Abstract
INTRODUCTION: Myelodysplastic Syndromes (MDS) are caused by progressive clonal dominance of mutated hematopoietic stem cells. While assessment of mutations in peripheral blood (PB) is an established standard for monitoring patients with some diseases such as chronic myeloid leukemia, the sensitivity and precision of PB screening and monitoring for somatic mutations in patients with MDS and related conditions such as clonal cytopenias of undetermined significance (CCUS) is less certain. Multiple published guidelines require bone marrow (BM) evaluations for the above, an invasive procedure not often performed outside of clinical trials. We compared somatic mutations and variant allele frequencies (VAFs) in paired PB and BM samples to assess the utility of PB results as a surrogate for diagnosis and monitoring. METHODS: The National MDS Natural History Study (National MDS Study) is an observational longitudinal cohort study of cytopenic patients suspected of having a diagnosis of MDS. All patients have PB and BM sampled when entering the study, prior to formal diagnosis. For this study, paired PB and BM baseline DNA samples from 36 patients were assessed for somatic mutations. Only subjects with minimum of one variant detected in their BM were included in this study. DNA was extracted from PB using an automated FlexSTAR (AutoGen) and from BM cell pellets using a QIAamp DNA Mini Kit (QIAGEN). Targeted exon sequencing of 96 genes was performed using a NovaSeq 6000 at a mean coverage of >1,200X and a mean breadth (bases covered at ≥100X) of >99.9%. Reads were aligned against a patched version of the build GRCh38 using BWA-MEM. VarScan2 was used to detect single nucleotide variants and short insertions/deletions with a minimum VAF of 2% and 5%, respectively. Resulting variants from 53 genes were manually reviewed to retain likely disease-causing variants. Correlation analysis between PB and BM VAFs was performed using Pearson correlation and linear regression. To confirm paired samples were sourced from the same subject, the Somalier software using a set of highly polymorphic SNPs across different ancestries was utilized to ensure a high level of relatedness. RESULTS: The 36 patients included 10 (28%) with MDS, 2 (6%) with MDS/MPN, 1 (3%) with AML (<30% blasts), and 23 (64%) with CCUS. The median age was 73 years (Table 1). Five of 36 patients had circulating PB blasts. Among the 191 mutations identified, the most common were TET2 (27%), SRSF2 (13%), DNMT3A (9%), SF3B1 (8%), and ASXL1 (8%). Correlation analysis of mutations (n=180) shared between paired PB and BM samples indicated a linear relationship between VAFs detected in PB and BM (Pearson correlation r=0.95, Figure 1), with a slight increase in variance observed with increasing VAFs. The deviation from the perfect (grey line) to the observed (yellow line) linear fit indicated that PB VAFs underestimated VAFs in BM by approximately 23% overall (Figure 1). For example, a BM VAF of 0.20 translated to a mean PB VAF of 0.16 based on the regression fit. The lower sensitivity was confirmed when looking at the number of total missed variants (present in BM, absent in PB): 9 of 101 (9%). Two mutations were detected in PB, but not BM. These mutations and diagnoses were 1) ETNK1 (VAF 0.026) in a patient with MDS-EB-1; and 2) CBL (VAF 0.021) in a patient with MDS with isolated del5q. Next, we assessed the relationship between percent of detected mutations and minimum VAF that was detected using linear regression and found that the higher the minimum VAF, the lower the percent of missed mutations. For example, for a 0.1, 0.2, and 0.3 minimum VAF, 11%, 9%, and 8% of BM mutations were missed, respectively. There was no difference in concordance between MDS and CCUS. SUMMARY & CONCLUSIONS: PB can be used to reliably identify somatic mutations in patients with suspected or established MDS and related conditions. Linear correlation between PB and BM VAFs showed, unsurprisingly, that BM mutations with higher VAFs are more likely to be detected. If PB is positive, this method of variant detection may be used potentially to diagnose and monitor. However, if negative (or if VAF is less than 0.10), bone marrow genomics are still necessary to exclude detectable clonal mutations. For quantitative predictive tools that use VAFs, PB VAFs would need to be corrected to improve predictive results; our estimated linear regression equation could be used to guide this in a larger prospective study.
