The Prognostic Impact of BCL2 Protein Expression in Acute Myelogenous Leukemia Varies with Cytogenetics¹ Free

Complex pathways exist for the regulation of the cell cycle and apoptosis. (1, 2, 3, 4). Abnormalities in the expression or function of key proteins in either of these regulatory pathways could confer proliferative advantages to leukemic cells and, thereby, contribute to leukemogenesis. Most antileukemic chemotherapy functions via induction of apoptosis; consequently, sensitivity or resistance of leukemic blasts to chemotherapy may correlate with the apoptotic potential of a cell (5). With regard to current therapy, AML³ can be divided into favorable and unfavorable groups based on certain presenting characteristics. Patients with favorable AML tend to be younger; have cytogenetic abnormalities, including inv(16), t(8:21), t(15:17), and diploid or insufficient metaphases (FIPC); and lack an AHD. Those with unfavorable AML have cytogenetic abnormalities, including −5, −7, +8, 11q23, Ph¹, or miscellaneous changes (UC); tend to be older; and are more likely to have an AHD. Molecular explanations to account for the differences in response to current therapy are lacking, with the exception of differences in the expression of MDR1 (6), but differences in expression and function of cell cycle and apoptosis regulating proteins between patients with favorable and unfavorable leukemias could be involved.

There is significant redundancy in the pathways for the induction and regulation of apoptosis. One of the key regulators is the BCL2 family (3) of proteins, consisting of antiapoptotic proteins like BCL2, BCL-X_L, Bag1, and Mcl1 and proapoptotic members like Bax, Bad, Bak, Bid, Bcl-X_s, and so on. These proteins are known to dimerize and affect the transmembrane potential in mitochondria in pro- or antiapoptotic ways, depending on dimerization patterns and relative quantities (3, 7). This, in turn, regulates whether cytochrome c is contained within the mitochondria or escapes to the cytoplasm, where it can activate Apaf-1, which, in turn, cleaves caspase-9, thereby initiating the activation of apoptosis (8, 9).

BCL2 is the best characterized of these proteins, and its role in the pathogenesis and prognosis of AML has been studied previously (10, 11, 12, 13). Although the methods of measuring and categorizing patients have varied, in general, these studies have found that patients with higher levels of BCL2 have a lower remission rate, an inferior survival, or both. Most of the patients involved in these studies were participants in cooperative group trials; consequently, they contained a high percentage of younger patients with favorable types of leukemias. Paradoxically, the report by Lauria et al. (12) noted that only 7 of 39 patients with high BLC2 had UC. Karakas et al. (13) found that patients with AML arising after MDS were more likely to have undetectable BCL2 mRNA, compared to patients with de novo AML (30 versus 12%), and fewer had strong expression (51 versus 64%). Additionally, they found that patients with no expression of BCL2 had lower remission rates and higher relapse rates than those with intermediate levels of BCL2. Maung et al. (10) found lower BCL2 expression in all 7 patients with “secondary” AML but in only 2 of 11 primary AML cases. If BCL2 was functioning in the same pathogenic role in leukemias with both favorable and UC, then it would be logical to expect increased, not decreased, frequencies of high BCL2 expression in patients with UC or prior MDS because they traditionally have lower remission rates and inferior survival outcomes. The fact that BCL2 levels may be lower among patients with unfavorable leukemias suggests that BCL2 may be acting differently in different types of leukemia.

We, therefore, hypothesized that the role of BCL2 in the leukemogenesis and prognosis of favorable and unfavorable leukemias might be different. To study this, we measured the level of BCL2 by quantitative Western blot in 198 patients with newly diagnosed and untreated AML and examined the prognostic impact of BCL2 in both favorable and unfavorable prognosis leukemias.

