Computational protocol: Season of Sampling and Season of Birth Influence Serotonin Metabolite Levels in Human Cerebrospinal Fluid

Similar protocols

Protocol publication

[…] MA metabolite concentration distributions were verified for normality using the Kolmogorov-Smirnov test (p-values >0.05 were considered indicative of a normal distribution). Median concentrations per season (defined as starting on the 21st and ending on the 20th three months later, e.g. spring starting March 21st and ending June 20th) were computed. Non-linear quantile regression (nlqr) – which models the median values per month - was chosen before inspecting circannual metabolite concentration variability to evaluate the effects of season of sampling and season of birth on MA metabolite concentrations assuming a cosine-shaped relationship. (Median values were used because of uneven distributions of sampling and birth dates). To validate the cosine function and model concentration measurements per day non-parametrically, we used LOESS (locally weighted scatterplot smoothing), which offers the advantage of modeling data points without having to set a function according to which the data are described. The default span in R 2.12.1 ( of 0.75 was used. In the event three months with highest median MA metabolite concentrations were sequential within the same season, the non-parametric group comparison Kruskal-Wallis test was used to compare metabolite levels between that season and the other seasons taken together. Kruskal-Wallis test results with p-values<0.05 were deemed significant. Two covariates, age and sex, were included in all nlqr-models. Other potentially confounding factors (amount of CSF suctioned, type and timing of procedure, comorbidities (psychiatric and other), psychotropic medication and other medication, LP level, and height and weight of participants) were investigated for association with MA metabolites by means of univariate linear regression after multiple (n = 5) imputation of missing values in SPSS 18.0 (SPSS for Windows, SPSS Inc). Factors that showed a univariate association (p<0.05) and did not show collinearity with age or sex (Pearson's r<0.6) were additionally entered into the models (which was the case for LP level and weight in the HVA model). Using Pearson's r, it was checked whether storage time and metabolite concentrations and whether sampling and birth date correlated. Since we hypothesized circannual variation with either one or two zeniths and nadirs per year would apply to all three MA metabolites, nlqr-models were fitted for each of these two scenarios. We measured the fit of each model per MA metabolite and compared the deviances (residual sum of squares). Thus, two nlqr-models were compared for goodness-of-fit for each of the six analyses (season of sampling and season of birth being the predictors for each of the three MA metabolites) and only results for the best fitting model were reported. To determine whether birth and sampling dates independently contributed to MA metabolite concentrations, a third model including both sampling and birth months within one model was created. Model 1a (5-HIAA and MHPG, one peak): Model 1b (5-HIAA and MHPG, two peaks): Model 2a (HVA, one peak): Model 2b (HVA, two peaks): Model 3a (5-HIAA and MHPG, sampling and birth month within one model, one peak): Model 3b (5-HIAA and MHPG, sampling and birth month within one model, two peaks): Model 3c (HVA, sampling and birth month within one model, one peak): Model 3d (HVA, sampling and birth month within one model, two peaks): In which:metabolite = concentration of MA metaboliteβ1 = baseline levelβ2 = amplitude (A)0.5236 = coefficient of t = 2π/12 (one cosine period in radians divided by the number of months per year; “x 2” added for the two-peaks model)t = month of sampling or birth; t1 = month of sampling; t2 = birth monthβ3 = phase shiftβ4, β5, β6, β7, β8, and β9 = covariates' coefficients (except for in model 3, where β4 and β5 are amplitude and phase shift, respectively)For each 1-peak model showing a significant amplitude the tmax (month during which a level is at its maximum), the tmin (month during which a level is at its minimum), and predicted maximum (PCmax) and minimum (PCmin) concentrations were computed:and the predicted concentration at tmax:and the predicted concentration at tmin:In each of the best fitting models, the effect of season of sampling or season of birth was considered significant when the amplitude (A) was significantly different from zero. The nlqr significance level was Bonferroni corrected and set at 0.05/6 = 0.00833 (since three tests for both season of sampling and season of birth were performed). All data analyses were performed with the statistical software package SPSS 18.0 (SPSS for Windows, SPSS Inc) and R 2.12.1 ( […]

Pipeline specifications

Software tools Quantreg, COSINE
Applications Miscellaneous, Protein interaction analysis
Organisms Homo sapiens
Chemicals Dopamine, Homovanillic Acid, Hydroxyindoleacetic Acid, Norepinephrine, Serotonin