The Research Centre of the University of Montreal Teaching Hospital (CR-CHUM)/Montreal Cancer Institute, Montreal, Quebec, Canada
Department of Medicine, University of Montreal, Montreal, Quebec, Canada
Centre de Recherche du Centre Hospitalier de l'Universite de Montreal Central University Hospital Research Center (CHUM)/Montreal Cancer Institute, 1560, rue Sherbrooke est, Montreal, Quebec, Canada H2L 4M1
In a previous microarray expression analysis, the authors identified candidate genes that were expressed differentially between ovarian tumors with low malignant potential and invasive serous epithelial ovarian tumors. Among them, the apoptosis-related candidate genes tumor necrosis factor-related apoptosis-inducing ligand (TRAIL), caspase 8 (CASP8), FLICE-inhibitory protein (FLIP), and cytochrome C (CYC) were identified.
For the current study, the authors conducted immunohistochemical analyses of a tissue array comprised of 235 serous tumors of different grades and stages to evaluate whether there was differential protein expression for these candidates and for the 4 death cell receptors of Trail: Dr4, Dr5, DcR1, and DcR2.
All proteins except DcR1 and DcR2 had significantly differential expression levels between grade 0 tumors (low malignant potential) and grade 2 and 3 tumors. Trail also showed differential expression between grade 0 tumors and grade 1 tumors. When all tumors were compared, the expression levels of Trail, Dr4, Dr5, DcR1, and Flip differed significantly between early-stage and advanced-stage disease. High Dr5 expression was associated with a poor prognosis in patients who had invasive tumors and in the subgroup of patients who had grade 3 tumors. Furthermore, the combinations of 2 proteins (Trail and Dr5, DcR2 and Cyc, Flip and Dr5, Flip and DcR2, DcR1 and Dr5 or Dr4 and Flip) revealed an association with patient prognosis.
Epithelial ovarian cancer (EOC) is a silent disease that usually is diagnosed at an advanced stage and, currently, represents the most fatal gynecologic malignancy.1 EOC tumors are classified by different histopathologies, and the serous type is the most frequent.2 Serous EOC tumors can be subdivided into those with low malignant potential (LMP) or borderline and invasive (TOV) tumors. In addition to a higher level of cellular atypia, the major morphologic criteria distinguishing these 2 classes is the presence of stromal invasion in TOV tumors, although at most microinvasion is observed in some LMP.3–5 Patients who have LMP tumors have an excellent 5-year survival rate (90–95%) compared with patients who have TOV tumors (30–40%). The degree of differentiation in malignant EOC defines the tumor grade, with LMP tumors (grade 0) being highly differentiated and TOV tumors classified as well-differentiated (grade 1), moderately differentiated (grade 2), or poorly differentiated (grade 3). The volume and extent of tumor spread define the clinical stage, varying from I to IV, with stage I limited to 1 or both ovaries, stage II associated with pelvic extension, stage III spreading into the abdominal cavity, and stage IV presenting with distant metastases.6, 7
After surgical removal of the tumor mass, patients with EOC usually receive a first line of platinum-based combined chemotherapy (for review, see Agarway and Kaye8). However, even if 80% of patients respond initially to treatment, most patients have recurrent disease and develop treatment resistance, with different latencies indicating the need for a better understanding of the disease and the identification of new therapeutic targets or treatment modalities. Some factors have been identified in drug resistance, such as the expression of multidrug resistance proteins, hypoxia in the tumor microenvironment, and resistance to apoptosis by overexpression of survival factors and down-regulation of death- signaling proteins.8
In a previous study, we defined molecular profiles that distinguished serous LMP tumors from grade 3 TOV tumors that included members of the tumor necrosis factor (TNF)-related apoptosis-inducing ligand (Trail) signaling pathway. Trail (Tnfsf10, Apo2L) is a secreted protein that induces apoptosis after binding to its receptors Dr4 (Trail-R1, Apo2, Tnfrsf10A) and Dr5 (Trail-R2, Trick2, Killer, Tnfrsf10B). Two other receptors, DcR1 (Trail-R3, Trid, Tnfrsf10C) and DcR2 (Trail-R4, Trundd, Tnfrsf10D), allow the binding of Trail but are unable to transduce the apoptotic signal, thus acting as decoy receptors.9 Trail was seen as a promising therapeutic agent, because it induces apoptosis in cancer or transformed cells but not in normal cells,10–13 whereas an in vitro study of EOC cell lines revealed higher sensitivity in normal cells than in cancer cells.14 It was suggested that this resistance to Trail-induced apoptosis by normal cells occurred in the presence of a higher level of decoy receptors at their surface compared with cancer cells.15 Preclinical studies of recombinant Trail in animal models have demonstrated that it is a potential therapeutic target for breast, colon, and ovarian cancer13, 14, 16, 17 (for review, see Debatin and Krammer9). Trail also synergized with cytotoxic drugs and radiation to achieve antitumor activity in various cancers, including malignant gliomas; melanoma; leukemia; and cancers of the breast, colon, and prostate (for review, see Debatin and Krammer9).