Between October 1, 1991, and July 15, 1995, 376 patients with newly diagnosed, untreated AML were seen at The University of Texas M. D. Anderson Cancer Center; peripheral blood samples were obtained prior to the initiation of therapy from 218 patients. Emergency initiation of therapy at night or on weekends, before research samples could be obtained, and absence of circulating blasts were the predominant reasons for nonaccrual. Eight patients opted for no therapy and were excluded from analysis, and insufficient sample material was available for another 12, leaving a sample size of 198. Clinical characteristics of these patients are summarized in Table 1. The median follow-up of patients who were alive on the study was 123 weeks at the time of this analysis. Samples for analysis were obtained during regularly scheduled diagnostic evaluations as part of protocols approved by the Human Subjects Committee of The University of Texas M. D. Anderson Cancer Center. Not surprisingly, the response and survival of patients who were not sampled were inferior to those of the study cohort.

Patients received induction therapy according to institutional protocols. Therapy consisted of HDAC-based regimens combined with idarubicin alone, fludarabine alone, or both, as described previously (14). Filgrastim (granulocyte colony-stimulating factor) was given to 144 patients per existing clinical protocols. Maintenance therapy was administered for 6 or 12 months and consisted of standard-dose ara-C alternating with lower dose versions of the induction regimen, as described previously (14). All 20 patients with APL received idarubicin plus all-trans-retinoic acid, as reported previously (15). One patient underwent allogeneic transplant and remains alive >1 year after transplant.

Western blotting was carried out using cell lysates derived from the Ficoll separation-generated mononuclear fraction of peripheral blood of 198 newly diagnosed patients with AML and 16 normal individuals. The percentage of blasts in the samples ranged from 1 to 100%, with average and median percentages of 53 and 56%, respectively. In FAB M4 and M5, potentially leukemic monocytes were not counted as blasts. As described previously (16, 17), whole-cell lysates from 5 × 10⁵ cells were electrophoresed through an 8–12% SDS-polyacrylamide gradient gel. Each gel run included a low expressing BCL2 control cell line (K562), a strongly expressing BCL2 control cell (Y79 ATCC HTB 18), two to three PBMC samples from normal individuals, and molecular weight markers. Protein was transferred to Immobilon polyvinylidene difluoride membrane (Millipore, Bedford, MA) using a semidry transfer apparatus at 0.8 mA/cm² for 1.5 h. The membrane was blocked in Tris-buffered saline with 0.05% Tween-20 and 3% nonfat dry milk (Blotto) at 4°C for 4 h and then exposed overnight to an anti-BCL-2 monoclonal antibody (clone 124; DAKO Corp., Carpinteria, CA) at a 1:2000 dilution in Blotto at 4°C. As part of a broader analysis and to verify the presence and quality of the protein (specifically, lack of degradation) in each sample, we also included other antibodies in the antibody cocktail: anti-RB monoclonal antibody MAB1 (Triton Biologicals, Alameda, CA; and later Ciba-Corning, San Diego, CA) at a 1:300 dilution; antiactin monoclonal antibody clone AC-40 (Sigma Immunochemicals, St. Louis, MO) at a 1:200 dilution; and anti-PCNA monoclonal antibody clone 19F4 (Boehringer Mannheim, Indianapolis, IN) at a 1:1200 dilution. Subsequently, the membranes were washed twice in Blotto, exposed to sheep antimouse IgG conjugated to horseradish peroxidase (1:2000) for 1 h, washed in Blotto and Tris-buffered saline with 0.05% Tween-20, and then exposed to chemiluminescence mixture for 1 min, according to the directions of the manufacturer (Amersham, Arlington Heights, IL). Films were then exposed at intervals of 15 s to 2 min until maximum saturation of the film had occurred. A representative blot is shown in Fig. 1.

Films with exposures in the linear range were analyzed by densitometry (Molecular Dynamics, Sunnyvale, CA). All patients were analyzed in duplicate, and all samples produced technically interpretable and reproducible Western blot assays. To normalize for variation in antibody concentration or time of exposure, we normalized the BCL2 signal from the patient against the BCL2 signal of the control cell line Y79. Results are expressed in terms of this ratio.

