Results

Pretreatment genetic characteristics

Sequencing of diagnostic (pre-induction) samples showed that high-risk genetic characteristics were common in this older cohort. Somatic mutations that indicate evolution from antecedent MDS, which are associated with poor outcome,^[4] were present in 127 of the 295 patients (43.1%); TP53 mutations and FLT3-internal tandem duplications (FLT3-ITDs) without concurrent NPM1 mutations, both of which reflect adverse molecular risk, were each present in 33 (11.2%); and NPM1 mutations without concurrent FLT3-ITDs, which are associated with favorable transplantation outcomes in younger patients, were present in 36 (12.2%; Figure 1).^[4,12,16] FLT3-ITDs with concurrent NPM1 mutations were present in 29 patients (9.8%). Tyrosine kinase inhibitors were used after transplantation in 22 patients with FLT3-ITDs.

Enlarge

Figure 1. Mutations present at the time of diagnosis. Shown are all mutations present at the time of AML diagnosis in the 295 patients in the cohort. Every patient is represented in an individual column, whereas genes and other AML features are listed in rows. Mutations are sorted by molecular ontogeny,⁴ with TP53 mutations listed first in green, secondary-type mutations (implying evolution from an antecedent MDS) in blue, and all other mutations (pan-AML/de novo type) below. NPM1 mutations and internal tandem duplications in FLT3 (FLT3-ITD) are in red, and DDX41 mutations are in gold. Complex cytogenetics are shown (top) above the TP53 mutations. The proportion of patients in the cohort with each alteration is reported on the right.

Germline mutations that predispose to development of leukemia can present in older adults. We identified putative germline DDX41 mutations (median VAF = 0.48) resulting in start-loss (p.M1I) or premature termination codons in 16 of 295 patients (5.4%).^[17,18] Of those, 13 also had a second DDX41 mutation (median VAF = 0.09), most commonly affecting the somatic hotspot p.R525H.^[19] Five patients had germline rare variants in genes that regulate telomere maintenance (TERT [n = 2], TERC [n = 2], and RTEL1 [n = 1]), which have been linked to myeloid leukemogenesis in adults and NRM after transplantation.^[20,21] As expected, we did not identify likely pathogenic variants in genes associated with leukemia predisposition syndromes that typically presents earlier in life, such as SAMD9, SAMD9L, or SBDS.^[22-24]

Clinical and genetic determinants of posttransplant LFS

To understand the relative impact of diagnostic and remission molecular status on posttransplant LFS in this cohort, we first developed an integrated model for LFS that included both genetic and nongenetic baseline characteristics. The overall workflow for the study is shown in Figure 2A. LFS and OS at 3 years after transplant for the full cohort were 42.1% and 46%, respectively, and the 3-year cumulative incidence of relapse was 38% (supplemental Figure 1). The 3-year incidence of NRM was 29%, which was higher than the observed NRM in the RIC arms of clinical trials in younger (CTN 0902; 4%)^[13] and older ( CTN 0502; 15%)^[8] patients with AML, but similar to NRM in retrospective studies of older patients who have undergone HCT (range, 30% to 40%).^[9,14,25-27]

Enlarge

Figure 2. Development of an integrated model for LFS after transplantation. (A) The study workflow used to develop a prognostic model that included both genetic and nongenetic factors in the full cohort of 295 individuals. We additionally assessed whether the presence or absence of mutations at remission further refined the baseline mode in the 192 individuals with available remission samples. (B) LFS of individuals in the low (n = 35; green), intermediate (n = 113; blue), high (n = 71; orange), and very high (n = 77; red) risk groups. Cumulative incidence of relapse for the same groups (C), and the incidence of NRM (D). See supplemental Tables 8 to 10 for variables defining low, intermediate, high, and very high risk groups.

