FLT3-ITD Measurable Residual Disease Monitoring in Acute Myeloid Leukemia Using Next-Generation Sequencing

Simple Summary FLT3-ITD monitoring is essential in AML management, and there is a great need for sensitive monitoring methods. We present a simple and easily applicable ITD-tracing algorithm optimized for MRD monitoring based on the NGS method. Our assay is sensitive to 0.001% and has a superior performance over conventional fragment analysis. For AML patients undergoing allo-HSCT, our assay showed that the MRD assessed before and after HSCT were significantly associated with a risk of relapse and a poor overall survival, respectively. This report highlighted the prognostic value of serial MRD monitoring using a sensitive method in a clinical setting of AML patients with FLT3-ITD. Abstract The in-frame internal tandem duplication (ITD) of the FMS-like tyrosine kinase 3 (FLT3) gene is an important negative prognostic marker in acute myeloid leukemia (AML). FLT3-ITD monitoring is essential for patients at relapse or those receiving FLT3-targeted therapies. Fragment analysis (FA) is commonly used to detect and quantify FLT3-ITDs; however, detecting low-burden FLT3-ITDs after a treatment is challenging. We, therefore, developed a customized, next-generation sequencing (NGS)-based FLT3-ITD assay that includes a new ITD-tracing algorithm, “SEED”, optimized for measurable residual disease (MRD) monitoring. NGS-SEED showed an enhanced sensitivity (0.001%) and has a superior performance over conventional fragment analysis. We further investigated the prognostic impact of MRD analyzed by NGS-SEED in AML patients who underwent allogeneic hematopoietic stem cell transplantation (HSCT). Our assay showed that the MRD assessed before and after HSCT were significantly associated with a risk of relapse and a poor overall survival, respectively, in a time-dependent analysis. Thus, this report highlighted the prognostic value of serial MRD monitoring using a sensitive method in a clinical setting of AML patients with FLT3-ITD.


Introduction
The in-frame internal tandem duplications (ITDs) of the gene encoding FMS-like tyrosine kinase 3 (FLT3) have been found in approximately 25% of acute myeloid leukemia (AML) cases [1]. Approximately 70% of these duplications are located in the juxtamembrane domain and consistently increase the FLT3 tyrosine kinase activity. FLT3-ITDs with allelic ratios ≥ 0.5 have been associated with an adverse prognosis [2,3]. In addition, FLT3-ITD monitoring is essential for patients at relapse and for those receiving FLT3-targeted therapy [4][5][6]. Fragment analysis (FA) using capillary electrophoresis has been the standard for detecting and quantifying FLT3-ITDs [7]. This assay has a high clinical utility because it provides reliable, rapid results. However, the detection of lower-burden FLT3-ITDs (<2%) after induction chemotherapy or hematopoietic stem cell transplantation (HSCT) is hampered by PCR bias [7][8][9].
Significant progress has been made in applying next-generation sequencing (NGS) to the detection and monitoring of FLT3-ITD [10][11][12][13][14], However, the detection of FLT3-ITD by NGS has been hindered by its unstable nature and complexity of the FLT3-ITD. The precise quantification of ITD burdens by NGS remains challenging because of the inherent biases and limitations of the standard clinical NGS pipelines. Multiple bioinformatics tools have been developed, but their performance is suboptimal for use without an adjustment [15,16].
To address these issues, we have developed a customized NGS-based FLT3-ITD assay optimized for the monitoring of measurable residual disease (MRD) with an enhanced sensitivity and specificity. We carefully evaluated the analytical performance of the NGSbased FLT3-ITD assay and applied it to track residual disease using clinical samples under morphological remission pre-and post-HSCT. We further described the clinical impact of the monitoring of MRD using the NGS-based FLT3-ITD assay.

Samples for Performance Evaluation
This study included clinical samples from 46 FLT3-ITD positive acute leukemia patients (45 with AML and one with T/myeloid mixed phenotype acute leukemia) who were diagnosed and received allo-HSCT at the Catholic Hematology Hospital from April 2013 to July 2020 (Table 1). A total of 211 BM samples were selected for the NGS-based FLT3-ITD assay, including those at diagnosis (n = 46), pre-HSCT (n = 79), post-HSCT at 1 month (n = 37), 3 months (n = 24), and 6 months (n = 25). Negative control samples were prepared from 10 patients with AML who were FLT3-ITD negative by FA (Supplementary Table S1). To evaluate the analytical performance of the NGS-based FLT3-ITD assay, we used 30 bp FLT3-ITD containing reference DNA (IVS-0017 Polyclonal Control DNA, InvivoScribe, Technologies, San Diego, CA), five clinical samples containing various sizes of ITDs (39, 78,  108, 156, and 206 bp, Supplementary Table S2), and their products (n = 74) serially diluted using FLT3 wild-type DNA (IVS-0000 Clonal Control DNA, InvivoScribe, Technologies, San Diego, CA, USA).

