Oxytocin normalizes the implicit processing of fearful faces in psychopathy: a randomized crossover study using fMRI

Adults with antisocial personality disorder with (ASPD + P) and without (ASPD – P) psychopathy commit the majority of violent crimes. Empathic processing abnormalities are particularly prominent in psychopathy, but effective pharmacological interventions have yet to be identified. Oxytocin modulates neural responses to fearful expressions in healthy populations. The current study investigates its effects in violent antisocial men. In a placebo-controlled, randomized crossover design, 34 violent offenders (19 ASPD + P; 15 ASPD – P) and 24 healthy non-offenders received 40 IU intranasal oxytocin or placebo and then completed an fMRI morphed faces task examining the implicit processing of fearful facial expressions. Increasing intensity of fearful facial expressions failed to appropriately modulate activity in the bilateral mid-cingulate cortex in violent offenders with ASPD + P, compared with those with ASPD – P. Oxytocin abolished these group differences. This represents evidence of neurochemical modulation of the empathic processing of others’ distress in psychopathy.


A. Oxytocin-mechanism of delivery to brain
As oxytocin is a relatively large peptide, it does not readily cross the blood-brain barrier when given intravenously [1].Hence, in experimental studies in humans, oxytocin has been delivered by the intranasal route.This is a convenient and safe mode of delivering the hormone, and has been used in this way for decades, with minimal problems: oxytocin produces no detectable subjective changes in recipients, produces no reliable side-effects, and is not associated with adverse outcomes when delivered in doses of 18-40 I.U.for short term use in controlled research settings [2].Intranasal oxytocin increases oxytocin levels in the CNS in animals [3,4] and humans [5], and a considerable volume of research has shown that that it exert effects on brain as measured by fMRI ( [6,7]).However, the mechanism by which intranasally-delivered oxytocin exerts its effects on brain remains is poorly understood and remains a point of contention [1,8].
The clearest model of this mechanism has been outlined by Quintana et al [9,10].This proposes that intranasally delivered oxytocin, absorbed through the nasal mucosa, can reach the brain via cranial nerves: the olfactory nerve via olfactory sensory neurons located in the mucous layer, and the trigeminal nerve via trigeminal ganglion cell fibers, which are also located close to the surface of the nasal cavity.The model also asserts that intranasally-delivered oxytocin may also exert effects on CNS after entering the systemic circulation pathways via the nasal mucosa.First, it is absorbed into systemic circulation via blood capillaries located underneath the membrane of the nasal cavity.
It may then: i) cross the blood-brain barrier in very small amounts-as has been demonstrated in rodents [11] and/or ii) exert effects through afferent feedback mechanisms to the CNS from receptors within peripheral organs [9].A subsequent study comparing intranasal and intravenous delivery showed that oxytocin dampened amygdala activation in response to emotional faces in the intranasal condition only [12].However, a more recent ASL study supports a model of effects via the systemic circulation only.This showed that both intranasally and intravenously delivered oxytocin exerted similar effects on cerebral blood flow in the oxytocinergic network (as described above) [13].
At present, a consensus on the precise mechanism of intranasally-delivered oxytocin on the CNS remains elusive.

B. Oxytocin-dose and timing
At the outset of this study, while the basic physiology of oxytocin was established [1] and a theory for mechanism of action had emerged [14], questions remained about the optimal way to manipulate the oxytocin system experimentally.Firstly, there was no consensus on optimal dose, with a range of doses being commonly used-from 8 I.U. to 40 I.U.[15].Further, whether any dose effect was linear or more complicated (e.g.inverted-U-shaped) had not been established [16].Also, there were several unresolved issues regarding nasal spray formulation, such as nasal spray viscosity, liposolubility, and ionisation (and these continue to exist [10]).Finally, a standardised method of delivering oxytocin intranasally was not available-i.e., different studies used different types of spray, and different timing of delivery of dose.In the context of these uncertainties, I chose to use 40 I.U., the highest clinically applicable safe dose administered to human volunteers, and followed a local protocol developed by Paloyelis et al [17], which had recently shown successful activation of the oxytocinergic network.Some studies have since investigated outstanding issues.Firstly, several studies have investigated the optimal dose of oxytocin.One fMRI study using a facial emotion paradigm in 116 healthy men compared both varying doses (12, 24, and 48 I.U.) and dose latencies (15-40, 45-70, and 75-100 minutes) of oxytocin in order to identify the most robust effects on amygdala reactivity [18].Effects were most prominent with the 24 I.U.dose and at a time window between 45 and 70 minutes after administration (though neural effects were also evident at the 48 I.U.dose).However, two randomised fMRI studies in healthy subjects by Quintana et al suggested that a lower dose of oxytocin-8 I.U.-was more effective [12,14].Notably, these studies delivered oxytocin using a breath-powered device, which is thought to overcome the barrier to delivery posed by the nasal valve [19].A further recent study using a dose-response design (9, 18 and 36 IU), demonstrated that intranasal oxytocin-induced changes in local regional cerebral blood flow (rCBF) in the amygdala at rest, and in the covariance between rCBF in the amygdala and other key hubs of the brain oxytocin system, follow a doseresponse curve with maximal effects for lower doses [20].
In summary, while doses of 40 I.U.or similar have shown consistent effects on neural markers, the optimal formulation, timing, and dosage for intranasal oxytocin in clinical and experimental neuropsychiatric research remains unknown, and is an ongoing challenge for future work in this area (see [10] for an up-to-date review).

