ANATOMIC PATHOLOGY Original Article

Transcription

1 ANATOMIC PATHOLOGY Original Article The Combined Determination of Proliferative Activity and Cell Density in the Prognosis of Adult Patients With Supratentorial High-Grade Astrocytic Tumors ROBERT KISS, PhD, 1 OLIVIER DEWITTE, MD, 2 CHRISTINE DECAESTECKER, PhD, 3 ISABELLE CAMBY, PhD, 1 LAURENCE GORDOWER, 1 KATTY DELBECQUE, MD, 1 JEAN-LAMBERT PASTEELS, MD, 1 JACQUES BROTCHI, MD, 2 AND ISABELLE SALMON, MD 4 Tumor growth represents the ratio between cell gain (number of mitoses per unit of time, ie, proliferative activity) and cell loss (number of cell deaths during the same unit of time). While in adults, proliferative activity parallels the level of malignancy in astrocytic tumors and therefore represents a useful diagnostic marker, cell loss has never been concomitantly assessed in tumors of this type. We hypothesize that cell density assessable on histologic slides represents the ratio between cell gain and cell loss. This hypothesis concerns only the diffuse type of astrocytic tumors. Proliferative activity (assessed by MIB1 antigen immunostain) and cell density were thus quantitatively assessed by means of a cell image processor in a series of 54 supratentorial astrocytic tumors of adult patients, which included 15 astrocytomas (ASTs), 18 anaplastic astrocytomas (ANAs), and 21 glioblastomas (GBMs). The results show that proliferative activity and cell density were highly correlated (P=.003) and that both correlated with histopathologic grade. The patients with a high-grade astrocytic tumor (ie, ANA or GBM) that exhibited a low level of In adults, astrocytic tumors represent the most common primary tumors of the central nervous system and include low-grade tumors, ie, astrocytomas From the Laboratoire d'histologie, Faculte de Medecine; ^Service de Neurochirurgie and Service de Pathologie, Cliniques Universitaires de Bruxelles; 3 lnstitut de Recherche Interdisciplinaire et de Developpements en Intelligence Artificielle (IRIDIA); Universite Libre de Bruxelles, Brussels, Belgium. R.K. and l.c. are Research Associate and Senior Research Assistant, respectively, with the Fonds National de la Recherche Scientifique (FNRS, Brussels, Belgium). Supported by grants awarded by the Fonds de la Recherche Scientifique Meciicale, Belgium. Manuscript received June 25, 1996; revision accepted September 27,1996. Address reprint requests to Dr Kiss: Laboratory of Histology, Faculty of Medicine, Free University of Brussels, 808 route de Lennik, 1070 Brussels, Belgium. proliferative activity but high cell density survived for significantly shorter periods than did patients with a tumor that exhibited low proliferative activity and low cell density (P=.002). The patients with a high-grade astrocytic tumor that exhibited high proliferative activity and high cell density survived for significantly less time than did the patients with a tumor that exhibited high proliferative activity but low cell density (P<.05). A marked difference in survival periods was observed between the patients with a high-grade astrocytic tumor that exhibited a low level of proliferative activity and low cell density and the patients with a tumor that exhibited a high level of proliferative activity and high cell density (P<.001). The concomitant determination of proliferative activity and cell density seems likely to enable determination of the few adult patients who have high-grade astrocytic tumors and who will survive for a considerable period (several years). (Key words: Astrocytic tumor; Proliferative activity; Cell density; Cell loss; Patient survival; Feulgen staining; Quantitative measurements; Image cytometry.) Am J Clin Pathol 1997,107: (ASTs), and high-grade tumors, ie anaplastic astrocytomas (ANAs) and glioblastomas (GBMs). 1-2 Highgrade astrocytic tumors kill more than 90% of adult patients fewer than 5 years after the initial diagnosis. 3 However, a few patients with high-grade astrocytic tumors survive more than 5 years. Using current prognostic analysis, it has not been possible to identify the latter group effectively. 3,4 The prognosis for astrocytic tumors is established by means of classic criteria. The most useful of these criteria are patient age and histologic tumor grading. 5,6 However, recent knowledge of the molecular mechanisms transforming normal astrocytes into tumor cells 7 (see Discussion) enables the use of other prognostic criteria for patients with high-grade astrocytic tumors identified by histopathologic diagnosis. The increase in the level of malignancy from ASTs through ANAs to GBMs is associated with an accumulation of 321

2 322 ANATOMIC PATHOLOGY Original Article genomic aberrations, 7 which, in turn, is associated with an increase in cell proliferation. The proliferation rate of an astrocytic tumor is thus related to its level of malignancy. This is the reason that the level of proliferative activity is determined increasingly more frequently in addition to histopathologic grading However, as emphasized by Coons and Johnson, 15 ' 16 the regional heterogeneity expressed by astrocytic tumors has an adverse effect on the usefulness of proliferative activity analysis. The independent variation in proliferative activity and histologic features suggests that the use of multiple analyses may allow a more accurate prediction of clinical behavior of astrocytic tumors in adult patients. 