Pleomorphic xanthoastrocytoma

2033

Pleomorphic Xanthoastrocytoma What Do We Really Know about It?

Caterina Giannini, M.D., Ph.D.1 Bernd W. Scheithauer, M.D.1 Peter C. Burger, M.D.2 Daniel J. Brat, M.D., Ph.D.2 Peter C. Wollan, Ph.D.3 Bolek Lach, M.D.4 Brian P. O’Neill, M.D.5 1

Department of Pathology, Mayo Clinic, Rochester, Minnesota.

2

Department of Pathology, Johns Hopkins Hospital, Baltimore, Maryland.

3

Department of Biostatistics, Mayo Clinic, Rochester, Minnesota.

4

Department of Pathology, Ottawa Civic Hospital, Ottawa, Canada.

5

Department of Neurology, Mayo Clinic, Rochester, Minnesota.

Presented in part at the 73rd annual meeting of the American Association of Neuropathologists, Pittsburgh, Pennsylvania, June 11–16, 1997. Address for reprints: Bernd W. Scheithauer, M.D., Department of Pathology and Laboratory Medicine, Mayo Clinic, 200 1st Street S.W., Rochester, MN 55905. Received July 30, 1998; revision received December 3, 1998; accepted December 23, 1998. © 1999 American Cancer Society

BACKGROUND. Pleomorphic xanthoastrocytomas (PXA) may recur and demonstrate aggressive clinical behavior with a mortality rate between 15% and 20%. To the authors’ knowledge, no histopathologic features currently are known to reliably predict recurrence or tumor progression. METHODS. The study was based on 71 cases with available information regarding clinical and therapeutic data and follow-up. Diagnostic features included cellular pleomorphism, giant and/or xanthic cells, eosinophilic granular bodies, desmoplasia, and leptomeningeal involvement. The mitotic index (MI), the presence of necrosis, and endothelial proliferation were recorded in all primary resection specimens. RESULTS. The study included 35 females and 36 males, age 26 6 16 years (mean 6 standard deviation). Approximately 98% of tumors were supratentorial, with 49% in the temporal lobe. Seizures were the presenting symptoms in 71% of patients. Extent of tumor removal was macroscopic total resection in 68% of cases and subtotal resection (STR) in 32% of cases. Postoperative radiotherapy, alone or with chemotherapy, was administered in 29% and 12.5% of cases, respectively. The recurrence free survival rates (RFS) were 72% at 5 years and 61% at 10 years, whereas overall survivals rates (OS) were 81% at 5 years and 70% at 10 years. In univariate analysis, the extent of resection was the single factor associated most strongly with RFS (P 5 0.003), followed by MI (P 5 0.007) and atypical mitoses (P 5 0.04). Necrosis was not found to be significant. The extent of resection and MI were confirmed as independent predictors of RFS by multivariate analysis. MI (P 5 0.001), atypical mitoses (P 5 0.02), and necrosis (P 5 0.04) were associated with OS by univariate analysis. In multivariate analysis, only MI was an independent predictor of survival. Information regarding MIB-1 labeling index and the use of adjuvant therapy was too limited to explore their prognostic significance confidently. CONCLUSIONS. The study confirms that PXA is an astrocytic tumor with a relatively favorable prognosis. MI and extent of resection appear to be the main predictors of RFS and OS. Given the slow growth of the tumor, more studied cases and longer periods of follow-up will be essential to confirm our findings regarding prognostic factors affecting this unusual tumor. Cancer 1999;85:2033– 45. © 1999 American Cancer Society.

KEYWORDS: glioma, pleomorphic xanthoastrocytoma, natural history, pathology, prognostic factors, neurooncology.

P

leomorphic xanthoastrocytoma (PXA), a new addition to the recent 1993 World Health Organization (WHO) classification of tumors of the central nervous system, was first described by Kepes and coauthors in 1979.1,2 This morphologically distinctive cerebral glioma occurred primarily in young subjects, showed a predilection to su-

