CLINICAL AND BIOLOGICAL IMPLICATIONS OF ANCESTRAL AND NON-ANCESTRAL IDH1 AND IDH2 MUTATIONS IN MYELOID NEOPLASMS

Abstract

Mutations in isocitrate dehydrogenase 1/2 (IDH1/2^MT) are drivers of a variety of myeloid neoplasms. As they yield the same oncometabolite, D-2-hydroxyglutarate, they are often treated as equivalent, and pooled. We studied the validity of this approach and found IDH1/2 mutations in 179 of 2119 myeloid neoplasms (8%). Cross-sectionally, the frequencies of these mutations increased from lower- to higher-risk disease, thus suggesting a role in clinical progression. Variant allelic frequencies indicated that IDH1^MT and IDH2^MT are ancestral in up to 14/74 (19%) vs. 34/99 (34%; P=0.027) of cases, respectively, illustrating the pathogenic role of these lesions in myeloid neoplasms. IDH1/2^MT was associated with poor overall survival, particularly in lower-risk myelodysplastic syndromes. Ancestral IDH1^MT cases were associated with a worse prognosis than subclonal IDH1^MT cases, whereas the position of IDH2^MT within clonal hierarchy did not impact survival. This may relate to distinct mutational spectra with more DNMT3A and NPM1 mutations associated with IDH1^MT cases, and more ASXL1, SRSF2, RUNX1, STAG2 mutations associated with IDH2^MT cases. Our data demonstrate important clinical and biological differences between IDH1^MT and IDH2^MT myeloid neoplasms. These mutations should be considered separately as their differences could have implications for diagnosis, prognosis, and treatment with IDH1/2^MT inhibitors of IDH1/2^MT patients.

INTRODUCTION

Mutations in isocitrate dehydrogenases 1 and 2 (IDH1^MT and IDH2^MT) are implicated in the development of numerous types of cancer, including glioma, chondrosarcoma, cholangiocarcinoma, angioimmunoblastic T-cell lymphoma and certain myeloid neoplasms¹. Within myeloid malignancies, IDH1/2^MT are present in a significant proportion of acute myeloid leukemia (AML) and, while less common, in myelodysplastic syndromes (MDS) and myeloproliferative neoplasms (MPN)^{2, 3}.

IDH1/2^MT impart a gain of function by causing single amino acid changes in the active sites of the enzymes. Whereas (wild-type) IDH1/2^WT convert isocitrate and NADP⁺ to α-ketoglutarate (αKG) and NADPH, mutant IDH1/2 convert NADPH and αKG to NADP⁺ and D-2-hydroxyglutarate (D-2HG)⁴. D-2HG and αKG have similar structures; resultant D-2HG accumulation in IDH1/2^MT cells inhibits many αKG-dependent dioxygenases, such as TET2 DNA hydroxyl demethylases^{5, 6} and Jumanji domain-containing histone demethylases (e.g., KDM6A/UTX and KDM3B/JMJD1B). Dysfunction of these enzymes is considered to be responsible for global DNA hypermethylation, inhibition of differentiation, and preservation of stemness^{7, 8}. In addition, D-2HG activates, rather than inhibits Egg Laying Defective Nine (EGLN), which induces HIF1α degradation and growth factor-independent proliferation^{9, 10}.

As direct evidence that IDH1/2^MT are oncogenic, introduction of IDH2^R140Q or IDH2^R172K into 10T1/2 mesenchymal progenitor cells yielded an AML-like disease in mice^11–13. IDH2^R140Q was also necessary for AML maintenance in mice¹², indicating that IDH1/2^MT may be useful as therapeutic targets. These findings motivated the development of specific inhibitors of IDH1/2-mutant enzymes^14–18. The currently available inhibitors, AG-221 that targets IDH2^R140Q and AG-120 that targets IDH1^R132H, restrict the production of D-2HG and thereby induce differentiation of AML cells ex vivo¹⁵. Both agents have shown promising preliminary results in phase I clinical trials¹⁹.