15,16 In addition to the regional heterogeneity in astrocytic tumors (intratumoral heterogeneity), intertumoral heterogeneity also occurs. In fact, our experience is based on a wide range of patients with high-grade astrocytic tumors who have been treated in our hospital since 1982, and this experience reveals that patients with similar clinical characteristics (eg, age, histopathologic grade, neurologic performance status), and who exhibit similar proliferative activity survive for different periods. This may be partially explained by the fact that proliferative activity is only one of the two biologic mechanisms that regulate tumor growth. Indeed, tumor growth is the balance between cell gain (number of mitoses per unit of time) and cell loss (number of cell deaths during the same unit of time). Proliferative activity relates to cell gain only. Thus, a tumor with a high level of proliferative activity but a high cell loss grows slowly. In contrast, a tumor with a relatively low level of proliferative activity but no cell loss grows more rapidly. Except for a study by Mundle et al, 17 who used an admirable but tedious (and inapplicable in a diagnostic routine) technique for estimating cell loss in leukemia, the estimation of cell loss is not used to assess the proliferation of tumors in general or of astrocytic tumors in particular. The aim of our study was to present an original method that makes assessment of proliferative activity and cell density (the balance between cell gain and cell loss) possible. Our series included 54 supratentorial astrocytic tumors in adult patients (AST, 15; ANA, 18; GBM, 21). Cell density was assessed on 5- um-thick Feulgen-stained sections by means of the Delaunay triangulation and Voronoi paving mathematical techniques included in the software used by the cell image processor. 18 Proliferative activity was assessed by means of the proliferation index 14 and the anti-mibl antibody, 19 which enables detection of the Ki-67 antigen. 20 Only the diffuse type of astrocytic tumors was considered. Cell density, as measured in this paradigm, may not simply reflect cell loss. Indeed, cell density, as measured in our system, may also reflect properties of intercellular adhesion and nuclear-cytoplasmic ratios. Cellular adhesion may, in turn, be affected by the extracellular matrix (as modified by tumor cells), cell surface receptors, and the formation of intercellular junctions or glial networks. However, the relevance and potential importance of the method and the derivative data are not changed because the parameter of cell density may not be directly related only to the ratio of cell gain to cell loss. Histopathological MATERIALS AND METHODS Diagnosis and Clinical Data The 54 patients included in the study underwent surgery at the Erasmus Hospital between 1984 and All 54 astrocytic tumors were infiltrating cerebral astrocytomas. The histopathologic diagnosis was determined as previously described, 18,19,21 ie, in accordance with the classification proposed by Burger et al, 1 ' 5 Kleihues et al, 2 and VandenBerg. 21 Our series of 54 astrocytic tumors included 15 ASTs, 18 AN As, and 21 GBMs. Of the 39 patients with high-grade astrocytic tumors (ie, ANA or GBM), 34 died because of the astrocytic tumor. Five patients with a high-grade astrocytic tumor (ANA, 3; GBM, 2) were still alive at the time of the analysis. Of the 15 patients with AST, 7 were still alive at the time of the analysis, but 8 had died. The 28 males and 26 females ranged in age from 16 to 83 years. All patients with high-grade astrocytic tumors had undergone surgery and had received postoperative radiation. The ages of patients with high-grade astrocytic tumors ranged from 42 to 83 years. Proliferative Activity Determination Anti-MIBl Antibody by Means of the Specimen preparation. Two paraffin blocks containing formalin-fixed tissues were available for each of the 54 astrocytic tumors under study. Six histologic slides were taken from each block. All these slides except the fifth were 5 um thick; the fifth was 80 um thick. In each series of six slides from one block, the first and the sixth were hematoxylin-eosin stained for diagnostic purposes and for the delineation of the tumor area to be analyzed. The "area of interest" was marked on the AJCP«irch 1997

3 KISS ET AL 323 Short and Long Survival in Astrocytic Tumors of the Adult first histologic slide and then traced onto the three subsequent slides (Nos. 2-4), which had been cut serially from the same block. The fifth slide was used following the method subsequently described (see "Feulgen-stained samples: determination of the proliferation index and cell density"), which made it possible to obtain DNA histograms. The sixth slide was the control for the method, ie, to ascertain that the blocks contained tumor tissue of the same type as that diagnosed on Slide 1. Immunohistochemical staining and quantitative measurements. As detailed elsewhere, 20 Slide 2 of each block was immunostained using the MIBl antibody (monoclonal, Immunotech SA, code No. 0505, clone MIB-1, dilution 1/50). Slide No. 3 served as a negative control (absence of a primary antibody) for Slide 2. Quantitative proliferative activity measurements were done using a SAMBA 2005 cell image processor (Alcatel-TITN, Grenoble, France) coupled to a JVC (KY15) color camera (JVC, Brussels, Belgium) and a Leitz (Diaplan) microscope (Leitz, Brussels, Belgium) with a 40x magnification lens (numerical aperture, 1.30). On each pixel, the tissue-integrated optical density (IOD) relating to the hematoxylin counterstaining and the specific immunohistochemical staining was computed on 256 densitometric levels in two distinct color channels. The IOD corresponds with the sum of the optical density values of each pixel of the nucleus. Immunohistochemical positivity for the MIBl antibody (Slide 2) was defined by the integrated optical values in excess of the mean plus two standard deviations of the corresponding negative control slides (Slide 3). The mean optical density (MOD), which relates to immunohistochemical staining intensity, was calculated on each cell nucleus. This MOD value was obtained by dividing the integrated optical density value for the MIBl immunostain of a given cell nucleus by its area. This MOD value calculation made possible avoidance of problems related to heterogeneity distribution in nuclear sizes. In each area of the tumor analyzed, the percentage of proliferating cells (ie, MIBl-immunopositive