2034

CANCER May 1, 1999 / Volume 85 / Number 9

perficial growth, and had a relatively favorable prognosis. Prior to that time, such tumors were often classified among giant cell glioblastomas given their ominous histologic features, which include marked cellular pleomorphism, nuclear atypia, and the presence of bizarre, multinucleate giant cells. Due to the rich reticulin network that characterizes superficial portions of PXA, some had initially been considered mesenchymal tumors and were designated fibrous xanthomas or xanthosarcomas by the same author.3 Finally, some had been included in the now defunct category of “monstrocellular sarcoma.” Unlike giant cell glioblastomas and sarcomas, however, PXAs are associated with much longer recurrence free and overall survivals. Since its initial description, over 100 PXAs have been reported, most as single cases or small series.4 – 67 Consequently, data regarding its natural history and prognostic factors are fragmentary. Nonetheless studies generally confirm the clinical and pathologic observations of the original series. It is of note that PXA may recur and that 15–20% undergo progressive anaplastic transformation. Such tumors are less pleomorphic and are more diffusely infiltrative.62,68 The 1993 WHO classification of brain tumors recognizes the occurrence of Grade 3 (anaplastic) transformation in PXAs but does not provide minimal criteria permitting their identification.1 Indeed, a careful review of published cases leaves the reader without a definition permitting prospective identification of “anaplastic PXA,” an aggressive variant prone to recurrence and/or a fatal outcome. Furthermore, although there is agreement that surgical resection is the treatment of choice, the role of adjuvant therapies is presently unclear.3 Herein, we studied a large series of patients with PXA in order to better define its clinicopathologic features and natural history. Our findings were compared with an analysis of previously reported, adequately documented cases. Having evaluated clinical data at initial presentation and the various pathologic features generally thought to signify aggressive behavior, we found that mitotic index IMI) was the main histologic predictor of recurrence and survival. Use of the term “PXA with anaplastic features” is recommended, followed by a clear statement of just which features, such as increased mitotic activity, necrosis, or other indicate correlating with potentially more aggressive clinical behavior were found. We purposely chose not to use the term “anaplastic (Grade 3I) PXA,” because that designation may provoke inappropriately aggressive treatment. Our results, although based upon the largest series of patients presently available, might still

not prove to be definitive. More cases and longer follow-up will be necessary to confirm or refute these findings.

MATERIALS AND METHODS Patient Identification The 71 cases studied were derived from the database of the Mayo Clinic tissue registry and from the consultation files of two of the authors (B.W.S., P.C.B.). Their inclusion was based on availability of original hematoxylin and eosin (H&E)-stained sections and/or representative paraffin embedded tissue obtained at primary surgery. Histologic features upon which the diagnosis of PXA was made included the original criteria of Kepes et al. as well as those of the 1993 WHO classification of tumors of the nervous system.1,2 Thus, PXA is a moderately cellular astrocytoma composed of cells of varying size and shape, giant and lipidized examples being characteristic. Often superficial in location and partly occupying the leptomeninges, PXA invariably shows some infiltration of underlying brain parenchyma and exhibits a tendency to perivascular (Virchow-Robin) space involvement. Reticulin stains demonstrate a meshwork surrounding individual or clusters of tumor cells, especially in superficial and leptomeningeal portions of the tumor. Small numbers of lymphocytes and plasma cells are usually present. Glial fibrillary acidic protein (GFAP) immunoreactivity is seen within the cytoplasm of greatly varying numbers of tumor cells. Although it was not stressed in the original description of PXA or in the WHO monograph, granular bodies of varying size, texture, and eosinophilia are a regular feature of this tumor.69

Clinical Features Data recorded included patient age and gender, presenting symptoms e.g., seizure or increased intracranial pressure duration of symptoms, tumor location, neuroimaging appearance (e.g., presence of a cyst), and mode of treatment, including type and extent of surgery as well as adjuvant radiotherapy and/or chemotherapy.

Pathologic Studies In all cases, routinely processed tissues included original and/or newly prepared 5 mm, H&E-stained sections. All were separately examined by two neuropathologists (C.G., B.W.S.), the diagnosis of PXA being confirmed independently. Specifically recorded were the presence of: cellular pleomorphism with multinucleate giant cells, spindle cells, small cells, and vacuolated (xanthic) cells; granular bodies, whether coarse and intensely eosinophilic or delicate and pale;

Natural History and Prognosis of PXA/Giannini et al.

2035

TABLE 1 Pleomorphic Xanthoastrocytoma: Clinical Summary Current series Variable Total number Male:female ratio Age2 Presenting symptoms Seizures (%) Seizure duration (yrs)2 Location (%) Temporal .1 lobe, including temporal Frontal, parietal, occipital Other Cystic (%) Follow-up (yrs)2

No.a

Value

No.a

Value

71 1:1 26 6 16

117

121 1.2:1 20 6 13

56 37

56

Literature review

98 71 7.6 6 8.5 49 10 17 5.6 48 5 6 4.8

52

67 6.2 6 8.4

98 97

45 22 11 5.8 58 5.7 6 5.4

FIGURE 1.