Genomic studies of molecular landscapes in human cancer have frequently combined IDH1^MT and IDH2^MT as a single functional group despite physiological differences: IDH1 is localized in the cytosol and IDH2 in the mitochondrial matrix. Remarkably, the spectrum of cancers and their subtypes differ in the distribution of IDH1/2^MT. IDH1^MT predominates in glioma (95%), chondrosarcoma (95%) and cholangiocarcinoma (80%), whereas the IDH1^MT:IDH2^MT ratio is more balanced or even skewed towards a higher frequency of IDH2^MT in AML¹. In addition, only IDH2^MT have been reported in angioimmunoblastic T-cell lymphoma, osteosarcoma and gastric cancer, again suggesting that there are pathophysiologic differences between IDH1^MT and IDH2^MT. Biochemical investigations have shown that the specific amino acid substitutions IDH1^R132H, IDH1^R132C, IDH2^R140Q or IDH2^R172K differ in D-2HG production potency. IDH2^R172K is the most potent, followed by IDH1^R132H, IDH1^R132C and IDH2^R140Q. There are more than 60 different αKG-dependent dioxygenases that can theoretically be inhibited by D-2HG with likely distinct IC₅₀ values for D-2HG. Thus, different IDH1/2^MT inhibit different sets of αKG-dependent enzymes and this may partially explain differing distributions of IDH1/2^MT between cancers¹.

To investigate the clinical impact of IDH1/2^MT in myeloid neoplasms, we studied a cohort of 2119 patients with myeloid neoplasms. We performed whole-exome/targeted multi-amplicon sequencing on the samples obtained from these patients from different institutions. We compared the mutational landscapes of IDH1^MT and IDH2^MT samples, their clinical associations, overall patient survival, and clonal hierarchies. Our aim was to provide insights into IDH1/2^MT myeloid malignancy pathogenesis, especially with respect to the clonal architecture of IDH1/2^MT cases, and to determine whether IDH1 and IDH2 mutations should be grouped or considered separately, particularly with respect to the potential benefits of IDH1/2-mutant inhibitors in various clinical contexts.

METHODS

Patient population

Blood and bone marrow samples were obtained from 2119 patients diagnosed with lower-risk MDS (868) and higher-risk MDS (536), defined per World Health Organization classification; secondary AML (sAML; 153); myeloproliferative neoplasms (MPN; 63); MDS/MPN (165); or primary AML (pAML; 334). From this cohort, 418 samples from 409 patients were subjected to whole-exome sequencing (WES). Furthermore, 1815 samples from 1761 patients were tested for a subset of genes (including IDH1 and IDH2; Supplementary Table S1) using targeted sequencing (TS). The total of 2179 samples listed above is greater than our total of 2119 samples because serial samples of 60 patients were included in the analysis. The sum of 409 patients subjected to WES and 1761 patients subjected TS is larger than 2119 because the WES and TS cohorts are partially overlapping.

Informed consent was obtained from patients according to protocols approved by the institutional review boards and in accordance with the Declaration of Helsinki. These patients were seen and treated at the Cleveland Clinic, the University of Tokyo and the Munich Leukemia Laboratory, diagnosis was confirmed on the basis of World Health Organization classification criteria. Patients with refractory anemia (RA), RA with ringed sideroblasts (RARS), refractory cytopenias with multilineage dysplasia (RCMD), RCMD with ringed sideroblasts (RCMD-RS) and MDS-unclassifiable (MDS-u) were classified as lower-risk MDS. Those with refractory anemia with excess blasts (RAEB-1 and RAEB-2) were classified as higher-risk MDS. Details of MDS/MPN and MPN subcategories are provided in Supplementary Table S2. Clinical details of the patients, including blood counts, demographics, and survival times, were obtained from medical records.

