Plasma anandamide concentrations are lower in children with autism spectrum disorder

Participants were N = 116 children (N = 60 children with ASD and N = 56 neurotypical control children), aged 3 to 12 years. Participants had been recruited as part of a previous study to investigate blood biomarkers and genetic variants in children with ASD [10]. Participants with ASD were primarily recruited through the Autism and Developmental Disorders Research Registry and by flyers posted in the Stanford University Autism and Developmental Disorders Clinic. Unrelated control participants were recruited through advertisements posted online or hardcopy in the surrounding community.

A comprehensive diagnostic evaluation was performed in children with ASD to confirm the accuracy of their existing diagnosis based on the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition, Text Revision (DSM-IV-TR) criteria [11]. This diagnosis was confirmed using research diagnostic methods (i.e., the Autism Diagnostic Instrument-Revised and the Autism Diagnostic Observation Schedule [12, 13]) by assessors trained by a research-reliable clinician. Study eligibility criteria were as follows. All participants were required to be (1) pre-pubertal, (2) in good medical health, (3) willing to provide a blood sample, and (4) capable of completing behavioral testing. Participants with ASD were included if they had a full-scale IQ of 50 and above. Control participants were included if they had an IQ score of 70 and above. Cognitive functioning was determined using the Stanford Binet 5th Edition [14]. ASD participant exclusion criteria included (1) a genetic, metabolic, or infectious etiology for ASD or (2) a DSM-IV-TR diagnosis of any severe mental disorder (e.g., schizophrenia, bipolar disorder). Participants using psychotropic medications were included if medications were stable for at least 1 week before blood collection. The most common medications in use by participants with ASD were stimulants (n = 10), selective serotonin reuptake inhibitors (n = 8), norepinephrine reuptake inhibitors (n = 5), and antipsychotics (n = 5). Controls were required to (1) be free of present or past neurological disorders; (2) be free of present or past psychiatric disorders on the basis of behavioral scales, a psychiatric evaluation, and, if needed, the Kiddie-Schedule for Affective Disorders and Schizophrenia for School-Aged Children [15]; (3) have no evidence of difficulty during gestation, labor, delivery, or immediate neonatal period, or abnormal neurological or developmental milestones; and (4) have no siblings with ASD.

Blood collection was performed between 10 am and 2 pm to control for any potential circadian rhythmicity in plasma AEA concentrations [16, 17]. Whole blood samples were collected into EDTA-treated vacutainer tubes and promptly centrifuged (1300×g at 4 °C for 10 min). The plasma fraction was aliquoted into polypropylene tubes and stored at − 80 °C until the morning of quantitative analysis. Plasma samples underwent lipid extraction with a modified salt-assisted liquid-liquid extraction (SALLE) [18]. During method optimization, comparison of traditional toluene liquid-liquid extraction to SALLE demonstrated that although both methods provided similar extraction yields (toluene vs SALLE, 85–90% vs 90–95%, respectively), SALLE was more reproducible, provided greater precision with decreased matrix effects, and had more efficient recovery. Commercially available stable isotope-labeled AEA-d8 (Cayman Chemicals; Ann Arbor, MI) was used as an internal standard (IS) for LC-MS/MS in creation of a calibration curve for quantitation of the analyte of interest, endogenous AEA. Plasma samples (100 μl/participant) were thawed in an ice bath (in less than 30 min), de-proteinized with 200 μl of acetonitrile containing 10 ng/ml internal standard solution mix and 50 μl of 5 M ammonium formate. Sample mixtures were then vortexed for 1 min before being spun at 13,000 rpm at 4 °C for 5 min. Organic layers from each sample were collected and transferred to autosampler vials in preparation for LC-MS/MS. Each sample was measured in triplicate. The samples could not be run as one group and, thus, were run in three sets.

