Mutational Landscape of the Transcriptome Offers a Rich Neoantigen Resource for Immunotherapy of Myeloproliferative Neoplasms

Background

Myeloproliferative Neoplasms (MPN) encompass several disease subgroups (ET, PV, and PMF) with distinct phenotypic and clinical features. Although the landscape of somatic mutations in MPN has been mapped in detail, the current therapies show limited efficacy in eliminating MPN cells. Immunotherapy for MPN have been suggested as an alternative to standard therapies, however, a broader view on the neoantigen landscape and MHC class I (MHCI) restriction of putative MPN neoantigens is missing.

Aims

We aimed to map the mutation and neoantigen landscape of MPN by analyzing whole transcriptome data and examine the applicability of targeted immunotherapy by analyzing co-occurrence of specific neoantigens with high affinity MHCI molecules in individual patients.

Methods

We have performed whole transcriptome sequencing of granulocytes for 113 MPN patients (32 ET, 17 PV, 55 PMF, and 9 post-MPN AML) and 15 controls. We implemented customized and established algorithms for genome wide fusion gene discovery, variant calling on 87 myeloid relevant genes, detection of splicing abnormalities and inferred these mutation classes directly from RNA-seq data for each patient. We extracted HLA genotypes from the same RNA-seq dataset. We examined the impact on protein sequence for each mutation class and predicted peptide affinities to MHCI. Top predicted peptide:MHCI interactions were validated in vitro using the UV-induced peptide exchange method.

Results

Following data processing and filtering, we identified a total of 13 gene fusions, 231 non-synonymous SNVs, and 21 Indels in 106/113 patients. Fusions were validated with Sanger sequencing, for SNVs and Indels, we re-sequenced DNA of 77 patients using a TruSight targeted sequencing panel. An unexpected high frequency of SF3B1 (0% ET, 14.6% PMF, 0% PV) hotspot mutations (K666N/R/T, K700E) was identified. We performed differential splice junction analysis comparing SF3B1 mutated and non-mutated patients and found 895 significantly differentially spliced junctions (750/895 were mediated through alternative 3'splicing). For each junction we calculated the percent splice in value (PSI) expressing the relative abundance of the alternative junction. We identified 20 junctions with retained intronic sequences with PSI>20%. For each of the 20 splicing defects we evaluated the impact of intron retention on the protein sequence. These splicing defects caused insertion of novel amino acids either in frame or through frameshifts. From all the mutations detected, we generated a virtual peptide library. In 113 patients, we detected 541 patient specific peptides predicted to bind patients' own MHCI proteins of which 102 had a predicted strong and 439 weak binding affinity. The recurrent CALR frameshifts and MPL-W515K/L/A mutations were also a rich resource for potential neoantigens due to their unique property to bind many shared common MHCI molecules (42, and 17 predicted for CALR and MPL, respectively). SF3B1 mutated patients had the highest number of predicted recurrent neoantigens with an average of 38 weak and 16 strong peptide/MHCI pairs. In total, we could identify 149 unique neoantigens covering 62% of MPN patients as a potential target in personalized cancer immunotherapy approaches. Of the 149 predicted unique neoantigens we tested 35 top ranking peptides for binding to recombinant MHCI monomers. We validated 70% of these peptides to be strong MHCI binders.

Conclusions

Our results provide a framework for systematic mining of neoantigens from different mutation classes using RNA analysis. The data presented in this study may serve as a resource for development of personalized vaccines or adoptive cell-base therapies, in particular for PMF patients positive for CALR, MPL and SF3B1 mutations.