Patients with chronic lymphocytic leukaemia (CLL) have a very variable clinical course and an increasing array of therapeutic options (O'Brien, 2001). These factors encourage further study of the pathophysiology of CLL and the development of reproducible, robust prognostic models for survival. The p53 tumour suppressor gene codes for a 53-kDa phosphoprotein transcription factor that is normally present in the cell nucleus and is involved in cell cycle arrest and apoptosis (Lane, 1992). Mutations of this gene are the most frequent genetic abnormalities found in human malignancies (Bartek et al, 1991). The protein product of a mutated p53 gene has a prolonged half-life within the cell nucleus, relative to wild-type p53 protein. This abnormally prolonged half-life enables the detection of p53 expression by immunohistochemistry (IHC) using antip-53 monoclonal antibodies (mAb) (Bartek et al, 1991). The detection of p53 has been associated with an adverse prognosis in a number of human malignancies and efforts are being made to develop therapeutic agents which restore the normal p53 function (Bullock & Fersht, 2001; Vousden, 2002). However, discrepancy between mutation of the p53 gene and IHC detection of p53 expression has been reported, and p53 expression, but not gene mutation, has been shown to be an independent prognostic factor in some tumours (e.g. malignant glioma) (Pollack et al, 2002). We thus conduced a study to assess: 1. the degree of abnormal p53 protein expression in patients with CLL, using bone marrow biopsies, (2) the degree of reproducibility of this detection by IHC between different investigators, and (3) the prognostic import of aberrant p53 expression. We also sought to create a model for survival in patients with CLL, based on p53 expression.
Summary. As the abnormal expression of p53 protein is prognostically significant in some human cancers, its significance in patients with B-cell chronic lymphocytic leukaemia (CLL) was assessed. Two investigators evaluated the percentage of bone marrow mononuclear cells that stained for p53, using biopsies stained with anti-p53 monoclonal antibody (DO-7), and graded the degree of staining (0, +, ++, +++). Samples from a cohort of 90 patients with CLL were studied (median age 60 years, range 30–89 years; 57 patients were (63%) previously untreated, 22 patients (24%) had received one or two prior regimens, 11 patients had received (12%) three to seven regimens. The overall percentage of cells positive for p53 staining was a median of 43 (range 1–88). No investigator effect was detected either in overall percentage cells rated p53 positive or on the degree of staining (Pearson's correlation coefficient 0·980, P-value < 0·001). A Cox proportional hazards model showed that the percentage of ++ and +++ p53-positive cells correlated with various prognostic factors in CLL (P < 0·0001). A multivariate model incorporating prior therapy, Rai stage, beta2 microglobulin (β2M) and p53 expression showed that only the percentage of p53-positive cells and β2M were predictive of survival, and enabled the development of a highly predictive model of survival based on these two parameters.
Patients and methods
Paraffin-embedded bone marrow biopsies from 90 patients with CLL and 12 patients with solid tumours without evidence of marrow involvement were used in this study. The study was performed according to an Institution Review-Board-approved clinical research protocol with written informed consent. The diagnosis of CLL was established based on morphological, immunological and molecular evaluation of peripheral blood and bone marrow. Immunological evaluation included flow cytometric analysis of leukaemic cells using CD19, CD5, CD20, CD23, CD11C, CD22, FMC-7, CD79b, CD3, CD4, CD8, kappa and lambda. Molecular studies included immunoglobulin and T-cell receptor genes as well as BCL-1 and BCL-2 rearrangement studies.
Immunocytochemistry. Decalcified, paraffin-embedded bone marrow biopsy samples were immunostained for p53 (DO-7; Dako, Carpinteria, CA, USA) using standard procedures (Cordone et al, 1998). The percentage of cells stained by p53 mAb were graded for staining (0, +, ++, +++) according to intensity, considering 0 as negative (Fig 1). The percentage of positive cells and degrees of expression were counted and graded by two independent investigators.
