Bone marrow microenvironment overlap between vexas and myelodysplastic syndrome demonstrated by targeted transcriptomic and artificial intelligence Free

Introduction: VEXAS (vacuoles, E1 enzyme, X-linked, autoinflammatory, somatic) syndrome is an inflammatory process that involves bone marrow (BM). Active inflammation is also a major feature of the microenvironment in the BM of patients with myelodysplastic syndrome (MDS). To explore the overlap between VEXAS and MDS, we used an artificial intelligence (AI) approach and compared the transcriptomic profile of the BM from VEXAS patients (N=59) with the BM from MDS patients (N=1021) and the BM of individuals without specific diagnostic abnormality or with CHIP (clonal hematopoiesis of indeterminate potential) (N=1030) (Normal).

Methods: RNA was extracted from the bone marrow samples from patients with VEXAS (N=59), MDS (N=1030), and Normal (1030). The RNA was sequenced by next generation sequencing (NGS) using a targeted RNA panel of 1600 genes. Hybrid capture sequencing library preparation was used, and RNA was quantified using transcript per million (TPM). To compare between the three groups, we used the RNA expression levels in an AI model based on Bayesian statistics and random forest. Bayesian statistics were used to rank the genes that distinguish between groups, then random forest was used to establish the signatures that distinguish between the groups. Two thirds of the samples were used for training and one third was used for testing the models.

Results: Using the AI model described above, the BM of patients with VEXAS was significantly different from normal. Only 8 genes (SF3A1, SMAD5, PRPF8, MCM3AP, NFYC, FGFR1OP, PPM1D, THRA) were adequate to distinguish VEXAS from normal with AUC of 0.910 (95% CI: 0.0867-0.953)(A). Except for PPM1D, all genes were at significantly higher levels in VEXAS while PPM1D expression was significantly lower in VEXAS. Three of the 8 genes are involved in RNA splicing and 3 in DNA damage repair. Of the tested 1600 genes, 453 genes showed significant (LogFDR <-3) difference in the level of expression between VEXAS and normal. In contrast, only 152 genes were significantly different (LogFDR <-3) between MDS and VEXAS. Using the same AI model, 40 genes were needed to distinguish VEXAS from MDS (AUC : 0.905, 95% CI: 0.0861-0.950)(B). All genes that were significantly abnormal in VEXAS as compared with normal were also significantly abnormal in MDS as compared with normal. The difference was in the same direction in that those with high expression in VEXAS vs. normal were also high in MDS vs. normal and vice versa.

Conclusions: These findings suggest that while the presenting symptoms in VEXAS are systemic inflammatory processes involving lungs, skin, blood vessels, and joints, there are significant abnormalities in bone marrow microenvironment mimicking the abnormalities seen in MDS. The demonstration that changes in splicing and DNA repair gene are the major molecular abnormalities that distinguish VEXAS BM as compared to normal suggests that therapeutic approaches targeting these abnormalities may help in ameliorating the clinical manifestation of this disease.