al. 2018, McLaughlin et al. 2019, Kraaijenvanger et al. 2020, Hosseini-Kamkar et al. 2023). We included brain regions associated with altered functional engagement following childhood abuse because individual phenotypic differences in processing social feedback could influence activation in brain regions shaped by the unique interpersonal nature of childhood abuse. ROIs included the anterior hippocampus, amygdala, and anterior insula, which have been implicated in threat detection and encoding (Phelps and LeDoux 2005, Strange et al. 2014), the dorsal anterior cingulate cortex (dACC), which integrates emotional salience and conflict monitoring (Botvinick et al. 2004), posterior cingulate cortex (PCC) and mPFC, which are implicated in social evaluative processing (Denny et al. 2012), and the putamen, implicated in feedback learning/motivation (Delgado et al. 2000).

Few studies test the relation between harsh parenting and brain function in parents of adolescent children. Thus, direct evidence linking harsh parenting to these ROIs is limited. To help evaluate specificity of results, we conducted sensitivity analyses of intergenerational transmission effects in brain regions less consistently implicated in task-based fMRI studies of childhood abuse (Teicher and Samson 2016, Hosseini-Kamkar et al. 2023), but with well-established roles in social processing (i.e. superior temporal sulcus, temporoparietal junction; Amodio and Frith 2006). Finally, the bilateral precentral gyrus was selected as a nonsocial control region due to its primary role as a motor region (Banker and Tadi 2026). This study provides a novel framework for understanding how phenotypic differences in social feedback processing may perpetuate or disrupt risk for intergenerational cycles of childhood abuse.

Methods

Participants

Participants (N = 56, 46.16 ± 7.49 years) included in analyses were female primary caregivers of 11–18-year-old children (Table 1), recruited through community advertisements in the greater Philadelphia area and from a database of parents whose children had previously completed studies in our lab. We limited recruitment to females based on concerns about diminished statistical power when controlling for or testing gender effects. The age range for participants’ children was chosen because this developmental stage is characterized by heightened parent-child conflict (Hadiwijaya et al. 2017, Mastrotheodoros et al. 2020), providing an optimal window to capture harsh parenting behaviors.

Table 1.

Sample demographics.

Participant mean age (SD)	46.16 (7.49)
Child mean age (SD)	15.50 (2.02)
Gender identity
Female	56
Male or other	0
Median yearly income (Q1–Q3)	$55,500 ($31,500–$89,750)
Race
White	32
Multi-racial	3
Black/African American	16
Asian/Pacific Islander	5
Ethnicity
Hispanic	7
Non-Hispanic	49
Childhood Abusea Score (SD)	26.18 (11.17)
Harsh Parentingb Score (SD)	1.48 (0.33)

^aChildhood Abuse scores derived from the Childhood Trauma Questionnaire.

^bHarsh Parenting scores derived from the Parenting Styles and Dimensions Questionnaire.

Eligible participants underwent screening via Zoom for exclusionary criteria (e.g. contraindications for fMRI, neurological or physical impairments, head injuries with loss of consciousness, etc.) before completing informed consent for procedures approved by the Temple University Institutional Review Board. Participants then attended one in-person visit where they underwent fMRI, completed questionnaires and were compensated $20 per hour for their time. Data for three additional participants were excluded from analyses due to falling asleep (n = 1) or excessive motion (fraction of censored motion >0.20; n = 2).

Measures

Childhood trauma

Childhood abuse was assessed using the self-report Childhood Trauma Questionnaire (CTQ-SF; Bernstein et al. 2003). The CTQ has two subscales: abuse (emotional, physical, and sexual domains) and neglect (physical and emotional domains). Each domain is measured with five items and uses a 5-point Likert scale (1 = Never to 5 = Very Often True). Consistent with prior studies of abuse (Sumner et al. 2014, Heleniak et al. 2016, McLaughlin and Sheridan 2016), to better characterize the sample, we applied thresholds derived by Walker et al. (1999) to the three abuse subscales: sexual abuse ≥8, emotional abuse ≥10, and physical abuse ≥8. Individuals at or above threshold on any three abuse subscales would have a high likelihood (ROC sensitivity and specificity ≥0.85) of being characterized by a clinician as having experienced significant abuse. Using these thresholds, 60% of the participants experienced significant abuse. Sensitivity analyses that include neglect are reported in Supplementary Materials.