Statistical methods. Univariate data were summarized using standard descriptive statistics, tabulation of categorical variables and histograms of numerical variables. Associations between categorical variables were assessed via crosstabulation and Fisher's exact test. Associations between pairs of numerical variables, including the percentage of p53 positivity for the two investigators, were assessed via scatter plots and Pearson's correlation coefficient. Unadjusted survival probabilities were estimated using the method of Kaplan and Meier. The Cox proportional hazards regression model was used to assess the ability of patient characteristics to predict survival, with ‘goodness of fit’ assessed by the Grambsch–Therneau test, Schoenfeld residual plots, martingale residual plots and likelihood ratio statistics (Grambsch, 1995; Grambsch et al, 1995). All scatter plots were smoothed using the lowest method of Cleveland. Owing to the high agreement between the two investigators, the average of their p53 positivity scores was used in the Cox model fits. Predictive variables in the Cox model were appropriately transformed, based on their smoothed martingale residual plots. Predictive variables in the Cox model were selected by performing a backward elimination with P-value cut-off of 0·05, then allowing any variable previously deleted to re-enter the final model if its P-value was < 0·05. The association between serum beta2 microglobulin (β2M) and p53 positivity in predicting survival was assessed using the final Cox model and also graphically using the co-plot method of Thall and Estey (2001). Models were also compared using likelihood ratio and Akaike information criterion (AIC) statistics. All calculations were carried out using the windows nt operating system in sas and splus, using the standard sas and splus functions, and the splus survival analysis package of Therneau (Venables & Ripley, 1999).
Reproducibility of the grading of p53 positivity
Bone marrow biopsies from 90 patients with CLL and 12 patients with solid tumours without bone marrow involvement were immunostained for p53. The staining intensity was graded as 0 for negative, + for faintly positive, ++ for moderately positive and +++ for strongly positive. Figure 1 shows examples of cells graded as positive using this scale. The percentage of cells expressing +, ++, or +++ p53 in the 90 bone marrow biopsies was separately assessed by two investigators. Bone marrow samples from the control 12 individuals showed no staining for p53. The overall percentage of cells that were positive for p53 staining was a median of 43%, range 1–88%. No investigator effect was seen, either in terms of the number of cells deemed positive for p53 expression or in the grading of positivity (Table I). The investigators' results were highly associated with each other (Pearson's correlation coefficient: 0·980, P-value < 0·001) (Fig 2).
|Variable||Mean ± SD||Median||Range|
|Age (years)||59·2 ± 11·7||59·6||30–89|
|White blood cells (× 109/l)||59·3 ± 71·7||25·7||2·4–333·9|
|Haemoglobin (g/dl)||12·4 ± 2·5||3·4||3·4–16·7|
|Platelets (× 109/l)||159·6 ± 80·0||157||12·0–473·0|
|β2M (mg/l)||3·5 ± 2·5||2·7||1·1–16·3|
|Lymphocytes in peripheral blood (%)||71·3 ± 20·5||73·0||4·0–99·0|
|Lymphocytes in peripheral blood (× 109/l)||45617·2 ± 57945·1||18779·0||611·0–242725·0|
|Bone marrow cellularity (%)||50·4 ± 27·3||40·0||15·0–100·0|
|Bone marrow lymphocytes (%)||64·4 ± 21·6||67·0||11·0–97·0|
|Splenomegaly (cm)||1·3 ± 3·5||0·0||0·0–19·0|
|Hepatomegaly (cm)||0·8 ± 2·1||0·0||0·0–10·0|
|% of cells with all staining (%p53total)||41·5 ± 21·6||43||1·0–88·0|
|% of cells with 3+ staining (% p53+++)||2·5 ± 5·7||0·0||0·0–27·0|
|% of cells with 2+ staining (%p53++)||6·0 ± 8·7||3·0||0·0–41·0|
|% of cells with 1+ staining (%p53+)||35·6 ± 21·5||34·5||0·0–90·0|
|% of cells with 3+ staining (%p53+++)||2·9 ± 7·0||0·0||0·0–40·0|
|% of cells with 2+ staining (%p53++)||5·6 ± 8·2||2·0||0·0–38·0|
|% of cells with 1+ staining (%p53+)||33·0 ± 18·7||33·0||1·0–78·0|
Correlation with aggressive features
The clinical and laboratory characteristics of the study cohort are summarized in Table I. Fifty-seven patients (63%) were previously untreated, 22 patients (24%) had received one or two prior regimens of therapy, 11 patients (12%) received three to seven regimens.