Pairwise associations between patient covariates were assessed graphically for pairs of numerical variables by examining scatterplots, by Wilcoxon-Mann-Whitney and Kruskal-Wallis (18) test statistics for categorical and continuous variables and by the Fisher exact test (19) and its generalizations (20) for pairs of categorical variables. Unadjusted survival and DFS analyses were performed using Kaplan-Meier plots (21). Unadjusted comparisons of survival and DFS between patient subgroups were made using the log-rank test (22). The Cox proportional hazards model (23) and its generalizations (24) were used to assess the ability of treatment indicators, BCL2, and other patient characteristics [age, sex, bilirubin, performance status, presence of an AHD (≥ 2 months of hematological abnormality), fibrinogen, hemoglobin, platelet count, white blood count, and albumin] to predict survival and DFS. Goodness-of-fit was assessed by the Grambsch-Therneau test (25), Schoenfeld residual plots, and martingale residual plots (24). All scatterplots were smoothed using the lowess method of Cleveland (26). Variables were transformed as appropriate based on these plots. Due to apparent threshold effects of BCL2 on survival and DFS indicated by goodness-of-fit analyses, BCL2 was replaced by the four-level categorical variable, indicating whether it was in its lowest quartile (BCL2 < 1.204), second quartile (1.204 ≤ BCL2 < 1.78), third quartile (1.78 ≤ BCL2 < 3.049), or fourth quartile (BCL2 ≥ 3.049). This was done to assess possible threshold effects of BCL2 on survival and DFS without searching for optimal cutoff points (27). Cox regression models for survival and DFS were obtained by first identifying any important interactive effects between BCL2, cytogenetics, treatment, and other patient characteristics, including these effects in an initial set of variables along with patient covariates and then performing a backward elimination with a cutoff P of 0.05. Variables were selected for inclusion in the model based on their previous recognition as important prognostic factors (28, 29, 30) and evidence of association with survival in this population. Variables showing a strong association with either BCL2 or cytogenetics were not included in the multivariate Cox model analyses to avoid colinearities. All computations were carried out on a DEC Alpha 2100 5/250 system computer (Digital Electronics Corporation, Nashua, NH) in Splus (31) and StatXact (Cytel Software Corporation) using both standard Splus functions and the Splus (StatXact 3 for Windows; Cytel Corporation, Cambridge, MA) survival analysis package of Therneau (32). Heteroscedasticity in BCL2 expression was assessed using Levene’s test (18).

All but one patient had detectable levels of BCL2, and a broad range of BCL2 protein expression was seen in all FAB types and cytogenetic groupings (Fig. 2, A and B, respectively). The dispersion of BCL2 expression was statistically similar for FAB types M1–M5 (P = 0.72; Levene’s test), and the range was broader than that observed with normal PBMCs (predominantly lymphocytes, which are known to express BCL2, and monocytes). There were too few patients with M6 or M7 FAB type for statistical analysis. In contrast, there was significantly different dispersion in BCL2 between different cytogenetic categories (P = 0.0002), with the greatest variation among those with UC (nested ANOVA, FIPC P, not significant; UC P = 0.00003).

One hundred thirty-two of the 198 patients had died at the time of analysis, with a median overall survival of 74 weeks (95% confidence interval, 51–98 weeks). Smoothed Martingale residual plots (Fig. 3) strongly indicated that BCL2 had a threshold effect on survival that interacted with cytogenetic category. Specifically, Fig. 3 shows that, among patients with FIPC, above a certain threshold, BCL2 was associated with an increased risk of death. In contrast, among patients with UC, values of BCL2 above a similar threshold were associated with an decreased risk of death. We, thus, represented BCL2 and cytogenetics taken together by the eight-category variable recording cytogenetics (UC versus FIPC) and the quartile of BCL2, as described earlier, in all subsequent analyses.

The Kaplan-Meier plots for each of these eight BCL2-cytogenetics categories (Fig. 4, top), strongly support the threshold effect noted above for the Martingale plots. By first fitting a Cox model that accounted for all eight categories (two cytogenetic categories FIPC versus UC and four quartiles of BCL2) and then collapsing categories with effect that did not differ significantly, we obtained five prognostic variables representing the joint effect of BCL2 and cytogenetic category on survival. It was readily apparent that FIPC patients with BCL2 expression in the lower three quartiles behaved similarly and that these patients with the best outcome may be used as a baseline group for comparison. Patients with FIPC and the highest quartile of BCL2 expression (FIPC-topqtr) had a significantly inferior survival experience (median survival, 78 weeks versus not reached for the baseline group, P = 0.008, RR = 2.18; Table 2). For patients with UC, there were highly significant increases in the median survival and decreases in the RR of dying (Table 2) with increasing BCL2 level, supporting the Martingale residual plot. These findings remain the same if only samples with >30% circulating blasts are analyzed or if patients with APL are excluded. Interestingly, 12 of 13 APL patient with lower BCL2 are alive, whereas 6 of 7 with high BCL2 have died (P = 0.0012). BCL2 level was not prognostic when all patients were considered together without stratification for cytogenetics.