In the current report, we focused on protein expression of Trail signaling members in ovarian cancer. By using immunohistochemistry (IHC), we assayed the expression levels of different candidate proteins on a tissue array that contained 235 serous tumor samples of different grades. Staining intensity was related to different clinical parameters, such as tumor grade, disease stage, and patient survival, to evaluate the usefulness of markers in the stratification of EOC tumors.
MATERIALS AND METHODS
Patients and Tissue Specimens
Tumor samples were collected after obtaining appropriate consent from patients who underwent surgery in the Division of Gynecologic Oncology at the Centre Hospitalier de l'Université de Montréal (CHUM). Histopathology, tumor grade, and disease stage, as defined by the International Federation of Gynecology and Obstetrics, were determined by an independent pathologist who reviewed and graded the tumor samples.7 Tissue selection criteria for this study were based on serous histopathology from chemotherapy-naive patients, and all samples were collected between 1993 and 2003. Clinical data were extracted from the Système d'Archivage des Données en Oncologie, which includes entries on tumor grade and stage; treatment and clinical outcomes, such as the progression-free interval, as defined according to the Response Evaluation Criteria in Solid Tumors18; and survival.
Two representative cores (0.6 mm in greatest dimension) from each tissue sample, after selection based on a review of the hematoxylin and eosin-stained slide, were arrayed on an empty paraffin block. The tissue array was composed of 56 LMP tumors (grade 0), 11 grade 1 tumors, 53 grade 2 tumors, and 115 grade 3 tumors (Table 1). Then the tissue array was sectioned, stained with hematoxylin and eosin, and received another pathology review to confirm its content.
Table 1. Description of the Serous Tissue Array
SD indicates standard deviation.
No. of tumors
Mean age ± SD y
49 ± 14
41 ± 15
60 ± 12
62 ± 10
Tumor residuum, cm
Mean ± SD, mo
52 ± 36
51 ± 49
13 ± 10
16 ± 12
Mean ± SD, mo
58 ± 33
56 ± 47
28 ± 14
28 ± 16
No. of censured patients
For IHC analysis, anticaspase-8 (anti-Casp8) p20 rabbit polyclonal antibody (sc-7890), anti-FLICE-inhibitory protein (anti-Flip) mouse monoclonal antibody (sc-5276), anticytochrome C (anti-CYC) mouse monoclonal antibody (sc-13156), anti-DcR2 goat polyclonal antibody (sc-11,638), anti-DcR1 goat polyclonal (sc-7193), anti-Dr4 goat polyclonal (sc-6823), and anti-TRAIL goat polyclonal antibody (sc-6079) were purchased from Santa Cruz Biotechnology (Santa Cruz, Calif). The specificity of the antibodies was tested by Western blot analysis.
The tissue arrays, cut into 5-μm sections, were stained by an immunoperoxidase method as described elsewhere.19 Tissue sections were heated to 60°C for 30 minutes, deparaffinized in toluene, and rehydrated in an ethanol gradient. After 3% H2O2 treatment, slides were submerged in boiling citrate buffer (0.01 M citric acid adjusted to pH 6.0) for 15 minutes, blocked with a protein blocking serum-free reagent (DakoCytomation Inc., Mississauga, Ontario, Canada), and incubated with the antibody for 60 minutes at room temperature. Tissues were incubated with either a secondary biotinylated antibody (DakoCytomation Inc.) or a rabbit antigoat biotin-conjugated antibody (1:300 dilution; sc-2774; Santa Cruz Biotechnology) for 20 minutes followed by incubation with streptavidin-peroxidase complex (DakoCytomation Inc.) for 20 minutes at room temperature. Liquid diaminobenzidine was applied to visualize the reaction (DakoCytomation Inc.), and nuclei were counterstained with hematoxylin. For negative controls, phosphate-buffered saline was used instead of the primary antibody. Protein expression was scored according to the extent (as a percentage of total malignant cells) and intensity (value of 0 for absence, 1 for low intensity, 2 for moderate intensity, and 3 for high intensity) of staining based on visualization. All slides were visualized independently by light microscopy at ×20 magnification. All slides were analyzed independently in a blind study by 2 independent observers, and the interrater agreement was >90%. When strong differences in scoring between the 2 observers occurred, the core was reevaluated to reach a consensus between the 2 observers.
