Unraveling Immunogenomic Features Germane to Pathobiology of Myelodysplastic Syndromes

Study of the immunologic features of cancer as well tumor immune surveillance enabled introduction of cancer immunotherapies. However, if compared to the dramatic changes obtained in solid tumors, comparable development of immune therapies in myeloid neoplasia (MN) has been lagging. Latest evidence suggests that throughout MDS evolution immune surveillance reactions may contribute to cytopenia by collateral inhibition of normal hematopoiesis, leading to emergence of leukemogenic clones eliciting a diverse immune responsiveness. Here, we aimed to unravel immunogenomic features of MDS pertinent to disease biologic and clinical behavior.

As cellular immune responses are most important in controlling neoplastic clones, we explored HLA and T cell receptor (TCR) immunogenomics of 101 diagnostic MDS samples [N=51 LR-MDS, N=32 HR-MDS, N=18 secondary AML-sAML)], including hypocellular MDS (hMDS, N=9) and MDS with co-occurring TLGL (N=5), wherein the contribution of immune forces appears to be a prominent feature.

Sequencing of HLA alleles revealed the presence of somatic mutations and losses in 14 patients (1 with subclonal clonal mosaicism) encompassing all spectra of the disease (20% in MDS-LGL, 10% in LR-MDS, 22% in hMDS, 10% and 28% in HR-MDS and sAML respectively). Both class I (20%) and II (80%) loci were affected. In addition to genetic adaptation to immune pressure, we have also studied HLA expression on myeloid blasts (CD45+/CD34+) by flow cytometry as an adaptive mechanism of HLA antigen attenuation. Benchmarking on HC, low (<25% of HC) expression of HLA-C and -DQ was noted (69% vs 27% and 66% vs 18%, respectively; both p<0.01). These findings prompted us to explore the presence of an underlying immunogenetic pressure contributing for such MN immune evasion/editing. Considering HLA evolutionary divergence (HED) a surrogate for immunopeptidome presentation capacity inherently linked to human ability to curtail cancer or incite autoimmunity, we noticed higher values of HED in HLA-C locus hMDS and LGL-MDS vs HR-MDS (p= 0.0068) and sAML (p= 0.0409).

Next, by deep TCR sequencing, we noticed a significantly higher degree of VB CDR diversity, as measured by the inverse Simpson index, in MDS patients compared to HC (p=1.5e-07). Notably, the maximal productive frequency, indicating the extent of T-cell expansion, was significantly higher in patients with hMDS and MDS-LGL compared to HR-MDS (p=0.0079). Additionally, intersample comparisons demonstrated a substantial overlap in TCR clonotypes among MDS cases, unlike the diversity seen in HC.

Next, we conducted an in-depth analysis focusing on the physiochemical properties of amino acids of TCR products, hypothesizing their influence on disease states and phenotypes. To test this, we compiled 36 continuous descriptors for naturally occurring amino acids, utilizing KIDERA factors, ST Scales, VHSE, and BLOSUM vectors. By employing a convolutional neural network (encoder-decoder architecture), we generated latent vector representations for each individual T-cell clone, capturing their inherent structural features. These clone representations were then aggregated per sample and weighted by the expansion size of each clone. Notably, significant differences in these latent VB CDR3 representations were observed between sAML/HR-MDS vs LR-MDS, but could not distinguish HC, mostly clustering within LR-MDS.

To further integrate structural features with additional HLA parameters, we developed a model incorporating such significant TCR latent features with HED scores, and the inverse Simpson index as a measure of TCR repertoire diversity. Through cross-validation and lasso regularization for feature selection, we found that HED of DPB1 and DRB1, along with structural features, had relevant significance in MDS clustering. With such parameters we then built a model demonstrating a marked improvement in classifying sAML/HR-MDS vs LR-MDS, compared to models focusing solely on HED and TCR repertoire diversity (cv-accuracy 0.85 vs 0.6).

In conclusion, HLA mutations/downmodulation and later immune evasion by HLA deficient clones may reflect the stigmata of an underlying immune pressure, possibly participating in suppression of non-dysplastic hematopoiesis. On the effector site, TCR structural features vary within the disease risk spectrum and may contribute to the clinical-molecular heterogeneity traditionally ascribed to MDS.