To identify gene mutations associated with LFS, we evaluated the 26 genes that were mutated in at least 10 patients (3%) in the study cohort. In univariable Cox models, mutations associated with inferior LFS included TP53 (HR 3.4; P < .001), JAK2 (HR 2.7; P < .001), FLT3-ITD without concomitant NPM1 mutation (HR 2.2; P < .001), and KRAS (HR 2.0; P = .004) (supplemental Table 6). The presence of an NPM1 mutation without concomitant FLT3-ITD was associated with prolonged LFS (HR 0.56, P = .002), as was the presence of a DDX41 mutation (HR 0.55; P = .036). We then generated a hierarchical model of molecular risk, in which unfavorable genetics included patients with TP53 or JAK2 mutations, or FLT3-ITDs without a concomitant NPM1 mutation (n = 63; 3-year LFS 6.7%); favorable genetics included patients without concomitant poor-risk mutations who had DDX41 or DNMT3A mutations, or NPM1 mutations without an FLT3-ITD (n = 95; 3-year LFS, 62%); intermediate genetics included all other patients (n = 137; 3-year LFS, 45%; P < .0001, for the 3-group comparison) (supplemental Figure 2).

Nongenetic factors related to the patient, disease, or transplant itself have been shown to influence LFS, NRM, and relapse.^[28,29] In univariable analysis, additional factors significantly associated with inferior LFS included pretransplant HCT comorbidity index (HCT-CI) ≥3 vs HCT-CI <3 (HR, 1.8; 95% confidence interval [CI], 1.3-2.7; P = .002), monosomal or nonmonosomal adverse karyotypes vs intermediate/favorable (HR, 4.4 for monosomal; 95% CI, 2.9-6.8; P < .001; HR 1.7 for adverse, nonmonosomal; 95% CI, 1.1-2.6; P = .013), CRi vs CR (HR, 1.8; 95% CI, 1.3-2.5; P < .001), and clinically defined secondary AML (sAML vs de novo AML; HR, 1.8; 95% CI, 1.3-2.4; P < .001) (supplemental Table 7).

To identify pretransplant factors that independently influence risk of death or relapse after transplantation in this cohort, we combined clinical and genetic variables in a multivariable frailty model adjusted for transplant center (Figure 2B). This model included the variables mentioned, as well as recipient age at the time of transplant, donor/recipient sex mismatch (male patient/female donor), intensive vs nonintensive induction therapy, white blood cell count at diagnosis, and receipt of consolidation therapy before transplant. Only molecular risk, karyotype, HCT-CI score, CRi, and clinically defined sAML retained significance and were included in the final model. Extended details are in supplemental Tables 8 to 13. Patients in the low-risk group (n = 35) had 3-year LFS of 86%, those in the intermediate-risk group (n = 113) had 3-year LFS 54% (vs low-risk; HR for death or relapse, 3.7; 95% CI, 1.8-11.3; P = .0004), those in the high-risk group (n = 71) had a 3-year LFS of 35% (HR, 5.6; 95% CI, 2.6-17.7; P < .0001), and those in the very high-risk group (n = 74) had a 3-year LFS of 9% (HR, 13.4; 95% CI, 6.3-42.2; P < .0001). The overall risk model reflects patients’ composite risk of relapse (Figure 2C) and NRM (Figure 2D).

Molecular genetics of CR

The presence of leukemia-associated gene mutations in CR has been associated with increased risk of relapse and inferior LFS after reduced-intensity allogeneic transplantation.^[13] To identify mutations present in remission at low abundance, we used a platform that incorporated duplex unique molecular identifiers, thereby enabling computational suppression of sequencing artifacts. By comparing remission with pretreatment time points, we defined each mutation as either persistent (pretreatment VAF ≥ 0.02 and detected in remission with ≥2 duplex read families) or emergent (pretreatment VAF = 0 or below 0.02 detection threshold and remission VAF ≥ 0.01) (supplemental Table 14). In total, we identified 352 persistent mutations, 326 of which (92.6%) met the VAF threshold of ≥0.001 proposed in recently published European LeukemiaNet guidelines for molecular MRD,^[30] and 100 emergent mutations.

