correspondence

Is ZAP70 still a key prognostic factor in early stage chronic lymphocytic leukaemia? Results of the analysis from a prospective multicentre observational study

Early studies revealed that ZAP70 is differentially expressed in IGHV-unmutated (UM) versus IGHV-mutated (M) chronic lymphocytic leukaemia (CLL) patients (Crespo et al, 2003) and that an increase in ZAP70 mRNA or ZAP70 protein expression could serve as a surrogate marker for IGHVUM status (Rosenwald et al, 2001). ZAP70 expression has been reported to be a valuable prognostic factor in CLL (Crespo et al, 2003; Morabito et al, 2009). Nevertheless, detection methods for ZAP70 remain to be fully standardized. Notably, the characteristics of ZAP70 make its detection rather difficult even by flow cytometry (FC) (Sloan-Lancaster et al, 1997; Best et al, 2006). Moreover, evidence that ZAP70 expression varies during disease course raises the issue of whether its expression can successfully predict outcome in early disease stages (Chaar et al, 2008). On this basis, we undertook a prospective multicentre study of 487 Binet A patients (O-CLL1 protocol, clinicaltrial.gov identifier NCT00917540) in whom ZAP70 expression was investigated at diagnosis and during follow-up using different detection methods. Diagnosis was confirmed by the study review committee according to FC analysis centralized at the National Institute of Cancer Laboratory in Genoa, where the determination of ZAP70 and CD38 expression and IGHV mutational status was carried out as previously described (Morabito et al, 2009). For ZAP70 determination, cells were stained for membrane CD19-PE, CD3PE-CY7 and CD5-APC (Becton Dickinson, Milan, Italy), fixed and permeabilized with Fix and Perm reagents (Caltag Laboratories, Buckingham, UK) and incubated with 1
5 lg of anti-ZAP70– FITC conjugate monoclonal antibody (clone 2F3.2 Upstate? Merck Millipore, Darmstadt, Germany). The percentage of ZAP70-positive CLL cells was determined with respect to two different setting controls. The first method was based on the signal obtained using an isotype-matched antibody as negative control (ZAP70 I/C). The second method used the expression of ZAP70 in normal T cells (ZAP70 T) as an internal positive control. ZAP70 expression was also determined as a mean fluorescence intensity ratio between T- and B-cells (ZAP70 T/B) (Rossi et al, 2010). Stained cells were analysed with a FACSCAlibur flow cytometer (Becton Dickinson). We internally validated our results by splitting the dataset into training and validation sets: the complete modelbuilding process applied to the training cohort to obtain the ª 2014 John Wiley & Sons Ltd, British Journal of Haematology

final prediction model was then repeated using the validation sample only, and validated (Tripepi et al, 2013). The two sets did not differ in terms of marker distribution and median follow-up. Receiver-operated characteristic analysis identified 40% as the most suitable cut-off value for ZAP70 I/C and ZAP70 T to distinguish IGVH-M from IGVH-UM patients, while 1
5 represented the best threshold ratio for ZAP70 T/B. Cases were also divided according to other proposed thresholds, such as 20% and 30% for ZAP70 I/C and ZAP70 T, and 2 and 2
5 for ZAP70 T/B) (Crespo et al, 2003; Morabito et al, 2009; Rossi et al, 2010). Figure 1 shows Forest plots of univariate Cox analyses of the three different ZAP70 detection methods for both training (Fig 1A) and validation (Fig 1B) sets. Specifically, in the training set, the ZAP70 I/C 20% threshold remained non-significant, the ZAP70 T threshold was significant at 40%, and the ZAP70 T/B was significant at ≥1
5 and >2. In the validation set, ZAP70 I/C 30% was no longer significant, ZAP70 I/C 40% and 40% ZAP70 T remained significant, and ZAP70 T/B ≥ 1
5 retained significance for both methods. Therefore, ZAP70 I/C 40%, ZAP70 T 40% and ZAP70 T/B ≥ 1
5 thresholds significantly predicted time to first treatment (TTFT) in both the training and validation sets. Furthermore, we stratified patients according to the number of positive scores for ZAP70 (i.e. ZAP70 I/C 40%, ZAP70 T 40% and ZAP70 T/B 1
5) obtained with the different methods (i.e. positive for 3, 2 or 1 methods), showing that the prediction power was more evident in those cases found positive by all three methods (Fig 1C). Overall, these data underscore the difficulty in providing reliable results across methods, mainly in the positive/negative grey zones. Finally, in a Cox multivariate analysis (model 1) in which ZAP70 I/C 40%, ZAP70 T 40% and ZAP70 T/B 1
5 thresholds were forced, the latter remained the sole independent variable maintaining association with TTFT [hazard ratio (HR) 0
55, 95% confidence interval (CI) 0
31–0
96, P = 0
037]. However, when all ZAP70 procedures were forced together with IGHV mutational status (model 2), ZAP70 T/B 1
5 lost its predictive power and IGHV mutational status remained the sole independent variable association with TTFT (HR 4
3, 95% CI 2
5–7
3, P < 0
0001), supporting the notion that this marker represents the most fitting variable to predict TTFT in early stage CLL. The stability of ZAP70 expression over time still remains a controversial issue (Crespo et al, 2003; Orchard et al, 2004;

doi:10.1111/bjh.13117

Correspondence

(A) ZAP70

HR

95% CI

P

I/C 20%

1·39

0·61–3·35

0·42

I/C 30%

1·98

1·12–4·05

0·04

I/C 40%

2·27

1·18–4·36

0·01

T 20%

2·05

0·78–5·27

0·25

T 40%

2·41

1·25–4·62

0·01

T/B ≥3

0·51

0·21–1·36

0·19

T/B ≥2

0·56

0·28–1·00

0·05

T/B ≥1·5

0·38

0·24–0·79

0·005

(B) ZAP70

HR

95% CI

P

I/C 20%

0·98

0·46–1·98

0·92

I/C 30%

1·48

0·76–2·91

0·26

I/C 40%

2·36

1·32–4·08

0·006

T 20%

0·79

0·43–1·72

0·59

T 40%

1·85

1·03–2·95

0·049

T/B >3

0·69

0·35–1·39

0·35

T/B >2

0·64

0·35–1·18

0·08

T/B >1·5

0·43

0·21–0·92

0·03

HR

95% CI

P

One method

0·86

0·38–1·95

0·7

Two methods

1·93

1·18–3·16

0·009

Three methods

3·57

1·98–6·43