Harsh parenting

Current harsh parenting was assessed using the self-report Parenting Styles and Dimensions Questionnaire (PSDQ-SF; Robinson 2001). The PSDQ quantifies three styles: authoritarian (i.e. harsh), authoritative, and permissive. We focused on harsh parenting (verbal hostility, corporal punishment/physical coercion, and nonreasoning punitive strategies) as it is most relevant to abuse risk (Gershoff 2002). Each strategy is measured by 4 items and uses a 5-point Likert scale (1 = Never to 5 = Always). Consistent with recommendations (Robinson 2001), the score for harsh parenting was averaged across strategies. Higher scores reflect greater tendency toward harsh parenting (range: possible = 1–5; observed = 1–2.3; Cronbach’s α = 0.78). One participant left one item blank; median level imputation was employed to generate their score.

fMRI-based social and monetary feedback tasks

During a Zoom interview, participants were told that they would complete a social evaluation task with peers who had already participated in the study, and a photograph of the participant was captured via screenshot. To reinforce deception, participants were told this image was edited to enhance quality and have an identical background to others. Participants were then told the photo was sent to age-matched peers who would indicate whether they would “like” or “dislike” the participant based on the photo. In reality, no other peers saw these photos or provided feedback.

During an in-person study visit, participants underwent MRI simulation to acclimate to the auditory and visual environment of the scanner and practice the tasks (see Supplementary materials 1.1–1.2). Participants next underwent fMRI scanning while completing social and monetary feedback tasks. Tasks were presented in separate runs, with the order counterbalanced across participants (see Supplementary material 1.3).

During the selection phase (Fig. 1A), participants were presented with 42 pairs of purported peers. For half of the trials, participants were instructed to select which peer in each pair liked them based on their photo (reward goal state, indicated by a green screen border), and for the other half they were asked to select which peer disliked them based on their photo (punishment goal state, indicated by a red screen border). Participants were told that, on each trial, both peers, one peer, or neither peer might provide feedback. Thus, feedback the participant received was only informative about the selected peer, and not the unselected peer.

Figure 1.

fMRI-based social and monetary feedback tasks. Note: Social and monetary task domains were administered in separate counterbalanced runs. 1A) Each pair of peers were gender-matched to be either two female faces or two male faces. 1B) Participants were told that there were three possible feedback scenarios for each trial: (1) both peers said they would like/dislike the participant; (2) one peer said they would like/dislike the participant while the other did not make a rating within the allotted time; or (3) neither peer rated the participant within the allotted time. Feedback was represented by green, red, and white circles, corresponding to correct selection of the face/door with the positive feedback/prize, the face/door with the negative feedback/loss, or incorrect selection of a face/door with no prize or loss, respectively. In both tasks, participants received equal amounts of positive, negative, and neutral feedback.

During the feedback phase (Fig. 1B), participants viewed feedback from the peer they selected. Participants saw one of three types of feedback: a green circle indicating correct selection of the peer who liked them (positive feedback), a red circle indicating correct selection of the peer who disliked them (negative feedback), or an open white circle indicating an incorrect selection of a peer that did make a rating within the allotted time (neutral feedback). If no peer was selected, no feedback was shown, and a prompt encouraged them to respond faster. This phase was the primary epoch of interest for fMRI-based analyses.

The monetary task was identical except participants saw pairs of doors and selected which door contained a $2 prize or a $1 loss. Participants were told that one of these trials would be randomly selected and added to their final payment.

