Deep learning methods for digital pathology analysis are an effective way to address multiple clinical questions, from diagnosis to prediction of treatment outcomes. These methods have also been used to predict gene mutations from pathology images, but no comprehensive evaluation of their potential for extracting molecular features from histology slides has yet been performed. We show that HE2RNA, a model based on the integration of multiple data modes, can be trained to systematically predict RNA-Seq profiles from whole-slide images alone, without expert annotation. Through its interpretable design, HE2RNA provides virtual spatialization of gene expression, as validated by CD3-and CD20-staining on an independent dataset. The transcriptomic representation learned by HE2RNA can also be transferred on other datasets, even of small size, to increase prediction performance for specific molecular phenotypes. We illustrate the use of this approach in clinical diagnosis purposes such as the identification of tumors with microsatellite instability.

No abstract

Summary Somatic mutations in IDH1/2 and TET2 result in impaired TET2 mediated conversion of 5-methylcytosine (5-mC) to 5-hydroxymethylcytosine (5-hmC). The observation that WT1 inactivating mutations anti-correlate with TET2/IDH1/2 mutations in AML led us to hypothesize that WT1 mutations may impact TET2 function. WT1 mutant acute myeloid leukemia (AML) patients have reduced 5-hmC levels similar to TET2/IDH1/2-mutant AML. These mutations are characterized by convergent, site-specific alterations in DNA hydroxymethylation, which drive differential gene expression more than alterations in DNA promoter methylation. WT1 overexpression increases global levels of 5-hmC, and WT1 silencing reduced 5-hmC levels. WT1 physically interacts with TET2 and TET3, and WT1 loss of function results in a similar hematopoietic differentiation phenotype as observed with TET2 deficiency. These data provide a novel role for WT1 in regulating DNA hydroxymethylation and suggest that TET2 IDH1/2, and WT1 mutations define a novel AML subtype defined by dysregulated DNA hydroxymethylation.

Although the majority of acute myeloid leukemia (AML) patients initially respond to chemotherapy, many patients subsequently relapse; the mechanistic basis for AML persistence following chemotherapy has not been delineated. Recurrent somatic mutations in DNA methyltransferase 3A (DNMT3A), most frequently at arginine 882 (DNMT3Amut), are observed in AML1–3 and in individuals with clonal hematopoiesis in the absence of leukemic transformation4,5. DNMT3Amut AML patients have an inferior outcome when treated with standard-dose daunorubicin-based induction chemotherapy6,7, suggesting that DNMT3Amut cells persist and drive relapse8. Here we show that Dnmt3amut induces hematopoietic stem cell (HSC) expansion, cooperates with Flt3ITD and Npm1c to induce AML in vivo, and promotes resistance to anthracycline chemotherapy. In AML patients, DNMT3AR882 mutations predict for minimal residual disease (MRD), underscoring their role in AML chemoresistance. DNMT3Amut cells show impaired nucleosome eviction and chromatin remodeling in response to anthracyclines, resulting from attenuated recruitment of histone chaperone SPT-16 following anthracycline exposure. This defect leads to an inability to sense and repair DNA torsional stress, which results in increased mutagenesis. Our studies identify a critical role for DNMT3AR882 mutations in driving AML chemoresistance, and highlight the importance of chromatin remodeling in response to cytotoxic chemotherapy.

TET2 converts 5-methylcytosine to 5-hydroxymethylcytosine (5-hmC) in DNA and is frequently mutated in myeloid ma-lignancies, including myeloproliferative neoplasms. Here we show that the level of 5-hmC is decreased in granulocyte DNA from myeloproliferative neoplasm patients with TET2 mutations compared with granulocyte DNA from healthy patients. Inhibition of TET2 by RNA interference decreases 5-hmC levels in both human leukemia cell lines and cord blood CD34 cells. These results confirm the enzymatic function of TET2 in human hematopoietic cells. Knockdown of TET2 in cord blood CD34 cells skews progenitor differentiation toward the granulo-monocytic lineage at the expense of lym-phoid and erythroid lineages. In addition, by monitoring in vitro granulomonocytic development we found a decreased granulocytic differentiation and an increase in monocytic cells. Our results indicate that TET2 disruption affects 5-hmC levels in human myeloid cells and participates in the pathogenesis of my-eloid malignancies through the disturbance of myeloid differentiation. (Blood. 2011;118(9):2551-2555)