cells) were assessed in three fields of 200 to 250 cells for each histologic slide. These three fields were selected randomly in the tumor section, ie, we did not focus only on the most proliferative, highly cellular areas of a tumor. Thus, the proliferative activity of each astrocytic tumor was determined on 1,200 to 1,500 cells when 2 blocks were available for the tumor and on 1,800 to 2,250 cells when 3 blocks were available. This large Staining Intensity 200 l _ da nd n cp Concentration Analysis of 227 cell(s) Labeling Index = 6.61% Staining Intensity a ^ QOI Concentration Analysis of 225 cell(s) Labeling Index : 26.67% FIG 1. Proliferative activity determination. A, A weakly proliferating anaplastic astrocytoma. B, A highly proliferating anaplastic astrocytoma. The determination of the proliferating cells was made by the MIBl immunostain (in 5 um-thick sections from formalin-fixed paraffin-embedded materials). This made possible the identification of the Ki-67 proliferating cell nuclear antigen that is expressed in all active parts of the cell cycle, but not in quiescent cells. The open squares located in the upper right part of the figure indicate MIBl-positive (proliferating) cells. Open squares in the bottom left indicate the MIBl-negative (nonproliferating) cells; these open squares overlap the black crosses that represent the negative control cells. The present quantitative analysis includes the determination of MIBl-positive cells in one (containing 227 and 225 cells, respectively) of the five fields analyzed for one of the two slides analyzed for each astrocytic tumor under study. (See Methods section for details.) Vol. 107 No. 3

4 324 ANATOMIC PATHOLOGY Article number of fields was analyzed to prevent serious problems due to local heterogeneity in astrocytic tumors. 15 ' 16 The blood vessels in an astrocytic tumor are often more cellular and proliferative than the surrounding tumor. These blood vessels were excluded in this analysis. Figure 1 shows the determination of proliferating cells, ie, MIBl-positive cells, vs nonproliferating cells in one of five fields examined in one of the two slides analyzed for a weakly (Fig 1, A) and a highly (Fig 1, B) proliferating anaplastic astrocytoma. sists of the computer transformation of the analogical Feulgen-stained images (Fig 2, A) into digitized Feulgen-stained nuclear images (Fig 2, B), and the subsequent transformation of these nuclear images into nodes representing the centers of gravity of the digitized nuclear images (Fig 2, C). In other words, each point of intersection (node) in the network illustrated in Figure 2,C corresponds with the barycenter of a cell nucleus. The software developed in our laboratory 18 makes it possible to calculate the area (in um 2 ) in which the number of cell nuclei (nodes) will be determined, to count the number of nodes in this area, and to determine the cell density by dividing the number of cell nuclei per area analyzed and multiplying this result by 10,000. Twenty cell density measurements were performed for each area described in "Specimen preparation." Thus, 40 to 60 measurements were performed per astrocytic tumor, depending on whether two or three blocks were available per tumor. Statistical analysis. The results are given as the mean ± SD and ± SEM. The statistical comparisons of the data were performed by means of the Fisher F test (one-way analysis of variance) after determination of the equality of variances by the Bartlett test and the normal distribution fit of the data by the % 2 test. The sets of ranks were tested with the nonparametric Kendall coefficient of concordance. Statistica/DOS software (Statsoft, Tulsa, Okla) was used for all statistical analyses. All the parameters quantitatively determined by means of the cell image processor were calculated without knowledge of the histologic data. RESULTS Relationship 'Among Proliferative Activity, Density, and Histopathologic Grades Cell When the MIB1 value distribution was plotted against the PI value distribution, statistical significance was achieved (P=.03; Fig 3, A). Whereas 54 astrocytic tumors were under study (AST, 15; ANA, 18; GBM, 21), the correlation could be completed for only 39 cases because the PI can be calculated on monomorphic DNA histograms only. Four ASTs, seven ANAs, and four GBMs exhibited polymorphic nuclear DNA content (data not shown). The level of statistical significance between the cell density and proliferative activity (as assessed by means of the MIB1 immunostain) value distributions was P=.003 (Fig 3, B). Subsequently, we used only the MIBl-related results because the MIB1 antigen immunostain made Feulgen-stained samples: determination of the proliferation index and cell density. Slide 4 was subjected directly to Feulgen-staining for measurement of cell density. Slide 5 (80 um thick) was subjected to a method described in detail elsewhere. 22 This method makes it possible to obtain single-cell nuclei suspensions (after pronase digestion) that are centrifuged onto glass slides (cytospins) and then stained by the Feulgen reaction. The nuclear DNA content of each astrocytic tumor under study was assessed by means of a DNA histogram computed on 500 to 1,000 Feulgen-stained cell nuclei. The computation was done with the same cell image processor as that described in "Immunohistochemical staining and quantitative measurements." The procedure for the measurement of nuclear DNA content is detailed elsewhere. 23 The proliferation index (PI) was calculated on each DNA histogram. The PI represents the percentage of cells engaged in the S phase of the cell cycle. Thus, whatever the ploidy level of the tumor, 14 the PI corresponds with the area under the curve between the two peaks Gl and G2. In accordance with the procedure we developed, each DNA histogram is computed on 50 IOD classes. In cases of diploid and hyperdiploid DNA histograms, the PI corresponds with the percentage of cells included in the five IOD classes after the Gl peak; for triploid and hypertriploid DNA histograms, the PI includes the percentage of cells in the six IOD classes after the Gl peak; and for tetraploid DNA histograms, the PI includes the percentage of cells in the seven IOD classes after the Gl peak. 14 The PI can be determined on monomorphic DNA histograms only because in polymorphic DNA histograms the Gl and S-cell populations overlap. 14 Therefore, the Pis were computed for only 39 of the 54 astrocytic tumors under study. The cell density was assessed by using a method described elsewhere. 18 Briefly, this method is based on the Delaunay triangulation and Voronoi diagrams restricted to two-dimensional studies. The method con- AJCP- irch 1997