Among patients with FIPC the CR rate was lower among patients with high BCL2 level, relative to those in the lower three quartiles, but this difference was not statistically significant (P = 0.19). Similarly, for those patients with UC, as the BCL2 level decreased, so did the remission rate, but again, this difference did not reach statistical significance (P = 0.20; Table 3). The same patterns were observed for relapse rate as well for both FIPC and UC patients. As BCL2 level increased, so did remission duration for UC patients (P = 0.04). High BCL2 was nearly a significant prognostic factor for remission duration for those with FIPC as well (P = 0.08; Fig. 4, bottom).

Recent studies have described a more favorable outcome for patients with a history of an AHD and higher BCL2 levels (33) Consequently, we evaluated whether the prognostic impact of BCL2 was affected by a history of an AHD. BCL2 level, in quartiles, was not prognostic of survival among patients with or without an AHD when all patients were considered simultaneously (P = 0.67 and 0.51, respectively). However, when stratified by cytogenetics, BCL2 level was only prognostic of outcome among those with no AHD (P = 0.02 for FIPC and P = 0.004 for UC patients). In contrast, for patients with a history of an AHD, BCL2 was not prognostic of survival for either FIPC (P = 0.43) or UC (P = 0.88) patients.

To investigate whether BCL2 level was an independent predictor of remission duration or survival, we performed univariate and multivariate Cox model analyses. The previously identified prognostic factors included were: age; sex; WBC and platelet counts; bilirubin, albumin, hemoglobin, and fibrinogen levels; FAB category; performance status; the presence of an AHD; cytogenetic abnormalities; treatment regimen (idarubicin-HDAC, n = 79; fludarabine-HDAC, n = 99; or all-trans-retinoic acid (ATRA)-idarubicin, n = 20); and BCL2 level. Because the prognostic impact of BCL2 interacted with that of cytogenetics, BCL2 and cytogenetics were included in the model as interactive variables. Patients with FIPC were divided into two groups, those with BCL2 levels in the lower three quartiles versus those in the top quartile. Those with UC were divided into three groups, based on whether their BCL2 expression was in the lower two quartiles, third quartile, or fourth quartile. The incidences of AHD or Zubrod performance status 3 or 4 were statistically similar between patients with FIPC and the lower three quartiles of BCL2 versus the top quartile and between the three groups of UC patients. Variables that were significant predictors in univariate analyses included age, sex, cytogenetics, performance status and AHD for survival, and hemoglobin for remission duration, along with the BCL2 terms.

Multivariate Cox models for survival and DFS that include treatment and the other patient characteristics in addition to BCL2-cytogenetics are summarized in Table 4. These models were obtained via the backward elimination procedure described earlier. The idarubicin-HDAC treatment group was used as the baseline group in the Cox model fits. White blood count and platelet count were omitted from the covariate set to avoid colinearity in determining this model due to their high associations with BCL2. Notably, treatment was of no prognostic significance (P = 0.60 for fludarabine-HDAC; P = 0.54 for idarubicin-ATRA) when added to the final model summarized in Table 4. These models confirm the pattern indicated by the preliminary analyses. Either considered per se or covariate-adjusted, patients with FIPC-topqtr, specifically FIPC and BCL2 > 3.049, had a RR that was nearly double that of patients with FIPC and a BCL2 < 3.0. In sharp contrast, the RR of death decreased as BCL2 increased among patients with UC, with significant differences between patients with BCL2 < 1.78 (lower two quartiles, RR = 4.31), 1.78 ≤ BCL2 < 3.049 (third quartile, RR = 2.45), and BCL2 ≥ 3.049 (fourth quartile, RR = 1).