Following the feedback tasks, participants answered a series of increasingly specific questions about their experience to assess deception (see Supplementary Material 1.4). Two participants were not deceived. However, they did not differ from deceived participants in their childhood abuse, neural activation, or harsh parenting, and results remained consistent when these participants were omitted. Thus, as with prior studies using these tasks, participants were included in final analyses (Quarmley et al. 2019, Jarcho et al. 2022, Wyngaarden et al. 2024).

fMRI data acquisition

Neuroimaging data were acquired using a 3-Tesla Siemens MAGNETOM Prisma MRI scanner (20-channel head coil). Each participant completed two functional runs (one per task), each consisting of 173 functional image volumes with 52 multiband (acceleration factor 2) interleaved axial 3 mm slices (in-plane resolution = 3 × 3 mm) acquired using a T2*-weighted echo-planar sequence (TR/TE = 1750/29 ms, flip = 74°; FOV = 240 mm, matrix = 80 × 80). To facilitate anatomical localization and co-registration of functional data, a high-resolution interleaved structural scan was acquired (sagittal plane) using a T1-weighted magnetization-prepared spoiled gradient-recalled echo sequence (TE/TI = 2.17/1150 ms, flip = 8°; FOV = 224 mm, matrix = 224 × 224, in-plane resolution, 1 × 1 mm). Stimuli were projected onto a screen and viewed via a head coil-mounted mirror. Responses were made and recorded using fMRI-compatible button boxes (Celeritas, 5 Fiber Optic Button Response Units).

fMRI preprocessing

Data was converted from DICOM to BIDS format (HeuDicov version 0.5.4). Preprocessing was conducted using AFNI (Analysis of Functional and Neural Images). For each participant, standard preprocessing steps were implemented with AFNI’s preprocessing pipeline tool, afni_proc.py. These steps included despiking, slice-timing alignment, nonlinear warping (SS warper) to MNI space, co-registration, smoothing to 6-mm full-width half maximum (FWHM), spatial normalization to standard space, scaling, and resampling, resulting in 3 mm³ voxels.

Individual-level analyses

Preprocessed data were analyzed using separate general linear models (GLMs) for social and monetary tasks. For each task, separate regressors were created for each temporal phase (selection and feedback), with two goal states for the selection phase (reward and punishment) and four feedback outcomes for the feedback phase (reward: positive, neutral; punishment: negative, neutral). Analyses focused only on the feedback phase (reward: positive > punishment: negative) to facilitate comparison of activation during valenced feedback outcomes. This approach emphasizes the neural differentiation of emotionally valenced conditions, rather than relative responses to neutral feedback, which would increase model complexity and potentially introduce variability unrelated to affective valence. All other regressors were treated as regressors of no interest. Additional regressors of no interest included six motion parameters (x, y, z, pitch, roll, yaw) and linear drift.

Individual-level regression analyses were performed using AFNI’s 3dDeconvolve function, generating a β coefficient and t-statistic for each voxel and regressor. Temporally adjacent TRs with a Euclidean-norm motion derivative > 0.5 mm were omitted from the model via censoring (percent censored; M = 2.31%, SD = 8.76%). Whole-brain percent signal-change maps were generated by dividing signal intensity at each voxel by the mean voxel intensity, then multiplying by 100.

Data analytic plan

All analyses were conducted with AFNI (Cox 1996, Cox and Hyde 1997) and R/RStudio (Posit Team 2021, R Core Team 2022). Neural activation during positive vs. negative feedback was extracted from ROIs using AFNI’s 3dROIstats. Core linear model assumptions were evaluated for all regression analyses, and all ROI analyses were corrected for multiple comparison (see Supplemental Materials 2.1.1–2.1.3 and 3.1.1–3.1.5).

Focal ROIs hypothesized to moderate relations between childhood abuse and risk for harsh parenting. Focal ROIs were selected based on meta-analyses and systematic reviews linking them to neural alterations associated with childhood abuse (Hart and Rubia 2012, Heany et al. 2018, McLaughlin et al. 2019, Kraaijenvanger et al. 2020, Hosseini-Kamkar et al. 2023), and social feedback processing (Vijayakumar et al. 2017, Feng et al. 2021, Martins et al. 2021, Mwilambwe-Tshilobo and Spreng 2021): anterior hippocampus, putamen, amygdala, dACC, PCC, mPFC, and anterior insula (Fig. S1).

