Association of Pediatric Acute-Onset Neuropsychiatric Syndrome With Microstructural Differences in Brain Regions Detected via Diffusion-Weighted Magnetic Resonance Imaging

This case-control study investigates whether diffusion-weighted imaging shows microstructural abnormalities across the brain regions of children with pediatric acute-onset neuropsychiatric syndrome.


Introduction
The diagnostic considerations of childhood-onset obsessive-compulsive disorder (OCD) have evolved substantially since the late 1980s. Early prospective studies 1,2 of children with OCD revealed that a subgroup experienced an unusually abrupt onset (<48 hours) and a relapsing-remitting course of neuropsychiatric symptoms, notably after streptococcal pharyngitis. In parallel, research on Sydenham chorea, a neurological sequela of group A streptococcal infection, found that 50% to 70% of affected children experienced obsessions and compulsions before, during, or after onset of chorea. [3][4][5] These findings led to the hypothesis that Sydenham chorea might share a common pathophysiology with acute-onset OCD even when chorea is absent, 5 thus forming the basis of a new diagnostic entity, namely, pediatric autoimmune neuropsychiatric disorders associated with streptococcal infections (PANDAS).
The new PANDAS classification was subsequently met with skepticism. Concerns were raised in particular about the lack of clear distinction between PANDAS and other neuropsychiatric diagnoses, as well as the strength of the association between OCD and group A streptococcal infection. [6][7][8][9][10] To address these issues, a group of clinicians and scientists convened in 2010 to discuss all possible cases of acute-onset OCD regardless of cause. 11 Participants unanimously agreed that acute onset was the key distinguishing clinical feature and proposed the more inclusive term of pediatric acuteonset neuropsychiatric syndrome (PANS). Since the development of this diagnostic entity, a number of clinics and research programs have further characterized the many comorbid symptoms that can suddenly start along with OCD, including sleep disturbances, emotional dysregulation, eating restriction, pain and sensory disturbances, urinary symptoms (enuresis and polyuria), cognitive and behavior regression, anxiety, transient psychotic symptoms, oppositionality, impulsivity, irritability and rage. [12][13][14][15][16][17][18][19][20][21][22][23] By definition, PANS includes not only PANDAS but also early, abrupt-onset OCD associated with other infectious and noninfectious triggers. 11,24 A comprehensive diagnostic workup must be performed to rule out other known differential diagnoses, such as Sydenham chorea, Tourette syndrome, and systemic lupus erythematosus. Current recommended diagnostic testing targets inflammatory biomarkers given the link between OCD disorders and autoimmunity observed in epidemiological studies. [25][26][27][28][29][30][31] Symptoms of PANS and PANDAS are hypothesized to result from crossreactive antibodies that breach a compromised blood-brain barrier and damage neuronal tissues in the basal ganglia and amygdala. 32 However, some studies 33,34 have reported poor predictive value (17%-40% for positive predictive value and 44%-74% for negative predictive value) and low test performance (15%-60% sensitivity and 28%-92% specificity) for standard PANS diagnostic evaluations, including inflammatory markers such as erythrocyte sedimentation rate, C-reactive protein, and antibody titers. There remains a pressing need for reliable biomarkers to improve diagnostic accuracy and validate the underlying pathophysiology of PANS.
Diffusion-weighted magnetic resonance imaging (DWI) could potentially serve as a noninvasive tool for assessing microstructural differences of the brain in children with a suspected inflammatory cause of abrupt-onset OCD. Although prior research has identified volumetric and inflammatory changes in the basal ganglia, [35][36][37] no studies to our knowledge have yet assessed cerebral blood flow (CBF), which may be associated with local inflammation, or the mean diffusivity or apparent diffusion coefficient (ADC), which measures the magnitude of water molecule diffusion within tissue. In the setting of neuroinflammation, ADC is expected to increase because of water molecules diffusing freely throughout the extracellular space. [38][39][40] Herein, a retrospective, cross-sectional analysis was conducted of regional brain volume, ADC, and CBF differences across various brain regions in patients with PANS and in pediatric control participants. It was hypothesized that affected children would exhibit higher diffusivity compared with control participants in the deep gray matter (eg, thalamus, basal ganglia, and limbic structures) given the predominant symptoms of OCD, emotional dysregulation, and sleep disturbances. We expected limited volumetric or CBF differences because of patient group heterogeneity in the timing of imaging after an acute flare in neuropsychiatric symptoms.

Methods
The study was approved by the Stanford University Institutional Review Board. Written informed consent was obtained from all parents, and assent was obtained for all patients in the study. A separate institutional review board protocol allowed for medical record review and analysis of magnetic resonance imaging (MRI) data under a specific protocol for the control group. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.