Most patients (79.7%; 153 of 191 with diagnostic mutations) had at least 1 persistent mutation, and 150 of 153 had at least 1 mutation with VAF ≥ 0.001. The rate of persistence and the size of the persistent mutant clone varied by gene. Genes that are typically mutated in founding clones^[31,32] were more likely to have persistent mutations (Figures 3A; supplemental Figure 3; supplemental Table 15) present in high abundance (Figure 3B), including DNMT3A (55 of 66 persisted; 83.3%; median VAF = 0.064), TET2 (47 of 61; 77.0%; median VAF = 0.106), SF3B1 (6 of 7; 85.7%; median VAF = 0.025), U2AF1 (12 of 15; 80.0%; median VAF = 0.097), ASXL1 (30 of 38; 78.9%; median VAF = 0.084), and TP53 (20 of 26; 76.9%; median VAF = 0.044). In contrast, genes typically mutated in subclones were less likely to have persistent mutations, and mutations that persisted were present in low abundance, including CEBPA (0%), NPM1 (9.3%; median VAF = 0.003), WT1 (17.6%; median VAF = 0.003), FLT3-ITD (19.0%; median VAF = 0.004), FLT3-TKD/JMD (22.0%; median VAF = 0.001), KRAS (22.3%, median VAF = 0.022), and NRAS (25.0%; median VAF = 0.010). MRD assessed by multiparameter flow cytometry was available for 87 patients with remission molecular assessments; of 26 patients who were flow MRD positive, 21 (80.7%) were molecular MRD positive. Conversely, of 61 patients who were flow MRD negative, 32 (52.5%) were molecular MRD positive (supplemental Table 16).

Enlarge

Figure 3. Characteristics of remission mutations. (A) Results of Fisher’s exact test for clearance or persistence of each gene, plotting the odds of clearance on the x-axis and significance, expressed as the negative log of the uncorrected P value, on the y-axis. Shown are mutations that are more likely to persist at remission (left) and those that are more likely to clear (right). (B) VAF of remission mutations. Median VAF is denoted by the dashed lines, and the range is indicated by the red violin plot for each gene. A version of this figure that excludes the 26 mutations with VAF < 0.001 is in supplemental Figure 3. (C) The number (left y-axis) and VAF (right y-axis) of mutations present at remission that were not detected at diagnosis, in descending order of frequency. VAFs are represented by blue dots and the total number of mutations is represented by the bars.

By comparing baseline and remission samples, we found that 66 of 192 patients (34.4%) had mutations that were newly detectable after treatment. The spectrum of emergent mutations was consistent with prior studies that reported clonal hematopoiesis after cytotoxic therapy^[33-35] and most commonly involved DNMT3A (23 mutations in 20 patients), TET2 (15 mutations in 13 patients), PPM1D (12 mutations in 10 patients), and TP53 (8 mutations in 8 patients) (Figure 3C).

Sole persisting DNMT3A, TET2, and ASXL1 mutations have been reported to have no impact on relapse risk and have been considered to reflect clonal hematopoiesis rather than frank residual leukemia.^[13,36] In older patients, however, ASXL1 mutations have been associated with evolution from prior MDS, whereas DNMT3A and TET2 mutations have not been.^[4,5] We found that DNMT3A and TET2 mutations commonly persisted in remission without other mutations (35 of 89 [39.3%] patients with DNMT3A or TET2 mutations at diagnosis remitted to only DNMT3A or TET2 mutations after treatment). In contrast, only 6 of 36 (16.7%) ASXL1 mutations persisted without other concomitant mutations, which were most commonly in MDS-associated genes such as SRSF2, RUNX1, and STAG2. Based on their different patterns of persistence and ontogenetic associations, we considered persistent sole DNMT3A or TET2 mutations as a group (DT), separate from the persistence of mutations in any other gene, including ASXL1 (MRD-positive).