Patient age distribution at initial diagnosis. The tumor affects predominantly children and young adults, 62% of whom are less than age 25 years. The distribution of previously reported cases is quite similar.

a

No. indicates the number of patients for whom information was available when this is different from the total number. 2 Data represent mean 6 standard deviation.

Rosenthal fibers; reticulin pattern and desmoplasia; calcification; inflammatory cell aggregates; growth pattern, including leptomeningeal extension and parenchymal infiltration; mitotic activity and MI; endothelial proliferation; and necrosis with or without pseudopalisading. The MI was defined as the maximum number of mitotic figures present in 10 high power fields (HPF) at 3400 magnification. All counts were performed meticulously by one of us (C.G.), with particular efforts made to distinguish karyorrhectic debris from mitotic figures. Also recorded was the number of HPF examined before encountering the first mitosis. At high magnification (3400), a single microscopic field of the Olympus BH-2 microscope (Olympus, Tokyo, Japan) fitted with a D plan 40 (0.65, 160/0.17) objective represented an area of 0.12 mm2. Immunohistochemical staining for GFAP, in most cases performed at the time of the original diagnosis, was repeated only in the cases wherein paraffin blocks and/or unstained slides were available. A detailed study of the immunophenotype of PXA is the subject of a separate publication.70 In 29 cases, immunostains for MIB-1 antigen were performed on 6 mm paraffin sections mounted on silanized slides using MIB-1 antisera (Immunotech, Westbrook, ME; monoclonal, dilution 1:100). To reduce variation, all specimens were stained simultaneously in a single batch, the procedure having been described in detail previously.71 MIB-1 labeling indices (MIB-1 LI) were expressed as percentage of nuclear area stained in areas of maximum labeling.

Statistical Analysis Recurrence free survival and overall survival were calculated from the date of first surgery to the date of second surgery or to death or last follow-up, respectively. Recurrence free survival and overall survival distributions, formulated by using the Kaplan–Meier method, were examined for equality with log-rank tests. To assess the strength of association between recurrence free/overall survival and the potential prognostic factors studied, Cox proportional hazards modeling was used both individually (univariate model) and jointly (multivariate model). The variables examined included patient age and gender; tumor site; presence or absence of a cyst; presenting symptoms (seizure vs. other); extent of tumor resection (gross or subtotal resection [GTR] vs. STR); presence or absence of parenchymal infiltration, a small cell component, mitoses, and necrosis; MI 0, ,5, or $5 mitoses per 10 HPF; presence or absence of atypical mitoses; number of HPFs examined before encountering the first mitotic figure; and MIB-1 LI.

Literature Review The 121 cases subject to analysis were derived from a review of cases reported to May 1997. For comparison purposes, these data are presented side by side with those derived from the study of our patient population. The data are incomplete in some instances, and the actual prevalence of a given feature is presented in association with the number of cases in which the information was available (denoted in parentheses).

2036


FIGURE 2. Histologic features of pleomorphic xanthoastrocytoma (PXA). Typically superficial in location, it variably involves leptomeninges, infiltrates underlying parenchyma, and extends into perivascular spaces (A). Cellular pleomorphism is the rule and includes spindle cells (B) as well as mononucleated and multinucleated giant cells, occasionally with vacuolated cytoplasm (C). Granular bodies, intensely eosinophilic or pale, are an almost constant finding (D). Reticulin staining surrounds individual or clustered tumor cells (E); glial fibrillary acidic protein staining varies in extent but is often intense (F). In some cases, mitotic figures and/or necrosis are present (G,H). D and H are reproduced with permission from Giannini and Scheithaur.69

RESULTS Clinical Findings Clinical data regarding our study population are summarized in Table 1. The 71 patients included 35 females and 36 males. Their ages varied considerably (mean 6 standard deviation [SD], 26 6 16 years; median, 22 years; range, 5– 82 years). Figure 1 demonstrates age distributions at diagnosis compared with that of previously reported cases. Presenting symptoms were known in 56 patients (79%) and consisted primarily of seizures (71%), often

of long duration (mean 6 SD, 7.6 6 9.2 years; median, 3 years). The tumors were supratentorial in 98% of the cases, one cerebellar example having been reported previously.46 The most common single location was the temporal lobe (49%), followed by the parietal (17%), frontal (10%), and occipital (7%) lobes. Of the 9 of 71 tumors that involved more than one lobe, the contiguous temporal lobe was affected in 7 (10% of total). Of the 56 cases in which neuroimaging data were available, 48% of the tumors had a cystic component.