5 BQSS ET AL 325 Short and Long Survival in Astrocytic Tumors of the Adult 11.^oo 1 I i j. +. t.j. A_ FIG 2. Schematic illustration of the cell density. A, Assessed by using a cell image processor on 5-um-thick Feulgen-stained histologic sections. The method, which relies on the use of the mathematical Delaunay triangulation and Voronoi paving techniques, consists of replacing the digitized Feulgen-stained nuclear images (B, black ovals) with nodes representing their centers of gravity (C). The cell density corresponds with the number of nodes divided by the area analyzed and multiplied by 10, > 75 C 70 > 65 g.60 o 55 C 50 g C Ml...,.,,,!..., A ^ Proliferative Activity (%) (MIB1 immunostain) ;...#... r*f^r :!:- *-» '*'» J * : " I : "~» ;-.-." "1. --^Z:?:::::»::::::;:::::::: ; I^* "i " " i!' "!. :...:::;::: ^..^... i i Proliferative Activity (%) (MIB1 immunostain) FIG 3. Relationship (A) between the proliferation index and proliferative activity (see Fig 1) values, and (B) between cell density (see Fig 2) and the proliferative activity values in a series of 54 astrocytic tumors in adults, including 15 astrocytomas, 18 anaplastic astrocytomas, and 21 glioblastomas. Cell density is expressed in arbitrary units. it possible to determine the proliferative activity in all the astrocytic tumors independent of DNA ploidy level, while PI determination did not. Because the goal of our study was to investigate whether cell density determination contributes additional prognostic information to measurements of proliferative activity, the results relate only to the patients for whom we obtained complete clinical follow-up data evidencing astrocytic tumor-induced death. The patients still alive at the time of the analysis were not included in the analysis of the survival data. The values for the mean proliferative activity (Figure 4, A) and cell density (Figure 4, B) were significantly lower in the AST group than in the ANA and GBM groups (P<.01). In contrast, the mean proliferative activity (see Figure 4, A) and the cell density (see Figure 4, B) values were similar in the ANA and GBM groups (P>.05). -~i Vol. 107 No. 3

6 326 ANATOMIC PATHOLOGY Original Article Figure 4C illustrates the survival period (elapsing between the first histopathologic diagnosis and patient death) in the AST (n = 8), ANA (n = 15), and GBM (n = 19) groups of patients who died of their disease. While the survival period in the GBM group seemed to be lower than in the ANA group, this difference was not statistically significant (P>.05). In contrast, patients in the AST group who died of their disease survived significantly longer than patients in the ANA and GBM groups (P<0.01; see Figure 4C). TST AST "GBJT GBM FlG 4. Histopathological groups. Development of the mean (black squares), the standard error (hatched rectangles) and the standard deviation (open rectangles) values from astrocytomas (AST, n = 8) through anaplastic astrocytomas (ANA, n = 15) to glioblastomas (GBM, n = 19) for proliferative activity (A), cell density (B, expressed in arbitrary units), and survival period (C). The values for proliferative activity, cell density, and survival period were calculated on a closed set of patients, eg, patients who had died by the time of analysis. A - C : Does Determination of Cell Density Contribute Additional Prognostic Information to the Determination of Proliferative Activity? Figures 5A and 5B illustrate the distributions of proliferative activity and cell density values in the 39 ANAs and GBMs under study. Of the 39 patients with these tumors, 34 had died of their disease by the time of the analysis. Their survival period distribution is illustrated in Figure 5C. Of the five patients with highgrade astrocytic tumors who were alive at the time of the analysis, three had ANA and two had GBM. The value distributions (Figures 5A through 5C) made identification of various pathologic and clinical subgroups possible. For proliferative activity, highgrade astrocytic tumors exhibiting an MIB1 index less than 8% were labeled weakly proliferating, while those exhibiting a MIB1 index greater than 8% were labeled highly proliferating. The high-grade astrocytic tumors with a cell density less than 50 were labeled weakly dense, while those with a cell density greater than 50 were labeled highly dense. The patients who died before 50 weeks after the first diagnosis of the astrocytic tumor were classified as the low survival group (n = 16); those surviving between 50 and 100 weeks were classified as the intermediate survival group (n = 11); and those surviving more than 100 weeks were classified as the high survival group (n = 7). The cutoff values were chosen to obtain similar numbers of cases in each subgroup. Comparison of the three survival subgroups showed the mean proliferative activity values to be similar in the low and intermediate survival subgroups (P>.05); the mean values of the low and intermediate survival subgroups were significantly higher than the mean for the high survival subgroup (P<.01). In contrast with the observation for the value for development of proliferative activity, the mean cell density value was significantly lower in the high survival subgroup than in the low survival subgroup (P<.05), and the mean cell density value in the intermediate subgroup did not differ significantly from that in the high survival group (P>.05; Fig 6, B). Figure 7 illustrates the additional prognostic value contributed by determination of cell density to the determination of proliferative activity. Subgrouping the AJCP March 1997

