Frequent reconstitution of IDH2^R140Q mutant clonal multilineage hematopoiesis following chemotherapy for acute myeloid leukemia

While the full repertoire of recurrent genetic lesions associated with acute myeloid leukemia (AML) is now likely known, the pathogenic contribution of each remains incompletely understood. Some are compatible with otherwise normal hematopoiesis. For example, aging humans with normal blood counts often exhibit hematopoietic sub-clones with mutations in DNMT3A, ASXL1 or TET2, among others, even though only a small minority ever develops myeloid malignancy.^1–3 This suggests that such genetic lesions confer a competitive advantage to hematopoietic stem and progenitor cells (HSPC) to facilitate formation of a pre-leukemic stem cell pool receptive to secondary mutations for full-fledged leukemic transformation.⁴ Consistent with this, DNMT3A mutations often persist in blood many years after successful chemotherapy for AML.⁵ Conversely, other genetic lesions appear tightly associated with malignant transformation. For example, mutations in NPM1 are not seen with aging and their persistence following chemotherapy predicts for impending relapse.⁶

Mutations in IDH1 and IDH2 are found in 15-20% of patients with AML⁷ but are extremely rare drivers of clonal hematopoiesis in aging (~100-fold less frequent than mutations in DNMT3A, for example).^1–3 It is currently unclear whether IDH mutant (IDH^mut) clones persist in long-term clinical remission, and if so, what is their natural history, clinical fate, multilineage differentiation potential and relationship to other mutations. Establishing this is essential not only for biologic understanding, but also to establish therapeutic endpoints in patients treated with mutant IDH inhibitors.

To address this we studied 23 patients presenting with AML or refractory anemia with excess blasts (RAEB) associated with mutations in IDH1 (n=10) or IDH2 (n=13) (Table S1), all of whom achieved complete morphologic remission (CR) after one or two courses of induction chemotherapy. In each case there was reduction in IDH mutant allele frequency (MAF) by quantitative digital PCR (dPCR) following treatment.⁸ However, samples collected at CR1 (after course 1 (n=6) or 2 (n=17)) demonstrated that only 3/23 (13%) had an undetectable IDH mutation burden (sensitivity ≤0.1% MAF) (Figure 1a). Following consolidation chemotherapy IDH MAF variably decreased or increased (Figure 1b). Considering all patients at completion of chemotherapy, only 5/23 (22%) had undetectable IDH^mut alleles (Figure 1c). Thus IDH^mut clones typically persist at the end of induction chemotherapy in the majority of patients achieving morphologic CR and are not eradicated by consolidation chemotherapy.

Figure 1. IDH mutant allele frequencies in AML/RAEB patients following treatment.

(a) IDH MAFs at presentation and at complete morphologic remission (samples collected after course 1 (n=6) or 2 (n=17)). * indicates p<0.001 by paired t-test. Mean±SD MAFs are shown. (b) IDH MAFs post-induction (i.e. after course 2) and post-consolidation in patients receiving one or more cycles of consolidation chemotherapy (n=13). (c) IDH MAFs at completion of intensive chemotherapy in patients treated with chemotherapy alone (n=16) or immediately pre-transplant in those allografted (n=7). (d) IDH MAFs pre- and post-transplant (day 28, n=1, day 100, n=5). (e) BM IDH2 MAFs in four patients who relapsed following chemotherapy. Morphologic relapse was documented at the last time point shown. (f) BM IDH2 MAFs in six patients in sustained morphologic CR. Biobank identifiers are shown. (g) Total plasma 2-hydroxyglutarate levels in four patients from (f). (h & i) Hierarchical placement of IDH mutations, as determined by targeted next generation sequencing. Fish plots (h) illustrate exemplar patterns of IDH^mut acquisition. IDH^mut clonal hematopoiesis is shown in red with hypothetical mutations X and Y shown in blue and orange respectively. (i) Graph shows categorization of types of IDH^mut clones in 23 presentation samples. An IDH^mut clone was considered dominant if the IDH MAF was the highest of all detected mutations. Where the IDH MAF was proportionately 15% larger than that of any other mutation the clone was considered sole dominant; otherwise it was deemed co-dominant. Where the IDH MAF was 50-85% of that of the highest mutation, the IDH^mut clone was considered a major sub-clone; otherwise it was deemed a minor sub-clone.

