Background

Genomic sequencing technologies have advanced rapidly over the past decade. Whole-genome sequencing (WGS) is now used in the diagnosis of rare diseases, circulating tumor DNA (ctDNA) is used to monitor cancer recurrence, and CRISPR-based gene therapies are being administered to patients with sickle cell anemia and beta-thalassemia. However, these four areas—germline, somatic, cell-free DNA (cfDNA), and gene therapy quality control (QC)—have each developed their own independent accuracy criteria and validation systems. Despite sharing the common technical challenge of identifying low-frequency variants, the sensitivity thresholds and reporting criteria applied in clinical settings vary.

In 2019, NIST released a small variant benchmark, and in 2022, the T2T Consortium completed the remaining 8% of the human genome, adding approximately 200 million base pairs and 99 protein-coding genes. In 2023, the Human Pangenome Reference Consortium constructed a graph-based representation of 47 haploid genomes representing diverse populations. While the reference itself is rapidly becoming more sophisticated, the standards for applying it clinically remain fragmented, which is the starting point of this paper.

Key Findings

This Perspective, published in Nature and authored by 27 researchers including Euan A. Ashley of Stanford University and Jennifer A. Doudna of UC Berkeley, systematically examines the standard gaps across the four clinical domains.

Discrepancies in Variant Allele Frequency (VAF) Sensitivity. cfDNA testing demonstrates reliable sensitivity only at VAFs of 0.5% or higher, with results becoming unreliable below this threshold (Deveson et al., 2021). In contrast, off-target detection in gene-edited cells requires ultra-deep sequencing at a 0.1% VAF level (Cromer et al., 2022). This represents a separation of validation frameworks for what is essentially the same challenge: detecting low-frequency variants.

The Neglected Complexity of Clinically Relevant Genes. Wagner et al. (2022) newly characterized 273 medically important genes that were previously excluded from existing benchmarks due to their structural complexity. Standard pipelines may miss variants in these genes, making long-read sequencing and hybrid approaches essential.

Population Diversity and Equity Concerns. Citing warnings that the clinical application of existing polygenic risk scores (PRS) may exacerbate health inequalities (Martin et al., 2019), the authors emphasize that the transition to a pangenome graph reference is critical for improving the accuracy of variant calling in non-European populations.

The authors propose the establishment of a synchronized standard system encompassing the FDA's Laboratory Developed Test (LDT) guidance, NIST's Genome Editing Consortium guidelines, and the EMA's guidance on advanced therapy medicinal products. The core idea is to leverage existing infrastructure such as the PrecisionFDA benchmarking platform, GA4GH (Global Alliance for Genomics and Health), and ClinGen to integrate quality metrics across the four domains.

Significance and Implications

The standard integration advocated for in this paper is not merely an administrative exercise. If the criteria for monitoring off-target effects in CRISPR therapies are operated independently of the sensitivity thresholds for cfDNA testing, conflicting sequencing results may occur in the same patient. In donor-derived cell-free DNA (dd-cfDNA) monitoring after organ transplantation, the lack of consensus on the range of biological variation also leads to differences in rejection criteria among institutions.

From a technological perspective, next-generation platforms such as the transition from linear references to pangenome graphs, UMI (unique molecular identifier)-based error correction, and sequencing by avidity are rapidly entering clinical use. If these technologies are introduced without a validation framework, the risk of reproducibility crises will inevitably repeat. As new applications continue to emerge, as seen in the NIH's SMaHT (Somatic Mosaicism) initiative, reducing the time lag between technology and clinical validation remains the most pressing challenge in this field.