Noninvasive Prediction of Graft-Verus-Host Disease Following Allogeneic Hematopoietic Cell Transplantation by Gene Expression Profiling

Chronic graft-versus-host disease (cGvHD) and immunosuppressive medications for its treatment result in significant morbidity and mortality, limiting the clinical benefit of allogeneic hematopoietic cell transplantation (HCT). A noninvasive, diagnostic and prognostic test is needed to determine which patients have active cGvHD at present and which subset will have active cGvHD in the future.

We performed peripheral blood gene expression profiling to identify an array of transcriptional biomarkers that discriminate presence of active cGvHD requiring immunosuppression following allogeneic HCT. Peripheral blood mononuclear cells were prepared by Ficoll gradient and cryopreserved from 63 patients (median age 50 yrs; range 19–64) following allogeneic HCT (median 475 days post-HCT) on an IRB approved protocol. Samples were randomly selected into a training set (23 cGvHD and 19 without cGvHD) and test set (12 cGvHD and 9 without cGvHD) and processed on the Agilent whole human genome 44k microarray platform. Bioinformatics programs including AILUN, GeneSpring, SAM, and PAM were used to select a minimum gene-set (false discovery rate, FDR<5%) to predict cGvHD in the training set. Patient characteristics in the two groups were collected for confounder/interaction analysis across the following parameters: age, gender, disease histology, disease status at HCT, graft source, conditioning regimen, white blood count (WBC) at sample, and follow-up status including relapse and survival. Type, grade, and activity of acute GvHD and cGvHD at time of sample and at most recent follow-up were determined by two independent investigators by standard clinical criteria.

There is considerable heterogeneity among cGvHD patients, not attributable to multiple confounding factors, suggesting molecular heterogeneity in the development and progression of cGvHD. A minimum of 10 unique genes selected from SAM and PAM predict active cGvHD at the time of the peripheral blood sample with 95% sensitivity and 78% specificity in the training set, and 75% sensitivity and 78% specificity in the test set. Eight and four of ten genes are enriched in NK and CD4 T cells, respectively, and involved in the IL-10 signaling pathway (p<0.01). The 10 gene-set predicted the presence of active cGvHD at last follow-up (median 849 days post-HCT) with 85% sensitivity and 83% specificity.

Peripheral blood gene expression across a 10 gene-set shows strong correlation of the predictive probability of cGvHD and the clinical phenotype. Given reasonable diagnostic sensitivity and specicifity for prediction of future cGvHD, additional longitudinal studies of this cohort and a prospective cohort are underway to further improve the accuracy of this approach and predictive power of this gene-set.

No relevant conflicts of interest to declare.