Dissociative Part-Dependent Resting-State Activity in Dissociative Identity Disorder: A Controlled fMRI Perfusion Study

Background In accordance with the Theory of Structural Dissociation of the Personality (TSDP), studies of dissociative identity disorder (DID) have documented that two prototypical dissociative subsystems of the personality, the “Emotional Part” (EP) and the “Apparently Normal Part” (ANP), have different biopsychosocial reactions to supraliminal and subliminal trauma-related cues and that these reactions cannot be mimicked by fantasy prone healthy controls nor by actors. Methods Arterial spin labeling perfusion MRI was used to test the hypotheses that ANP and EP in DID have different perfusion patterns in response to rest instructions, and that perfusion is different in actors who were instructed to simulate ANP and EP. In a follow-up study, regional cerebral blood flow of DID patients was compared with the activation pattern of healthy non-simulating controls. Results Compared to EP, ANP showed elevated perfusion in bilateral thalamus. Compared to ANP, EP had increased perfusion in the dorsomedial prefrontal cortex, primary somatosensory cortex, and motor-related areas. Perfusion patterns for simulated ANP and EP were different. Fitting their reported role-play strategies, the actors activated brain structures involved in visual mental imagery and empathizing feelings. The follow-up study demonstrated elevated perfusion in the left temporal lobe in DID patients, whereas non-simulating healthy controls had increased activity in areas which mediate the mental construction of past and future episodic events. Conclusion DID involves dissociative part-dependent resting-state differences. Compared to ANP, EP activated brain structures involved in self-referencing and sensorimotor actions more. Actors had different perfusion patterns compared to genuine ANP and EP. Comparisons of neural activity for individuals with DID and non-DID simulating controls suggest that the resting-state features of ANP and EP in DID are not due to imagination. The findings are consistent with TSDP and inconsistent with the idea that DID is caused by suggestion, fantasy proneness, and role-playing.

Text S1: Image acquisition for the non-simulating control group The MRI data of the non-simulating control group (NS) were obtained at the University Hospital of Zurich with a 3-T Philips Ingenia whole-body magnetic resonance imaging equipped with a 15-channel head coil. Ingenia and Achieva use an identical software. The pulse sequence varied slightly from the one used for the DID patient group (DID) on the Achieva system regarding 1) background suppression labeling (Ingenia: 1710 ms/2860 ms; Achieva: 1680 ms/2760 ms) and 2) TE (Ingenia: 14 ms; Achieva: 12 ms). The pulse sequence for the acquisition of the T1-weighted image was identical on both systems.

Text S2: Controlling for inter-scanner variability
Scanner-to-scanner variability of activation makes it difficult to say if group differences (i.e., DID-NS; NS-DID) in fMRI response patterns are genuine results or just reflect scanner differences. Previous studies have already addressed this issue in the context of multicenter studies involving hardware and software differences across sites [1,2,3]. These studies have shown that the major sources of variability of fMRI measurements are based on the different scanner manufacturers and the scanners' field strengths. As all participants have been measured on a 3-T Philips system, these major scanner effects can be ignored in the present study. It is important to note that in Tahmasebi et al. (2012), scanner effects could only explain 2% of the total variance of the fMRI response. This observation is in line with studies showing that the inter-subject variance involves a higher part of the total variance on fMRI measurements compared to between-scanner variance [2,4].
Nevertheless, scanner dependent changes regarding the SNR can have an impact on the brain activation patterns [3,5,6,7]. In the present follow-up study, we checked statistically if the perfusion differences between the patient group (DID), measured on the Achieva system, and the non-simulating control group (NS), measured on the Ingenia system, are confounded by SNR differences of the two systems. For this purpose, we followed the approach of Friedman and colleagues, who used a particular type of SNR, i.e. the signal-tofluctuation-noise-ratio (SFNR) per subject as a covariate of no interest in the statistical model [3]. Friedman et al. (2006) showed that this statistical adjustment significantly reduced scanner effects on the effect size of the MRI signal in a study including data measured on high-and low-field scanners from different vendors. For SFNR, the "signal" (S) is defined as the mean intensity in a region over time, whereas "fluctuation noise" (FN) is the standard deviation in this region of the same time series after a 2 nd order polynominal detrending [3].
The SFNR is calculated on a voxel-wise basis as S divided by FN (S/FN).
For the present control analysis, SFNR images were calculated using an in-house  Table S1 and discussed in Text S3. For a direct visual comparison, the statistical parametric maps of the t-tests performed with and without the SFNR adjustment are contrasted in "glass brain" renderings in the Figure S1.
As the SFNR is related to activation effects [3,8,9], the SFNR is typically used as a sensitivity measurement of imaging systems. Three t-tests were performed (SPSS) in order to check for significant differences in the SFNR obtained on the same scanner (i.e., Achieva: DIDanp-DIDep) and on two different scanners (i.e., Achieva versus Ingenia: DIDanp-NS, DIDep-NS). DIDanp, DIDep, and NS were all treated as independent groups in order to have the same statistical power for all three tests. It might be speculated that inter-scanner differences in SFNR are higher compared to intra-scanner differences. P-values were Bonferroni corrected and set at p<0.0167, one-sided). Results are depicted in Figure S2 and discussed in Text S3. Table S1: Resting-state regional cerebral blood flow (rCBF) differences between DID patients and non-simulating controls. The median SFNR in GM per subject was entered as covariate of no interest in the statistical model. MNI

A)
Figure S1: "Glass brain" renderings in saggital and coronal view comparing the statistical parametric maps for the t-tests A) with and B) without SFNR covariate adjustment. p<0.00167, cluster-size threshold = 10 voxels. Text S3: Discussion of inter-scanner variability SFNR estimates have been included as covariate of no interest in the statistical model to revoke potential scanner effects [3]. The results depicted in Figure S1 indicate that our major group differences are not confounded by inter-scanner variability. The comparison of Table   S1 and Table  p<.00167, kE= 9). The hippocampus has been described as a part of the scene construction network [10,11,12] in the Discussion section. Despite the slighly weaker hippocampal effect, we stick to the interpretation that NS engaged in imagination of future and past events, as in NS compared to DID patients, the remainder of the previously observed areas still showed increased perfusion after the statistical control with the SFNR values (see Figure S1), in particular the frontal polar cortex mediating future thinking [13,14,15,16] and the occipital cortex involved in imagining past events [17].
The results of the t-tests depicted in Figure S2 are in line with the assumption that inter-scanner-differences are not the major source of the observed group differences. Taken together, after controlling for scanner variation in SFNR, the previous findings outlined in the Discussion section largely remain intact and the essence of our group differences does not change. Our data are in line with previous studies demonstrating that technical changes are not the major source of variability and support the feasibility of comparing data measured on identical scanner systems [1,2,4]