The LC-MS/MS system was a TSQ Vantage triple quadrupole mass spectrometer coupled with an Accela 1250 HPLC system (Thermo Fisher Scientific, San Jose, CA). Baseline separation was achieved with gradient elution from an Acquity BEH C18 column (150 mm × 2.1 mm, 1.7 μm particle size) (Waters, Millford, MA). Calibration curve linearity was validated in spiked plasma from 0.25–10,000 pg/μl. The lower limit of AEA quantification was 50 fg on column. The total LC-MS/MS run was 8 min in duration, and samples were maintained at 4 °C throughout. Mobile phases were 0.1% formic acid (A) and acetonitrile containing 0.1% formic acid (B). Linear gradient conditions were as follows: 0–1 min, 50% B; 1–2 min ramp to 98% B; 2–4 min, 98% B; 4–5 min ramp back to 50% B; and 5–8 min equilibration at 50% B. The heated electrospray source was operated in positive ion mode. The triple quadrupole mass spectrometer was operated in selected reaction monitoring (SRM) mode for mass/charge transitions specific to AEA (m/z: 348.3 > 287.4, 203.4, 269.2, 91.0) and AEA-d8 (m/z: 356.25 > 294.3, 252.1, 206.1). Chromatograms were processed using Xcalibur software as well as visually inspected for inconsistencies.

Intrinsic to the consideration of AEA quantification is the “endogenous level challenge,” which relates to the difficulty encompassed by efforts to minimize contributions of sample matrix (e.g., plasma) effects through the use of a blank matrix for preparation of calibration standards in LC-MS/MS quantitation [19, 20]. It is known that for analytes of interest, such as AEA, that will be present at basal conditions, availability of a true blank matrix is rare. Thus, several strategies have been proposed to overcome this confound. The following strategy is a variation of the “authentic analyte in authentic matrix” approach and is related to the “standard addition” approach [19, 20]. For the calibration curve, peak area ratio of spiked unlabeled to labeled analyte (AEA/D8-AEA) was determined in buffer (i.e., phosphate-buffered saline, termed PBS) and extracted “blank pooled plasma” (i.e., from non-affected, healthy adult plasma) and used to construct a linear regression equation: y = m (x) + b, where y is equal to the peak area ratio of spiked analyte/internal standard, m is equal to the slope of the calibration curve, x is equal to the concentration of analyte, and b is equal to the y-intercept of the calibration. Calibration curves were fitted using different weighting schemes (unweighted, 1/x, or 1/x²) and over different ranges of spiked AEA concentrations (minimum 0.1 pg/μl; maximum between 20 and 10,000 pg/μl prepared in plasma, equivalent to 0.5 pg to 50 ng on column). As experimental samples produced calculated AEA concentrations near the lower end of this range, the curve that generated the lowest average relative error at the lower end of the spiked concentration range was retained (range 0.2–20 pg/μl, weighted by 1/x²). Unknown sample AEA concentrations were corrected for the endogenous presence of the analyte in plasma, estimated as the negative X-intercept of the calibration curve. The lowest limit of quantification (LLOQ) in extracted plasma was defined as signal-to-noise ratio of 10 to 1 and was 50 fg for AEA. Additionally, the triplicate measures for each participant were highly consistent, with intra-class correlation (ICC) coefficients within each sample set between 90.2 and 94.8%. High ICC coefficients provide evidentiary support for the reliability of calculated AEA concentrations [21].

Study data were managed using REDCap [22] and analyzed using JMP V.13 (SAS Institute, Cary, NC, USA). Four participants were excluded from analyses because their samples either produced highly inconsistent concentrations between replicates (N = 1F with ASD) or extremely elevated AEA outlier concentrations were detected (i.e., three standard deviations above mean; N = 3 controls: 2M, 1F). A logistic regression was performed to test whether mean AEA concentrations predicted group membership (i.e., ASD vs. control), using a generalized linear model with a binomial error distribution and logit link function. All models included the following blocking factors: age, full-scale IQ, ethnicity, gender, sample set, and sample order within set (as described below).