Panel Based Error Corrected Next Generation Sequencing versus Flow Cytometry to Detect Measurable Residual Disease in Acute Myeloid Leukemia

A 35 gene error corrected next generation sequencing (NGS) panel was created using single molecule molecular inversion probes with applicability to 83% of acute myeloid leukemia (AML). We accrued 201 patients of adult AML treated with conventional therapy, in morphological remission and evaluated measurable residual disease using NGS (NGS-MRD) as well as multiparameter flow cytometry (FCM-MRD) at post induction (PI) and consolidation (PC) time points. A total of 344 mutations were detected [median VAF of 0.95% (0.76% after excluding mutations in DNMT3A, TET2, ASXL1)] during assessment of MRD. Nearly 71% of patients harbored PI NGS-MRD (and 40.9% harbored PC-MRD). Patients harboring NGS-MRD had a significantly higher cumulative incidence of relapse (CIR), inferior overall survival (OS) and relapse free survival (RFS) as compared to NGS-MRD negative patients at PI and PC time points. NGS-MRD was predictive of inferior outcome in intermediate cytogenetic risk and demonstrated potential in favorable cytogenetic risk AML. Patients who cleared PI NGS-MRD (and stayed negative) had a significantly improved survival as compared to patients who became negative subsequently indicating that kinetics of NGS-MRD clearance was of paramount importance. NGS-MRD identified over 80% of cases identified by flow cytometry at PI time point whereas FCM identified 49.3% identified by NGS. Both FCM and NGS MRD were important in predicting outcome however, PI NGS-MRD emerged as the most important independent prognostic factor predictive of inferior outcome. We demonstrate that panel based NGS-MRD is highly predictive of outcome and advantageous when compared to FCM-MRD in AML treated with conventional therapies.