As summarized in Table II, there were significant direct correlations between the percentage of p53-positive staining cells and serum β2M, lower haemoglobin levels and age, when the percentage of ++ or +++, but not + were considered. Similar correlations were observed when we combined the ++ and +++ grades of staining (Table II). Combining the ++ and +++ grades also simplified counting of cells and increased reproducibility. These data suggest that cells with faint p53 staining (+) were not clinically relevant, and these were not considered in subsequent prognostic models. There was significant correlation between the percentage of ++ and +++ p53-expressing cells and advanced Rai stage disease (P = 0·029) (Fig 3). There was no correlation between p53 expression and whether patients were previously treated or not (P = 0·1).
|%p53++||%p53+++||%p53++ and %p53+++||β2M||Platelets||Hb||WBC||Lymphocytes||Age|
|%P53+||0·22*||− 0·053||0·18||0·064||− 0·13||− 0·014||0·038||0·075||− 0·071|
|%P53++||0·74**||0·99**||0·20||− 0·056||− 0·28**||0·17||0·018||0·20|
|%P53+++||0·82**||0·25*||− 0·070||− 0·34**||0·11||− 0·055||0·28**|
|%P53++ and %P53+++||0·22*||− 0·07||− 0·31**||0·16||0·015||0·250*|
|β2M||− 0·54**||− 0·67**||0·19||0·089||0·287*|
|Platelets||0·36**||− 0·18||− 0·18||− 0·315**|
|Hb||− 0·30**||− 0·15||− 0·357**|
When we correlated p53 expression and survival, the Cox regression model showed that the percentage of cells showing grade ++ or +++ p53 expression was predictive of patient survival (P < 0·001). Similar results were obtained when grades ++ and +++ were combined (P < 0001). As expected, β2M and Rai stage were also highly predictive of survival (P < 0·00001 and 0·0004 respectively). In a multivariate analysis incorporating β2M, Rai staging, prior therapy and p53 expression (++ and +++), only p53 expression, staging and β2M remained predictive of survival (Table III). Patients with CLL can also be dichotomized according to p53 expression. Recursive partitioning and regression tree (RPART) analysis demonstrated that patients with ≥ 19% p53-positive cells (++ and +++) had significantly shorter survival (P < 0·003) (Fig 4).
|Parameter estimate||Standard error||P-value||Relative risk|
|%P53++ and %P53+++||0·043%||0·013||0·0012||1·04|
|Prior Rx (yes versus no)||0·37||0·66||0·57||1·45|
Development of a prognostic model
Univariate and multivariate analyses showed that β2M and level of p53 expression were the most important significant predictors of survival in CLL. As illustrated in Fig 5, showing co-plots of Kaplan–Meier estimates of patient survival, an increasing percentage of p53 had a negative impact on survival, independent of serum β2M levels, and serum β2M levels impacted survival at the lower levels of p53 expression, but not at higher levels. Therefore, we developed an algorithm that enabled better stratification of patients with CLL for therapy, and prediction of outcome based on p53 expression, β2M level and whether the patient has been previously treated or not (Table IV). This approach can be adapted as a simple computer program that can be used with personal computers or personal digital assistants (PDA). The prediction of patients' survival time in Table IV was performed via the use of a visual basic for applications (VBA) computer program developed within microsoft®excel. The core construction and analysis was based on the Cox proportional hazard model given in Table III. To develop this VBA application, we first obtained base hazard rate and parameter estimates using the sas® proportional hazards regression, PHREG, procedure. These estimates were then used to obtain estimated probabilities of surviving from one to 24 months given the patient's covariates. The program makes use of these estimated probabilities to create a graphical representation of a given patient's predicted survival curve. The program also provides 6-month, 1-year and 18-month survival probabilities. The graphical interface allows the user to input patient characteristics in a user-friendly manner. The application program can be obtained via E-mail by sending an electronic request to: firstname.lastname@example.org.
