First-Trimester Non-Invasive Prediction of Preterm Birth Using Cell-Free DNA Fragmentomics

Technical Summary: First-Trimester Non-Invasive Prediction of Preterm Birth Using Cell-Free DNA Fragmentomics

Problem Statement
Preterm birth (PTB), defined as delivery before 37 weeks of gestation, remains a leading cause of neonatal morbidity and mortality globally, with limited capacity for early risk identification in asymptomatic women. Current first-trimester prediction strategies relying on maternal history, cervical length, and biochemical markers exhibit modest sensitivity and specificity, particularly in nulliparous women. While cell-free DNA (cfDNA) fragmentomics has shown promise in detecting pregnancy complications, existing approaches for PTB prediction have required additional assays beyond routine Non-Invasive Prenatal Testing (NIPT), such as paired cfRNA sequencing or promoter profiling, which increase costs and limit clinical applicability. There is a critical need for a method that leverages existing routine NIPT data to accurately predict spontaneous PTB without additional blood collection or sequencing.

Methodology
The study employed a prospective, multicenter, nested case-control design involving 286 pregnant women across various clinical sites in Vietnam.

• Cohort: The study included 82 spontaneous PTB cases and 204 term controls. Samples were collected during routine first-trimester NIPT (median gestational age: 12 weeks), prior to the onset of clinical signs of preterm labor. Medically indicated preterm deliveries were excluded to focus on spontaneous PTB.
• Sequencing: Maternal plasma cfDNA was extracted and subjected to whole-genome sequencing on the AVITI platform to a depth of approximately 20 million paired-end reads (2 × 75 bp) per sample.
• Feature Extraction: Five distinct classes of fragmentomic features were extracted from the sequencing data:
  1. End Motifs (EM): Relative frequencies of all 256 possible 4-mer genomic sequences at the 5' ends of fragments.
  2. Fragment Length (FLEN): Size distribution of aligned fragments.
  3. Nucleosome Distance (ND): Profiles derived using the Griffin reference panel.
  4. Copy Number Alterations (CNA): Derived from read-depth analysis across 1,000-bp bins.
  5. Joint Features (FLEN × EM): Combined fragment-length and end-motif representations.
• Model Development: The cohort was split into a training set (n=228; 65 PTB, 163 term) and a validation set (n=58; 17 PTB, 41 term). L1-regularized (LASSO) logistic regression classifiers were developed for each feature class. Hyperparameters were optimized via 5-fold cross-validation in the training set, and models were independently evaluated on the validation set without retraining.

Key Results

• Feature Discriminative Power: Among the five feature classes, the 4-mer end-motif (EM) profiles exhibited the most pronounced differences between spontaneous PTB and term controls. Specifically, PTB samples showed enrichment of C*** motifs (e.g., CCAA, CAAA) and depletion of G*** and T*** motifs compared to controls. In contrast, fragment length, nucleosome distance, and copy number alteration features showed only modest or inconsistent differences.
• Classifier Performance:
  • Training Cohort: The EM-based classifier achieved a median cross-validated AUC of 0.980, significantly outperforming FLEN (0.712), CNA (0.698), FLEN × EM (0.648), and ND (0.587).
  • Validation Cohort: The EM-based classifier maintained high performance with an AUC of 0.970 (95% CI, 0.912–1.000). At a specificity threshold of >90%, the model achieved a sensitivity of 94% (95% CI, 78–100%).
  • Comparison: Other feature classes performed substantially lower in the validation set (e.g., CNA AUC 0.731, ND AUC 0.602). Notably, combining FLEN with EM features resulted in lower performance (AUC 0.759) than EM alone, suggesting that low-discriminative features introduced noise.

Key Contributions and Claims
The paper claims that end-motif signatures derived from routine first-trimester NIPT data can accurately identify pregnancies at risk of spontaneous preterm birth. The primary contributions are:

1. Repurposing Existing Data: The study demonstrates that accurate PTB prediction (AUC=0.970) can be achieved using only the cfDNA sequencing data already generated for routine aneuploidy screening, requiring no additional blood sampling, library preparation, or sequencing.
2. Superiority of End Motifs: The research identifies 4-mer end-motif composition as the dominant fragmentomic discriminator for first-trimester PTB, outperforming fragment length, nucleosome positioning, and copy number alterations.
3. Clinical Feasibility: By integrating with the established global NIPT infrastructure (which performs over 10 million tests annually), this approach offers a scalable solution for early risk stratification, potentially enabling timely interventions such as progesterone supplementation or cervical pessary placement.

Significance and Limitations
The authors position this work as a proof-of-concept study demonstrating a substantive improvement in predictive accuracy and translational feasibility compared to previous methods that required multimodal assays or later gestational sampling. The high performance is attributed to the biochemical specificity of end motifs, which reflect the cleavage preferences of extracellular endonucleases and encode a signature of systemic and placental processes.

However, the paper explicitly acknowledges limitations:

• Generalizability: The cohort consisted exclusively of Vietnamese (Kinh ethnicity) participants. External validation in geographically and ethnically diverse populations is required.
• Sample Size: The total cohort size (n=286) is modest relative to the 256-dimensional feature space, raising the possibility of optimistic performance estimation. The authors caution that the high AUC should be interpreted with care pending replication in larger independent cohorts.
• Exploratory Nature: No formal sample size or power calculation was conducted.

The study concludes that while the analytical framework is promising and extensible to other placental-mediated complications, prospective multi-ancestry validation and integration with established clinical predictors are necessary before clinical implementation.