The anterior hippocampus, amygdala, and PCC were defined using the Harvard-Oxford probabilistic atlas (Desikan et al. 2006, Makris et al. 2006), thresholded at 50% probability. The hippocampus was segmented into three tertiles along its long axis. Only the anterior tertile was used, as prior research indicates that signal from the middle tertile reflects a mixture of anterior and posterior hippocampal processing (Kerr et al. 2007, Poppenk et al. 2013). The putamen was defined via the Oxford-GSK-Imanova Structural Striatal Atlas (Tziortzi et al. 2011). The dACC was defined by combining supracallosal anterior cingulate cortex and the mid cingulate cortex masks from the AAL3 atlas (Rolls et al. 2020). This dACC mask was then bisected at y > 9, which is the posterior boundary from prior work showing dACC activation to social rejection (Jarcho et al. 2016). The anterior insula mask was generated using a Neurosynth meta-analysis of the term “anterior insula” and applying a threshold of 10 to isolate activation in the anterior region only. The mPFC was defined from a coordinate-based meta-analysis on childhood abuse (10 mm spheres excluding white matter centered at x = ±10, y = 60, z = 10; Hosseini-Kamkar et al. 2023).

ROIs used in sensitivity analyses

One set of sensitivity analyses was performed to ensure effects did not generalize to brain regions implicated in social processing more broadly. ROIs were selected based on prior research linking them to social perception and mentalizing (Amodio and Frith 2006) and their less consistent association with childhood abuse (Teicher and Samson 2016, Hosseini-Kamkar et al. 2023): STS and rTPJ. A second set of sensitivity analyses was performed to ensure effects did not generalize to brain regions without relations to social processing. These analyses included an ROI implicated in motor function: the bilateral precentral gyrus (Banker and Tadi 2026) (Fig. S1). The rTPJ was defined using a connectivity-based rTPJ parcellation derived from the FSL TPJ atlas (Mars et al. 2012). We combined anterior TPJ, posterior TPJ, and TPJ inferior parietal lobule clusters from a 50%-thresholded summary image. The FreeSurfer Desikan-Killiany cortical parcellation distributed with AFNI (Cox 1996) was used to define the STS (bilateral banks) and precentral gyrus (bilateral banks).

Task effects

To determine if focal ROIs or those selected for sensitivity analyses exhibited task-specific effects, we conducted two sets of one-sample t-tests comparing activation during positive vs. negative feedback to zero in social or monetary tasks.

Neural moderators of the relationship between childhood abuse and harsh parenting

Two sets of linear regression analyses were conducted for each focal ROI to quantify the extent to which neural response to positive vs. negative (social or monetary) feedback moderated the relation between childhood abuse and harsh parenting. For significant interactions, simple slopes analyses were performed to elucidate directionality of effects. Model assumptions of linearity, normality, and homoscedasticity were met and there were no multivariate outliers across models (see Supplementary material 2.1.1–2.1.2 and 3.1).

To reduce the number of statistical tests, follow-up Pearson correlations tested whether ROIs that moderated the association between childhood abuse and harsh parenting reflected a coherent network. A combined mask was then created from significantly correlated ROIs, and network activation during the social task was tested as a moderator of the association between childhood abuse and harsh parenting. To evaluate task specificity, we repeated the moderation analysis for the monetary task and used a linear hypothesis test to compare the condition-specific abuse-by-network interaction terms (Hayes 2022). A G*power sensitivity analysis indicated 80% power to detect a medium interaction effect (f² = 0.15). Given the parallel task design, and shared Positive > Negative contrast, social and monetary network activation were expected to be correlated, which can reduce precision for the coefficient contrast and require greater power (Vatcheva et al. 2016). Accordingly, we treated this cross-task comparison as an exploratory analysis. To aid interpretation, supplementary Pearson correlations were performed between childhood abuse, harsh parenting, and activation in the social and monetary network masks.