By contrast, of seven patients receiving hematopoietic stem cell transplant (HSCT) (including one transplanted in CR2), IDH^mut alleles were undetectable in bone marrow (BM) in five (71%) by day 100 (Figure 1d). Of these, three survive with undetectable IDH^mut alleles 15, 22 and 51 months post-transplant. One died in CR1 14 months post-transplant of procedure-related complications with undetectable IDH1^R132C, and the fifth suffered graft failure, low-level reconstitution of IDH1^R132C-mutant hematopoiesis and died 22 months post-transplant. Of the two others, one died of complications early post-transplant and the other failed to clear IDH2^R172K following HSCT, heralding early relapse. Thus, in contrast to chemotherapy, allogeneic HSCT is effective in durable eradication of IDH^mut hematopoiesis.

We next evaluated outcomes of 16 patients treated with chemotherapy alone for whom more than six months’ follow up was available (including one subsequently transplanted in CR2). Three of six IDH1^mut patients relapsed following a progressive rise in MAF, whereas three exhibit persistent low level IDH1 mutant clones (<1% MAF) and remain in CR1 after 7, 22 and 24 months’ follow up respectively (Figure S1). Four of 10 IDH2^mut patients also relapsed, again following a progressive rise in MAF (Figure 1e). In contrast, six IDH2^mut patients (all with IDH2^R140Q mutations) remain in CR after 9-52 months’ follow up with normal or near-normal blood counts (Table S2) despite persistent and substantial IDH2^R140Q mutant clones, which in some cases dominate hematopoiesis (Figure 1f). As expected, plasma D-2-hydroxyglutarate (D-2-HG) levels mirrored MAFs during CR (Figure 1g).⁸ Thus, IDH2^R140Q mutant hematopoietic clones frequently persist and may predominate in patients with normal or near-normal hematopoiesis following successful chemotherapy for AML.

We next addressed the hierarchical placement of IDH mutations versus other known drivers in myeloid cancer using targeted next-generation sequencing in the 23 presentation samples (Table S1). We identified 109 high confidence somatic variants in 26 genes (median 4 per sample; range 2–9; Figure S2; Table S3) and confirmed the presence of all IDH mutations, with concordant MAFs versus dPCR (Figure S4). The IDH^mut clone was dominant in 12/23, a major sub-clone in 7/23 and a minor sub-clone in 4/23 cases (Figures 1h, 1i, S4, S5, Table S3). Where dominant, the IDH MAF was similar to that of co-occurring mutations in 8/12 cases precluding identification of the ancestral mutation. However, in 4/12 cases (BB93, BB161, BB187 and BB484) the IDH MAF was larger, suggesting that mutations in IDH1 or IDH2 can predate mutations in SRSF2, JAK2 and DNMT3A. In 11 cases where the IDH^mut clone was sub-clonal, the dominant clone exhibited mutations in DNMT3A (n=6), SRSF2 (n=5) or STAG2 (n=1), indicating that in these instances mutations in IDH1 or IDH2 were secondary events. Sub-clonal placement of IDH mutations had no clear impact on the likelihood of persistence post-chemotherapy.

Sequencing of remission samples identified persistence in CR1 of mutations in DNMT3A, RUNX1, JAK2, ASXL1 and RAD21, in addition to those in IDH1 and IDH2. NPM1 and FLT3-ITD mutations were not detected in remission samples (Figures S4, S5, Table S3). Illustrative of this, five IDH2^R140Q mutant patients who remain in long-term CR presented with a concomitant NPM1 mutation (Table S3). In each case the NPM1 mutation has remained undetectable despite persistence of the IDH2^mut clone (Figures 2a-c, S6), which exhibited additional persisting concomitant mutations in two patients (in SRSF2, BCOR and GATA2; Figures 2c, S6). By contrast, relapsing IDH^mut patients exhibited recrudescence of mutations in NPM1 (BB120, BB475) or RUNX1 (BB85, BB235, BB350) as determined by Sanger sequencing or digital PCR (data not shown). Thus mutations in SRSF2 and DNMT3A may precede or follow those in IDH1 or IDH2, whereas FLT3-ITD or mutations in NPM1 appear secondary and associated with the presence of frank leukemia.

Figure 2. Multilineage contribution of IDH2^R140Q mutant clonal hematopoiesis.