Patient Selection
In this case-control study, consecutive patients evaluated in the Stanford University PANS multidisciplinary clinic from September 3, 2012, to March 30, 2018, were retrospectively reviewed.
The study population was identified using the following inclusion criteria: patients were aged 4 to 18 years, met criteria for PANS (eAppendix in the Supplement), and received 3 Tesla (T) DWI before initiation of anti-inflammatory or immunomodulatory therapy. Patients with a clinical profile that better matched a diagnosis of Sydenham chorea (ie, clinically significant chorea) or Tourette syndrome (ie, waxing and waning OCD and tics) were excluded. The presence of neuropsychiatric symptoms and neurological signs at the time of MRI was investigated through review of medical records and prospectively administered patient questionnaires. The Children's Global Assessment Scale (CGAS) is a clinician's rating of a patient's psychological and social function ranging from scores of 1 to 100, with higher scores indicating better functioning. 41 Clinicians in this study received training on assigning CGAS scores and recorded scores at the end of each clinical encounter. The CGAS scores were collected in a research database, and missing scores were obtained retrospectively.
Members of the medical imaging team (E.S.M., N.D.F, and K.W.Y) who were blinded to the disease and hypothesis under study performed the image analysis and excluded images with inadequate data, such as motion artifacts and unidentified bright objects, and structural pathologies, such as masses or cysts. Among 60 consecutive patients who met study entry criteria, 6 patients were excluded by blinded investigators because of imaging or motion artifacts, 3 patients for major pathologies, and 17 patients for suboptimal atlas image registration. In total, 34 patients with PANS before initiation of treatment were compared with 64 pediatric control participants. optimize age matching with the patient group. Rigorous quality control was similarly performed on control MRI. The control population has been previously described. 42

MRI Acquisition
Patients underwent MRI at varying points during their disease course relative to an initial or recurrent symptomatic flare-up. For all individuals, 3T MRI was obtained with an 8-channel head coil on a single MRI scanner (Discovery 750W; GE Healthcare). Pseudocontinuous arterial spin labeling (ASL) MRI was performed using the methods described by Dai et al. 43 Briefly, this vendor-supplied ASL MRI was performed using a pseudocontinuous labeling period of 1500 milliseconds (ms), followed by a 1500-ms postlabel delay. Whole-brain images were obtained with a 3-dimensional, backgroundsuppressed, fast spin-echo stack-of-spirals method. Multiarm spiral imaging was used, with 8 arms and 512 points obtained on each arm (bandwidth, 62.5 kHz), yielding a 3-mm 2 in-plane spatial resolution and a 4-mm section thickness. A high level of background suppression was achieved using 4 separate inversion pulses spaced around the pseudocontinuous labeling pulse. The acquisition time was approximately 5 minutes for this sequence, which also included proton density images required for CBF quantification. For a graphic setup of the ASL, the sagittal image was used for alignment after the 3-plane localizer. Postprocessing was performed using an automated reconstruction procedure according to the microsphere methods described by Buxton et al. 44 Other pseudocontinuous ASL MRI parameters were repetition time of 4632 ms, echo time of 10.5 ms, 24-cm field of view, and 3 excitations.
In addition, echoplanar whole-brain DWI was obtained in all patients, with repetition time of 1500 ms, echo time of 37 ms, 90°flip angle, 24 × 24-cm field of view, 128 × 128-pixel acquisition matrix interpolated to a 256 × 256-pixel matrix, 44 sections with 4-mm slice thickness, no skip, and 2 diffusion weightings of b = 0 seconds/mm 2 and b = 1000 seconds/mm 2 , for which diffusion gradients were obtained in 3 directions and averaged. Derived from DWI, the ADC has demonstrated high reproducibility and was performed as part of routine institutional neuroimaging. 45

Image Processing
A custom image processing pipeline was used in this work to extract quantitative values of regional brain volume, ADC, and CBF, previously described by Forkert et al. 42 In brief, rigid registration was used for motion correction of the DWI data set with and without diffusion weighting before calculation of the quantitative ADC parameter map using the equation described by Stejskal and Tanner. 46 For volumetric, regional diffusion, and CBF analysis, the Montreal Neurological Institute 152 brain atlas 47 was nonlinearly registered to the DWI data set using a concatenated affine and b-spline transformation and maximization of the mutual information metric. The resulting nonlinear transformation was used to warp the Harvard-Oxford subcortical brain regions, as defined in the Montreal Neurological Institute atlas space, to the individual-specific brain anatomy. Brain regions included in the atlas were the cerebral white matter, cerebral cortex, lateral ventricles, thalamus, caudate, putamen, pallidum, hippocampus, amygdala, nucleus accumbens, and brainstem. Three experienced, blinded observers (E.S.M., N.D.F., and K.W.Y.) performed strict quality control of all patient and control MRI registrations, excluding those with suboptimal overlap between reference and patient brain regions. The aligned brain atlas regions were then used to measure the regional brain volume, median ADC, and median CBF values for corresponding brain structures in each hemisphere. The lateral ventricles were only included in the volumetric assessment.

