Serum thyroid-stimulating hormone and interleukin-8 levels in boys with autism spectrum disorder

The study protocol was submitted by The Johnson Center for Child Health and Development (Austin, TX) and approved by the Austin Multi-Institutional Review Board (AMIRB). All methods employed in the study were carried out in accordance with the relevant guidelines and regulations. Informed consent was obtained from the parents or legal guardians of all subjects prior to their participation in this research.

The initial study participants used to measure analytes on the RBM platform consisted of 30 boys with ASD and 30 typically-developing (TD) boys, ages 2–8 years. An additional group of 13 ASD and 9 TD samples (boys, ages 2–8 years) were included in the subsequent analysis on the MSD platform to increase the sample size. Subjects were either recruited directly from The Johnson Center clinic, or through the use of informational study flyers circulated around Austin, TX. Written informed consent was received from the parent or guardian of all subjects prior to enrollment. Briefly, the psychiatric, medical, and family histories of all participants were obtained. For the ASD group, the subjects were assessed by a psychologist trained in research reliability using both the Autism Diagnostic Observation Schedule (ADOS) and the Autism Diagnostic Interview–Revised (ADI-R). Clinical diagnosis was made based on these data and overall clinical impression using DSM-IV criteria. For this particular study, subjects with a diagnosis of Asperger’s Syndrome or Pervasive Developmental Disorder––not otherwise specified, were excluded. For the TD group, all subjects underwent a developmental screening using the Adaptive Behavior Assessment System-Second Edition (ABAS‐II) that was assessed by the psychologist. TD subjects were excluded if their score on the ABAS-II suggested possible abnormal development and the need for further evaluation. TD subjects were also excluded if they had a first- or second-degree relative diagnosed with ASD. Subjects diagnosed with a genetic, metabolic, or other concurrent physical, mental, or neurological disorder were excluded, as were subjects that were currently taking psychiatric medications (or had taken psychiatric medications within the last 3 months prior to enrollment). All subjects were healthy with no reported illnesses for 3 weeks prior to participation in the study.

Due to the high degree of phenotypic heterogeneity in ASD, we further sub-characterized ASD subjects into three groups [16]: (i) those who were nonverbal, (ii) those with gastrointestinal (GI) concerns, and (iii) those with regressive autism. Subjects with ASD were defined as nonverbal if there was a complete absence of intelligible words at time of diagnostic assessment of autism. ASD subjects were classified as having GI concerns if they reported at least one of the following symptoms: (i) constipation; (ii) diarrhea; (iii) abdominal bloating, discomfort, or irritability; (iv) gastroesophageal reflux or vomiting; and/or (v) feeding issues or food selectivity. ASD subjects were classified as having no-regression if the child exhibited traits of autism from infancy, and regressive autism if they had typical early development and later lost function in language and/or social interactions (based on questions probed in the ADI-R). The correlations between protein levels, phenotypic sub-groups, and clinically relevant quantitative traits from the ADOS were analyzed.

A fasting blood draw was performed on healthy children between the hours of 8–10 a.m. Blood was collected in a 3.5 ml serum separation tube (SST; Vacutainer System; Becton-Dickinson) using standard venipuncture technique. The blood was gently mixed in the SST by five inversions and then stored upright for clotting at room temperature for 10–15 min. Blood was then spun immediately after the clotting time in a swing bucket rotor for 15 min at 1100–1300 g at room temperature. Serum was removed immediately after centrifugation and transferred into coded cryovials in 0.5 ml aliquots. Aliquots of serum were immediately placed upright into storage boxes in a −20 °C freezer for up to 6 h. Samples were then transferred to a −80 °C freezer for long-term storage.

Sample aliquots were coded and shipped on dry ice to Myriad Rules-Based Medicine (RBM; Austin TX) for evaluation using DiscoveryMAP 175+ for quantitative immunoassay of inflammatory molecules and hormones. This multianalyte Luminex profiling platform examined over 175 protein analytes [17, 18]. Final data were reported as the absolute concentrations in the serum. A total of 30 ASD and 30 TD male serum samples were analyzed. Some coded duplicate samples were also run and used to analyze analyte measurement performance. Analyte measurements that showed >15% variance were excluded from the data analysis.

We sought to replicate the serum biomarker proteins identified on the RBM platform by subsequently analyzing proteins on a Meso Scales Discovery (MSD) platform [19] run in-house. Compared with the traditional ELISA approach, the MSD platform shows greater sensitivity and is able to reliably detect different proteins across a broad dynamic range of concentrations [20]. The assay is based upon electro-chemiluminescence technology by using specific capture antibodies coated at corresponding spots on an electric wired microplate. This platform was used to measure TSH in 43 ASD boys and 39 TD boys, and IL-8 in 36 ASD boys and 35 TD boys; the two proteins showing the greatest percent difference in the ASD and TD samples run on the RBM platform. We measured the two proteins in samples that were run in duplicate accordingly to the manufacturer’s protocol. Any duplicate value with >15% variance was removed from the final data analysis, and every plate was run with a standard concentration curve.