C. Morphed Faces Task-full details
Participants were instructed to lie as still as possible during the entire task.They were provided with a small keypad, and asked to place their index finger on button 1 and their middle finger on button 2. They were presented, on a small screen within the scanner, with images of male and female faces expressing fearful and happy expressions.They were asked to identify the sex of the face presented, by pressing button 1 for female and 2 for male.All images were of Caucasian adults (50% female) drawn from well-validated images in the Pictures of Facial Affect Series (Ekman, 1976).To allow for analysis of parametric modulation, photos displaying each target emotion were morphed with a photo of the same face displaying a neutral expression in 4 different gradients (40%, 60%, 80%, 100% of the target emotion) to produce a total of 32 unique images (4 individuals (2 men, 2 women) x 2 emotions (fearful, happy) x 4 intensities).Images were rapidly presented in a series of 50ms frames.Stimulus presentations were followed by a fixation point, which was on screen for a jittered duration of 1250-4250ms.Each subject was presented with a total of 80 fearful and 80 happy expressions, with a total task duration of 9 minutes, 56 seconds.Neural responsivity to facial emotions is typically investigated using fMRI and two-dimensional images of facial emotions.
The paradigm used may be explicit, requiring participants to identify the emotion displayed (for an example in antisocial populations see (Contreras-Rodriguez, Pujol et al.

2014)
) .However, one potential limitation of these paradigms in studies on individuals with ASPD+/-P is their reliance on the veracity of the subjects' responses.This is particularly relevant to research in ASPD+/-P, as the tendency to lie and manipulate are core features of ASPD+P, and may also be present to some degree in ASPD-P.Such traits may extend into the experimental domain, leading to spurious findings, for example if subjects choose not to co-operate with the task, or provide wilfully incorrect answers.In contrast, implicit tasks of facial emotion processing seek to eliminate the potential for behavioural responses that do not give a true reflection of performance.
These tasks are designed to focus subjects' attention on an unrelated process, for example, identifying the biological sex of the face shown, while measuring brain responsivity on fMRI (or electrophysiological measures) as the outcome measure.A further consideration is the intensity of facial emotional expressions, which has been shown to affect their neural processing in studies in healthy subjects (Lin, Mueller-Bardorff et al. 2016, Wang, Yu et al. 2017).
The Morphed Faces task enables the exploration of modulated intensities of fearful facial expressions, and has been previously utilized in a study of conduct disordered children [21].The use of morphed-face stimuli carries particular advantages.At the behavioral level, use of subtly varying morphs within an emotion class increases stimulus novelty, a factor previously shown to influence brain response to emotional faces [22].Increasing variability in emotion displays aims to reduce the habituation associated with repeated viewing of identical face-emotions [22,23].At the neural level, use of a gradient of intensity overcomes two limitations of prototypical '100%' face-emotion displays: correlation of the degree of neural activation with the intensity of emotion expressed [24,25] and differential engagement of components of face responsive networks, including limbic regions, with increasing emotional intensity [26].Use of morphed-face stimuli allows modelling of linear changes along a continuum and thus examination of a "dose-response" curve of neural activity changes with increases in facial emotion [23,27].

D. Lack of findings in amygdala
The lack of significant findings in the amygdala was a somewhat surprising finding.A large-scale 2009 meta-analysis suggested that the amygdala plays an important role in neural processing of fearful faces [28].In antisocial populations, differential amygdala responsivity has been thought to underpin behavioural differences in responsivity to distress cues in CD+CU (vs CD-CU) and ASPD+P (vs ASPD-P).For instance, youth with conduct problems and high CU traits demonstrated reduced amygdala responses to fearful expressions compared to those with low CU traits [29][30][31][32], while youth with CD+CU [21,33] and adults with ASPD+P [34] demonstrate reduced amygdala responses to fearful expressions compared to healthy controls.In contrast, youth with conduct problems and low CU traits (similar to CD-CU) showed increased amygdala responses to fearful faces (compared to youth with no conduct problems [35]), and individuals with heightened reactive aggression (similar to ASPD-P) have been shown to have amygdala hyperactivity in response to facial expressions of fear [36].However, changes in amygdala reactivity may be difficult to detect due to the relatively small size of the region [37,38] and stimulus-correlated signal fluctuation in nearby veins [39].A recent systematic review of neuroimaging studies in antisocial populations has also cast doubt on the prominence of the role of the amygdala [40].Specifically, this study revealed a high proportion of null findings, a disproportionate number of positive findings from low powered studies, and peak coordinates of reduced amygdala activity not primarily falling within the anatomical bounds of the amygdala.Recommendations for future work included rigorous labeling of significant clusters of voxels, large-scale studies for adequate power, and a shift in focus to neural networks, as opposed to discrete regions.

