The impact of oxytocin administration on brain activity: a systematic review and meta-analysis protocol

The neuropeptide oxytocin (OT) has attracted significant scientific and lay interest for its role in social cognition and behaviour [1, 2]. For example, a single administration of OT has been shown to modulate the perception of social cues [3], motivate in-group cooperation [4], increase gaze to the eye region of faces [5], and reduce anxiety [6, 7]. Due to these reported cognitive and behavioural effects, researchers have begun investigating OT’s potential to treat psychiatric conditions, such as autism spectrum disorders, schizophrenia, and social anxiety disorder in a number of clinical trials (for a review, see [8]).

Although the modulatory effects of OT on social behaviour and cognition have been demonstrated repeatedly [8, 9], the mechanisms are poorly understood [10, 11]. To better elucidate these behavioural and cognitive effects, researchers have investigated the neural correlates of OT’s effects using brain-imaging tools such as functional magnetic resonance imaging (fMRI). Converging evidence from this field suggests the amygdala—a key brain region involved in the processing of emotional [12] and social stimuli [13]—is an important target of OT administration [14‐18]. Although the amygdala has received significant research interest, other areas of the brain have also implicated in OT’s response [19].

Interest in the use of fMRI to understand the effects of OT has been increasing exponentially in the past decade (Fig. 1), with a total of 115 publications using the keywords “oxytocin” and “fMRI” published between 2004 and 2014 (although only a minority of these publications specifically assess the impact of OT administration on fMRI outcomes). Early work primarily investigated neural activity during emotional task processing after OT administration [14, 16, 20]; however, recent research has begun to investigate resting state activity and connectivity [21, 22]. Research is yet to synthesize studies that explore resting state neural activity, which is important for understanding neural modulation with OT regardless of task, particularly within the context of heterogeneity in task design. Meta-analysis provides a robust statistical method of synthesizing effect sizes across studies and is a valuable tool for clarifying past findings.

Various methods are available for the meta-analysis of brain-imaging data [23]. Prior research synthesizing OT fMRI data has adopted a meta-analytic approach using effect size signed differential mapping [24]. This approach uses extracted fMRI peak coordinates to create an estimated map of all possible brain region coordinates, which are then synthesized. However, this approach can bias effects of interest estimates where individual studies have reported a relatively high number of peaks when compared to others by chance. Additionally, it is difficult to analyse and interpret the impact of key study design characteristics such as dose and sex using this approach, unless a large number of studies are available. The impact of these problems on effect size-based meta-analysis results is difficult to determine within the existing limitations of the literature; therefore, meta-analysis methods that produce interpretable consistency measures are likely to provide guidance for future experimental investigation in this area. Given this is the ultimate goal, the Multi-Level Kernel Density Analysis (MKDA) coordinate-based approach is suitable [25] in that it provides clear interpretability and is unbiased by number of within-study peaks (see more details below). However, an updated version of effect size signed differential mapping—Seed-based d Mapping (SDM)—has been released [26], which can combine reported coordinates and t-maps, which are statistical parametric maps displaying the t statistic after estimation of the experimental parameters of interest for a given study. The t statistic is estimated at each voxel—the small volumetric unit of the statistical parametric map (e.g. 2 mm³)—for the whole brain, which in the context of performing a meta-analysis allows estimation of effect sizes at each voxel. By increasing the number of t-maps entered into the analysis, this approach has increasing sensitivity while reducing false positives [26]. The use of t-maps is preferable, as the inclusion of studies with reported coordinates requires the statistical estimation of unreported voxels. However, this approach relies on the availability of t-maps (e.g. as supplementary material or by direct researcher request). Therefore, if studies have a significant number of usable t-maps available (i.e. which will significantly increase sensitivity and decrease false positives), we will employ SDM in the first instance. Once data is extracted from eligible studies, maps of d values [27] and their variances are then created for meta-analysis (see below for further synthesis details).