7 KISS ET AL 327 Short and Long Survival in Astrocytic Tumors of the Adult 34 high-grade astrocytic tumors on the basis of their proliferative activity and cell density values (see Figures 5, A and 5, B), ie, according to whether they were weakly proliferating/weakly dense (WPI-WCD group, n = 5), weakly proliferating/highly dense (WPI- HCD, n = 20), highly proliferating/weakly dense (HPI- WCD, n = 4), or highly proliferating/highly dense (HPI-HCD, n = 5) shows that the survival progressively Distribution of MIB1 Percentages Distribution of Cell Density Values Distribution of Survival Periods FlG 5. Distribution of the proliferative activity (A), the cell density (B), and the survival period (C) in 39 supratentorial high-grade astrocytic tumors in adults, including 18 anaplastic astrocytomas and 21 glioblastomas. Of the 39 patients, 34 patients died by the time of the analysis (C). The prognostic value of the proliferative activity and cell density was evaluated on this closed series of 34 patients. decreased from the WPI-WCD subgroup to the WPI- HCD subgroup and the HPI-WCD subgroup to the HPI-HCD subgroup (Fig 7). The following levels of statistical significance were obtained for the paired comparisons: WPI-WCD vs WPI-HCD, P<.01; WPI-HCD vs HPI-WCD, P>.05; HPI-WCD vs HPI-HCD, P<.05; and WPI-WCD vs HPI-HCD, P<.001. The astrocytic tumor subgrouping described herein was independent of histopathologic tumor grade. The WPI-WCD subgroup included three ANAs and two GBMs, the WPI-HCD subgroup included eight ANAs and 12 GBMs, the HPI-WCD subgroup included two ANAs and two GBMs, and the HPI-HCD subgroup included two ANAs and three GBMs. DISCUSSION As emphasized by Ganju et al, 12 the ability to divide subsets of patients with glial neoplasms into prognostic groups is limited because only a few clinical and pathologic variables can be used to do so. Furthermore, the determination of prognostic variables in astrocytic tumors of the adult is not easy because of the heterogeneous biologic character of these tumors. This heterogeneity is caused by at least two distinct phenomena. The first phenomenon is summarized by Gillapsy et al, 24 who report that increasing evidence suggests that in mammals, astrocytes are a heterogeneous family of cells that share certain properties, but differ in lineage and biochemical and functional aspects. These authors state that GBMs arising from glial precursors may also represent a family of related, but distinct, cell types. The second phenomenon concerns the molecular events that transform normal astrocytes into malignant cells, and is summarized by Collins and James, 7 who report that the earliest events in glioma progression include a loss of genetic information from the long arms of chromosomes 13 or 22 or the short arm of chromosome 17, for which targeting of the p53 tumor suppressor gene has been indicated. The loss of a single complement of type-i interferon genes from 9p and loss of genetic information from 19q are seen in rumors with an intermediate grade. 7 Events associated with the highest grade of astrocytic tumors include the loss of the second type-i interferon gene complement and of genetic information from chromosome 10 and gene amplification, most commonly that coding for the epidermal growth factor receptor. 7 The accumulation of these genomic aberrations leads to a deregulation of the cell cycle and a parallelism between an increase in malignancy and an increase in proliferative activity. This is why, in addition to Vol. 107 No. 3

8 328 ANATOMIC PATHOLOGY Original Article histopathologic grading, the proliferative activity measurement, 8-16 and, to a lesser degree, the DNA ploidy level are increasingly used for the determination of the malignancy level of astrocytic tumors in adult patients. However, Coons and Johnson 15 state that because of the regional heterogeneity, multivariate analysis of proliferative activity, DNA ploidy level, and histopathologic grade must be done to allow more accurate predictions of survival for patients with a glioma. We had previously used multivariate analyses to integrate the information contributed by conventional prognosticators (such as patient age and histopathologic tumor grade) with that contributed by biologic prognosticators, including proliferative activity and DNA ploidy level. Using the polynomial function of the Cox model, we showed that in addition to patient age and histopathologic tumor grade, the DNA ploidy level is also a potent independent prognosticator for astrocytic tumors of the adult 25 if the DNA ploidy level is determined by using the DNA histogram type and not the DNA index. Six ploidy levels were used in the assessment of DNA ploidy level determined by the DNA histogram type. Those levels included diploid, hyperdiploid, triploid, hypertriploid, tetraploid, and polymorphic nuclear DNA content patterns. 26 This proposal to use six distinct levels in the nuclear DNA content was recently validated by using a theoretical model. 27 Of the six ploidy levels, hypertriploid astrocytic tumors were associated with a low proliferative level of activity when compared with nonhypertriploid astrocytic tumors. 14 This feature partly explains why the adult patients with hypertriploid astrocytic tumors survived significantly longer than the patients with nonhypertriploid astrocytic tumors. 25 Unfortunately, hypertriploidy only occurs in 10% to 15% of supratentorial astrocytic tumors in adults, 25,26 making this prognosticator useful for only a minority of patients with an astrocytic tumor. Therefore, the second method to determine new prognostic factors for astrocytic tumors was designed for eventual use in a larger population of patients. This second method uses two complementary computerassisted methods. 