Report

Noninvasive blood tests for fetal development predict gestational age and preterm delivery

See allHide authors and affiliations

Science  08 Jun 2018:
Vol. 360, Issue 6393, pp. 1133-1136
DOI: 10.1126/science.aar3819
  • Fig. 1 Sample collection timelines from the Denmark, University of Pennsylvania, and University of Alabama cohorts.

    Squares, inverted triangles, and lines indicate sample collection times, delivery dates, and individual women, respectively.

  • Fig. 2 Application of cfRNA measurements to predict gestational age.

    (A) For each gene, cfRNA transcript count measurements are shown over the course of gestation. Each point represents the mean cfRNA value ± SEM for either 31 women or 21 women (the latter denoted by †). The antepartum period is highlighted in gray. Placental, immune, and fetal liver–specific genes are highlighted in blue, green, and orange, respectively. (B) Heat map of the Pearson correlation coefficient for each gene pair shows that placental, immune, and fetal liver–specific cfRNA [same group colors as (A)] measurements are highly correlated with each other [median Pearson correlation r = 0.79 (placenta), 0.79 (immune), 0.74 (fetal liver); P < 10−14]. Placental and fetal liver–specific genes also show a weak degree of cross-correlation (r = 0.47, P < 10−15). Gene order matches order shown in (A), omitting genes denoted by † in (A). (C) Cross-validated random forest (RF) model predicts time to delivery from sampling time point (R = 0.91, P < 10−15, n = 21) for training cohort. (D) Cross-validated random forest model predicts time to delivery from sampling time point (R = 0.89, P < 10−15, n = 10) for validation cohort. (E) Distribution of difference in weeks between observed and predicted gestational age at delivery using cfRNA measurements from the second (T2), third (T3), or both (T2 & T3) trimesters (left to right) versus using ultrasound measurements from the first trimester (T1). AU, arbitrary units.

  • Fig. 3 Application of cfRNA measurements to predict risk of spontaneous preterm delivery.

    (A) Heat map of the z-scores for 38 differentially expressed genes identified using cfRNA-seq (P < 0.001, exact test, likelihood ratio test, and quasi-likelihood F test) shows that genes distinguish women who delivered spontaneously preterm from women who delivered at full term. The two groups of women were separated using hierarchical clustering. (B) Means ± SD for differentially expressed genes validated using qRT-PCR in the discovery [University of Pennsylvania (I) and Denmark (II)] and validation [University of Alabama (III)] cohorts. *P < 0.05, **P < 0.01, ***P < 0.0005 (Fisher exact test). (C) Receiver operating characteristic curves for classifier designed to separate women who deliver spontaneously preterm from women who deliver at full term for both the discovery cohort (University of Pennsylvania and Denmark, AUC = 0.86) and the validation cohort (University of Alabama, AUC = 0.81).

  • Table 1 Comparison of gestational age estimates using cfRNA and ultrasound.

    Distribution of difference between estimates of gestational age, which assume delivery at 40 weeks gestation, and observed gestational age at delivery listed for four distinct methods, where n indicates the number of women included. Gestational age was estimated using cfRNA measurements from the second (T2), third (T3), or both (T2 and T3) trimesters and ultrasound measurements from the first trimester (T1).

    MethodΔ [observed – expected delivery date (weeks)] (%)
    < –2–1 to –2± 1+1 to +2> +2
    cfRNA (T2, n = 28)50183200
    cfRNA (T3, n = 31)06232942
    cfRNA (T2 and T3, n = 31)196451020
    Ultrasound (T1, n = 31)02648233

Supplementary Materials

  • Noninvasive blood tests for fetal development predict gestational age and preterm delivery

    Thuy T. M. Ngo, Mira N. Moufarrej, Marie-Louise H. Rasmussen, Joan Camunas-Soler, Wenying Pan, Jennifer Okamoto, Norma F. Neff, Keli Liu, Ronald J. Wong, Katheryne Downes, Robert Tibshirani, Gary M. Shaw, Line Skotte, David K. Stevenson, Joseph R. Biggio, Michal A. Elovitz, Mads Melbye, Stephen R. Quake

    Materials/Methods, Supplementary Text, Tables, Figures, and/or References

    Download Supplement
    • Materials and Methods 
    • Supplementary Text
    • Figs. S1 to S5
    • Tables S1 to S3
    • References 

Stay Connected to Science

Navigate This Article