Patients
Of the 46 patients with acute leukemia and FLT3-ITD, 42 received intensive chemotherapy to induce remission. Thirty-four patients (73.9%) were treated according to the standard protocol, which consisted of 3 + 7 idarubicin (IDA) and cytarabine (Ara-C). Seven patients (15.2%) were treated with 3 + 7 daunorubicin plus Ara-C. One patient with acute promyelocytic leukemia received all-trans retinoic acid (ATRA) plus IDA. Decitabine was used as an initial treatment regimen in two patients of an advanced age (67 and 68 years). Low-dose Ara-C plus etoposide was administered to one patient (64 years) due to infectious complications. One older patient (71 years) with multiple comorbidities participated in the ASP2215 study to receive gilteritinib as an induction regimen. Eleven (23.9%) adult patients received FLT3 inhibitors as salvage chemotherapy (n = 5, quizartinib via expanded access program; n = 5, azacitidine + sorafenib; n = 1 gilteritinib) or induction chemotherapy (n = 1 gilteritinib via ASP2215 study). Two pediatric patients were treated with sorafenib in combination with IDA/Ara-C at relapse after the HSCT (Supplementary Figure S1). Nine patients were enrolled in randomized, placebo-controlled clinical trials for the FLT3 inhibitors; therefore, they were excluded from a further survival analysis in this specific variable. This study was carried out in accordance with the Declaration of Helsinki and was approved by the Institutional Review Board (IRB) of Seoul St. Mary's Hospital (KC22RISI0206).

NGS-Based FLT3-ITD Assay
We developed an NGS-based FLT3-ITD assay to detect FLT3-ITD and optimized the bioinformatics pipeline to monitor MRD. The primer set covering FLT3 exons 14-15 (forward-chr13:28,608,315-28,608,296, reverse-chr13:28,608,129-28,608,108) was designed to produce a 208 bp unique amplicon. The library was prepared using at least 50 ng of gDNA, which was measured using a Nanodrop spectrophotometer (Nanodrop, Wilmington, NC, USA). Paired-end sequencing reads (150 bp) were generated using a NextSeq sequencer (Illumina, San Diego, CA, USA). The 300 bp-sized long paired reads were sequenced using the MiSeq platform (Illumina, San Diego, CA, USA) to analyze longer ITDs 17. Negative results were confirmed using 400 ng of gDNA. The average depth of coverage was 829,568 per sample, and the minimum depth was 157,123.

SEED Algorithm for MRD of FLT3-ITD
Most ITDs typically have a short nucleotide sequence, termed a spacer, between the duplication and its origin [17]. Therefore, the entire insertion segment is composed of a spacer and duplication. We assigned a unique SEED sequence for each ITD, encompassing the junctional region of insertion and its origin, based on the structural features. The entire procedure of the sequencing data, including the alignment, recalibration, and SEED algorithm, is illustrated in Supplementary Figure S2. We hypothesized that the SEED sequence would reflect the true allelic burden more precisely than the full ITD sequence. The size of the SEED sequence was set to 12-20 bp to optimize the detection sensitivity and specificity. To designate the SEED sequence, a Binary Alignment Map (BAM) file was visualized in the Integrative Genomics View (IGV) for the inspection of the ITD. Insertion or soft-clipped bases that showed >3 reads of identical sequences out of >10,000 reads were considered to be true ITDs. The variant allele frequency (VAF) values were calculated as the proportion of SEED-containing reads to the total read count. The command script that utilizes SAMtools to count the SEED reads from a BAM file is as follows: > samtools view [bam.file] | grep GAAATCAACGGGAAACTCC | wc -l The results were compared with those from FA and those analyzed by open-source tools for the detection of FLT3-ITD, including ITDseek (version 1.2, https://github.com/ tommyau/itdseek (accessed on 26 August 2020)) [18], getITD (version 1.5.13, https:// github.com/tjblaette/getitd (accessed on 26 April 2021)) [19], and Pindel (version 0.2.5b8, https://github.com/genome/pindel (accessed on 26 August 2020)) [20].