Apoptosis was assayed with anticleaved Casp3 antibody (no. 9661; Cell Signaling Technology, Beverly, Mass) according to the manufacturer's protocol. All sections were scored according to the extent (as a percentage of total malignant cells) and intensity (value of 0 for absence, 1 for low intensity, 2 for moderate intensity, and 3 for high intensity) of staining based on visualization. Slides were treated as described above (see IHC). All slides were analyzed in a blinded manner by 2 independent observers.
The association between IHC staining intensity and clinical variables (tumor grade, stage, progression-free interval, and patient survival) was analyzed by using the Mann-Whitney U test. Comparisons of grades were performed between grade 0, grade 1, grade 2, and grade 3. The stages were divided into 2 groups: early (stages I–II) and advanced (stages III-IV). A Spearman correlation test was performed to evaluate the correlation between expression levels of the different proteins. The significance of the proteins in predicting the survival of patients with EOC was analyzed by using Kaplan-Meier survival curves coupled to the log-rank test. The threshold of intensity used in the log-rank test to discriminate between the low- and high-intensity groups was based on the median intensity of TOV tumors (grades 1–3). Patients with follow-up <18 months were not included in this analysis. Statistical analysis was performed with SPSS software (version 11; SPSS Inc., Chicago, Il) with statistical significance set at P < .05.
Staining of Apoptosis-related Protein in Tumor Tissue
We previously performed molecular profiling of 6 LMP tumor tissues and 12 TOV tumor tissues by using Affymetrix GeneCHIP HuGeneFL microarray (Santa Clara, Calif).20 A subset of differentially expressed candidate genes was implicated in death receptor apoptosis, including TRAIL, CASP8, FLIP, and CYC. To evaluate alterations at the protein level of these candidates in EOC, we undertook IHC using tissue microarray composed of 235 tumors of serous histopathology (Table 1). Because TRAIL was identified as a candidate gene, we were interested in determining whether the protein levels of its receptors (Dr4, Dr5, DcR1, and DcR2) also were modulated in EOC. Representative cores from each grade of tumor (grades 0–3) stained with the 8 different antibodies are presented in Figure 1. Trail staining was observed in the cytoplasm of epithelial cells and also diffusely in stromal cells. Staining was located mainly in the cytoplasm but was also associated with a cytoplasmic membrane staining in assays of Dr4, DcR1, and DcR2; whereas Dr5 presented additional nuclear staining. We noted exclusive cytoplasmic staining for Flip, Casp8, and Cyc, and the latter was distinguished by granular staining, consistent with the expected mitochondrial localization of this protein.
Association of Staining Intensity With Histologic Grade
Next, we attempted to determined whether protein expression was associated with tumor grade (Table 2). Statistically significant differences in protein expression, defined by the extent and intensity of staining, were observed for all proteins with the exception of DcR1 and DcR2 when grade 0 tumors (LMP) were compared with grade 2 and 3 tumors. Dr5, DcR1, Casp8, and Cyc had higher expression levels in grade 2 and 3 tumors compared with the levels in grade 0 tumors (P < .01); whereas Trail, Dr4, and Flip had lower expression levels in higher grade tumors (P ≤ .01) (Table 2). Trail and Flip were the only proteins that exhibited differential expression between the 2 most differentiated groups of tumors (grade 0 > grade 1; P = .03 and P < .001, respectively) (Table 2). Statistically significant higher protein expression with increasing grade was observed when comparing grade 1 tumors with grade 2 and/or grade 3 tumors for Cyc (P = .02 and P = .01, respectively) and for Casp8 (P = .05 for grade 2 and 3 tumors), whereas lower expression levels of Trail were observed (P = .04 for grade 3 tumors) (Table 2). Grade 2 tumors expressed significantly higher levels of Trail and Dr4 compared with grade 3 tumors (Table 2).