The RBM data were analyzed using random forest methods. Random forest analysis was developed as an ensemble learning method that utilizes a classification tree as the base classifier [21]. Hundreds of training and test sets, of 15 subjects/group, were analyzed to determine the importance of a panel of analytes to correctly identify ASD subjects. For the MSD data, differences between the ASD and TD groups were analyzed with Mann-Whitney U tests. For comparing the accuracy of the two analytes for predicting ASD vs. TD, we used cut scores and area under the curve (AUC) analyses. Diagnostic accuracy and AUC were computed by ROC (receiver operation characteristic) curves using SPSS V23; the optimum probability cutoffs were determined using mathematical formulas in Microsoft Excel™ to maximize accuracy and the perpendicular distance from the 45 degree line of equality. The p < 0.05 level was considered to be statistically significant for analyses using the MSD platform (i.e., for TSH and IL-8 assays).

Regression analyses for ASD subjects were conducted using the R lavaan package, which fits models using full information maximum likelihood estimation that makes use of all available data. Thus, data from all ASD subjects were included in each model [22] (n = 43 for TSH and n = 36 for IL-8). Protein levels were regressed on each of the ADOS subdomain scores and phenotypic sub-grouping to examine whether levels of TSH and/or IL-8 were related to a clinical measure of ASD and comorbidities. Prior to fitting regression models, IL-8 was log-transformed to reduce the positive skew; the transformed distribution was approximately normally distributed and met guidelines for covariance matrix-based models [23].

A total of 184 analytes were measured in serum samples from 30 ASD boys and 30 TD boys on the RBM Luminex platform; however, 51 analytes were undetectable, and 23 exhibited >15% spot-to-spot variance, and were therefore omitted from analysis. The undetectable proteins likely represent faulty antibodies, as these proteins have been detected with previous versions of RBM's Discovery Map. Eleven of the remaining 110 serum proteins measured, shown in Table 1, were selected based upon a random forest analysis, and 5 of the analytes were significantly different between ASD and TD at the p ≤ 0.04 level (uncorrected for multiple comparisons). Using a random selection of 15 ASD and 15 TD subjects for a training set and the same number of subjects for a test set, 100 comparisons of data indicated that the random forest test set could accurately confirm ASD vs. TD samples with an average area under the ROC curves of 0.761. Those proteins with the highest importance for confirming a diagnosis of ASD include TSH, stem cell factor, monocyte chemotactic protein 4, ferritin, and IL-8.

Using 43 ASD samples, we found TSH levels to be 30% lower vs. TD samples (n = 37): 1.42 ± 0.08 (mean ± SEM) and 2.04 ± 0.16 mIU/l, respectively, (p < 0.0056; Mann-Whitney U test). TSH levels were within the normal range for this hormone for children of the age studied (http://​www.​mayomedicallabor​atories.​com/​test-info/​pediatric/​refvalues/​) (see Fig. 1). For the prediction of ASD, TSH levels were 94% sensitive and 60% specific. The area under the ROC curve was 0.674, p = 0.006. We measured IL-8 levels in 36 ASD boys, and it was 16% higher vs. TD boys (n = 35): 12.17 ± 0.52 (mean ± SEM) and 10.52 ± 0.51 pg/ml, respectively, (p < 0.0306; Mann-Whitney U test) (see Fig. 1). For the prediction of ASD, IL-8 levels were 94% sensitive and 55% specific. The area under the ROC curve was 0.652, p = 0.023.

In order to determine whether the accuracy in confirming ASD vs. TD is enhanced by an analysis using a combination of analytes, we analyzed the prediction accuracy using both TSH and IL-8 analytes (Table 2). Here, we used samples from subjects that were run for both analytes on the MSD platform, which resulted in 18 ASD and 20 TD boys. The predictive accuracy for TSH alone was 76%, and for IL-8 alone, it was 74%. Using the two analytes together, the predictive accuracy was 82%, with 89% sensitivity and 75% specificity. ASD cases were predicted, using cut scores, as having TSH levels below 1.8 mIU/l and IL-8 levels above 10.3 pg/ml. The area under the ROC curve for the model using the two analytes was 0.842 ± 0.067 SEM (p < 0.001).

TSH was regressed on each of the ADOS subdomain scores. The following three domains exhibited a significant negative correlation whereby higher scores (more severe symptoms) in the subdomains were associated with lower levels of TSH: social interaction (z = −2.61, p = 0.009), communication + social interaction (z = −2.12, p = 0.034), and stereotyped behavior and restrictive interests (SBRI) (z = −2.28, p = 0.023). There was not a significant relationship between TSH and the ADOS communication subdomain (z = −0.55, p = 0.581).

IL-8 was regressed on each of the ADOS subdomain scores. Among ADOS subdomains, there were no significant relationships between IL-8 and communication (z = 0.16, p = 0.871), social interactions (z = −0.15, p = 0.877), communication + social interactions (z = −0.06, p = 0.953), or SBRI (z = 0.75, p = 0.455).

The percentage of children with ASD who were nonverbal was 50%. The percentage of children with ASD displaying GI issues was 85%. Regressive autism was seen in 63% of the study group. There were no significant relationships between either TSH or IL-8 and the autism sub-groups. For TSH: nonverbal (z = −0.51, p = 0.609), GI concerns (z = −0.14, p = 0.890), and regression (z = −1.12, p = 0.265). For IL-8: nonverbal (z = −0.20, p = 0.843), GI concerns (z = −0.21, p = 0.833), and regression (z = −0.49, p = 0.624).