Statistical Analysis
The detection limit was determined to be 95% probit using probit regression analysis. The MRD positivity of FLT3-ITD was only determined for the samples at morphological remission (BM blasts < 5%). When multiple ITDs were detected in one sample, the sum of the ITDs was used. If multiple follow-up tests were performed at the same time point, the highest VAF value was selected to determine the MRD level. The MRD status at remission, combined with the clinical data, was examined for the prognostic value. We analyzed the predictive power of clinical and mutational factors for relapse, non-relapse mortality (NRM), event-free survival (EFS), and overall survival (OS). We defined EFS as the time from the HSCT to relapse or death, and OS as the time from the HSCT to death from any cause. The medical records were tracked until September 2021. The EFS and OS were estimated using the Kaplan-Meier method. A competing risk analysis was performed to estimate the probability of a cumulative incidence of relapse (CIR) and NRM. The CIR was compared across groups using the Gray test and cmprsk module in R [21,22]. The EFS and OS were compared using the Cox proportional hazards regression. The post-HSCT MRD and GVHD status at different time points were evaluated as the time-dependent covariates. Receiver-Operating Characteristic (ROC) analysis was performed to determine the optimal cut-off values for predicting relapse, ESF, and OS. For multivariate analysis, adverse variables with a p-value of <0.05 in the preceding univariate analysis were entered into a proportional hazard model for the sub-distribution of a competing risk or Cox proportional hazard model. Then, we fitted a joint model of longitudinal and time-to-event data to account for the informative dropout of the MRD measurements at 4 longitudinal time points (pre-HSCT, post-HSCT at 1 month, 3 months, and 6 months) due to relapse using the R package JM [23,24]. Statistical analyses were performed using MedCalc version 19.1.7 (MedCalc Software, Ostend, Belgium), SAS version 9.4 (SAS Institute Inc., Cary, NC, USA), and R software version 4.1.3 (R Foundation for Statistical Computing, Vienna, Austria).
We then compared the results from the NGS-SEED with those from the FA for 211 clinical samples. The NGS-SEED detected 126 FLT3-ITDs in 114 samples, while the FA detected only 60 ITDs in 55 samples ( Figure 1C). Fifty-three types of ITD were identified, ranging from 18 to 206 bp (median 48 bp), positioned between chr13:28,608,102 and chr13:28,608,313 (Supplementary Table S3). Seven patients had two ITDs, one of which emerged during the follow-up and was not detected by the FA due to its very low burden (VAF, 0.014%). We found another interesting case with two different ITDs that were not distinguished by the FA owing to their similar sizes (24 and 27 bp). The quantitative detection of FLT3-ITD correlated well between the NGS-SEED and FA, with a coefficient of 0.92 (p < 0.001, 95% CI: 0.90-0.94, Figure 1D). The VAFs from the NGS-SEED were lower than those from the FA (p < 0.001, Figure 1E).

FLT3-ITD Assessment after Treatment
The

FLT3-ITD Assessment after Treatment
The with a median VAF of 0.375%, 0.041%, and 0.009%, respectively. The post-HSCT MRD was identified by the NGS-SEED in 6 (16.7%) with a median VAF of 0.19% at 1 month, and post-HSCT at 3 months and 6 months. MRD was detected in three (15.8%) and four (33.3%) patients with median VAF values of 0.01% and 0.15%, respectively. None of these were detected by the FA (<2%) (Figure 2).
In the ROC curve analysis, pre-HSCT MRD > 0.1%, and post-HSCT MRD > 0.001% showed the highest area under the curve (AUC) for predicting relapse (AUC = 0.730, 0.747, respectively), and post-HSCT MRD > 0.001% and > 0.1% were the optimal classifiers for EFS and OS, respectively (AUC = 0.661, 0.672, respectively) ( Supplementary Table S5). Then, we performed a multivariate analysis using the optimal MRD factors and potential confounders. As shown in Table 3, pre-HSCT MRD > 0.1% was significantly associated with relapse, while post-HSCT MRD > 0.1% was the only significant predictor for the OS.