Sensitivity analyses

To evaluate specificity of the moderation effects, we conducted several sensitivity analyses. First, we ran parallel moderation models using the STS and rTPJ as ROIs implicated in social processing more generally, and the precentral gyrus as a nonsocial ROI. Second, to contextualize results, an exploratory analysis tested whether similar moderation patterns emerged when childhood neglect was used in analyses instead of abuse.

Results

Task effects

Social task

One-sample t-tests determined positive compared to negative social feedback elicited greater engagement in putamen (Fig. 2A; t(55) = 3.91, p_corrected < .01). No other ROIs, or the combined average of ROI activation, exhibited significant task-related effects (p’s > .27).

Figure 2.

ROI task effects during social and monetary feedback. Note: Boxplots show median and IQR, violin plots show full distribution of data. *Indicates Bonferroni-corrected P < .05; ** indicates P < .01; *** indicates P < .001.

Monetary task

One sample t-tests determined positive compared to negative monetary feedback elicited greater engagement in the mPFC (Fig. 2B; t(55) = 3.03, p_corrected < .05). A similar pattern was observed in the putamen (t(55) = 2.41, P = .019) and amygdala (t(55) = 2.17, P = .034) but did not survive FDR correction. No other ROIs exhibited significant task-related effects (p’s > .17).

Neural moderators of the relationship between childhood abuse and harsh parenting

Social task

Differences in neural response to positive compared to negative feedback moderated the relation between childhood abuse and harsh parenting in the anterior hippocampus, putamen, amygdala, dACC, PCC, and mPFC (Table 2; Fig. 3). Simple slopes analyses indicated that childhood abuse severity was associated with greater harsh parenting, specifically for participants who exhibited greater activation during negative vs. positive social feedback (i.e., slopes at −1SD; see Table 2). No moderating effects were found for anterior insula.

Table 2.

Interaction statistics for social feedback activation (positive vs. negative) moderating the relation between childhood abuse and harsh parenting.

ROIs	Main effects				Interaction effects					Simple slopes
Focal ROIs	β	SE	t	P-value	Int. β	Int. SE	Int. t	Int. P-value	FDR corrected P-value	−1SD P-value (Pos < Neg)	+1SD P-value (Pos > Neg)
Anterior hippocampus	−0.02	0.13	−0.12	.90	−0.22	0.10	−2.16	.036	.04*	.011	.969
Putamen	−0.09	0.13	−0.68	.50	−0.31	0.12	−2.58	.013	.02*	.006	.434
Amygdala	−0.11	0.13	−0.85	0.40	−0.27	0.12	−2.25	.029	.04*	.007	.863
dACC	−0.11	0.13	−0.86	0.40	−0.42	0.13	−3.12	.003	.01*	.002	.140
PCC	0.04	0.13	0.30	0.76	−0.39	0.13	−2.96	.005	.01*	.003	.273
mPFC	0.10	0.13	0.84	0.41	−0.40	0.13	−3.16	.003	.01*	.001	.261
Anterior insula	−0.18	0.14	−1.30	0.20	−0.17	0.17	−1.04	.302	.30	–	–
Sensitivity ROIs
TPJ	0.00	0.14	−0.02	0.98	−0.13	0.11	−1.16	.250	.38	–	–
STS	0.12	0.15	0.84	0.40	−0.10	0.11	−0.88	.383	.38	–	–
Precentral Gyrus	−0.06	0.14	−0.44	0.66	−0.11	0.11	−1.02	.311	.38	–	–

FDR corrected P values.

^* P < .05.

Figure 3.

Neural moderators of the relation between childhood abuse and harsh parenting during social and monetary feedback. P values are FDR corrected * indicates a significant simple slope at P <.05; ** indicates P <.01. (Blue border = focal ROI, red border = sensitivity ROIs).

Monetary task

Differences in neural response to positive compared to negative monetary feedback did not moderate the relation between childhood abuse and harsh parenting in any ROIs (Table 3; Fig. 3).

Table 3.