Introduction
Acute Myeloid Leukemia is a disease characterized by heterogeneous biology resulting in varying clinical outcomes including relapse. 1,2 There are limited novel treatment options such as targeted therapies using FLT3 or IDH2 inhibitors and most patients are treated based on the presumptive risk of relapse. 3,4 This risk adapted therapy typically considers standard cytogenetic and molecular variables as recommended by the European LeukemiaNet. 5 [6][7][8][9][10][11][12][13][14] Although universally applicable, FCM-MRD suffers from suboptimal ability to predict relapse in AML compared to precursor B lineage acute lymphoblastic leukemia. A diverse array of sensitive molecular methods have been used to detect MRD in AML such as real time PCR 15 and droplet digital PCR. 16 These are useful for monitoring of individual gene mutations such as AML with mutated NPM1 17 and chimeric gene fusions such as RUNX1-RUNX1T1. 18 Next generation sequencing (NGS) is a promising tool for sensitive MRD monitoring and has been used successfully to monitor NPM1 19,20 , RUNX1 21 and FLT3 22 mutations as well as chimeric gene fusions. 23,24 However, these methodologies can be utilized in only a subset of patients that harboured these mutations at diagnosis and this strategy discounts for clonal heterogeneity and evolution which are relevant to the pathogenesis of AML.
DNA based focussed target enrichment strategies (gene panels) are an attractive solution to detect MRD using NGS (NGS-MRD) in a mutation rich disease such as AML. 25,[26][27][28] However, short read sequencers are inherently prone to base calling errors limiting variant calling at 3-5% variant allele . CC-BY 4.0 International license It is made available under a perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint fraction (VAF). 29 Although acceptable for diagnostic molecular pathology, this is undesirable assay performance for the detection of MRD. Error corrected sequencing involves the physical incorporation of random oligonucleotides or unique molecular identifiers (UMI) at the library preparation stage prior to amplification of DNA. This allows us to tag individual DNA molecules with an unique molecular fingerprint. 30,31 Such approaches have been used for myelodysplastic syndromes 32 and for pre transplant MRD monitoring of AML as demonstrated by Thol and colleagues. 33 Thol utilized a sensitive patient specific mutation tracking approach using UMI based MRD detection. Although applicable to a broad spectrum of AML mutations, a tailored approach poses logistical and regulatory hurdles towards prospective MRD testing especially for early MRD timepoints.
In this study, we have evaluated the clinical utility of error corrected NGS to detect MRD in AML using single molecule molecular inversion probes (smMIPS). 30,34 Each smMIP contains an 8 bp UMI and binds to a single molecule of DNA. Using consensus sequence-based variant calling we can detect somatic mutations including small indels in a sensitive manner. We demonstrate that error corrected NGS-MRD at early timepoints in therapy is significantly predictive of outcome in patients of AML treated with conventional therapies. Furthermore, we systematically compare multicolour FCM-MRD with error corrected NGS-MRD and assess the clinical utility of these two assays in a cohort of AML.
. CC-BY 4.0 International license It is made available under a perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint  9,20 Somatic mutations at diagnosis were evaluated using a smMIPS based 50 gene myeloid panel as described previously by our group. 35 Of these, the smMIPS MRD panel (see below) was applicable to 83.2% patients  is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted August 25, 2020. . https://doi.org/10.1101/2020.08.23.20180372 doi: medRxiv preprint compared using log-rank test. Multivariate analysis was performed using the Cox proportional-hazards regression analysis that considered FCM-MRD and NGS-MRD.
Separate models were constructed for post induction and post consolidation MRD time points. Grey test was used to compare the cumulative incidences of relapse (CIR) and nonrelapse mortality (NRM) using "cmprsk" module in R. 36 The same module was used to generate representative graphs. Positive predictive value (PPV) and negative predictive value (NPV) were calculated as described. b. Bioinformatics: Reads were demultiplexed, trimmed, paired end assembled and mapped to the human genome (build hg19). Singleton reads (originating from one UMI) were discarded, and consensus family based variant calling performed using tools described in supplementary methods. We then created a site and mutation specific error model to ascertain the relevance of an observed variant at each site. 34 Criteria for variant calling using the smMIPS MRD assay are described in supplementary methods. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted August 25, 2020. . https://doi.org/10.1101/2020.08.23.20180372 doi: medRxiv preprint c. MRD for FLT3-internal tandem duplications (ITD): FLT3-ITD were detected using a novel one-step PCR based NGS assay (see Supplementary Table 3). Variants were detected using a recently described algorithm for the accurate detection of FLT3-ITD. 37 d. Orthogonal detection of MRD in NPM1 mutated AML: NPM1 mutations were additionally tracked using an ultrasensitive orthogonal NPM1 MRD assay. 20

DETECTION OF MRD IN USING MULTICOLOUR FLOW CYTOMETRY (FCM-MRD)
FCM-MRD was detected using a previously described 10 colour two tube MRD assay. 9,20,35,38 This approach uses a combination of leukemia associated immunophenotype and difference from normal approaches to detect MRD in AML.
. CC-BY 4.0 International license It is made available under a perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

PATIENTS:
The median follow-up of the cohort was 42.3 months. The median OS was 35.9 months (95%CI-27.2 to 42.8) for the entire cohort. The median RFS was 21.6 months (95%CI-17.0 to 28.9) months. Additional patient characteristics can be seen in Table 1. 2. NGS BASED AML MRD: a. Assay Characteristics: We describe an NGS-MRD approach that is applicable to 83.2% of all AML. For patients in morphological CR (n=319) this approach was applicable to 83.4% (n=266). Of these, MRD could be assessed in 201 cases. A co-occurrence plot indicating interactions of mutations tracked by NGS-MRD, prior to therapy, can be seen in Figure 1A is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted August 25, 2020. . Nearly 71% (n=139; 70.9%) of patients harboured MRD at the end of induction and 40.9% (n=52) at the end of consolidation. Patients harbouring MRD had a significantly higher CIR as compared to MRD negative patients at PI time point as seen in Figure 2A, Table 2.
For patients who were PC NGS-MRD positive a similar observation for CIR was made ( Figure 2B, Table 2). The presence of NGS-MRD at the end of induction was associated with inferior OS and RFS ( Figure 2C,E) as detailed in Table 3 is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted August 25, 2020. . figure 11). There was no genetic difference observed between these two groups (supplementary figure 12).

d. Orthogonal detection of NGS-MRD in NPM1 mutated AML:
Orthogonal comparison of MRD detection in NPM1 mutated AML was performed in 75 patients (23.2% of all MRD samples; Supplementary Methods, supplementary figure 13).

