Abstract
Morphology has been the pillars of the MDS diagnosis. In the genetic era, the subjectivity of pathologic evaluation in the absence of integration with mechanistic correlations is a clear limitation of current diagnostic schemes. Well-known genotype/phenotype associations, including e.g., linking of SF3B1 mutations to RARS or JAK2/SF3B1 mutations to RARS-T, suggest that somatic events can shape morphologic features and vice versa. However, the complexity of morphologic and genetic changes precludes identification of many consequential genotype/phenotype associations. As objective image analytic tools are being developed, this project aims at determining the most mutually predictive relationships between combinations of morphologic features and genomic changes.
We have analyzed 1,079 MDS patients for somatic CNVs and mutations in a targeted panel of 33 genes frequently mutated in myeloid cancers. Our stringent bioanalytic pipeline removed artifacts, SNPs and errors. We applied this pipeline to discovery (2/3) and validation (1/3) cohorts.
Bone marrow morphological alterations were mapped to binary features called by an independent pathologist in a blinded fashion based on uniformly defined criteria occurring in >10% of cells. A total of 10 such features were investigated. For instance, myeloid, erythroid and megakaryocytic dysplasia occurred in 54%, 70% and 72% of patients, respectively. 89% had at least one cytopenia, 57% multiple cytopenias, and 50% had at least one myeloproliferative feature (e.g., monocytosis). In addition, all cases were partitioned in accordance with two risk groups defined by IPSS-R (lower risk <3.5 and higher risk >3.5).
NGS analysis identified 1,929 somatic mutations, but for proper correlation with morphologic definitions only mutations with a clonal burden >10% were used. Our initial univariate analysis yielded 52 significant associations (q<.1). For instance, myeloid dysplasia was more associated with ASXL1, NRAS,SRSF2, STAG2, and TET2 mutations. Univariate associative analyses failed to yield signatures due to the extensive presence of the heterogeneity in genotype profiles. Noting patterns of interdependence exhibited by the 24 individual morphologic features, an unsupervised cluster analysis based on the consensus clustering method was used to identify intrinsic patterns of co-occurrence and morphologic subtypes of MDS. A semi-supervised machine learning (ML) technique based on Bayesian partial exchangeability investigated the extent to which morphologic subtypes of MDS can be discriminated on the basis of patterns of mutation incidence and co-occurrence. To augment existing clinical models of prognosis, ML was applied to low risk patient MDS subtypes. A single model, selected and subsequently validated using our independent test set, yielded genetic signatures demonstrating morphologic orientation. An unsupervised cluster analysis revealed 5 discrete morphological MDS groups on the basis of 24 features (Fig.1a). High-risk subtypes were clustered in group 1 (G1), whereas low risk subtypes clustered in the other 4 groups, each of which had unique prominent morphological features (Fig.1b). G2 had pancytopenia, G3 had monocytosis, G4 had elevated megakaryocytes, and G5 had erythroid dysplasia. Exhibiting prognostic utility, overall survival differed significantly among these groups (Fig.1c). Mutational frequencies also differed between these groups, e.g., G1 frequently had mutations of RUNX1, TP53, and STAG2. 8 genetic signatures characterize the morphologic phenotypes of low risk MDS patients (S1-8; Fig.1d). Signature-A (SA) was enriched for TET2 mutations, SB had the co-occurrence of TET2 and SRSF2 mutations, whereas SG was enriched in SF3B1 mutations. Deviation in survival (p<0.026) demonstrated the prognostic value of these genetic signatures among low risk patients. The genetic signatures were predictive of specific morphological features (Fig.1e). SB dominated G3 (73%), and SC were frequent for G4 (57%). Among 18 morphologic/genetic associations identified in the discovery set, 9 were confirmed in the validation cohort. For instance, we validated that SB was enriched in monocytosis and TET2 mutations; while SD in anemia and SF3B1 mutations. This is the first comprehensive analysis that links somatic molecular lesions to morphological patterns to identify distinct phenotype/genotype associations.
Disclosures
Nazha: Jazz Pharmacutical: Research Funding; Abbvie: Consultancy; MEI: Other: Data monitoring Committee; Novartis: Speakers Bureau; Daiichi Sankyo: Consultancy; Tolero, Karyopharma: Honoraria; Incyte: Speakers Bureau. Sekeres:Celgene: Membership on an entity's Board of Directors or advisory committees; Millenium: Membership on an entity's Board of Directors or advisory committees; Syros: Membership on an entity's Board of Directors or advisory committees. Hobbs:Amgen: Research Funding; SimulStat Inc.: Consultancy. Maciejewski:Novartis: Consultancy; Alexion: Consultancy.