Sequencing technology

For WES, DNA was obtained from tumor and germline paired samples. The 50 Mb of protein-coding sequences was enriched from total genomic DNA by liquid-phase hybridization using SureSelect (version 4) (Agilent Technology, Santa Clara, CA), followed by massively parallel sequencing with HiSequation 2000 (Illumina, San Diego, CA). Somatic mutations were called as previously described^{20, 21}, using the GATK algorithm (Broad Institute). To minimize false positives and focus on the most prevalent or relevant somatic events, we implemented a rational bioanalytic filtering approach and applied heuristic bioanalytic pipelines. For confirmation of somatic mutations, we analyzed paired germline DNA from CD3⁺ lymphocytes. For TS of a specific panel of genes we applied multi-amplicon deep sequencing (TrueSeq; Illumina) to frequently-affected exons of 60 selected genes. The sequencing libraries were generated according to an Illumina paired-end library protocol and subjected to deep sequencing on MiSeq (Illumina) instrumentation according to standard protocol. Copy number information at the locus of each mutation was assessed as previously reported²². For samples sequenced in Munich, 104 known or putative mutational gene targets in MDS were examined for mutations from the cohort using massively parallel sequencing (Illumina) or SureSelect (Agilent) captured target sequences²³. High-probability oncogenic mutations were called by eliminating sequencing/mapping errors and known/possible SNPs based on available databases and frequencies of variant reads. Genomic copy number status was calculated by directly enumerating corresponding sequencing reads in each exon.

Analysis of clonal architecture

Variant allelic frequencies (VAF) were calculated as the fraction of mutated reads of the total number of reads of a certain gene. VAFs were adjusted using copy number information at the locus of each mutation. We recapitulated the clonal architecture of a patient using these copy number-adjusted VAFs. Ancestral vs subclonal events were determined using a copy number-adjusted VAF difference between two events, with a higher VAF indicating ancestral origin. We used an VAF threshold of 5% (absolute) to reliably discriminate ancestral from subclonal events. Events below this threshold were considered to be of undeterminable ancestry. This threshold was chosen based on previous studies²⁴ and statistical calculations of our own data, based on the average depth of sequencing in our samples.

Statistical analysis

Comparisons of proportions were performed using the χ² and Fisher’s exact tests and differences in values and in ranks were assessed by Student t tests and Mann-Whitney U tests, respectively. Cox models were used to identify correlates with overall survival. Kaplan-Meier curves were generated to graphically depict survival differences. Throughout, 2-sided tests were used with significance defined as α<.05. These analyses were performed using SPSS, GraphPad Prism 6 and the R statistical programming language.

RESULTS

Clinical characterization of the IDH1/2 mutation

Among the 2119 patient samples tested (Table 1), we identified IDH1 and IDH2 mutations in 78 (4%) and 101 (5%) cases, respectively, in a mutually exclusive fashion. Of the IDH1^MT cases, IDH1^R132C (49%, n=36) and IDH1^R132H (38%, n=28) were the most common; 12% and 1% harbored IDH1^R132X and IDH1^T98I, respectively. Amongst IDH2^MT cases, IDH2^R140Q (85%, n=84) was the most frequent mutation, followed by IDH2^R140X (5%, n=5), IDH2^R172K (9%, n=9) and IDH2^R172W (1%, n=1) (Figure 1A). IDH1^MT patients were younger than IDH1/2^WT patients (P=.01), an observation similar to that reported for IDH1^MT glioblastoma patients²⁵. Across the spectrum of MDS, the frequency of IDH1/2^MT increased with progression, as IDH1^MT occurred in 2% of lower-risk MDS cases, in 3% of higher-risk MDS cases and in 7% of sAML cases. The average number of mutations in each of these categories as determined by WES did not increase as much with 21, 30, and 27, respectively (Supplementary Figure S1). IDH2^MT frequency increased as MDS progressed (in 2% of lower-risk MDS, in 7% of higher-risk MDS and in 6% of sAML cases). These findings suggest that in IDH1/2^MT MDS cases, a high proportion of IDH1/2^MT are associated with progression to advanced disease. In addition, IDH1 and IDH2 mutation frequencies were relatively lower in MDS/MPN and MPN patients, and showed the highest frequency in pAML patients (Figure 1B). While IDH2^MT were enriched for a normal karyotype, IDH1^MT patients more often had a deletion of chromosome Y in comparison with IDH1/2^WT patients. Other clinical characteristics were similar between IDH1/2^MT and IDH1/2^WT patients (Table 1; Supplementary Table S3). More detailed clinical information on IDH1MT and IDH2MT patients with myeloid malignancies is shown in Supplementary Tables S4 and S5.

Table 1.