Discussion
MRD is one of the most powerful independent prognostic markers for AML [13,[25][26][27]. With the emergence of targeted agents, the role of the monitoring of MRD in the management of AML is growing. The recent guidelines recommend the use of MRD with a detection limit lower than 0.1% [13,26,28]. For FLT3-ITD, the current recommended standard method for detection and quantification is FA, but its sensitivity is challenging, at approximately 2% [9]. Recently, multiple NGS-based assays for the assessment of FLT3-ITD have been introduced, but they still require an adjustment for a clinical implementation owing to their limited analytical performance [15][16][17]29].
In the current study, we developed a novel assay using the SEED sequence to analyze the FLT3-ITD MRD using the NGS data. In our assay, the initial ITD was searched using the IGV window, and thereafter, a specific SEED sequence to the ITD was assigned and further queried from the entire sequencing data. This strategy improved the analytical accuracy by preserving the mutated SEED sequence reads, even if there were polymorphisms and/or random sequencing errors.
We have shown that the NGS-SEED detects all types of ITDs, including long ITDs, when other assays frequently misidentify or misread their positions. In our analysis of 211 clinical samples, low-burden mutations were additionally detected using NGS-SEED in 59 FA-negative samples. The NGS-SEED showed an excellent linearity across ITDs of various sizes and achieved a detection limit of approximately 0.001%. Tracing the ITDspecific sequence of SEED was also advantageous for distinguishing separate clones with similar ITD sizes. Moreover, the NGS-SEED algorithm can be applied to other FLT3-ITD NGS datasets using other primer sets.
Using technical accomplishments, we further investigated the prognostic impact of MRD according to various cut-offs and time points. The NGS-SEED had a significant impact in both pre-and post-HSCT. We monitored the MRD pre-HSCT and post-HSCT at 1, 3, and 6 months. Eighty percent of our patients harbored persistent pre-HSCT MRD, and 86% of them were detected by NGS-SEED. It appeared that the best pre-HSCT MRD cut-off to predict relapse was 0.1%, with 37% of patients showing an MRD above it. The significance of pre-HSCT FLT3-ITD MRD as a risk factor for relapse have been shown by other groups using NGS methods. Hourigan CS et al. found that 9 out of 10 patients with FLT3-ITD MRD with VAF 0.03-0.97% relapsed after HSCT, while 1 out of 7 patients with undetectable MRD relapsed [30]. The recent study by Loo et al. described the prognostic impact of FLT3-ITD MRD at a 0.001% cut-off, which is much lower than our result [31]. The optimal NGS-MRD cut-off for the relapse prediction has not yet been defined, so a further investigation is required.
After allo-HSCT, the FLT3-ITD MRD was decreased to a measurable level by NGS alone. As a time-dependent covariate, post-HSCT MRD > 0.1% was significantly associated with poor OS in the multivariate analysis. In addition, patients with post-HSCT MRD had a significantly higher CIR rate. In the subsequent analysis using the joint model of longitudinal and time-to-event data, we could validate the prognostic impact of the longitudinal MRD value, which has been rarely described before [32].
These results collectively indicate that NGS-based FLT3-ITD MRD assays should be considered for serial use in pre-and post-HSCT [33]. Considering that the sensitivity of NGS-based FLT3-ITD assays in this and previous studies was nearly 0.001-0.005% [19,34], it is relevant to obtain an accurate MRD detection near the significant cut-off.
This study had several limitations. Given the retrospective nature of our analysis, we could not control for the considerable heterogeneity in patient characteristics, disease status, and type of treatment. In addition, the statistical power of this study was limited by a small sample size. Therefore, our conclusions should be further confirmed in a lager prospective cohort. The manual ITD searching process of NGS-SEED requires further improvement before it can be widely applied. A dedicated user interphase and software is now underdevelopment to enhance the reproducibility and shorten the turn-around time of this method. Despite these limitations, this study clarified the significance of pre-HSCT FLT3-ITD MRD as a risk factor for relapse. In addition, post-HSCT MRD is a strong risk factor for a poor survival.

Conclusions
In this study, we demonstrated the prognostic impact of the sensitive FLT3-ITD MRD using SEED-NGS. Then, we carefully proposed the optimal cut-off level for FLT3-ITD MRD in patients with AML who underwent allo-HSCT according to various time points. This report also highlighted the prognostic value of serial MRD monitoring using a sensitive method in a clinical setting of AML patients with FLT3-ITD. Further studies incorporating targeted therapies in large cohorts of patients may potentiate the significance of FLT3-ITD MRD and broaden its application.

Institutional Review Board Statement:
This study was carried out in accordance with the Declaration of Helsinki and was approved by the Institutional Review Board (IRB) of Seoul St. Mary's Hospital (KC22RISI0206) for studies involving humans.
Informed Consent Statement: Patient consent was waived because of the retrospective nature of the study and the anonymous analysis.
Data Availability Statement: The datasets generated during and/or analyzed during the current study are available in the NCBI SRA repository (BioProject ID:PRJNA880286).