Molecular pathology selected abstracts

Create PDF

Editors: Donna E. Hansel, MD, PhD, chair of pathology, Oregon Health and Science University, Portland; Richard D. Press, MD, PhD, professor and director of molecular pathology, OHSU; James Solomon, MD, PhD, assistant professor, Department of Pathology and Laboratory Medicine, Weill Cornell Medicine, New York; Sounak Gupta, MBBS, PhD, senior associate consultant, Mayo Clinic, Rochester, Minn.; Tauangtham Anekpuritanang, MD, molecular pathology fellow, Department of Pathology, OHSU; Fei Yang, MD, assistant professor, Department of Pathology, OHSU; and Andres Madrigal, MD, molecular genetic pathology fellow, Department of Pathology, OHSU.

Application of cell-free DNA in molecular diagnosis of vascular malformations

November 2020—Sporadic vascular malformations are congenital malformations of arteries, veins, capillaries, or lymphatic vessels, or a combination of these, and are associated with significant morbidity. The majority of them are caused by postzygotic somatic pathogenic variants in oncogenes in the PI3K-MTOR and RAS-MAPK pathways, including within PIK3CA, TEK, MAP2K1, BRAF, and KRAS. Investigators have assessed whether therapeutic agents targeting these pathways should be used to augment or replace traditional surgical management. But because these somatic variants are restricted to cells within the tissue of vascular malformations (VM), it is necessary to conduct genetic testing on the surgically resected tissue to qualify patients for trials of targeted therapies. Approximately 10 percent of cell-free DNA (cfDNA) originates from endothelial cells. Given the intimate contact of the VM tissues with the circulatory system, the authors proposed that mutation analysis of noninvasive cfDNA could be used for molecular diagnosis of these lesions. To test this hypothesis, they performed sensitive droplet digital polymerase chain reaction on cfDNA from the plasma samples of patients affected with three subtypes of VM: extracranial arteriovenous malformations (AVM), venous malformations (VeM), and lymphatic malformations (LM). All of the patients had somatic mutations that were previously confirmed using the surgically resected VM tissue. The plasma cfDNA analysis detected the expected mutations in two out of eight patients with AVM and one out of three patients with VeM but none of the 26 patients with isolated LM. The authors rationalized that the poor connectivity of LM tissue to blood could have caused the false-negatives in patients with these malformations. Therefore, they hypothesized that the lymphatic fluid drawn from the cystic structure might be a better diagnostic sample. In the seven LM patients who had a cyst fluid sample available, the expected mutations were detected by the cfDNA analysis, with some cfDNA samples having a variant allele fraction (VAF) greater than their corresponding tissue VAF. The authors further extended the cfDNA analysis to cyst fluid collected from LM patients who did not have surgical resection and subsequently detected somatic variants in four out of five people. They proposed that larger volume cfDNA samples potentially could increase diagnostic yield in AVM and VeM. The study was limited by a small sample size and the heterogeneity in plasma and cyst fluid collection. The authors concluded that this study supports further investigation of cfDNA-based genetic testing in the plasma or cystic fluid of patients who have vascular malformations, especially in patients with lymphatic malformations for whom the VM tissue-based analysis is insufficient as a result of significant intralesional heterogeneity.

Zenner K, Jensen DM, Cook TT, et al. Cell-free DNA as a diagnostic analyte for molecular diagnosis of vascular malformations [published online September 4, 2020]. Genet Med. 2020. doi:10.1038/s41436-020-00943-8

Correspondence: Dr. James T. Bennett at jtbenn@uw.edu

Performance of mutation pathogenicity predictors using variant datasets

In silico tools for predicting the functional impact of genetic variation are used routinely in clinical diagnostics to help classify genetic variants, especially missense variants. Traditional tools, such as the commonly used SIFT and PolyPhen-2, use inter-species conservation, physicochemical distances of exchanged protein residues, and x-ray crystallographic structures to make functional predictions. Clinical guidelines for variant classification, such as the American College of Medical Genetics (ACMG) guidelines, recommend using consensus-based approaches in which the evidence is applicable only when a concordance is reached among a majority, if not all, of the prediction tools used. Recently developed meta-predictors, such as REVEL, GAVIN, and ClinPred, are machine-learning–based algorithms that integrate information from a large number of primary sources and apply quantitative weight to maximize accuracy, with the intent that a single meta-predictor might be able to replace the aforementioned manual consensus-based practice. The authors conducted a study in which they compared the performance of these three bioinformatic meta-predictors, along with SIFT and PolyPhen-2, in two sets of high-quality pathogenic and benign variants. One was an open dataset compiled from online public databases, including HGMD and ClinVar for pathogenic variants and gnomAD for benign variants. The other was clinical datasets compiled from the Deciphering Developmental Disorders and UK Association for Clinical Genomic Science (ACGS) clinical databases, with variants manually curated according to the ACGS/ACMG guidelines. In both variant datasets, the meta-predictors demonstrated a greater sensitivity and specificity than the traditional SIFT and PolyPhen-2 tools, which were used individually or in combination following the consensus-based approaches. Among the three meta-predictors, REVEL performed slightly better. However, the performance of all meta-predictors was substantially worse in the clinical dataset than in the open dataset, showing a reduction in specificity from a range of 0.95 to 1.00 in the open dataset to a range of 0.25 to 0.60 in the clinical dataset. The sensitivity remained largely constant (about 0.90) in both datasets. This phenomenon was not unexpected and possibly illustrated the problem of overfitting as the machine-learning algorithms were often trained on data available from large online public databases that possessed extra features that did not necessarily contribute to the variant classification. The clinical datasets discussed herein, being independent and novel, might provide a more representative assessment of the performance of these tools. In summary, this study indicates that the meta-predictors of variant pathogenicity offer advantages over the consensus-based approaches and may contribute to the universal long-term goal of improving computational tools to assist in routine next-generation sequencing data interpretation.