Analysis of Imaging Data
Using atlas-based MRI analysis, regional brain volume, diffusion, and CBF were measured in the cerebral white matter, cerebral cortex, thalamus, caudate, putamen, pallidum, hippocampus, amygdala, nucleus accumbens, and brainstem. An age and sex-controlled multivariable analysis of covariance (MANCOVA) was used to test the null hypothesis that the mean values of MRI parameters for patients are equal to those of control participants. Only a subset of patients (n = 25) had CBF In addition to the statistical analysis described above, plots with age-related 5th, 10th, 25th, 50th, 75th, 90th, and 95th quantile curves were generated for each parameter, investigated using local piecewise regression analysis described by Sakov et al 48 based on the control group data for each brain structure and parameter. Corresponding data points for the patient group were also plotted for visual assessment. The plots were generated using the R statistical software package, version 3.2.2 (The R Foundation for Statistical Computing).  Table 1). 41 Patients underwent MRI at different time points after symptomatic presentation, with patients with new acute-onset psychiatric symptoms receiving imaging within days of initial onset and some patients receiving imaging years after initial onset (mean time after initial onset, 404 days; range, 9 days for an acute-onset case to 4097 days for a chronic case). The detailed selection process, including patient and control inclusion and exclusion criteria, is shown in  To assess whether patients with subtle choreiform movements or twitches differed from those without these movements, 3 secondary MANCOVA analyses were performed with regional brain volume, median ADC, and median CBF parameters as dependent variables, respectively. Statistically  [11.94]; P < .04). These differences were also smaller in magnitude compared with those between patients with PANS with subtle choreiform movements or twitches and control participants. No differences in the univariate regional brain volume and CBF analyses reached statistical significance.

Qualitative Analysis
The results of the statistical evaluation were also confirmed by visual analysis of the plots generated using local piecewise regression analysis. Overall, no difference between patients with PANS and control participants is obvious for regional brain volume or CBF values analyzed (eFigure in the Supplement). However, the plots clearly show statistically significantly increased ADC values across  brain regions, especially in the deep gray matter (thalamus, basal ganglia, amygdala) and nucleus accumbens (Figure 2).

Discussion
It was expected that patients with PANS would exhibit increased regional diffusion across thalamic, basal ganglia, and limbic structures compared with control participants given the hypothesized inflammatory pathophysiology and cardinal psychiatric symptoms. 37,49,50 This study identified increased diffusion patterns across patients in all brain regions analyzed, with the most pronounced differences in the deep gray matter, including the thalamus, basal ganglia (caudate, putamen, and pallidum), and amygdala. Clinically, the most affected areas align with the type of neuropsychiatric symptoms seen in PANS. Inflammatory changes to the amygdala may impair emotional regulation, leading to sudden-onset emotional lability and irritability. [50][51][52] Basal ganglia disruption may result in motor dysregulation and obsessive-compulsive symptoms. 53 Thalamic abnormalities are associated with these alterations given its central role in mediating interconnectivity with multiple regions, such as the basal ganglia and amygdala, 54,55 as well as its influence on sleep-wake cycles. 56 Prior studies 35,36 of MRI in patients with recent-onset PANDAS and Sydenham chorea noted increased volume in the basal ganglia during acute presentation compared with control participants.
Anecdotal evidence has also shown that basal ganglia size might correlate with the disease course because striatal volume tends to normalize during remission. 57 In the present study, no volumetric differences were found between patients with PANS and control participants, likely because most patients were in a chronic disease state ( as pediatric OCD, that are well described in the neuroimaging literature. [60][61][62] Because of skepticism around the inflammatory cause of PANS, a recent commentary recommends against laboratory or neuroimaging tests in children with mild to moderate, acuteonset, psychiatric-only presentations. 63 However, our MRI findings revealed substantial cerebral diffusion abnormalities among patients with and without subtle choreiform movements or twitches, as well as across a wide range of CGAS scores, a validated measure of impairment. 41 Further research should consider leveraging advanced regression models to understand which MRI parameters lead to the largest improvements in diagnostic accuracy for PANS. Given natural variation in MRI quality and atlas registration, future studies may also explore less rigid quality control restrictions in the patient selection process to expand the addressable population.

Limitations
Our study has several limitations. Because this investigation was a retrospective study, MRI was obtained at different time points in the disease course, so the high number of patients with chronic symptoms in this study may have introduced bias. However, all patients selected were homogeneous in having abrupt onset or relapse of obsessions and compulsions in childhood and were experiencing disabling psychiatric symptoms at the time of MRI. Furthermore, this study included a broad range of CGAS scores in the patient sample, suggesting notable diffusion abnormalities across the spectrum of disease severity. However, this metric only considers a child's global functioning over the past week and does not take into account the child's baseline function before PANS. Statistically, we also note that the effect size or partial η 2 of the MANCOVA tests may be considered large based on benchmarks in the literature, 64 but these measures will need to be reassessed with future MRI studies. 65 In addition, radiological findings herein may have been biased by the variable time from last flare onset to MRI acquisition. The influence of disease state on cerebral diffusion changes is unknown. For the present analysis, the well-established adult Montreal Neurological Institute atlas was registered to the MRI of patients with PANS and control participants. Alternatively, age-appropriate atlases for pediatric brains could have been more appropriate, but results may not be directly comparable with the use of multiple atlases. 66 Another limitation of this study is the omission of trends in laterality given that data were merged from both the left and right hemispheres