|% p53 (++ and +++)*||β2M†||% survival ≥ 12months (95% confidence interval)|
|Previously untreated||Previously treated|
|0||1·5||96·0 (92·2, 100·0)||92·4 (84·3, 100·0)|
|2·7||94·6 (89·5, 100·0)||89·7 (80·0, 100·0)|
|5·9||87·7 (75·7, 100·0)||77·5 (62·4, 97·1)|
|3||1·5||95·5 (91·3, 100·0)||91·4 (82·5, 100·0)|
|2·7||93·9 (88·4, 99·7)||88·4 (77·9, 100·0)|
|5·9||86·2 (73·2, 100·0)||74·9 (59·3, 95·4)|
|23||1·5||90·3 (82·2, 99·3)||82·1 (66·3, 100·0)|
|2·7||87·0 (76·7, 98·6)||76·2 (59·1, 98·2)|
|5·9||72·0 (51·7, 100·0)||52·8 (35·0, 81·4)|
As options for therapy of advanced CLL become more complex, the need for prognostic systems based on objective end-points increases (O'Brien, 2001). In this study, we examined the reproducibility of assessment of abnormal p53 expression by IHC in CLL and its prognostic import. P53 is a transcription factor that is activated by DNA breaks (Levine, 1997). When activated, it can induce cell apoptosis or cell cycle arrest. Thus p53 is critical for either the repair or death of cells with DNA damage and thus acts against the development of abnormal cell clones (Lane, 1992). If p53 malfunctions, e.g. becomes mutated or deleted, its role in protecting the genome is compromised, genetic damage can be transmitted and neoplastic clones develop (Lane, 1992; Levine, 1997). Thus it is regarded as a pivotal tumour suppressor gene. It is mutated in over 50% of all human malignancies. The rate of p53 gene mutation has been reported to be relatively low in CLL, being more frequently mutated in people with rapidly progressive or variant disease (Gaidano et al, 1991; Fenaux et al, 1992; Gandini et al, 1994; De Angeli et al, 2000). Mutations in the p53 gene are associated with patients who have CLL with both cytotoxic drug resistance, which is not multidrug resistance (MDR) mediated, and a poor response to purine analogues, which are the most active anti-CLL agents available (el Rouby et al, 1993; Dohner et al, 1995; O'Brien et al, 2001). Detection of p53 gene mutations is complex, relatively expensive and difficult to integrate into routine practice. Unlike wild-type p53, mutated p53 has a prolonged intracellular half-life and thus becomes detectable by IHC (Gannon et al, 1990; Lepelley et al, 1994). Thus IHC detection of p53 has been considered a marker of p53 gene mutation (Lepelley et al, 1994; Levine, 1997). However, p53 gene mutation and protein detection may be discordant in the lymphoproliferative disorders, particularly CLL (Aguilar-Santelizes et al, 1994). In a recent report by Pollack et al (2002) in children with high-grade gliomas, p53 expression, as detected by IHC, was independently prognostic, while mutations in the p53 gene were not.
As IHC methodologies are readily used in daily practice, we first sought to evaluate whether IHC methodology was reproducible for p53 detection and grading of expression. The study data indicated that the independent investigators were highly concordant in their overall rating of cellular p53 positivity (Table I). The grading of p53 expression on a three point scale (Fig 1) was also highly reproducible between the different investigators (Tables I and II). Thus, this technique could be incorporated into practice and is suitable for the generation of reproducible data in both the clinical and research settings. The study data demonstrated that the percentage of cells with strong positivity for p53 expression (++ and +++) can be combined and counted together, but should be distinguished from faint staining (+) cells. These data enabled the development of a model for the prediction of survival in CLL patients based on β2M and p53 expression.
The study data would indicate that p53 positivity, as assessed by IHC, is a very important independent prognostic variable in patients with CLL. These data concur with those recently reported by Cordone et al (1998), in their analyses of p53 expression in the peripheral blood mononuclear cell population in patients with CLL. These investigators reported that expression of p53 protein, as measured by IHC (deemed positive if ≥ 1% of lymphocytes showed ‘strong’ nuclear staining), was strongly associated with p53 gene mutations, advanced disease, progressive disease, refractoriness to therapy and reduced survival. Univariate analyses of prognostic features in the study population confirmed the importance of β2M, prior therapy and Rai stage (O'Brien et al, 1993; Zwiebel & Cheson, 1998). The level of p53 expression joined these established factors as being a significant predictor of survival. On multivariate analysis, the factors that retained prognostic significance were serum β2M, number of lines of prior therapy and degree of p53 expression. Based on this analysis, a VBA computer program was developed within microsoft®excel. This program, which can be electronically downloaded as detailed above, enables the development of estimated probabilities of survival upon entry of a patient's relevant covariates. The application program can facilitate other investigations into the reproducibility of a p53-based prognostic schema for patients with CLL and/or its further clinical and research application.
Further studies are necessary to determine whether the grading system for the degree of p53 expression described in this study will retain its value as a prognostic indicator when applied to peripheral blood lymphocytes. The data reported by Cordone et al (1998), summarized above, would indicate that this is likely to be so. Other investigators have recently reported other mechanisms of disruption of p53 function in CLL other than mutation or deletion, including mutation of regulatory genes, such as ATM (Lin et al, 2002). Relationships between p53 dysfunction, CD38 expression and IgV(H) mutation, all adverse prognostic factors, are being explored (Lin et al, 2002). Prognostic schemas based on these molecular abnormalities are likely to be developed and to significantly impact on the therapy of patients with CLL (Oscier et al, 2002). Whether the relatively simple IHC technique will be an accurate enough reflection of the prognostic import of the various pathways leading to abnormal p53 function in CLL will be the subject of future studies.