Clinical characterization of the IDH1/2 mutation in 2119 patients

	IDH1MT (N=78)	p-value	IDH2MT (N=101)	p-value	IDH1/2WT (N=1940)	Total (N=2119)
Age (median)	60	0.01	65	0.58	66	66
Sex*
Male	46 (59%)		64 (63%)		1154 (59%)	1264
Female	30 (38)		33 (33%)		761 (39%)	824
Diagnosis
Low-risk MDS	21 (2%)	0.0049	15 (2%)	0.0001	832	868
High-risk MDS	18 (3%)	0.5947	37 (7%)	0.0186	481	536
sAML	11 (7%)	0.0239	8 (6%)	0.6869	134	153
MDS/MPN	2 (1%)	0.08	7 (4%)	0.85	156	165
MPN	0	0.176	2 (3%)	0.76	61	63
pAML	26 (8%)	0.0001	32 (10%)	0.0001	276	334
Karyotype**
Normal	46 (59%)	0.2789	65 (64%)	0.0138	1010 (52%)	1121 (53%)
Aberrant	27 (35%)	0.2917	29 (29%)	0.0164	792 (41%)	848 (40%)
Complex	5 (6%)	1	7 (7%)	1	136 (7%)	150 (7%)

^*Sex NA: IDH1^MT: 2, IDH2^MT: 4, IDH1/2^WT: 23

^**Karyotype NA:IDH1^MT: 0, IDH2^MT: 0, IDH1/2^WT: 2

Overview of clinical characteristics of the IDH1^MT, IDH2^MT and IDH1/2^WT patients in the study cohort. Percentages in the diagnosis rows are calculated as percentage of patients with a certain diagnosis with IDH1^MT or IDH2^MT. Percentages in the karyotype rows are calculated as percentage of IDH1^MT/IDH2^MT patients with a certain karyotype of their myeloid neoplasm. P values are calculated using the Student’s t test (Age) or Fisher’s Exact test (all other characteristics).

Figure 1. Breakdown of IDH1/2 mutations among disease types and specific amino acid substitutions.

(A) Frequencies of IDH1 and IDH2 mutations in various myeloid neoplasms. (B) Pie chart showing the percentages of the specific IDH1/2 mutational amino acid substitutions in the cohort.

Abbreviations: pAML, primary acute myeloid leukemia; sAML, secondary acute myeloid leukemia.

Molecular characterization of IDH1 and IDH2-mutated myeloid neoplasms

To characterize the molecular features of patients with myeloid neoplasms with IDH1^MT or IDH2^MT, we analyzed associations between IDH1/2^MT and other mutational events (Figure 2A; Supplementary Table S6). Overall, NPM1 mutations were more frequent in both IDH1^MT and IDH2^MT cases, compared with IDH1/2^WT patients (Figure 2B and Supplementary Tables S6-8). Other somatic events were not significantly associated with both IDH1^MT and IDH2^MT cases compared with IDH1/2^WT cases.

Figure 2. Mutational spectrum and most and least frequently co-occurring mutations in IDH1/2-mutated cases.

(A) Mutational spectrum and clinical spectrum of IDH1^MT cases (left half) and IDH2^MT cases (right half) with myeloid neoplasms. (B) Prevalence of co-occurring mutations in IDH1/2^MT cases, compared with IDH1/2^WT cases.

Abbreviations: lrMDS, low-risk MDS; hrMDS, high-risk MDS; sAML, secondary acute myeloid leukemia; MDS/MPN, myelodysplastic syndromes/myeloproliferative neoplasms; pAML, primary acute myeloid leukemia; smut, somatic mutation; del, deletion; hemimut, hemizygous mutation.

