Early Detection of Malignant and Premalignant Peripheral Nerve Tumors Using Cell-Free DNA Fragmentomics

Abstract

Purpose: Early detection of neurofibromatosis type 1 (NF1)-associated peripheral nerve sheath tumors (PNST) informs clinical decision-making, enabling early definitive treatment and potentially averting deadly outcomes. In this study, we describe a cell-free DNA (cfDNA) fragmentomic approach that distinguishes nonmalignant, premalignant, and malignant forms of PNST in the cancer predisposition syndrome, NF1.

Experimental design: cfDNA was isolated from plasma samples of a novel cohort of 101 patients with NF1 and 21 healthy controls and underwent whole-genome sequencing. We investigated diagnosis-specific signatures of copy-number alterations with in silico size selection as well as fragment profiles. Fragmentomics were analyzed using complementary feature types: bin-wise fragment size ratios, end motifs, and fragment non-negative matrix factorization signatures.

Results: The novel cohort of patients with NF1 validated that our previous cfDNA copy-number alteration-based approach identifies malignant PNST (MPNST) but cannot distinguish between benign and premalignant states. Fragmentomic methods were able to differentiate premalignant states including atypical neurofibromas (AN). Fragmentomics also adjudicated AN cases suspicious for MPNST, correctly diagnosing samples noninvasively, which could have informed clinical management.

Conclusions: Novel cfDNA fragmentomic signatures distinguish AN from benign plexiform neurofibromas and MPNST, enabling more precise clinical diagnosis and management. This study pioneers the early detection of malignant and premalignant PNST in NF1 and provides a blueprint for decentralizing noninvasive cancer surveillance in hereditary cancer predisposition syndromes.

Figure 1.

Size-selected copy number in cell-free DNA identifies MPNST but does not resolve premalignant tumor states. A, Highest per-participant size-selected ichorCNA copy-number–derived tumor fraction in each clinical state for initial (I) and validation (V) cohorts. B, Significance levels of validation cohort size-selected tumor fractions compared with validation MPNST in leave-one-out Wilcoxon rank-sum tests, expressed as −log₁₀P values. C, Fragment-length densities for cfDNA in patients with NF1 with PN, AN, and MPNST. Inset represents a magnified view between 90 and 150 base pairs. Colors represent clinical diagnosis.

Figure 2.

Fragmentomic features differentiate benign, premalignant, and malignant PNST. A, Study schema. Participants consisted of patients with imaging- and biopsy-proven MPNST or AN, established patients with PN, and healthy donors. Plasma was collected for fragmentomic analysis. AN and MPNST tissue DNA, when available, was clinically sequenced using the TSO 500 targeted oncology panel. cfDNA was extracted from plasma and underwent WGS with fragmentomic profiles assessed by fragment end motifs, bin-wise fragmentomic profiles, and NMF fragment signatures. Models for each feature type were trained on OVO comparisons with resultant features input into a logistic regression model with 10 repeats of five-fold CV. Optimal thresholds were calculated from ROC analysis of models’ predicted scores using Youden J-statistic. The results were correlated with clinical diagnoses and outcomes. B, Heatmap of fragmentomic features. Samples are grouped by diagnostic cohort (MPNST, AN, PN, and healthy) with samples in each cohort ranked from the highest to lowest size-selected CNA-derived tumor fraction. Rows are grouped by fragmentomic feature type. CV, cross-validation; OVO, one-versus-one; TSO 500, TruSight Oncology 500. (A, Created with BioRender.com.)

Figure 3.

Bin-wise fragmentomic analysis reveals distinct profiles of healthy controls, PN, AN, and MPNST. A, Heatmap of principal component eigenvalues of fragmentation profile features differentiating AN from MPNST. The relative importance of the features is represented at the right (bin-wise short/long ratio changes) and top (chromosomal arm changes) of the heatmap. Red feature importance lines indicate features or principal components associated with MPNST, whereas blue feature importance lines are associated with AN. B, Ratio of short to long fragments in 5-Mb bins across the genome in healthy volunteers and patients with PN, AN, and pretreatment MPNST, and patients receiving treatment for their MPNST. C, Bin-wise fragmentomic scoring and every 3-month surveillance MRI tumor volumes for a patient with AN diagnosed by biopsy, suspicious for malignant transformation given the rate of tumor growth. Fragmentomic scores for healthy vs. AN (green circle), PN vs. AN (purple circle), and AN vs. MPNST (orange circle) along with discrimination thresholds (horizontal black lines) from a paired blood draw are on the left y-axis. Tumor volumes are on the right y-axis, with 26% growth over 1 year indicated. D, OVO ROC AUCs of logistic regression models with 10 repeats of fivefold cross-validation performed over the bin-wise short/long ratio and chromosomal arm z-score data (See “Materials and Methods”). Healthy and disease states included within each comparison are shown along x- and y-axes, with AUC indicated both numerically and by heat level. S/L, short to long.