Table 2. Statistical Analyses of Candidate Expression Determined by Immunohistochemistry and Associated With Tumor Grade
Association of Staining Intensity and Clinical Stage
Because clinical staging is used to describe tumor extension, we assessed whether the staining intensity of apoptosis-related proteins was associated with stage. We grouped clinical stages I and II as early-stage tumors (21 LMP tumors and 14 TOV tumors) and clinical stages III and IV as advanced-stage tumors (35 LMP tumors and 174 TOV tumors) (Table 1). When all tumors (grades 0–3) were analyzed together, we observed significantly higher levels of Dr5 expression in advanced-stage tumors (P = .03) compared with the expression levels of Trail, Dr4, DcR1, and Flip, which were higher in early-stage tumors (P = .01, P = .03, P = .03, and P < .001, respectively) (Table 3). When TOV tumors were considered separately, DcR1 and Flip showed higher expression levels in early-stage tumors compared with advanced-stage tumors (P = .01 and P < .001, respectively). However, because of the small number of early-stage tumors and the strong association with tumor grade, the observed differential expression may have reflected the influence of grade rather than stage.
Table 3. Expression of Apoptosis-related Protein Determined by Immunohistochemistry and Association With Disease Staging
Because one of the hallmarks of cancer is resistance to cell death, we assessed whether our samples exhibited differential indices of apoptosis. One of the methods of evaluating apoptosis linked to signaling from cell death receptors, such as Trail receptors, is to measure the level of activation of the effector Casp3. We tested apoptosis levels by IHC using an antibody specific to cleaved Casp3 (Fig. 2) and visually scored staining as low, moderate, or high. Levels of apoptosis varied in each tumor grade (Table 4); and although, no statistically significant difference was observe, we did note a tendency toward lower activated caspase-3 expression in LMP tumors compared with grade 2 and 3 TOV tumors (P = .08 and P = .09). Next, we assessed whether the level of apoptosis, measured by the presence of cleaved Casp3, was correlated with the expression patterns observed with Trail signaling pathway members. Using a Spearman correlation test, we were unable to establish a significant correlation between any Trail signaling members and the level of cleaved Casp3 (data not shown).
Table 4. Analysis of Apoptosis Monitored by Staining Intensity of Cleaved Capsase-3
Cleaved caspase-3 staining
Grade 0 samples
Grade 1 samples
Grade 2 samples
Grade 3 samples
None or weak
Association of Apoptosis-related Proteins and Patient Survival
Then, we assessed whether the apoptosis-related proteins in our study could predict patient outcomes. We used the median staining intensity in TOV tumors for each protein as the threshold to perform survival analyses based on Kaplan-Meier curves coupled with a log-rank test. We observed that only high expression of the receptor Dr5 was associated with poor survival in patients with invasive tumors (P = .03) (Fig. 3A, Table 5). Furthermore, because tumor grade was related to survival in patients with EOC, we performed an independent survival analysis for each tumor grade. Although no markers were associated with survival in the group of grade 2 tumors, high expression of Dr5 was associated with a poor patient prognosis in the subgroup of patients with grade 3 TOV tumors (P = .03) (Fig. 3B, Table 5).