However, when IDH1^MT cases were compared with IDH1/2^WT cases, they were enriched for DNMT3A (P<.0001), PHF6 (P=.0006) and FLT3 (P=.0195) mutations, whereas IDH2^MT cases were enriched for ASXL1 (P<.0001), SRSF2 (P<.0001), RUNX1 (P=.0034) and STAG2 (P<.0001) compared with IDH1/2^WT cases. NPM1 mutations occurred more frequently in IDH1^MT patients than in IDH2^MT patients (P=0.045), whereas IDH2^MT patients were enriched for ASXL1 (P=0.016), SRSF2 (P=0.007) and STAG2 (P=0.019) mutations compared with IDH1^MT patients. Cohesin complex mutant carriers (STAG2, RAD21, SMC3) were more frequent in IDH1^MT (P=.025) and IDH2^MT (P<.0001) cases compared with IDH1/2^WT cases. Conversely, SF3B1 and TET2 mutations occurred less frequently in IDH1^MT (P<.0001 and P=.023) and in IDH2^MT (P<.0001 and P<.0001) cases than in IDH1/2^WT cases. Although we observed a negative correlation between IDH1 and TET2 mutations, TET2 mutations, if present, were less infrequent in IDH1^MT patients than in IDH2^MT patients (Figure 2C).

Analysis of ancestry and clonal architecture of IDH1/2-mutated patients

Using copy number-adjusted VAF, we reconstructed the clonal architecture of IDH1/2^MT patients to establish whether an IDH1^MT or IDH2^MT was an ancestral or subclonal mutation. Clonal hierarchy was further confirmed and refined by serial analyses, performed in 60 exemplary cases (Figure 3C). IDH1^MT and IDH2^MT were ancestral in 19% and 34% of IDH1/2^MT patients and subclonal in 55% and 45%, respectively (Figure 3A). As we used a cut-off value of 5% (absolute) in VAF difference, the ancestral vs. subclonal status of some IDH1/2^MT remained “undetermined” when there were small VAF differences between IDH1/2 mutations and other mutations (see Methods). A lower proportion of IDH1^MT cases were of ancestral origin compared to IDH2^MT cases (P=.027). We also observed differences in the clonal succession between specific IDH1/2^MT variants (Figure 3B). Whereas IDH1^R132H and IDH1^R132C mutations were ancestral in equal frequencies (22% and 21%), other IDH1 mutations (IDH1^T98I (n=1), IDH1^R132L (n=1) and IDH1^R132S (n=5)) occurred only as subclonal events. Similarly, IDH2^R140Q mutations were ancestral in 39% of IDH2^R140Q-mutated cases, whereas other IDH2^R140 mutations (IDH2^R140L (n=2) and IDH2^R140W (n=3)) were always subclonal. IDH2^R172 mutations were ancestral in only 10% of IDH2^R172-mutated patients. In the different types of myeloid neoplasms we did not observe differences in the distribution between ancestral vs. subclonal mutations in IDH1/2 and there was no difference in the mean VAF of IDH1^MT vs IDH2^MT (Supplementary Figure S2A-C). We observed higher IDH1/2^MT VAFs in patients with ancestral IDH1/2^MT than subclonal IDH1/2^MT (Figure 3C and Supplementary Figure S2D). IDH1/2^MT VAFs were highest in patients with IDH1/2^MT of undeterminable ancestry, which may reflect a higher disease burden in patients when IDH1/2^MT cooperate with other mutations. WES analyses revealed that IDH1^MT and IDH2^MT were ancestral in 1/21 (5%) and 2/24 (8%) of IDH1/2^MT cases, respectively, whereas TS suggested that IDH1^MT and IDH2^MT were ancestral in 13/53 (25%) and 32/75 (43%) of cases (P=.0002). This difference probably reflects missed ancestral mutations in the TS samples (Supplementary Table S9). The clonal architecture of representative IDH1/2^MT patients (out of 60 studied) that were serially sequenced is shown in Figure 3D and Supplementary Figure S2E.

Figure 3. Analysis of clonal architecture of IDH1/2-mutated cases.

(A) Percentages of IDH1/2^MT cases in which the IDH1/2 mutation was an ancestral event, a subclonal event or an event of undeterminable ancestry, based on the variant allelic frequency of the IDH1/2 mutations and other co-occurring mutations. (B) As in (A), but with specific IDH1/2 amino acid substitutions. (C) Mean variant allelic frequencies of IDH1 mutations and IDH2 mutations in IDH1/2^MT cases in which the IDH1/2 mutation is an ancestral or subclonal event, or of undeterminable ancestry. (D) Fish plots of serially sequenced IDH1/2^MT patients.