Interaction statistics for monetary feedback activation moderating the relation between childhood abuse and harsh parenting.

ROIs	Main effects				Interaction effects					Simple slopes
Focal ROIs	β	SE	t	P-value	Interaction β	Int. SE	Int. t	Int. P-value	FDR corrected P-value	−1SD P-value (Pos < Neg)	+1SD P-value (Pos > Neg)
Anterior hippocampus	−0.01	0.14	−0.08	0.94	−0.11	0.14	−0.83	.41	.48	–	–
Putamen	−0.30	0.13	−2.32	0.02	−0.12	0.13	−0.94	.35	.48	–	–
Amygdala	−0.08	0.14	−0.57	0.57	−0.13	0.12	−1.05	.30	.48	–	–
dACC	−0.03	0.14	−0.23	0.82	−0.21	0.14	−1.49	.14	.48	–	–
PCC	0.03	0.13	0.25	0.80	−0.12	0.13	−0.96	.34	.48	–	–
mPFC	−0.11	0.13	−0.84	0.40	−0.20	0.13	−1.55	.12	.48	–	–
Anterior insula	−0.20	0.14	−1.41	0.16	−0.08	0.12	−0.61	.54	.54	–	–
Sensitivity ROIs
TPJ	−0.01	.14	−0.09	0.92	−0.07	0.12	−0.59	.56	.83	–	–
STS	−0.06	.15	−0.37	0.71	−0.03	0.12	−0.22	.83	.83	–	–
Precentral Gyrus	−0.02	0.13	−0.18	0.86	−0.13	0.13	−1.00	.32	.83	–	–

FDR corrected P values.

^* P < .05.

Combined mask of correlated ROIs

Significant Pearson correlations emerged across 5 of 6 ROIs implicated in the relation between childhood abuse and harsh parenting (anterior hippocampus amygdala, dACC, PCC, and mPFC). The putamen was significantly correlated with 3 of 5 ROIs and was retained for the network mask (see Table S1). This broader network moderated the relationship between childhood abuse and harsh parenting (β = −0.42; SE = 0.12, t = −3.61, P < .001). Simple slopes analyses confirmed a strong relationship between childhood abuse severity and increased harsh parenting, but only for participants exhibiting greater network activation during negative vs. positive social feedback (i.e. slope at −1SD; P < .001). For the monetary task, although activation in the focal ROIs was correlated (see Table S2), network activation during the receipt of positive vs. negative monetary feedback did not moderate the relationship between childhood abuse and harsh parenting (β = −0.20; SE = 0.13, t = −1.47, P = .148). Follow-up exploratory analysis directly comparing condition-specific (social vs. nonsocial) moderation effects was not significant (F(1,50) = 0.03, P = .87). See Table S3 for Pearson correlations between childhood abuse, harsh parenting, and activation masks.

Sensitivity analyses

Differences in neural response to positive compared to negative feedback in the rTPJ, STS, and precentral gyrus did not moderate the relation between childhood abuse and harsh parenting in either task (Tables 2 and 3). Analyses that tested the extent to which brain function moderated the relation between neglect (rather than abuse) and harsh parenting did not yield results that survived multiple corrections (see Supplementary material 3.3).

Discussion

This study provides a novel contribution to understanding the risk for intergenerational transmission of abuse. We found that the association between childhood abuse and harsh parenting was stronger among individuals with higher reactivity to negative (vs. positive) social feedback in brain regions implicated in both social processing and early childhood abuse. This pattern failed to emerge in brain regions canonically linked to social (STS, rTPJ) or motor (precentral gyrus) processing, but not childhood abuse. Likewise, brain function during the processing of nonsocial, monetary, feedback failed to moderate the relation between childhood abuse and harsh parenting. This provides some evidence for domain specificity of effects beyond valenced-related processing. Taken together, results suggest that altered processing of social feedback may serve as a novel marker of risk for the intergenerational transmission of abuse.