Table 5. Survival Analysis Performed Using a Kaplan-Meier Curve Coupled to a Log-rank Test
Combination of Apoptosis-related Proteins to Predict Patient Survival
We evaluated whether tumors that exhibited certain patterns of protein expression were associated with tumor aggressiveness and influenced prognosis. We coupled Kaplan-Meier curves with a log-rank test to assess the significance of proteins in defining patient prognosis. The results of all protein analyses were combined in a pair-wise manner in which the 4 different combinations of expression levels were considered: high/high, high/low, low/high, and low/low. One combination presented a statistically significant association (Dr4 and Flip; P = .02) (Fig. 4A), whereas 4 others showed a trend toward an association with patient prognosis (DcR1 and Dr5; P = .06 [Fig. 4B]; Trail and Dr5; P = .09 [Fig, 4C]; Flip and Dr5; P = .09 [Fig. 4D]; and DcR2 and Cyc; P = .08 [Fig. 4E]). When all 4 combinations were compared, only Flip and DcR2 failed to show an association (P = .23) (Fig. 4F), although this result reflects the inability of markers to stratify patients in the early part of the curve (before 18 months). Subsequently, we refined the analysis to include only the 2 curves that defined the highest difference, and these selected, pair-wise combinations were compared further (Fig. 5). Tumors that expressed high levels of Dr4 and low levels of Flip conferred a worse prognosis compared with all other combinations of expression of these 2 markers (P = .006, P = .01, and P = .04) (Fig. 5A-C, respectively). Patients who presented with tumors that expressed low levels of both DcR1 and Dr5 had a better prognosis compared with patients who presented with tumors that expressed high levels of both DcR1 and Dr5 (P = .07) (Fig. 5D) and also compared with patients who presented with tumors expressed low levels of DcR1 and high levels of Dr5 (P = .02) (Fig. 5E). Tumors that expressed high levels of Flip and low levels of Dr5 conferred a better prognosis compared with tumors that expressed low levels of Flip and high levels of Dr5 (P = .02) (Fig. 5F). Tumors that expressed low levels of both DcR2 and Cyc were associated with longer survival than tumors that expressed high levels of DcR2 and low levels of Cyc (P = .04) (Fig. 5G). Patients who had tumors that expressed high levels of Trail and low levels of Dr5 had longer survival than patients who had tumors that expressed low levels of Trail and high levels of Dr5 (P = .04) (Fig. 5H). Finally, although comparisons of all combinations of Flip and DcR2 were not significant, we did note that the combination of high expression of Flip and low expression of DcR2 seemed to confer a better prognosis compared with all other combinations of these 2 markers (P = .09, P = .10, and P = .03) (Fig. 5I–K, respectively).
For this report, we conducted a study on Trail signaling using tissue microarray and focused on the protein expression of Trail, its receptors (Dr4, Dr5, DcR1, and DcR2), other members of the pathway (Casp8, Cyc), the level of apoptosis in EOC, and correlated their expression with clinical factors, such as tumor grade and prognosis. In EOC, Trail signaling not only may influence the growth of cancer cells per se, but it may determine immunologic responses and could reflect the patient sensitivity or resistance to chemotherapy. Thus, it is the interplay of all of these factors that will influence disease endpoints and, ultimately, patient prognosis.
We demonstrated that Trail (Tnfsf10, Apo2L) was overexpressed in LMP tumors compared with every grade of invasive tumor, it was underexpressed in grade 3 TOV tumors compared with grade 1 and 2 TOV tumors, and it was overexpressed in early-stage disease versus advanced-stage disease when all tumors were included in the analysis. However, when LMP and TOV tumors were considered independently, no such association was observed. These results are consistent with those in the literature showing either Trail overexpression in early-stage EOC compared with advanced-stage EOC, or no association with disease stage.21, 22 However, at the RNA level, TRAIL was overexpressed in poorly differentiated (grade 2 and 3) tumors compared with well-differentiated (grade 1) tumors.23 This discordance between protein levels and RNA levels was reported previously in breast cancer24; it also was reflected in our microarray analysis, in which grade 3 TOV tumors expressed higher levels of TRAIL than LMP tumors, and in another published microarray study in which TOV tumors were compared with normal ovarian surface epithelium cells.25 In colon cancer, at the protein level, results similar to ours were reported in which Trail expression was lower in adenomas compared with adenocarcinomas.26
We observed the usual cytoplasmic and membrane staining of all Trail receptors, whereas nuclear staining also was observed for Dr5. Although it was unexpected, nuclear staining for a Trail receptor previously was reported in colon tumors.26 We observed that grade 2 and 3 TOV tumors had higher expression of Dr5 compared with LMP tumors but did not demonstrate the differential expression of DcR2 and DcR1 compared with LMP tumors. Our results are consistent with an investigation of colon cancers in which Dr5 levels were higher in carcinoma cells compared with normal cells, and no difference was observed for DcR1 or DcR2.26