In subclonal IDH1/2^MT cases, we investigated the corresponding ancestral events. These analyses showed heterogeneity, with 19 and 20 different ancestral mutations in 35 and 38 cases with subclonal IDH1^MT or IDH2^MT, respectively. Subclonal IDH1^MT were most often preceded by an ancestral DNMT3A mutation (30%; Supplemental Figure S2F), whereas ancestral RUNX1 mutations most frequently preceded a subclonal IDH2^MT (17%; Supplemental Figure S2G). We observed a RUNX1 mutation preceding a subclonal IDH1^MT only once. Hierarchical clonal analyses of neoplasms with indeterminable IDH1/2^MT clonal architecture (small VAF differences between IDH1/2 mutations and other mutations) may provide insight into which mutations co-operate (in the case of enrichment of co-occurring mutations; Figure 2) or compete (in the case of mutual exclusivity with co-occurring mutations) with IDH1/2^MT. DNMT3A mutations were most frequently the “main competitor” of IDH1^MT and IDH2^MT of undetermined ancestry (Supplemental Figure S2H-I). We did not observe a single RUNX1 mutation that competed with IDH2^MT, indicating that VAF differences between ancestral RUNX1 mutations and subclonal IDH2^MT, and vice versa, are rather large.

Prognostic effect of IDH1/2 mutations in myeloid neoplasms

In our cohort, IDH1/2^MT were associated with worse overall survival (Figure 4A). In subset analyses, we observed that IDH1/2^MT were significantly associated with reduced median overall survival in lower-risk MDS patients (41 vs. 66 months, P=.03), but not in higher-risk MDS (34 vs 30 months, P=.91), sAML (26 vs. 19 months, P=.74) or pAML patients (20 vs. 16 months, P=.79; Figure 4B-E). This finding further illustrates the role of IDH1/2^MT in progression to a more malignant disease. Patients with ancestral IDH1^MT tended to have a worse survival than patients with subclonal IDH1^MT (23 vs. 32 months, P=.09; Figure 4F), whereas there was no survival difference between patients with ancestral vs. subclonal IDH2^MT (33 vs. 30 months, P=.35; Figure 4G).

Figure 4. IDH1/2 mutations and associations with overall patient survival in myeloid neoplasms.

(A) Survival data of IDH1/2^MT patients with myeloid neoplasms. (B-E) Subset survival analyses in low-risk MDS (B), high-risk MDS (C), secondary AML (D) and primary AML (E). (F) Survival data of IDH1^MT patients with myeloid neoplasms, stratified on the ancestral importance of the IDH1 mutation. (G) as in (F), but with IDH2 mutations. (H) Subset survival analysis in low-risk MDS patients, with IDH1 and IDH2 mutations shown separately. (I) As in (F), but with IDH1 mutations in low-risk MDS patients.

Abbreviations: 1°, ancestral genetic event; 2°, subclonal genetic event; ?°, genetic event of undeterminable ancestry.

The association between IDH1^MT or IDH2^MT and overall survival was investigated separately in low-risk MDS to assess the impact on subsequent outcomes in early disease. Compared with IDH1/2^WT patients, we found that IDH2^MT patients (30 vs. 66 months, P=.003), but not IDH1^MT patients (42 vs. 66 months, P=.64), had a worse prognosis (Figure 4H). Because IDH1^MT were less frequently ancestral than IDH2^MT in lower-risk MDS patients (22% vs. 40%), we analyzed whether ancestry determines to what extent IDH1/2^MT impacts overall survival. Indeed, patients with ancestral IDH1^MT had a worse survival than patients with myeloid neoplasms that had a subclonal IDH1^MT (23 vs. 42 months, P=.05) or IDH1/2^WT patients (23 vs. 66 months, P=.006; Figure 4I). Ancestral IDH2^MT may also be associated with worse prognosis, compared with subclonal IDH2^MT. However, definitive conclusions from this subset analysis were hindered by the limited number of cases (Supplementary Figure S3).