23 The first method uses the digital image analysis of Feulgen-stained nuclei, making possible the computation of 15 morphonuclear and 8 DNA ploidy-related parameters. The second, which is the decision-tree technique and which forms part of the Supervised Learning Algorithms, enables the most discriminatory parameters to be determined. This method makes possible the identification of two subgroups of adult patients with ANAs, ie, a subgroup of ANA patients with a survival period similar to that of I Weak Intermediate High Survival Period-Related Groups Weak Intermediate High Survival Period-Related Groups FIG 6. Development of the mean (black squares), standard error (hatched rectangles), and standard deviation (open rectangles) values relating to proliferative activity (A) and cell density (B, expressed in arbitrary units) in three groups of 34 adult patients who died of high-grade (anaplastic astrocytomas and glioblastomas) astrocytic tumors. The three groups included Weak (survival <50 weeks, n = 16), Intermediate (survival between 50 and 100 weeks, n = 11), and High (survival >100 weeks, n = 7). The cutoff values defining the survival periods were chosen, on the basis of the distribution values illustrated in Figure 5C, to obtain a similar number of patients in each of the three groups. the patients with ASTs and a second subgroup of patients with ANAs with a survival period similar to that of patients with GBM. 23 The method that we developed 23 resembles the method developed by Scheithauer et al, 12 who used a classification and regression tree using clinical, pathologic, flow cytometric, cytogenetic, and molecular genetic variables to show new potential prognostic factors for patients with glioma. For example, these AJCP March 1997

9 KISS ET AL 329 Short and Long Survival in Astrocytic Tumors of the Adult authors showed that multivariate classification and regression tree analysis identified several subsets of patients with different prognoses, and in one of these subsets, ie, the patients younger than 66 1/2 years with grade 4 astrocytic tumors, the survival period was longer for those with aneuploid tumors than for those with nonaneuploid tumors. These results parallel our results, which show a survival bonus for patients with hypertriploid tumors (a specific subset of aneuploid tumors) as compared with patients with nonhypertriploid tumors (including diploid, ie, nonaneuploid tumors). 25 These results show that computer-generated variables analyzed by multivariate analyses or algorithms relating to artificial intelligence contribute valuable prognostic information on high-grade astrocytic tumors in adults. However, these methods are sophisticated and are not available for all laboratory routines. Thus, these methods are not available for neuropathologists who must make diagnoses and offer prognostic information to clinicians and neurological surgeons. A large gap exists between a way to apply these computer-assisted methods and the routine use of the anti-mibl antibody. The anti-mibi antibody makes possible the determination of the proportion of proliferating cells on formalin-fixed paraffin-embedded histologic sections. 19 This proportion of cells exhibits the Ki-67 antigen, which is a human proliferating cell nuclear antigen that is expressed in all the active parts of the cell cycle, but not in quiescent cells. 28 It may be assumed that the determination of proliferative activity in an astrocytic tumor may provide the neuropathologist with valuable diagnostic, and even prognostic, information. However, this proliferative activity concerns only one of the two biologic compartments (cell gain, or proliferative activity, and cell loss, or cell death, eg, apoptosis), that regulate tumor growth. A number of methods are useful in the determination of the cell loss compartment but are too complex or too tedious to be applied in pathological routines. 29 In contrast, the method that we propose, which associates the determination of the proliferative activity and cell density, could easily be applied in diagnostic routines. In this article, we describe a method used to validate our working hypothesis. This method uses two types of software, the first computing the proliferating cell rate and the second the cell density. Both are computed by means of a cell image processor. The determination of the proliferating cell rate is usually undertaken without specialpurpose machines in most nonspecialty laboratories, that is, by the direct semiquantitative counting of MIBl-positive or Ki-67-positive cells. The cell density may also be evaluated semiquantitatively by using inexpensive graphic tables on microphotographs or directly on histologic cuts. Our findings lead us to the following conclusions. The decrease of cell density within a tumor could theoretically be explained by two opposite mechanisms: greater dispersion of viable cells, resulting in faster spreading of the tumor, or cell loss, resulting in slower tumor growth. In our study, it seems that tumors with lower cell density were more often found in patients with longer survival than were tumors with high cell density. Therefore, a decrease of cell density in a given astrocytic tumor may relate to cell loss rather than to an increase in spreading. Whatever their histopathologic grade (ie, AST, ANA, or GBM), if the proliferative activity (cell gain compartment) and cell density are high, then the tumor growth rate will be high because the cell loss will be low (due to the high density). In contrast, if the proliferative activity (cell gain compartment) is high and the cell density is low, then the cell loss factor will be high. In this case, tumor aggressiveness will be overestimated by the determination of proliferative activity alone. The