More severe childhood abuse was associated with higher risk for harsh parenting, but only among caregivers who exhibited greater neural response to negative versus positive social feedback in anterior hippocampus, putamen, amygdala, dACC, PCC, and mPFC. These results align with prior work showing that parents who interpret their child’s behaviors as more negative or threatening are more likely to respond with hostile or punitive parenting strategies (Beckerman et al. 2018, Rodriguez and Silvia 2022), and other data that show low quality social relationships moderate intergenerational transmission of abuse (Schofield et al. 2013). Conversely, greater neural response to positive versus negative social feedback did not buffer the association between childhood abuse and harsh parenting. This may be because our task measured acute responses to brief positive social feedback, whereas protective factors in prior intergenerational work are typically characterized as stable relational resources (e.g. safe, stable, and nurturing relationships; Schofield et al. 2013). Thus, sensitivity to negative relative to positive social feedback may be a better marker of vulnerability to intergenerational abuse transmission, while resilience is better characterized by sustained supportive relationships. Future studies should test whether neural sensitivity to positive social feedback covaries with relationship quality and contributes to the protective associations observed for high-quality relationships.

The brain regions that moderated the link between childhood abuse and risk for harsh parenting are both sensitive to early childhood abuse and implicated in detecting, interpreting, and regulating responses to socially salient information. For instance, the anterior hippocampus and amygdala help identify and encode emotionally significant cues, especially those involving threat or safety (Phelps and LeDoux 2005, McCrory and Viding 2015). Likewise, dACC supports conflict monitoring and is engaged when feedback is unexpected or socially threatening (Botvinick et al. 2001, Eisenberger et al. 2003). The mPFC and PCC index the personal relevance of social information and help integrate feedback with self-focused processing (Buckner et al. 2008, Andrews-Hanna et al. 2011). Finally, the putamen contributes to reinforcement learning and updating expectations following social feedback (O’Doherty et al. 2004, Daw et al. 2006). Although greater response to negative compared to positive feedback across this network may reflect enhanced sensitivity to negative social cues, diminished sensitivity to positive cues, or both, a consistent pattern of response biased toward greater relative neural engagement to negative feedback suggests a phenotypic response that predisposes risk for intergenerational transmission of abuse. This response may reflect or contribute to biases in social learning. However, because the current study used one-shot social feedback, it is not well-suited to test this relation. Future studies that enable caregivers to learn from iterative instances of positive and negative peer feedback are needed to isolate the specific neurocognitive mechanisms at play.

This study has several strengths. First, while many studies test how brain function in children is altered by caregiver abuse, few studies have assessed the relation between brain function and childhood abuse among adult caregivers, and none test the extent to which brain function moderates current parenting style. Given the profound consequences of parenting style on children (Baumrind 1999, Pinquart 2017), more attention should be paid to understanding the neurocognitive mechanisms that characterize caregivers who engage in harsh parenting. Second, the inclusion of well-matched social and nonsocial feedback tasks enhanced interpretability of our findings by enabling a test of domain specificity. The moderating effect of neural response was observed for social, but not monetary, feedback. However, exploratory analyses showed direct comparison of moderating effects across social and monetary tasks were not significant. Larger studies that are adequately powered to test cross-task differences are needed before definitive statements about specificity can be made. Third, sensitivity analyses helped rule out alternative interpretations of the results. Specifically, effects failed to generalize to childhood neglect, or to brain regions more generally implicated in processing social information, such as the STS and rTPJ, or to other regions such as precentral gyrus. This supports the idea that phenotypic differences in processing social feedback could exert their effects on brain regions shaped by the unique interpersonal nature of childhood abuse. Although longitudinal work is needed to test this idea, current results provide an important extension to prior research on social transmission pathways.