MRD associations with baseline characteristics and posttransplant outcomes

The impact of molecular MRD on relapse risk after RIC HCT was defined based on the detection of remission mutations without knowledge of baseline genetic characteristics.^[13] Therefore, we sought to determine whether pretreatment factors were associated with the likelihood of MRD-positive CR. In multivariable logistic regression analysis, genetic ontogeny (secondary vs de novo ontogeny,^[4] odds ratio [OR], 7.9; 95% CI, 3.6-17; P < .0001; and TP53 vs de novo, OR, 7.7; 95% CI, 2.3-26; P = .001; and high-risk cytogenetics vs low risk OR, 5.7; 95% CI, 1.2-26.6; P = .027) were each associated with MRD positivity (Figure 4A; supplemental Table 17). For patients with MRD-positive remissions, remission clonal abundance, defined by the mutation with highest VAF per patient, was median VAF = 0.021 for patients with de novo genetic ontogeny, 0.042 for those with secondary ontogeny, and 0.113 for those with TP53 ontogeny (supplemental Figure 4). Most patients with DDX41 mutations at diagnosis had MRD-negative remissions.

Enlarge

Figure 4. Clinical and genomic associations of molecular remission states. (A) Co-occurrence of nongenetic factors and specific persistent mutations for the 192 individuals in the remission cohort, grouped by type of molecular measurable residual disease (MRD). Patients who cleared all diagnostic mutations and patients with only persistent DNMT3A or TET2 mutations (together considered MRD negative) are in the left and middle respectively, and patients with other persistent mutations (MRD positive) are on the right. Remission DNMT3A, TET2, and ASXL1 mutations are shown at the top, with other classes of remission mutations shown next. Diagnostic genetic variables associated with molecular remission states are listed next, including DDX41 mutations (shown in gold) and molecular ontogeny (de novo in gray, secondary in blue, and TP53 in green). Patients who relapsed are shown in black and those who died are in red. (B) Distribution and co-occurrence of treatment-emergent mutations in patients with different patterns of molecular persistence. Patients who cleared all diagnostic mutations and patients with only persistent DNMT3A or TET2 mutations are in the left and middle, respectively, whereas patients with MRD-positive remissions are on the right.

Posttreatment-emergent mutations may reflect either outgrowth of chemoresistant leukemic subclones or expansion of clonal hematopoiesis that is clonally unrelated to the leukemia itself.^[32,37] In patients who cleared diagnostic mutations or had only persistent DT mutations, treatment-emergent mutations were most commonly in DNMT3A and TET2 (13 of 20; 65%; Figure 4B). In contrast, the majority of newly detected TP53 and PPM1D mutations (15 of 18; 83%), as well as the majority of newly detected mutations in other genes (27 of 31; 87%) occurred in patients who also had persistent diagnostic mutations.

In univariable analysis, patients with MRD-positive remission had inferior LFS compared with those who had MRD-negative remission or had sole persistent DNMT3A/TET2 mutations (HR for death/relapse vs MRD negative/DT, 1.58; 95% CI, 1.07-2.32; P = .021; Figure 5A). These results were similar when reassigning 2 of 153 patients from MRD^positive to MRD^negative and 1 patient from DT to MRD^negative based on the provisional VAF threshold of 0.001 proposed in the European Leukemia Network guidelines for molecular MRD^[30] (supplemental Figure 5). The difference in LFS was driven by a higher rate of relapse in the MRD-positive group (at 3 years, 42.5% vs 20.6% for MRD negative/DT; P = .0006). The rates of relapse were unchanged when the definition of MRD negative was expanded to include sole persistent DNMT3A, TET2, or ASXL1 mutations (DTA; supplemental Figure 6). There was no significant difference in NRM based on MRD status (24.7% vs 17.4%, P = .24). When accounting for overall pretransplant risk, however, MRD positivity had no independent effect on LFS (Figure 5B). Three-year LFS was similar in patients with MRD-positive remission compared with MRD-negative/DT remission across risk groups: low risk (100% vs 86.7%; P = .39), intermediate risk (58.3% vs 53.6%; P = .75), high risk (37.7% vs 45.6%; P = .68), and very high risk (10.2% vs 12.5%; P = .93).

Enlarge

Figure 5. LFS according to baseline risk and MRD status. LFS according to baseline risk and MRD status. (A) Unadjusted LFS for all remission cohort patients (n = 192) according to MRD status. (B) LFS for remission cohort patients according to baseline risk status: low (n = 25); intermediate (n = 76); high (n = 40); or very high (n = 50). Patients with MRD-positive remissions are in red, and patients with MRD-negative remissions (all mutations cleared or DT only) are in blue.