Graphs show IDH2^R140Q versus NPM1 MAFs at the indicated time points and in the indicated cell populations in (a) BB161 and (b) BB287. Remission samples were collected 20 and 18 months following presentation, respectively. (c) MAFs for the somatic mutations identified by targeted next generation sequencing at the indicated time points following presentation in patient BB161. (d) and (e) show allelic discrimination dPCR plots for plucked single colonies isolated following 14 days of culture of CD34⁺ BM HSPC collected 40 months (BB161) and 18 months following presentation (BB287). Plasmid standards containing 0%, 50% and 100% mixes of DNA sequences coding for IDH2^R140Q & IDH2^WT are shown. Control samples from IDH2^WT and IDH2^R140Q mutated bulk patient samples are also shown (n=2 for each). (f) Allelic discrimination dPCR plot for individual colonies derived from single-sorted phenotypic HSCs (CD34⁺38^-90⁺45RA^-Lin^-) cultured in individual wells in methylcellulose for 14 days from BB287 17 months following presentation. BFU-E, burst forming unit erythroid; CFU-GM – granulocyte/macrophage colony forming unit; RFU, relative fluorescent units.

To determine whether post-chemotherapy IDH2^R140Q clones contribute to multilineage hematopoiesis, we isolated by flow-sorting phenotypically defined high purity cell populations from the BM and blood of two individuals in prolonged CR (Figure S7). In both patients neutrophils and monocytes were predominantly derived from IDH2^R140Q mutant clones, as was BM erythropoiesis and myelopoiesis (Figures 2a,b). The B-lineage was derived from an IDH2^R140Q clone predominantly in BB287 and to a lesser extent in BB161. The T-lineage was not involved. The immunophenotypic BM hematopoietic stem cell (HSC) compartment exhibited 20-35% involvement with IDH2^R140Q mutant cells, with higher proportionate involvement of the downstream multipotent progenitor compartment in both cases. To interrogate further their functional potential, CD34⁺ HSPCs from deep remission time points from both patients were cultured in methylcellulose clonogenic assays. In keeping with our in vivo data, 96% and 50% respectively of individually isolated colonies (including both erythroid and granulocyte/macrophage colonies) were IDH2-mutated (Figures 2d,e), as were 60% of colonies derived from single flow-sorted immunophenotypic HSCs from BB287 (Figure 2f). Similar analyses in a patient with persistent IDH1^mut hematopoiesis (BB355) 22 months after presentation demonstrated involvement of neutrophils and monocytes, but not lymphocytes (Figure S8).

In particular, our data demonstrate dynamic reconstitution of IDH2^R140Q-mutated, functionally normal, clonal hematopoiesis following successful chemotherapy for AML. Such clones may on occasion exhibit a concomitant SRSF2^P95R mutation. Our observations provide compelling in vivo evidence for the pre-leukemic nature of IDH2^R140Q mutations in view of their multilineage differentiation potential. These data extend insight into recent observations that immunophenotypic HSCs carry IDH mutations at diagnosis in 80% of IDH^mut AMLs,⁹ whilst xenografts from 36% of IDH^mut AML patient samples generated non-leukemic multilineage grafts.¹⁰

One unique feature of IDH^mut AML is the generation of D-2-HG. This metabolite was present at high levels in the plasma of patients with persistent IDH^mut clonal hematopoiesis (Figure 1g), an observation which counters the concept that D-2-HG is directly transforming, as has recently been suggested.¹¹ Nevertheless, caution is mandated for the long-term outcome of these patients which to date remains unclear.

Importantly, our findings indicate that approaches tracking IDH mutations (or D-2-HG as their surrogate) will not reliably detect minimal residual leukemia, in contrast to recent claims.^12,13 Further, total eradication of IDH^mut clones (in particular IDH2^R140Q-mutant clones) as a therapeutic goal, for example using specific pharmacological inhibitors, while desirable, may not be necessary to achieve prolonged disease-free survival given that IDH mutations and leukemic disease are not tightly linked in many cases. That said, it remains unclear whether our conclusions extend to patients with IDH1^R132 or IDH2^R172 mutations because in our cohort persistence or recurrence usually heralded leukemic relapse, and the three cases with persistent IDH1^mut hematopoiesis display low MAFs. This hints at discrepant biology associated with different IDH mutations, in keeping with reported biochemical differences.¹⁴ More broadly, our findings add IDH2^R140Q to DNMT3A, TET2 and other mutations as mediators of clonal hematopoiesis of indeterminate potential,¹⁵ at least after chemotherapy for established AML.