Despite its strengths, this study has several limitations. First, both childhood abuse history and harsh parenting behaviors were assessed via self-report, which may be biased by social desirability and retrospective inaccuracies (Baldwin et al. 2019). Prospective methods that do not rely on self-report would strengthen findings. Second, 60% of caregivers experienced significant abuse. This is higher than the ∼24.1% reported in a large community sample of women (Walker et al. 1999). In contrast, harsh parenting scores were relatively low in our sample (M = 1.48, SD = 0.33) compared to other community samples (M = 1.76–2.04, SD = 0.38–0.52; Burke et al. 2019, Rutledge et al. 2019, Delvecchio et al. 2020). Thus, it is important to characterize this as a sample at high risk for abuse transmission. Replicating this study in caregivers with higher expression of harsh parenting is needed. Third, although the composite abuse score showed high internal reliability, specific abuse subtypes may exhibit distinct associations with neural response and parenting style. Future work should test subtype-specific models to clarify whether moderation differs across emotional, physical, and sexual abuse. Fourth, the cross-sectional design precludes definitive conclusions about causal relationships between childhood abuse, neural mechanisms, and harsh parenting. It is plausible that engaging in harsh parenting, or a current family environment that is stressful/high conflict, could result in altered neural response to social feedback. Longitudinal studies are needed to test the directionality of these relations. Finally, our findings may not generalize to male caregivers, and intergenerational transmission processes may occur through partner dynamics not measured in this study. For example, caregivers with abuse histories may select more violent or unpredictable partners, exposing their child to harsh parenting through secondary caregivers.

Despite these limitations, our findings identify a promising neural moderator of intergenerational risk that warrants replication in larger, preregistered studies.

Taken together, our findings suggest that altered neural response to negative social feedback may shape vulnerability to the intergenerational transmission of abuse. These results provide a promising and underexplored target for future intervention research. For example, clinicians could test interventions that reduce negative social feedback salience and strengthen emotional regulation during social interactions. These interventions might focus on real-time emotional regulation during social interactions, helping to reframe potentially threatening feedback. Given that current interventions aimed at reducing the intergenerational transmission of maltreatment have only had modest success (see Euser et al. 2015, van IJzendoorn et al. 2020 for meta-analyses) identifying new intervention targets remains an urgent public health priority. The present study provides a strong foundation for novel lines of future research aimed at clarifying how neurocognitive processes may contribute to cycles of abuse or foster resilience across generations.

Author contributions

Camille Rose Johnston (Conceptualization [lead], Data curation [lead], Formal analysis [lead], Funding acquisition [lead], Investigation [equal], Methodology [equal], Project administration [equal], Software [lead], Supervision [equal], Validation [equal], Visualization [equal], Writing—original draft [equal], Writing—review & editing [equal]), Ronan M Cunningham (Formal analysis [equal], Project administration [equal], Software [supporting], Supervision [supporting], Visualization [supporting], Writing—original draft [supporting], Writing—review & editing [lead]), Isabelle M Moger (Data curation [supporting], Formal analysis [supporting], Writing—original draft [supporting], Writing—review & editing [equal]), OgheneTejiri V Smith (Data curation [equal], Project administration [equal], Supervision [supporting], Writing—review & editing [supporting]), Katie Dow (Project administration [lead], Supervision [lead], Writing—review & editing [supporting]), Anna Kilmer (Project administration [supporting], Writing—review & editing [supporting]), William John Mitchell (Methodology [supporting], Software [equal], Writing—review & editing [supporting]), Vishnu Murty (Conceptualization [supporting], Methodology [supporting], Supervision [supporting], Writing—review & editing [supporting]), and Johanna Jarcho (Conceptualization [supporting], Formal analysis [supporting], Funding acquisition [equal], Investigation [equal], Methodology [supporting], Resources [lead], Supervision [equal], Writing—original draft [supporting], Writing—review & editing [equal])

Supplementary material

Supplementary material is available at SCAN online.

Conflicts of interest

The authors declare the following financial interests/personal relationships which may be considered as potential competing interests: Johanna M. Jarcho reports financial support was provided by National Institute of Mental Health. Camille R. Johnston reports financial support was provided by National Institute of Mental Health. All other authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Funding

This work was supported in part by the National Institute of Mental Health (JMJ: R01 MH132727, R21 MH137759; CRJ: F31 HD113512).

Data availability

Data and analysis code can be found on OSF (https://osf.io/2efkn/).