Disease phase, transplant donor type, donor recipient match, age, and interval from diagnosis to transplantation (the EBMT risk score) are recognized variables that affect transplant outcomes for chronic myeloid leukemia (CML), but do not entirely account for the heterogeneity in outcomes. We have previously applied a probabilistic method to a large CML microarray gene expression dataset, and found a 6-gene signature of disease phase that discriminated between early and late chronic phase (CP). The combined expression of all 6 genes could be represented as a probability score where values closer to 0 are more similar to CP and values closer to 1 are more similar to blast crisis (BC). Moreover, in 17 accelerated phase (AP) CML patients, the 6-gene probability score was associated with outcomes after transplantation. We thus hypothesized that genes predictive of CML progression could be used to predict outcomes after transplantation regardless of disease phase. We derived 6 additional models (i.e. gene sets) from our CML microarray data, each consisting of 6–10 genes (total of 35 genes), that are also highly predictive of CML progression. These gene sets were derived using a novel network-driven approach aimed to identify genes that are functionally related to genes in pathways that are known to be associated with CML. We then examined expression of the genes in these models using quantitative PCR in bone marrow samples from 213 patients (176 CP, 23 AP, and 14 BC remission patients) prior to myeloablative allogeneic transplantation. GUSB was used as an endogenous control to correct for RNA integrity. Transplants occurred between 1993 and 2007 and a majority of patients did not receive prior tyrosine kinase inhibitor therapy. For CP patients, gene expression for all genes and models was independent of white blood cell and blast count. Among 176 CP CML patients, 45 patients died and 24 patients relapsed by last contact, leading to 1-year and 5-year estimates of overall survival of 85% and 78%, respectively, and 1-year and 5-year estimates of relapse of 7% and 12%, respectively. In CP patients we found not only that the expression of the original six-gene model (NOB1, DDX47, CD101, LTB4R, SCARB1, SLC25A3) was associated with a trend towards increased relapse, but that another model (RALGDS, LASP1, G6PD, ADRBKI, LRPPRC, PSMA1) was statistically significantly associated with an increased risk of relapse. In CP patients we found that an increase of 0.2 in the 6-gene probability score correlated with an increase in relapse of 46% (HR=1.46 (1.06-2.02, p=.02)) after adjustment for EBMT risk score (Figure 1a ). Lastly, we also found that, individually, several of our progression-associated genes were statistically significantly associated with overall survival (G6PD and CAMK1D (Figure 1b )), relapse (RAC2 and ADRBK1), and non-relapse mortality (G6PD, CIQBP, and CAMK1D). In conclusion, these data suggest that gene expression prior to therapy is associated with treatment outcomes even after considering the contribution from known risk factors. These data provide evidence that a molecular signature associated with disease progression when detected in CP patients drives outcomes after transplantation. Given that all treatment outcomes are dependent on phase, it is possible that the expression of these genes prior to tyrosine kinase inhibitor therapy may also predict response.
Oehler:Pfizer: Research Funding. Radich:Novartis: Consultancy, Honoraria, Research Funding; Pfizer: Consultancy, Honoraria; Bristol-Myers Squibb: Consultancy, Honoraria.
Asterisk with author names denotes non-ASH members.