TL;DR
This paper introduces a novel combined Bayesian and haplotype-based method for non-invasively predicting fetal genotypes from maternal plasma DNA. This approach significantly improves accuracy in identifying single nucleotide variations (SNVs) and insertions/deletions (InDels), paving the way for more reliable prenatal diagnosis of monogenic disorders.
What Problem Does This Paper Address?
Non-invasive prenatal diagnosis (NIPD) offers a low-risk way to detect genetic disorders during pregnancy. However, existing methods often struggle with accurately inferring fetal genotypes, especially at heterozygous loci or when parental haplotype information is incomplete. This limitation hinders the ability to diagnose a wide range of monogenic diseases, particularly those not caused by common mutations or those with de novo variants. Furthermore, some methods require proband DNA, which may not be available, or have limited accuracy in complex genetic scenarios. This study aims to overcome these limitations by developing a more robust method that leverages both maternal plasma cell-free DNA (cfDNA) and parental genomic DNA.
Methods
The researchers developed a two-pronged approach:
- Bayesian Model: This model uses maternal plasma cfDNA and parental genomic DNA to infer fetal genotypes. It calculates the probability of different maternal and fetal genotype combinations based on read depth, fetal fraction (FF), and Mendelian inheritance laws.
- Haplotype-Based Method: This method leverages long-range phasing of parental genomic DNA to reconstruct parental haplotypes. This is crucial for accurately determining which parental allele is inherited by the fetus, especially at heterozygous loci.
The core innovation lies in combining these two methods. The Bayesian model is used for certain loci (e.g., homozygous maternal, heterozygous paternal), while the haplotype-based method handles more complex scenarios (e.g., heterozygous maternal and paternal). This synergistic approach aims to fill in the gaps left by each individual method, leading to a more comprehensive and accurate fetal genotype prediction. They also employed specialized algorithms like Sequential Probability Ratio Testing (SPRT) and "closest-variant" algorithms to infer parental contributions.
Datasets
The study utilized samples from two groups of participants:
- Healthy Pregnancies: Five couples with healthy singleton pregnancies were recruited. Their maternal plasma cfDNA and parental genomic DNA were sequenced. Umbilical cord blood was collected post-delivery for validation.
- High-Risk Pregnancies: Five additional couples whose fetuses were at risk for specific monogenic diseases were recruited. Amniocentesis was performed for prenatal diagnosis, and parental genomic DNA was Sanger-sequenced to confirm disease variants.
The cfDNA was sequenced using whole-genome sequencing (WGS) and the parental genomic DNA using single-tube long fragment reads (stLFR) sequencing.
Key Insights and Ideas
- Synergy is Key: The most significant insight is that combining Bayesian inference with haplotype-based phasing dramatically improves fetal genotype prediction accuracy, especially for complex heterozygous loci where individual methods falter.
- Haplotype Resolution Matters: The use of stLFR sequencing allowed for longer-range haplotype phasing, which is critical for accurate inference, outperforming previous methods that relied on shorter haplotype blocks.
- Fetal Fraction is a Hurdle, But Not Insurmountable: While a higher fetal fraction (FF) generally leads to better accuracy, the combined method demonstrated impressive performance even at very low FF (as low as 4%), suggesting its potential for early pregnancy diagnosis.
- Beyond SNVs: The method's ability to accurately detect both SNVs and InDels is a significant step forward, as InDels are often challenging to call.
Key Results
The combined method achieved high accuracies in predicting fetal genotypes:
- SNVs:
- Maternal homozygous and paternal heterozygous loci (AAAB): 96.2 ± 5.8%
- Maternal heterozygous and paternal homozygous loci (ABAA): 96.2 ± 1.4%
- Maternal heterozygous and paternal heterozygous loci (ABAB): 87.2 ± 4.7%
- InDels:
- AAAB loci: 94.6 ± 1.9%
- ABAA loci: 80.2 ± 4.3%
- ABAB loci: 79.3 ± 3.3%
- Monogenic Disease Diagnosis: In five fetuses at risk for monogenic diseases, the approach detected pathogenic variants with an accuracy of 87.5%.
Limitations and Future Work
The authors acknowledge several limitations and suggest future directions:
- ABAB and InDel Accuracy: While improved, the accuracy for ABAB loci and InDels at AAAB and ABAB loci could still be enhanced.
- Improving the Bayesian Model: The Bayesian model could be further refined by incorporating additional features, such as fragment size and preferred ending positions of fetal cfDNA fragments, to better distinguish fetal from maternal reads.
- Scalability and Clinical Feasibility: The authors suggest that as genome-wide direct phasing becomes more accessible and cost-effective, their method could become more widely adopted. Further work on increasing clinical feasibility is also warranted.
- De Novo Variants and X-linked Inheritance: While the paper doesn't explicitly state limitations for these, previous Bayesian methods have struggled with de novo variants and X-linked inheritance, which might be areas for future investigation with this combined approach.
Summary updated