Statistical Analysis

To identify the candidate peptide epitope sequences for their subsequent use in the screening microarrays, discovery microarrays were first qualitatively assessed for regional artifacts (dust, lint, etc.) and staining abnormalities. Only regions with peptides exhibiting the highest median FIs and minimal spot aberrations were then quantitatively assessed for peptide immunoreactivity. Peptides with the highest median foreground intensities (≥600 FI) and a signal to noise ratio ≥5 for at least one of the pooled maternal samples within the discovery peptide microarrays were considered as highly reactive for this study and were thus included in subsequent screening peptide microarrays (Nagele et al., 2013). These cut-off values were selected based on cutoff ranges reported in similar autoantibody epitope mapping studies that also used PEPperPRINT microarray technology (Hamilton et al., 2015; Korkmaz et al., 2013).

The duplicate coefficient of variation (CV) was calculated for each peptide epitope, and peptides with a CV greater than 50% were set to missing. Peptides that were not bound by any pooled plasma samples in the discovery microarrays were identified and selected to serve as negative controls for the screening peptide microarrays.

Within the screening peptide microarrays, a peptide was determined to be positive (reactive) for a given maternal plasma sample if both of the following criteria were met:

The Chebyshev Inequality Precision Value (CI-p-Value), calculated with the red foreground median fluorescent data, was less than 0.05 for both spots for that peptide. The CI-p-Value is defined as

where Y_k is the observed FI for a peptide spot, s is the standard deviation of control spots on the array, and X¯ is the sample mean of control spots on the array(Love, 2006).

The CV between duplicate spots was less than 50%.

After determining the positive/negative status of individual peptides for each sample via CI-p-Values, we first excluded all peptides that were negative against 100% of the 85 maternal samples in an effort to select for robust peptide reactivity profiles. Peptides that were identified as positive for more than 5% of all maternal samples were considered to be immunodominant (Maksimov et al., 2012). To determine whether reactivity to the individual peptide epitopes of interest differed across maternal sample populations, the resulting positive/negative peptide reactivity data was then compared between maternal subjects. Given our relatively small sample size, we deemed it inappropriate to calculate statistical significance across maternal sample groups with either chi-squared test of independence or Fisher’s exact test at this time. Instead, odds ratios (ORs) with 95% confidence intervals (95% CIs) were calculated for each individual peptide in two distinct sets of preliminary comparative analyses. In the first set of calculated ORs (Set 1), individual peptide reactivity of all 85 maternal samples were compared across mothers of children with ASD and of mothers of TD children. These initial comparisons included maternal samples previously identified by western blot to be non-reactive (negative) to any of the seven protein antigens of MAR ASD. The second set of comparative analyses (Set 2) was calculated only in mothers previously determined via western blot to harbor autoantibodies specific the antigenic protein that corresponds to the peptide epitope of interest (ASD mothers, N = 11-20; TD mothers, N = 3-9). For example, only mothers that were determined to be reactive against LDH-A were included in the third set of OR calculations for the corresponding LDH-A peptide epitopes. A 0.5 continuity correction was applied to all OR calculations for observations with zero cell counts(Subbiah and Srinivasan, 2008). Statistical analyses were performed with PepSlide Analyser software (PEPperPRINT) and SPSS (Version 23, IBM Corporation, Armonk, NY). All graphs were creating using GraphPad Prism software (GraphPad Software, San Diego, CA).