We repeated these analyses using DFS in place of survival. One hundred forty-six of the 198 patients died or had relapsed at the time of analysis, with a median overall DFS of 51 weeks (95% confidence interval, 38–66 weeks). The Martingale residual plots and the Kaplan-Meier plots for the effects of BCL2 and cytogenetics on DFS indicated the same interactive effects as on survival (data not shown). The Cox model for DFS (summarized in Table 4) yielded results similar to those of the survival model, with the only substantive differences being that hemoglobin > 7.6 was a significant predictor of DFS (P = 0.022), whereas it was only marginally predictive of survival (P = 0.076), and the subgroup with FIPC-topqtr had marginally significantly worse DFS (P = 0.087).

Regardless of the end point or population included the BCL2 terms were always significant independent predictors of outcome, with the exception that FIPC-topqtr was only marginally significant for remission duration (P = 0.06). These analyses demonstrate that UC with lower BCL2, and FIPC with high BCL2 are each independent prognostic factors for overall and event-free survival and that UC with lower BCL2 is an independent predictor for remission duration.

Similar to other studies (10, 11, 12, 13, 34, 35, 36), expression of BCL2 was observed in nearly all samples, and the median level of BCL2 expression did not vary with FAB type or cytogenetics. The heterogeneity of expression was similar among FAB types but was higher among those with UC. Other studies (12, 13, 36) have reported similar median values for different cytogenetic categories but only limited data comparing the range of expression. APL cases were observed to have strong expression in a third of cases in this study compared to 0, 60, and 75%, reported by Campos et al. (11), Banker et al. (36), and Karakas et al. (13), respectively. All prior studies agree with this study, in that there is a wide degree of heterogeneity in the expression of BCL2 in AML. It is unlikely that the differences in expression are attributable to the contaminating lymphocytes and monocytes that would be present in the PBMC population after Ficoll separation. As shown in Fig. 2, normal PBMCs have a relatively narrow band of expression, whereas most of the samples from the leukemia patients were outside of this range. This suggests that the majority of the signal is coming from the leukemic cells.

Among leukemia patients with favorable prognostic factors (FIPC, younger age, and no AHD), high BCL2 expression was an adverse prognostic factor for both survival and event-free survival and was nearly significant as a prognostic factor for remission duration. This agrees with the findings of other studies (10, 11, 12, 13), which included primarily favorable patients. In the report by Karakas et al. (13), the adverse effect of high BCL2 was more prominent when patients over 60 or those with a prior MDS were excluded. The report by Lauria et al. (12) did not show an effect of BCL2 level on survival, although high BCL2 was associated with a lower CR rate. It is tempting to speculate that had patients with UC (who generally had lower levels of BCL2) been excluded a difference in survival could have been seen. In the setting of a “favorable” leukemia, high levels of BCL2 would allow a cell to escape or suppress apoptotic signals, including those induced by chemotherapy. A higher percentage of patients express higher levels of BCL2 at relapse than they do initially (36), although this difference was not always statistically significant (35). This could arise from either the induction of BCL2 expression in response to chemotherapy, as demonstrated by Lauria et al. (12), or by the selection for, and ultimate predominance of, initially rare high BCL2 expressing leukemia cells during a cycle of chemotherapy as demonstrated by Andreeff et al. (37). A model for leukemogenesis, in which a first change causes a reciprocal translocation that affects proliferation and differentiation, could be supplemented by other changes that affect apoptotic potential. Increased BCL2, arising from an undefined stimulus, would block apoptosis induction and confer additional advantages to the leukemic cell. BCL2 is known to block activation of caspases (4), and we have observed that high levels of uncleaved (the biologically inactive forms) of caspase-2 and -3 are adverse prognostic findings among AML patients, especially those with FIPC (38).