same may be said of a tumor with a low pro- J ) 0) g. ^T 150 I Astrocytic Tumor Groups FlG 7. The survival period elapsing between histopathologic diagnosis and death in four groups of patients (N = 34, see Figure 5, C) with high-grade astrocytic tumors. The four groups were defined on the basis of the proliferative activity and cell density values of the astrocytic tumors (see Figures 5, A and 5, B). According to these distribution values, a low level of proliferative tumor activity corresponded arbitrarily with an MIB1 index <8%, while a high proliferative activity corresponded with an MIB1 index >8%. A weak cell density corresponded arbitrarily with a value <50, while a high cell density corresponded with a value >50. Thus, by combining these two biologic variables, four groups of high-grade astrocytic tumors were obtained: weakly proliferating/weakly dense (WPI-WCD, n = 5), weakly proliferating/highly dense (WPI-HCD, n = 20), highly proliferating/weakly dense (HPI-WCD, n = 4), and highly proliferating/highly dense (HPI-HCD, n = 5). Vol. 107 No. 3

10 330 ANATOMIC PATHOLOGY Article liferating rate and a high density: the biologic aggressiveness of the tumor could be underestimated if only proliferative activity is determined. Regarding the hypothesis we made for the potential relation between cell loss and decreasing cell density, we experimentally validated such a hypothesis in a previous work. 30 The fact remains that cell density could also be influenced by cell size, degree of differentiation, and degree of infiltration of surrounding brain tissue. Indeed, cell density, may also reflect properties of intercellular adhesion and nuclear-cytoplasmic ratios. Cellular adhesion may, in turn, be affected by the extracellular matrix (as modified by tumor cells), cell surface receptors, and the formation of intercellular junctions and glial networks. However, the relevance and potential importance of the method and the derivative data are not changed because the parameter of cell density may not be directly related to only the ratio of cell gain to cell loss. All these possibilities are being experimentally tested in our laboratory. We determined in the present study (data not shown) and in a previous study relating to soft tissue tumors 18 that heterogeneity of cell size and nuclearcytoplasmic ratios in tumor sections did not significantly affect the measurements relating to cell density. The amount of variation present in individual cases is detailed elsewhere. 23,31 These reports 23,31 show that the amount of variation significantly changes with histopathologic grading. In contrast, in the present study, the amount of variation relating to the morphonuclear characteristics did not significantly affect the assessment of cell density (data not shown). Couldwell et al 9 recently demonstrated that under specific experimental conditions, the proliferative activity of glioma cell lines can increase, while the overall growth is decreasing. Our results show that proliferative activity assessed by using the anti-mibl antibody is significantly correlated with the activity assessed by using the PI determined on DNA histograms. The proliferative activity and the cell density variables parallel the level of malignancy. During the validation stage of our study, we considered low- and high-grade astrocytic tumors. In contrast, in patient survival studies, only the highgrade astrocytic tumors were included, because the low- and high-grade tumors have substantially different clinical behavior. 1,2,21 Furthermore, the highgrade astrocytic tumors are the most difficult for which reliable prognosis markers must be determined. 1,2,6,8,21 Because the major focus of our study was the 34 patients with high-grade astrocytomas for whom complete survival data were available, division of these patients into four groups based on combinations of proliferative activity and cell density resulted in three of the four groups having only four or five patients each. This is a small number of cases from which to draw conclusions. However, patients with high-grade astrocytic tumors who survive for several years are very rare. Our hypothesis is particularly useful when a discrepancy appears between proliferative activity and cell density values; consequently, the quantitative cell density measurement contributes additional prognostic information for proliferative activity determination. Indeed, our results relating to patient survival show that the patients with a high-grade astrocytic tumor, whose tumors exhibited a low level of proliferative activity but high cell density, survived for significantly shorter periods than did the patients whose tumors exhibited low proliferative activity and low cell density. Furthermore, the patients with high-grade astrocytic tumors that exhibited a high level of proliferative activity and high cell density survived for significantly shorter periods than did the patients with tumors that exhibited a high level of proliferative activity but low cell density. A marked difference was observed in the survival periods between the patients with highgrade astrocytic tumors that exhibited low proliferative activity and low cell density and the patients with tumors that exhibited high proliferative activity and high cell density. A combination of cell density and proliferation labeling index more accurately predicts patient survival in high-grade astrocytomas than either method alone. In other words, such a combination made possible the identification of different prognostic groups in adult patients with high-grade astrocytic tumors. In addition to a better evaluation of high-grade astrocytic tumor aggressiveness than by determination of proliferative activity alone, the concomitant determination of proliferative activity and cell density seems likely to make it possible to identify the few adult patients with high-grade astrocytic tumors who will survive for long periods (several years). The present study thus suggests that a subgroup of biologically low-grade astrocytic tumors should exist among histologically high-grade tumors. All background discussion and the cited hypothesis for this study concern only the diffuse type of astrocytic tumors. New experiments are now underway to investigate such a hypothesis in other glial tumor types. AJCP«irch 1997