Figure 4.

cfDNA fragment end composition distinguishes premalignant from malignant nerve sheath tumors. A, Specific 4-mer end motifs that best classify between clinical diagnosis pairs in OVO comparisons; (P values calculated using the Benjamini–Hochberg corrected t test). B, End-motif diversity scores do not differentiate cohorts; (Kruskal–Wallis H-statistic P value > 0.05), although (C) MDS in the leave-one-out Wilcoxon rank-sum test against MPNST approach (PN or AN) or surpass (healthy) significance of P < 0.05. D, Percent contribution of NMF-deconvolved end-motif profiles (F-profiles) in each plasma cfDNA sample by clinical state (P values calculated using the Tukey post hoc test after Bonferroni-corrected ANOVA). E, ROC curves comparing AN and PN using each fragment end method: best-performing individual end motif (ACCA; AUC = 0.69), motif diversity score (AUC = 0.52), and best distinguishing F-profile (F-profile 5; AUC = 0.70). MDS, motif diversity score; *, < 0.1; **, < 0.001; ***, < 0.0001.

Figure 5.

NMF decomposition of cfDNA fragment signatures distinguishes benign, premalignant, and malignant PNST. A, Fragment-length signatures inferred from two-component NMF decomposition of healthy and MPNST cfDNA samples (See “Materials and Methods”). B, Correlation between size-selected CNA-derived tumor fraction and NMF MPNST score. Circle colors denote samples correctly classified by NMF and tumor fraction (red), NMF only (blue), tumor fraction only (yellow), or misclassified by both NMF and tumor fraction (gray). The thresholds for detecting MPNST by tumor fraction (healthy–MPNST, 0.0305, See “Materials and Methods”) and NMF (healthy–MPNST, 0.5895, See “Materials and Methods”) are denoted by horizontal and vertical dashed lines, respectively. C, ROC AUC of logistic regression following 10 repeats of fivefold cross-validation using OVO plasma cfDNA NMF deconvolution scores as input. Healthy and disease states included within each comparison are shown along x- and y-axes, with AUC indicated numerically and by heat level. TF, tumor fraction.

Figure 6.

cfDNA fragment-length NMF adjudicates diagnostic challenges and granularly classifies across healthy and disease states. A, NMF of cfDNA fragment lengths correctly classifies MPNST initially misclassified as AN by tissue biopsy. This patient had core-needle biopsy of a PET-CT avid left scapular tumor. Histopathology of the needle biopsy was consistent with a premalignant AN. Given the morbid location and nonmalignant pathology, the patient underwent a narrow-margin resection of the tumor. By NMF, both prebiopsy plasma cfDNA and preresection plasma cfDNA were consistent with MPNST (closed circle). Histopathology of the surgical resection tissue revealed foci of high-grade MPNST within interfacing AN and PN. Due to the inadequate oncologic margins, the patient underwent adjuvant radiotherapy. Plasma cfDNA during adjuvant radiotherapy showed moderately decreased NMF scores just below the MPNST detection threshold (open circle). On day 117 postresection, the patient was found to have radiographic evidence of MPNST at the edge of the radiation field consistent with locoregional recurrence. Corresponding plasma cfDNA NMF was again consistent with MPNST. Size-selected CNA-derived tumor fraction from cfDNA also identified MPNST throughout the treatment course. MPNST detection cutoffs are indicated by dashed lines with closed circles above the line indicating liquid biopsy MPNST detection. B, Comparison of size-selected CNA-based and fragmentomic cfDNA liquid biopsy methods’ ROC AUC across all nonmalignant, premalignant, and malignant disease states. The inset legend indicates the utilized method. Data shown are 10 repeat fivefold cross-validated. L, left; TF, tumor fraction; XRT, radiotherapy.