FCM BASED AML-MRD:
The presence of FCM-MRD was associated with inferior OS, RFS and CIR at the end of induction and consolidation as detailed in Tables 2, 3  is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted August 25, 2020. . https://doi.org/10.1101/2020.08.23.20180372 doi: medRxiv preprint 13) had relapsed, three died due to non-relapse causes and four were alive at last follow up (supplementary table 5

MULTIVARIATE ANALYSIS:
Both FCM and NGS MRD were important in predicting outcome as seen in Table 3 is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted August 25, 2020. Previously Jongen-Lavrencic have demonstrated clinical utility of NGS to detect MRD in AML by using computational error correction to mitigate sequencing errors. 26 Such an approach although easy to implement discounts for batch effects and variability that occurs because of library clustering and batch dependent PCR artefacts. 31 In that context, to the best of our knowledge, this is the first study to determine the clinical importance of (error corrected, panel based) NGS-MRD in AML treated with conventional therapies. Although our NGS-MRD strategy works in a majority of AML, we were curious about the genetic basis of cases (n=65 out of 393) in which this strategy did not work. The is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted August 25, 2020. . https://doi.org/10.1101/2020.08.23.20180372 doi: medRxiv preprint cytogenetic and mutational landscape can be seen in supplementary figure 16. Nearly half of these patients did not harbour any mutation at diagnosis (n=29; 44.6%). Insight into rest of the cases revealed ASXL2 as a recurrently mutated gene (n=8,12.3%). Incorporation of ASXL2 in future iterations of our panel will as well as other UMI based RNA sequencing approaches 23,24 to monitor chimeric gene fusions will help in increasing the breadth of our approach.
Consistent with previous reports, we find that in some patients, mutations in DTA genes are present at high VAF at MRD time points (Figure 1) indicating an origin from an ancestral clone possibly originating from clonal haematopoesis. 26,33,39 We are in agreement with Thol and colleagues who indicated that mutations in TET2 could reliably be used as MRD markers, at least in a fraction of cases. 33 Unlike amplicon based 27,33 approaches, the advantage of a smMIPS based capture includes a stable panel which can be used across a spectrum of cases and no susceptibility to allelic skew or PCR induced errors prior to incorporation of the UMI barcode. Disadvantages of smMIPs include poor performance for GC rich genes such as CEBPA gene and inability to capture low yield or poor quality of DNA (a problem not infrequently seen with MRD samples). The library preparation process, is relatively low cost in nature and the overall process has a realistic, turnaround time of five to seven days. Our observation is that sensitivity in the clinic for most mutations is close to 0.1% VAF. A lower can be obtained for complex indels such as NPM1 and FLT3-ITD. Improvements with sensitivity may be possible through duplex sequencing based methods albeit at a much higher cost of sequencing. 29 Lastly, based on this data we find that mutations in NPM1, FLT3, NRAS, KIT, IDH1, IDH2, WT1,  is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted August 25, 2020. . https://doi.org/10.1101/2020.08.23.20180372 doi: medRxiv preprint . CC-BY 4.0 International license It is made available under a perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted August 25, 2020. . https://doi.org/10.1101/2020.08.23.20180372 doi: medRxiv preprint   Figure 1A: The interaction of mutations at baseline is demonstrated here using Fisher's Exact test. Co-occurrence is indicated in grey colour and mutual exclusivity is indicated in red. Figure 1B: The total number of mutations detected per patient and the number of such patients in the cohort is displayed. The total number of mutations in DNMT3A-TET2-ASXL1 genes is indicated here as a fraction. Figure 1C   is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted August 25, 2020. .