DISCUSSION

In this report, we describe the clinical and molecular characterization of IDH1^MT and IDH2^MT patients with myeloid neoplasms. The size of the cohort allowed for the most comprehensive analysis to date of molecular, morphologic and clinical features associated with IDH1/2 mutations, separately and combined. For the first time, a comprehensive analysis of clonal architecture distinguished ancestral from subclonal somatic lesions and determined differences in their clinical and biological impact. We demonstrate that IDH1^MT and IDH2^MT can occur as ancestral or subclonal defects. In a substantial proportion of cases, IDH1^MT (19%) and IDH2^MT (34%) represent ancestral lesions, but more often IDH1/2^MT follow other ancestral mutations as subclonal events, likely explaining higher percentages of subclonal events in advanced myeloid disease. When present in lower-risk/early MDS, IDH1/2^MT are associated with a poor prognosis, while in higher-risk myeloid neoplasms, prognosis could not be further stratified. Patients with ancestral IDH1^MT exhibit worse survival than those with a subclonal IDH1^MT mutation, particularly in lower-risk MDS. Such a difference was not found when ancestral vs. subclonal IDH2^MT were compared, indicating that IDH2^MT more rapidly dominate the clonal hierarchy.

Analyses of mutations that positively or negatively correlate with IDH1/2^MT in myeloid neoplasms have heretofore treated IDH1/2^MT cases as a single functional entity. In our cross-sectional analysis, this approach of pooling IDH1/2^MT was only adequate for their association with NPM1 mutations, and their mutual exclusivity with TET2 mutations, as these are the only lesions that were significant for both IDH1^MT and IDH2^MT separately. In addition, IDH1^MT cases were significantly enriched for DNTM3A and PHF6 mutations, and IDH2^MT cases for ASXL1, SRSF2, RUNX1 and STAG2 mutations. When IDH1/2^MT cases would be pooled, all these six mutations are significantly more frequently occurring in IDH1/2^MT vs. IDH1/2^WT cases. This would falsely suggest that they occur more frequently in both IDH1^MT and IDH2^MT cases, whereas the significant correlation is only true for either IDH1^MT or IDH2^MT cases. “pAML-associated” DNMT3A and NPM1 mutations occurred significantly more frequently in IDH1^MT cases than in IDH2^MT cases, whereas the reverse was true for “MDS/sAML/(MPN)-associated” ASXL1, SRSF2, STAG2 and RUNX1 mutations. Of note, there were no significant proportional differences with respect to diagnosis between IDH1^MT and IDH2^MT cases, with only a trend (P=0.051) towards higher-risk MDS patients among IDH2^MT cases than among IDH1^MT cases. Apparently, the functional cooperation between certain mutations (e.g. DNTM3A mutations and IDH1^MT or RUNX1 mutations and IDH2^MT) in the clonal hierarchy is driving the aforementioned positive correlations without resulting in disease phenotypes that are typical for these lesions.

The differences in at least some of the biologic/clinical features between IDH1^MT and IDH2^MT cases may stem from the differences in the biochemical consequences of these lesions. For instance, the various IDH1/2^MT differ in D-2HG levels they produce, which is likely due to the impact of the specific amino acid substitutions on the catalytic site and thereby the conversion rate of αKG to D-2HG. Whereas IDH1 is cytoplasmic, IDH2 is mitochondrial and αKG levels may not be completely interchangeable between these subcellular compartments, generating different conditions for IDH1^MT and IDH2^MT to synthetize D-2HG. Furthermore, nuclear DNA and histone demethylases (e.g. TET2, Jumonji) and cytoplasmic prolyl hydroxylases (EGLN) are considered to be drivers of IDH1/2^MT-induced oncogenesis¹, suggesting that D-2HG mainly functions oncogenically outside mitochondria. It is unknown to what extent D-2HG passes the mitochondrial membrane. Therefore, the impact of the compartmentalization of αKG and D-2HG on the downstream effects of IDH2^MT vs. IDH1^MT is unclear.

