Limitations for use of NGS to detect residual disease in AML
| Problem . | Significance . | Solution . |
|---|---|---|
| NGS AML panels used at diagnosis are unfit for AML MRD | Insufficient sample input/sequencing depth limits assay sensitivity. Insufficient error correction increases false-positive rate at low VAF. | NGS currently suited for research but not clinical use in AML MRD |
| ELN guidelines on use of NGS in AML MRD coming | ||
| Association between variant tracked and residual AML clone(s) | Variants closely linked to AML (eg, FLT3-ITD) often subclonal/unstable. “Stable” variants, such as those in DMNT3A, also seen in ARCH/CHIP. | Tracking a panel of gene regions in remission will likely improve predictive power, at increased cost. |
| Larger datasets will provide more information on which variants detected at MRD stage are the most associated with subsequent relapse. | ||
| Genetic clonal heterogeneity of AML = risk of false-negative tests | Variants detected at diagnosis are not necessarily present in the AML clone remaining after unsuccessful treatment responsible for relapse. | Tracking a panel of gene regions in remission will likely improve predictive power, at increased cost. |
| Deep profiling of diagnostic sample to screen for minor subclones (eg:, TP53) may have utility. | ||
| Error rates intrinsic to NGS = risk of false-positive tests | Traditional NGS approaches cannot reliably identify novel variants present at low VAF (<2-5%) with sufficient specificity (ie, many false positive variant calls mask rare true positive variant). | ECS using UMI consensus clustering and/or bioinformatic approaches, such as background error models (Figure 2D), are helpful. Low VAF variants seen in diagnostic sample or multiple surveillance samples more likely true variants. |
| Correlation of NGS results with other measures of AML MRD | Tests designed to detect residual disease in AML may classify patient sample differently based on modality used (qPCR vs flow cytometry vs NGS). No single test represents a “gold standard” for detecting, in patients in remission, those cells that will subsequently lead to AML relapse. | Studies designed to integrate information from different AML MRD tests performed on the same sample cohorts are underway. |
| Lack of uniform reporting standards | How many UMI reads needed to call a variant (3, 5)? How many distinct UMI read families per variant needed to call MRD? How many genomic equivalents as input? Standardized filtering, consensus clustering, and variant calling needed? What about controls, duplicates, platforms? | NGS currently suited for research but not clinical use in AML MRD |
| FDA guidance for MRD in hematological malignancies published in draft form. | ||
| ELN guidelines on use of NGS in AML MRD coming |
| Problem . | Significance . | Solution . |
|---|---|---|
| NGS AML panels used at diagnosis are unfit for AML MRD | Insufficient sample input/sequencing depth limits assay sensitivity. Insufficient error correction increases false-positive rate at low VAF. | NGS currently suited for research but not clinical use in AML MRD |
| ELN guidelines on use of NGS in AML MRD coming | ||
| Association between variant tracked and residual AML clone(s) | Variants closely linked to AML (eg, FLT3-ITD) often subclonal/unstable. “Stable” variants, such as those in DMNT3A, also seen in ARCH/CHIP. | Tracking a panel of gene regions in remission will likely improve predictive power, at increased cost. |
| Larger datasets will provide more information on which variants detected at MRD stage are the most associated with subsequent relapse. | ||
| Genetic clonal heterogeneity of AML = risk of false-negative tests | Variants detected at diagnosis are not necessarily present in the AML clone remaining after unsuccessful treatment responsible for relapse. | Tracking a panel of gene regions in remission will likely improve predictive power, at increased cost. |
| Deep profiling of diagnostic sample to screen for minor subclones (eg:, TP53) may have utility. | ||
| Error rates intrinsic to NGS = risk of false-positive tests | Traditional NGS approaches cannot reliably identify novel variants present at low VAF (<2-5%) with sufficient specificity (ie, many false positive variant calls mask rare true positive variant). | ECS using UMI consensus clustering and/or bioinformatic approaches, such as background error models (Figure 2D), are helpful. Low VAF variants seen in diagnostic sample or multiple surveillance samples more likely true variants. |
| Correlation of NGS results with other measures of AML MRD | Tests designed to detect residual disease in AML may classify patient sample differently based on modality used (qPCR vs flow cytometry vs NGS). No single test represents a “gold standard” for detecting, in patients in remission, those cells that will subsequently lead to AML relapse. | Studies designed to integrate information from different AML MRD tests performed on the same sample cohorts are underway. |
| Lack of uniform reporting standards | How many UMI reads needed to call a variant (3, 5)? How many distinct UMI read families per variant needed to call MRD? How many genomic equivalents as input? Standardized filtering, consensus clustering, and variant calling needed? What about controls, duplicates, platforms? | NGS currently suited for research but not clinical use in AML MRD |
| FDA guidance for MRD in hematological malignancies published in draft form. | ||
| ELN guidelines on use of NGS in AML MRD coming |
ARCH, age-related clonal hematopoiesis; CHIP, clonal hematopoiesis of indeterminate potential (an alternative term for ARCH); ECS, error-corrected sequencing.