11 KISS ET AL 331 Short and Long Survival in Astrocytic Tumors of the Adult REFERENCES 1. Burger PC, Scheithauer BW, Vogel FS, eds. Surgical Pathology of the Nervous System and Its Coverings. 3rd ed. New York, NY: Churchill Livingstone; Kleihues P, Burger PC, Scheithauer BW, eds. Histological typing of tumors of the central nervous system. World Health Organisation International Histological Classification of Tumors. 2nd ed. Berlin: Springer-Verlag; Chandler KL, Prados MD, Malec M, Wilson CB. Long-term survival in patients with glioblastoma multiforme. Neurosurgery. 1993;32: Kleinberg L, Wallner K, Malkin MG. Good performance status of long-term disease-free survivors of intracranial gliomas, lnt J Radiat Oncol Biol Phys. 1993;26: Burger PC, Vogel FS, Green SB, Strike TA. Glioblastoma multiforme and anaplastic astrocytoma: pathologic criteria and prognostic implications. Cancer. 1985;56: Devaux BL, O'Fallon JR, Kelly PJ. Resection, biopsy, and survival in malignant glial neoplasms: a retrospective study of clinical parameters, therapy, and outcome. / Neurosurg. 1993;78: Collins VP, James CD. Gene and chromosome alterations associated with the development of human gliomas. FASEB ]. 1993;7: Burger PC, Shibata T, Kleihues P. The use of the monoclonal antibody Ki-67 for the identification of proliferating cells: application to surgical neuropathology. Am J Surg Pathol. 1986;10: Couldwell WT, Weiss MH, Law RE, Hinton DR. Paradoxical elevation of Ki-67 labeling with protein kinase inhibition in malignant gliomas. / Neurosurg. 1995;82: Hoshino T, Ahn D, Prados MD, Lamborn K, Wilson CB. Prognostic significance of the proliferative potential of intracranial gliomas measured by bromodeoxyuridine labeling. Int I Cancer. 1993;53: Kim DK, Hoyt J, Bacchi C, et al. Detection of proliferating cell nuclear antigen in gliomas and adjacent resection margins. Neurosurgery. 1993;33: Ganju U, Jenkins RB, O'Fallon JR, et al. Prognostic factors in gliomas: a multivariate analysis of clinical, pathologic, flow cytometric, cytogenetic, and molecular markers. Cancer. 1994;74: Onda K, Davis RL, Shibuya M, Wilson CB, Hoshino T. Correlation between the bromodeoxyuridine labeling index and the M1B1 and Ki-67 proliferating cell indices in cerebral gliomas. Cancer. 1994;74: Salmon I, Kiss R. Relationship between proliferative activity and ploidy level in a series of 530 human brain tumors including astrocytomas, meningiomas, schwannomas and metastases. Hum Pathol. 1993;24: Coons SW, Johnson PC. Regional heterogeneity in the DNA content of human gliomas. Cancer. 1993;72: Coons SW, Johnson PC. Regional heterogeneity in the proliferative activity of human gliomas as measured by the Ki-67 labeling index. / Neuropathol Exp Neurol. 1993;52: Mundle S, Iftikhar A, Shetty V, et al. Novel in situ double labeling for simultaneous detection of proliferation and apoptosis. / Histochem Cytochem. 1994;42: Berthe JV, Goldschmidt D, Salmon 1, et al. Image cytometry analysis of Feulgen-stained nuclei in 72 lipomatous lesions including atypical lipomas and well-differentiated liposarcomas. Am ] Clin Pathol. 1996;106: Cattoretti G, Becker MH, Key G, et al. Monoclonal antibodies against recombinant parts of the Ki-67 antigen (MIB1 and M1B3) detect proliferating cells in microwave-processed formalin-fixed paraffin sections. I Pathol. 1992;168: Kruczynski A, Salmon I, Camby I, et al. The characterization of nuclear DNA content, the proliferative activity and the immunohistochemical expression of GFAP, VIM, LEU-7, S- 100, p53 and cathepsin D in primitive human glioblastoma multiformes (GBMs) versus human GBM cell lines grafted into the brains of nude mice, lnt ] Oncol. 1995;6: VandenBerg SR. Current diagnostic concepts of astrocytic tumors. / Neuropathol Exp Neurol. 1992;51: Kiss R, Salmon I, Camby I, Gras S, Pasteels JL. Characterization of factors in routine laboratory protocols which significantly influence the Feulgen reaction. / Histochem Cytochem. 1993;41: Decaestecker C, Salmon I, Camby I, et al. Identification of highversus lower-risk clinical sub-groups in the group of adult patients with supratentorial anaplastic astrocytomas. / Neuropathol Exp Neurol. 1995;54: Gillapsy GE, Miller RH, Samols D, Goldthwait DA. Antigenic and differentiative heterogeneity among human glioblastomas. Cancer Lett. 1993;68: Salmon I, Dewitte O, Pasteels JL, et al. Prognostic scoring in adult astrocytic tumors using patient age, histopathological grade, and DNA histogram type. / Neurosurg. 1994;80: Salmon I, Kruczynski A, Camby I, et al. DNA histogram typing in a series of 707 tumors of the central and peripheral nervous system. Am j Surg Pathol. 1993;17: Kiss R, Salmon 1, Kruczynski A, Camby 1, Pasteels JL, Van Ham P. Aneuploidy occurrence in human tumors, a logicalautomaton approach. Ann Cell Pathol. 1995;9: Gerdes J, Schwab U, Lemke H, Stein H. Production of a mouse monoclonal antibody reactive with a human nuclear antigen associated with cell proliferation. Int} Cancer. 1983;31: Darzynkiewicz Z, Bruno S, Del Bino G, et al. Features of apoptotic cells measured by flow cytometry. Cytometry. 1992;13: Kruczynski A, Yeaton P, Darro F, et al. Characterization by means of Voronoi paving of the influence of anti-hormone and/or anti-growth factor antibodies on the in vitro cell kinetics of human colorectal neoplastic cell lines. Int ] Oncol. 1996; 8: Salmon I, Rorive S, Camby I, et al. Stereotactic biopsies from astrocytic tumors: diagnostic information contributed by the quantitative chromatin pattern description. Anal Quant Cytol Histol. 1995;17: Vol. 107 No. 3