D-2HG functions as an oncometabolite that inhibits various αKG-dependent dioxygenases. Whereas the effects of D-2HG on DNA demethylase TET2, histone demethylase JumonjiC and the HIF1α degrader EGLN have been described thoroughly^{9, 26, 27}, there are in fact over 60 different αKG-dependent human dioxygenases involved in a plethora of cellular functions that may be inhibited, or possibly activated, as in the case of EGLN, by D-2HG. All these enzymes have specific IC₅₀ values of D-2HG for inhibition of αKG-dependent enzymes. Thus, each D-2HG concentration (i.e. each IDH1/2^MT variant) is expected to inhibit a specific subset of αKG-dependent dioxygenases and alter cellular functions in a variety of ways.

It has been proposed that the αKG-dependent DNA demethylase TET2 is one of the most important downstream targets of IDH1/2^{MT 26, 28}. D-2HG inhibits TET2 and results in a genome-wide DNA hypermethylation in IDH1/2^MT cancers that induces stemness and inhibits differentiation⁷. In our cohort, and that of others²⁹, IDH1/2 and TET2 mutations are mostly mutually exclusive, supporting the notion of similar cellular downstream effects. TET2 mutations, albeit rare, were more frequent in IDH1^MT cases (14%) than in IDH2^MT cases (4%). It is possible that IDH1^MT result in less TET2 inhibition than IDH2^MT and that a synergistic/additive effect of co-occuring IDH1 and TET2 mutations augments TET2 inhibition. Of note, TET2 is widely held to be a downstream element of the pathogenic cascade induced by IDH1/2^MT. Despite this, the profoundly distinct nosologic and morphologic spectra associated with TET2 and IDH1/2 mutations speak against this notion. Another putative important downstream target of IDH1/2^MT is UTX/KDM6A^{7, 27}. However, we did not observe mutual exclusivity between IDH1/2 and UTX mutations, suggesting that D-2HG-mediated UTX inhibition has a different effect than the UTX mutations that we observed. Notably, we did not observe a negative correlation between IDH1/2 and WT1 mutations, as previously reported³⁰.

Our study provides the first comprehensive subclassification of IDH1/2^MT cases with myeloid neoplasms based on their rank within clonal hierarchy, and thus their timing in clonal ontogeny. While cases initiated by IDH1/2^MT are a distinct subset of myeloid neoplasms, in many instances IDH1/2^MT are subclonal. The impact of the corresponding ancestral lesions, then, may more profoundly shape the individual neoplastic biology. This observation has clinical implications because, at least theoretically, therapeutic targeting of subclonal lesions is less likely to be curative as the ancestral clone cannot be eliminated. Remissions achieved in such cases thus have a greater likelihood of relapse, and durable clinical benefit may depend on combination therapies that also target the ancestral event(s). In addition, we report that ancestral IDH1/2^MT are related to poor clinical outcomes, whereas this association is weaker for subclonal IDH1/2^MT. These findings illustrate that patients with myeloid neoplasms carrying ancestral IDH1/2^MT are the best candidates for therapy with IDH1/2^MT inhibitors. Thus, sequencing that targets mutations frequently co-occurring with IDH1/2^MT, in particular those which are often ancestral may aid rational clinical treatment of IDH1/2^MT patients with IDH1/2^MT inhibitors.

IDH1/2^MT occur in a myriad of cancer types. Their biology in glioma and acute myeloid leukemia has been studied most intensively. The distribution of IDH1/2^MT differs between gliomas and myeloid neoplasms, and fundamental differences exist in IDH1/2^MT biology between these 2 tumor types. Whereas IDH1/2^MT are very early events in gliomagenesis and has even been proposed as a canonical ancestral event^{31, 32}, our findings show that this is in stark contrast to myeloid neoplasms, in which IDH1/2^MT are ancestral in a minority of cases. In addition, IDH1/2^MT associate with a strikingly prolonged overall survival in glioma^{25, 33}, whereas our study, and that of others³⁴, showed that these mutations are associated with worse prognosis in myeloid neoplasms.

In summary, we present results that demonstrate distinct differences between IDH1^MT and IDH2^MT. The position in the clonal hierarchy may be important for understanding the impact on the biology and clinical consequences of these mutations, and may refine future treatment of IDH1/2^MT myeloid neoplasms with IDH1/2^MT inhibitors. We conclude that the minority of patients with myeloid neoplasms carrying ancestral IDH1/2^MT are the best candidates for therapy with IDH1/2^MT inhibitors.