The unanticipated finding in this study is that the prognostic impact of BCL2 is reversed for patients with UC. For these patients, lower levels of BCL2 are associated with lower response rates, higher relapse rates, shorter remission duration, and, ultimately, inferior survival. A lower level of BCL2 was an independent prognostic factor for both survival and remission duration. Although this has not been recognized fully in prior reports, they contain corroborating data for this counterintuitive finding. In the study by Karakas et al. (13), among patients with a prior MDS, a group that historically has a high percentage of UC, those with high BCL2 levels had a better CR rate (63 versus 51%), and among patients over 60 years of age, the CR rate was higher for those with high BCL2 (54 versus 32%). Lepelley et al. (33), reporting on BCL2 levels in MDS patients also found that high BCL2 was a favorable prognostic finding. In the study by Maung et al. (10), none of five patients with secondary leukemia and lower levels of BCL2 expression achieved CR. However, the small sample size and low remission rate confound comparison to that study (only 2 of 17 treated achieved CR). No other study has analyzed data with UC patients separately so it is possible that similar trends exist in the other studies and was overlooked. Additionally, Charpin et al. (39) recently reported that intense BCL2 expression in breast tumors was significantly correlated with longer DFS and recurrence-free survival.

Why might higher levels of BCL2 be a good prognostic sign for patients with UC, while the reverse is true for patients with FIPC? These studies, as well as most others, have considered BCL2 in isolation. However, the function of BCL2 is dependent upon which other members of the BCL2 family that it dimerizes with (3) or upon the phosphorylation status of BCL2 (40, 41) and upon other upstream and downstream events.

Two small studies have looked at the ratio of BCL2 to Bax, a proapoptotic protein (42, 43). In both studies, patients with higher ratios had lower CR rates. A simple explanation is that the net effect of the BCL2 family may be proapoptotic, despite the high levels of BCL2 due to overexpression of other proapoptotic members. Another is that, in some UC patients, high BCL2 levels are a response to the stimulation of proapoptotic family members and that the cells need to maintain very high levels to prevent apoptosis. These cells presumably are living on the edge of apoptosis, and because BCL2 expression is near its maximum, it cannot be raised further in response to an apoptotic signal, so the cells die. In contrast, those cells with lower BCL2 levels do not require BCL2 for protection, and there is sufficient room for increases in BCL2 expression to occur in response to an apoptotic signal. A third possibility is that cells with lower levels may have developed other methods of avoiding apoptosis (i.e., high expression of Bcl-X_L; Ref. 44) or that downstream regulators of apoptosis are modified. Hence, they are independent of regulation by this family of proteins. Presumably, they would really have a high resistance to apoptosis, despite the low level of BCL2, which is not expressed because it is not needed. It is possible that, in these patients, increased levels of cleaved caspase-3 are inactivating BCL2 or converting it to a proapoptotic death effector (45). In contrast, cells with higher levels may have more intact apoptosis pathways and still be receptive to input from the BCL2 family. Finally, there is evidence that BCL2 activity may be regulated by phosphorylation (41). Although evidence of phosphorylated BCL2 was not apparent in this study, differences in phosphorylation might alter the relative amount of functional BCL2 and account for the differences in outcome. All of the studies published to date have measured BCL2 levels, but none have correlated levels with actual apoptosis rates. Because there may be discordance between BCL2 levels and apoptosis rates, it is imperative in future studies to measure multiple members of the BCL2 family simultaneously and to develop better methods to determine the net functional balance of the BCL2 family (37). Correlations with other apoptosis-related proteins and with caspase-2 and -3, bax, and PKCα are in progress in our group.

These findings lend credence to the idea that the two types of leukemia, favorable (young, FIPC, and no AHD) and unfavorable (UC, older, and prior AHD) may arise and be affected by separate mechanisms. BCL2 might have a different role in the pathogenesis and leukemic progression of each, thereby imparting a different prognostic implication to the same level of expression. This distinction could have important therapeutic implications as future therapies designed to induce apoptosis or block antiapoptotic proteins are developed. Antisense deoxyoligonucleotides against BCL2 have been studied in a clinical trial (46). This study would suggest that antisense BCL2 would have efficacy for patients with FIPC and high levels of BCL2 but would be unlikely to have efficacy for those with UC and lower levels of BCL2.

We thank Gisella Sanchez-Williams, Patty Reed, and Sherri Pierce for assistance with data management.