Statistical insights into the development of Non-Invasive Prenatal Diagnosis of Maternally-Inherited Monogenic Disorders

Prenatal testing using fetal genomic DNA obtained by amniocentesis offers an early, rapid and reliable diagnosis, however this invasive procedure presents small but significant risk for both fetus and mother. The deployment of technologies such as droplet digital-PCR (ddPCR) or NGS for analysis of cell-free fetal DNA, allows to set up various non-invasive prenatal tests detecting single-gene disorders. While already implemented in clinical routine for paternal and recurrent de novo point mutation exclusion, the non-invasive detection of single-gene disorders arising from maternal inheritance remains a challenge. For most of maternally-inherited monogenic disorders, we propose to non-invasively identify fetal status from an iterative collection of maternal plasma samples during pregnancy until conclusion. Each sample is split for two experiments: one involving the unknown fetal fraction and the other the unknown probability for an allele of interest to be mutant. We develop a new statistical model for locus copy numbers involved in each experiment that gives a proper statistical description of the biological experiment and derive a likelihood ratio test of the hypotheses “affected” against “unaffected” fetus identification. Our model agrees with Poisson model for droplets in ddPCR experiment and allows iterative pooling of results from each biological experiment. This model accounts for an analysed number of genomes equivalents that increases with plasma volume and hence benefits from iterative blood sampling. Consequently, discrimination rate is expected to increase along the iterations.

The NIPT performance is evaluated empirically for various blood sample sizes with or without the pooling strategy. Through the model, we investigate whether the procedure is robust to departure from biological assumptions and to probe bias. The strength of our method relies in a conservative control of statistical errors ensured by replacing the usual plug-in estimate of the fetal fraction, whose fluctuations are not controlled, by its lower bound estimation.