Study cohort

Ethics approval was obtained before commencement of the study (Mercy Health Human Research Ethics Committee, study number R03/23), and signed informed consent was obtained from all participants. Women attending the Mercy Hospital for Women (tertiary teaching hospital in Melbourne, Australia) for their first prenatal visit were invited to participate if an office ultrasound scan confirmed the presence of a viable fetus (cardiac activity) at 7-11 weeks of gestation. After the attending clinician confirmed that the fetus was live (cardiac activity seen on ultrasound scan), a single blood sample was collected, processed, and frozen on the same day. Gestational age was determined by measurement of the fetal crown-rump length with ultrasound scanning on the same day the blood samples were taken.

After 756 samples had been collected over 31 months, 57 cases of SGA (birthweight, <10th percentile for gestation according to percentile charts that were derived from the local Australian population ) and 71 gestation-matched control subjects (birthweight, 30-80th percentile; median, 50th percentile) were selected whose gestation ranged from 7-10 weeks. To describe cytokine profiles across each gestational week in normal pregnancies (30-80th percentiles), an additional cohort to the control group was included from whom blood samples were taken at 11 weeks of gestation (n = 15). This group at 11 weeks of gestation was not used in any comparisons with cytokine levels from SGA pregnancies. Exclusion criteria comprised known or suspected infection, fetal chromosomal aberration, and other concurrent pregnancy complications, such preeclampsia, gestational diabetes mellitus, and fetal macrosomia. Women with chronic diseases, such as hypertension or diabetes mellitus, were excluded.

Sample collection

Approximately 8 mL of venous blood was collected into vacuette serum clot separator tubes (Greiner Bio-One, Kremsmünster, Austria) without additives. The tubes were centrifuged at 3000 g for 10 minutes, and the serum was collected and stored at –80°C.

Multiplex assay

The Bio-Plex solution array system (Bio-Rad Laboratories, Hercules, CA) was used to simultaneously measure 27 cytokines, chemokines, and growth factors in these samples (Bio-Rad Laboratories, Hercules, CA), following the manufacturer’s instructions. The following analytes were measured: interleukin (IL)-1β, -1 receptor antagonist (IL-1ra), -2, -4, -5, -6, -7, -8, -9, -10, -12(p70), -13, -15, and -17; eotaxin; fibroblast growth factor–basic; granulocyte colony–stimulating factor (G-CSF); granulocyte macrophage colony–stimulating factor; interferon-γ; interferon-γ-inducible protein-10; monocyte chemoattractant protein (MCP)–1; macrophage inflammatory protein (MIP)-1α; MIP-1β; platelet-derived growth factor–BB; regulated on activation, normal T-cell expressed and secreted (RANTES); tumor necrosis factor–α, and vascular endothelial growth factor. All assays were performed with the technician blinded to clinical details, which were scrambled on the Bio-Plex plate and performed in duplicate. All sera were analyzed in 4 batches and had 1 freeze-thaw cycle (on ice) for aliquoting purposes.

Only a single serum sample from the selected cohorts was collected and screened. A standard curve was generated for each cytokine with a 5-parameter logistic regression curve fit; analyte concentrations were determined automatically with the Bio-Plex Manager software (version 4.01). All results below the detectable limit were recorded as zero; those results above the standard curve were assigned a value equal to the top standard.

Statistical analyses

Statistical analyses were performed with Statistical Package for the Social Sciences software (version 17.0; SPSS Inc, Chicago, IL). Demographic and clinical variables were compared with the use of the Students t test or the chi-square test for contingency tables; a probability value of < .05 was considered statistically significant.

To compare individual cytokines, a predefined algorithm was used ( Supplementary Figure 1 ). Raw data for each cytokine was tested for normality by the plotting of the standardized residuals and compared with the use of the 1-sample Kolmogorov-Smirnov test. Nonparametric data were transformed with either a square root or cube root function, with normality again tested with the Kolmogorov-Smirnov test.

For normally distributed data (or data that could be transformed to normality), the following protocol was used: One-way analysis of variance (ANOVA) was performed to determine whether cytokine concentrations varied with gestation, followed by Tukey honestly significant difference post-hoc comparison across each gestational week. For cytokines that did not vary with gestation, SGA and control groups were compared with the use of a 2-way ANOVA, with gestational age included as a fixed factor. For cytokines that varied significantly across gestational ages, a linear regression analysis was performed. If this was significant, a 2-way ANOVA was used to compare SGA and control groups, with gestational age included as a continuous variable.

For cytokine data that were not distributed normally (and could not be transformed to normality), the following procedure was adopted: The Kruskal-Wallis test was used to test for changes in serum concentrations across gestation, followed by a Mann-Whitney comparison across each gestational week. For cytokines that did not vary with gestation, a Mann-Whitney U test was used to compare levels between SGA and control groups. For cytokines that did vary with gestation, the data were converted to multiples of the median, and the 2 groups were compared with the use of the Mann-Whitney U test. Multiples of the median were calculated by the division of each value by the median of the controls within the same gestational age. In the case in which the median was zero and the multiples of the median values could not be generated, the raw data were rank transformed, and 2-way ANOVA was performed. Hochberg’s procedure was used to correct for multiple hypothesis testing when we compared cytokine concentrations between the control and SGA groups; a probability value of < .036 was considered statistically significant.

The cytokines that were found to differ significantly between the SGA and control groups were assessed as potential predictive biomarkers of disease with conditional logistic regression modeling. Those cytokines that were found to be significant in univariate modeling were entered into a multivariate binary logistic regression model with the use of backward elimination to investigate their predictive potential for SGA. Receiver operator characteristic curves were generated; the sensitivity, specificity, and predictive values for the “best” biomarker combination were calculated.