S23: Full details of preprocessing and individual level analyses
Functional MRI data were preprocessed and analyzed using Analysis of Functional NeuroImages (AFNI) software [41].Data from the first five repetitions were collected prior to magnetization equilibrium and were discarded.FMRI data were despiked, and volumes were censored if meeting a motion threshold of >1mm or if 10% of voxels within a volume met outlier criteria as determined by 3dTOutcount (none did).The anatomical scan for each participant was registered to the Talairach and Tournoux atlas [42] and each participant's functional EPI data were registered to their Talairach anatomical scan in AFNI.Functional images were motion corrected and spatially smoothed with a 6-mm full width half maximum Gaussian kernel.The data then underwent time series normalization and these results were multiplied by 100 for each voxel.The resultant regression coefficients are therefore representative of a percentage of signal change from the mean.
At the single subject level, regressors depicting each of the response types and nuisance motion regressors were then created by convolving the train of stimulus events and realignment parameters with a gamma-variate haemodynamic response function to account for the slow haemodynamic response.Regressors of interest included i) fear; ii) its parametric modulation by intensity; iii) happiness; iv) its parametric modulation by intensity.Linear regression modelling was then performed using the regressors described above plus regressors to model a first order baseline drift function.This produced a beta coefficient and its associated t-statistic for each voxel and each regressor.The modulated regressors were then taken forward for the group level analyses.
The parametric modulation of neural responses by fearful facial emotion intensity (regressor 2 above) data were then entered into a 3 Group (NO, ASPD-P, ASPD+P) vs 2 Condition (placebo, oxytocin) 3dMVM (ANOVA style computations) model for fearful expressions.As there were significant group differences in illicit drug use, established by urinary drug screening on the day (see Supplementary Table 1), and there was a feasible mechanistic basis for a consequent effect on neural responsivity, this was incorporated as a covariate.This model provides outputs for the overall effects of task, group, and condition.Within this framework, general linear tests were coded to assess differential effects of condition (oxytocin or placebo) between the groups.To investigate responsivity to the task itself, the overall intercept (F) was used.Post hoc pairwise comparisons were conducted to decompose these interactions by examining between and within group effects.Correction for multiple comparisons was performed using a spatial clustering operation in AFNI's 3dClustSim utilizing the autocorrelation function (-acf) with 10,000 Monte Carlo simulations for the whole-brain analysis.Spatial autocorrelation was estimated from residuals from the individual-level GLMs.The initial threshold was set at p = 0.001.This process yielded an extant threshold of k = 22 voxels for the whole brain (multiple comparison corrected; p < 0.05).

S34: Data Quality Control
Volumes were censored if there was 1mm motion across adjacent volumes.No participant in the final sample for the current study had >5% censored volumes.Two group (NO vs All ASPD) by two condition (placebo vs oxytocin) repeated measures ANOVAs (with post-hoc ASPD-P vs ASPD+P analysis) were carried out for five motion parameters: TRs above threshold, average motion per TR, maximum displacement during scanning, outcount, and maximum F. These revealed no significant group or condition differences (multiple comparison corrected).Data was also manually checked for quality using two main steps.Firstly, the alignment of each scan was checked by visually inspecting coronal, sagittal and axial images.Secondly, the functional (EPI) imaging sequences were inspected for each scan to screen out highly abnormal activation patterns which were likely to represent poor quality data.Following such quality checks, one subject was excluded due to excess subject head movement during the task.

Figure S1 .
Figure S1.Flowchart of recruitment and participation of offenders with antisocial personality disorder with or without psychopathy (ASPD+/-P).

Figure S2 .
Figure S2.Flowchart of recruitment and participation of non-offending health controls (NO).

Figure S5 .
Figure S5.Significant group (ASPD+P v ASPD-P) x condition (placebo vs oxytocin) interaction effect in left midcingulate cortex during fear processing.Color bar represents t statistic.

Figure S6 .
Figure S6.Individual beta values for group (ASPD+P v ASPD-P) x condition (placebo vs oxytocin) interaction effect in left midcingulate cortex during fear processing.Individual subjects' data plotted as dots.Means are indicated by horizontal bars.Error bars represent standard deviations.NO= non-offenders ASPD-P = violent offenders with antisocial personality disorder but not psychopathy.ASPD+P= violent offenders with antisocial personality disorder and psychopathy

Table S1 . Correlations between PCL-R facets and factors.
Cronbach's alpha for four facets was 0.79.*significant at p<0.001

Table S2 . Correlations between Reactive Proactive aggression Questionnaire (RPQ).
Cronbach's alpha for the two subscales was 0.905.*significant at p<0.001 Main effect of task: parametric modulation of neural responses by fearful facial emotion intensity in all subjects

Table S3 . Main effect of task-parameter details
*Did not survive clusterwise correction for multiple comparisons (22 voxel threshold).