One of the hallmarks of cancer is evasion from apoptosis.27 This process can be accomplished in different ways, including a reduction of death receptor, ligand, and effector levels or the up-regulation of antiapoptotic proteins by genetic and/or epigenetic changes. In EOC, no mutations in Trail receptors were observed.28 However, epigenetic silencing of TRAIL receptors was reported in different cancer types, including malignancies of the breast, lung, bladder, and cervix as well as lymphoma, leukemia, myeloma, neuroblastoma, and EOC29–32 (for review, see Debatin and Krammer9). Casp8, which is an effector of apoptosis, was inactivated by hypermethylation in a number of different tumors derived from neuroblastoma, brain tumors, Ewing sarcoma, and small lung cell carcinoma (for review, see Debatin and Krammer9). However, in EOC cell lines, no loss of expression of Casp8 was reported in Trail-resistant cell lines.33
We observed that LMP tumors exhibited higher protein levels of Flip than grade 1, 2, or 3 TOV tumors. Flip is overexpressed in several tumors (Burkitt lymphoma, pancreatic carcinoma, melanoma, and neuroblastoma), and this increased expression is associated with resistance to chemotherapy in many types of cancers, including EOC10, 34 (for reviews, see Debatin and Kramer,9 Fraser et al.,25 and Ozoren and El-Deiry36). Although a higher level of apoptosis was observed in our analysis of grade 2 and 3 TOV tumors compared with grade 0 tumors, the difference was not statistically significant (P = .08 and P = .09, respectively), in agreement with another study in which the apoptotic index in ovarian tumors was higher in high-grade tumors compared with low-grade tumors.37 In cell lines from different cancers, including EOC, sensitivity to Trail was associated with high expression levels of Dr4 and Dr510, 38 and reduced expression levels of DcR1 (for review, see Ozoren and El-Deiry36). Other studies, including those on EOC cell lines, have revealed that only Dr4 levels, and not Dr5 levels, are correlated with Trail sensitivity10, 33, 39, 40 and that the ratio of death and decoy receptors do not correlate with resistance to Trail.14
In the current work, we observed that high expression of Dr5 was associated with a worse patient prognosis either when TOV tumors of all grades were included or when only the subgroup of grade 3 TOV tumors was included, which is consistent with a previous report on lung cancer.41 In addition to its association with patient survival, Dr5 also was associated with tumor grade and disease stage, suggesting a role in EOC, although further molecular studies will be needed to address this point. We also demonstrated that, even if a protein alone is not associated significantly with patient prognosis, pairing its expression with another protein could result in a significant association. In this study, the combinations of Dr4/Flip, DcR1/Dr5, Flip/DcR2, Trail/Dr5, Flip/Dr5, and Cyc/DcR2 were associated with patient survival. It is noteworthy that the highest significance of any pair-wise combination appeared to involve combining Flip or Dr5 with another protein. We noted that Dr5 alone displayed greater significance for patient survival compared with all combinations that involved Dr5, although combinations that involved Dr4/Flip were as significant as Dr5 alone. These findings suggest that a strategy that sums the activities of different partners within a pathway may be more appropriate in designing nomograms for patient stratification. We have highlighted the advantage of looking at several proteins implicated in the same signaling pathway and the importance of holistic approach when dealing with a specific pathway, because the sum of the deregulation of each member is what generates the downstream effect of the pathway. In particular, this appeared to be reflected in the survival analysis; because the combination of 2 markers, instead of a protein alone, was required to significantly define patient prognosis.
In conclusion, to our knowledge, the current study is the first to simultaneously study both Trail and its 4 receptors in EOC. In this report, we demonstrated that EOC tumors present the deregulation of apoptosis-related proteins that also are related to tumor differentiation. This deregulation is accompanied with a trend toward higher apoptosis levels in grade 2 and 3 TOV tumors compared with LMP tumors. Although we demonstrate that only Dr5 by itself was associated significantly with patient prognosis, our results indicated that a combination of markers together can provide relevant information on patient prognosis, suggesting that the interplay of members along a common signaling pathway influence the ultimate outcome. It remains to be determined in a larger series whether these proteins can be used clinically as markers of disease progression, either by themselves or in combination with proteins from other pathways.
We are grateful to Louise Champoux, Lise Portelance, Manon de Ladurantaye, Marise Roy, and Stephanie Girard for technical assistance. We thank the gynecologic oncologists of the University of Montreal Central University Hospital Research Center (CHUM) for providing specimens. We are grateful to laboratory members for thoughtful discussions. We acknowledge the editorial assistance of Mr. Ovid Da Silva and the statistical assistance of Mr. Robert Boileau (Research Support Office, Research Center, CHUM)