Representational similarity analysis of neural patterns in childhood maltreatment

Surmounting evidence has connected region-specific neural alterations with childhood maltreatment (CM). Univariate region of interest (ROI) approaches indicate amygdala hyperactivity and atypical prefrontal activity in response to negatively valenced stimuli, and volumetric differences, which may underpin aberrant threat processing, emotion regulation, and executive functioning in these groups [1]. Such univariate analyses aggregate voxelwise data which, although ideal for measuring overall activity of ROIs, lack sensitivity of item-level activity and global network interactions. Therefore, they cannot provide information on how information processing occurs.

To complement this work, multivariate pattern analysis (MVPA) can be used to detect patterns across individual voxels. The computational technique presents the opportunity to spatially map the distribution of neuronal firing, uncover the organization and contribution of ROIs in response to information, and compare response profiles across contexts [2]. Representational similarity analysis (RSA) is one type of MVPA that assesses similaritites of one pattern to another to disentangle overlap in how categorical information is spatially encoded in the brain [3]. Hence, RSA offers the ability to investigate encoding in CM-exposed individuals under various conditions or in comparison to controls.

In one example, maltreated youth demonstrated increased differentiation in the left hippocampus for fear vs. neutral faces, with the strongest effect observed for physical abuse among maltreatment types [4]. Dissimilar left hippocampal representation for valenced stimuli challenges the theory that amygdala hyperactivity underpins heightened threat detection. It has been suggested that neural alterations have differing trajectories dependent on the type of CM (e.g., deprivation vs. threat). Yet it can be challenging to examine this heterogeneity with univariate approaches. RSA is additionally advantageous, as it can integrate multidimensional data such as behavioral measures. Barrett and colleagues applied intersubject RSA to measures of deprivation and threat, and structural development of the frontoparietal network, fronto-amygdala, and hippocampus circuit over an 18-month period in adolescent girls. They found that operationalizing similarity by a dimensional model of deprivation and threat was a better predictor of neurodevelopment than a cumulative risk model [5]. Together, these studies present novel evidence on CM-related alterations in global patterns of brain activation and structural development that could not otherwise be captured through univariate analyses.

We encourage exploration of the neural signatures of CM using multivariate analysis. RSA for fMRI data is a novel technique that could benefit from further validation for use in social neurosciences and can be conducted in adjunct to already supported analyses [2]. MVPA has previously shown to be effective for predictive modeling of treatment response in psychiatric patients [6]. Consequently, differentiating the neural impact of various CM types may help predict prognosis and target treatments. Furthermore, by identifying irregular neural representations at an early age, the opportunity to intervene during critical periods could mitigate the risk of developing associated maladaptive behaviors. The brain is a complex and interconnected system and accordingly, analysis on the neurobiological markers of CM ought to be multidimensional for early detection and efficient treatment.

Author contributions

MD was responsible for preparation and submission of the report with supervision and revision from AT. Both authors have contributed to the final manuscript.

Competing interests

The authors declare no competing interests.