A molecular score by quantitative PCR as a new prognostic tool at diagnosis for chronic lymphocytic leukemia patients

Abstract

Background: Several markers have been proposed to predict the outcome of chronic lymphocytic leukemia (CLL) patients. However, discordances exist between the current prognostic factors, indicating that none of these factors are totally perfect.

Methodology/principal findings: Here, we compared the prognostic power of new RNA-based markers in order to construct a quantitative PCR (qPCR) score composed of the most powerful factors. ZAP70, LPL, CLLU1, microRNA-29c and microRNA-223 were measured by real time PCR in a cohort of 170 patients with a median follow-up of 64 months (range3-330). For each patient, cells were obtained at diagnosis and RNA was extracted from purified CD19 cells. The best markers were included in a qPCR score, which was thereafter compared to each individual factor. Statistical analysis showed that all five RNA-based markers can predict treatment-free survival (TFS), but only ZAP70, LPL and microRNA-29c could significantly predict overall survival (OS). These three markers were thus included in a simple qPCR score that was able to significantly predict TFS and OS by dividing patients into three groups (0/3, 1-2/3 and 3/3). Median TFS were >210, 61 and 24 months (P<0.0001) and median OS were >330, 242 and 137 months (P<0.0001), respectively. Interestingly, TFS results were also confirmed in Binet stage A patients (P<0.0001). When compared to other classical factors, this score displays the highest univariate Cox hazard ratio (TFS: HR=9.45 and OS: HR=13.88) but also provides additional prognostic information.

Conclusions: In our hands, this score is the most powerful tool for CLL risk stratification at the time of diagnosis.

Figure 1. Kaplan-Meier survival curves for TFS/OS of RNA-based markers.

TFS and OS were plotted using Kaplan-Meier methods, and curves were compared with the log-rank test (n = 170). ZAP70 (A and F), LPL (B and G), CLLU1 (C and H), microRNA-29c (D and I) and microRNA-223 (E and J). All cut-offs were determined using ROC curve analysis. Further information can be found in Table 1.

Figure 2. qPCR combining ZAP70, LPL and microRNA-29c stratifies CLL patients in terms of TFS and OS.

TFS and OS, according to our qPCR score, were plotted with Kaplan-Meier methods for all Binet stages (A and B) and only Binet stage A (C and D). Curves were compared with the log-rank test (all Binet stages, n = 170; Binet stage A, n = 124). The hazard ratio (HR) was calculated with univariate Cox regression. Further information can be found in Table 1.

Figure 3. qPCR can fine tune prognosis in good or poor prognostic subgroups.

The qPCR score was applied to different subgroups (previously divided by other prognostic factors). TFS curves are shown for LPL (A, F), CD38 (B, G), LDT (C, H), β2-M (D, I), and cytogenetic abnormalities (E, J). The median TFS and patient numbers in the different groups are provided in Table 2. Groups of less than two patients are not represented in this figure.

Figure 4. Supplementary prognostic value of the qPCR score.

Each prognostic factor was used to divide the patient cohort in two different prognostic subgroups according to prognostic factors reported in Table 1. The qPCR score was then applied to all poor prognosis subgroups and good prognosis subgroups. (A) and (B) show the median TFS and median OS of the good prognosis subgroups, whereas (C) and (D) show the median TFS and median OS of the poor prognostic subgroups. The dotted line and error bar represent the mean and the SEM, respectively. Statistical differences were assessed using the Kruskal-Wallis test. Parts (E) and (F) show forest plots comparing the univariate Cox HR of the qPCR score with other prognostic factors for TFS and OS prediction, respectively. The hazard ratio (HR) of all variables was calculated by univariate Cox analysis and plotted with the 95% Cl on this forest plot. More details can be found in Table 2.