Natural History and Prognosis of PXA/Giannini et al. TABLE 2 Pleomorphic Xanthoastrocytoma: Frequency of Histologic Features Current series Variable

No.

Multinucleated giant cells Xanthic cells Granular bodies Eosinophilic Pale Rosenthal fibers Nuclear inclusions Reticulin Lymphocyte collections Calcifications GFAP reactivity

% 92 66 93 81 71 27 87 81 83 18 98

52

41

GFAP: glial fibrillary acidic protein. No. indicates the number of patients for whom information was available when this is different from the total number.

TABLE 3 Pleomorophic Xanthoastrocytoma: Histologic Findings at Initial Presentation Current series Variable Mitoses Absent ,5 3 10 HPF $5 3 10 HPF Endothelial proliferation Necrosis MIB-1 labeling index $2

No.

%

Literature review No.

%

86 37 45 18 0 11

28 2 28 80

7 13

29 21

—

HPF: high-power fields. No. indicates the number of patients for whom information was available when this is different from the total number.

Histologic Findings The characteristic morphologic features of PXA are illustrated in Figure 2. The frequency with which these were identified in the present series is illustrated in Tables 2 and 3. Despite meticulous examination of all available tissues, 26 tumors (37%) revealed no mitotic figures. Low level mitotic activity (,5 mitoses per 10 HPF) was identified in 32 tumors (45%), more than half having only 1 per 10 HPF. Levels of $5 mitoses per 10 HPF were noted in 13 cases (18%), the index being distributed uniformly between 5 per 10 HPF and 10 per 10 HPF. Of the 45 tumors that exhibited mitoses, the first

2037

mitotic figure was found in the first HPF examined in 9 cases (20%); in 22 cases (49%), it was found in the first 25 HPF; in 24 cases (53%), it was found in the first 50 HPF; in 32 cases (71%), it was found in the first 100 HPF; and, in 43 cases (96%), it was found in the first 200 HPF. A significant negative correlation was observed between the MI and the number of HPF counted before encountering the first mitosis (Spearman correlation coefficient, 20.487; P 5 0.017). Nonetheless, in some cases with mitotic indices ,5 mitoses per 10 HPF (3 of 9), the first mitosis was found in the first HPF examined, whereas, in 1 case with a high MI (9 mitoses per 10 HPF), the first mitosis was found only after study of 122 HPF. These results reflect regional heterogeneity of mitotic activity in PXA. With regard to proliferative activity and its comparability between the three groups (no mitotic activity; ,5 mitoses per 10 HPF; $5 mitoses per 10 HPF) comprising the study set, we compared the total number of HPF and found that the respective mean, SD, and range values were 181 6 84 (41–344), 254 6 126 (35–507), and 239 6 88 (100 –381). The numbers of HPF examined were significantly lower in the group without mitoses than in that with ,5 mitoses per 10 HPF (P 5 0.03), whereas there were no significant differences between the group without mitoses and that with $5 mitoses per 10 HPF and between the groups with #5 and $5 mitoses per 10 HPF(P . 0.05). This finding, together with the fact that there were a few cases in both the group without mitoses (3 of 26; 11%) and the one with ,5 mitoses per 10 HPF (3 of 32; 9%) in which less than 100 HPF were assessable (whereas, in all tumors with $5 mitoses per 10 HPF, more than 100 HPF were examined), suggests that a modest underestimation of both the number of cases showing mitotic activity as well as of the MI itself may have taken place. Although endothelial proliferation was not observed in any initial resection specimen, necrosis was noted in 8 cases (11%). Our results are summarized and compared with published data (Table 3). Surprisingly, published information regarding mitotic activity is scant; only two tumors were subject to a formal mitotic count.26,43 MIB-1 LIs performed in such a way as to reduce methodological variation (see above) were available in 29 cases (41%), and the results were as follows: mean 6 SD, 1.9 6 3.1%; median, 0.6%. In 23 cases (79%), the labeling index was low (#2%). It exceeded 2% only in 6 instances (21%), the range being from 2.9% to 14%.

2038


TABLE 4 Histologic and Follow-Up Data for 9 Patients with Pleomorphic Xanthoastrocytoma Histologic features Patient 69 P R1 R2 21 P R1 65 P R1 67 P R1 57 P R1 14 P R1 R2 R3 40 P R1 3 P R1 32 P R1

Age

Gender

MI

Necrosis

EP

Follow-up

7 11 15

F

2 0 1

2 2 2

2 2 2

Recurrence 4 yrs Recurrence 4 yrs Alive and well 0.4 yr

23 38

M

2 11

2 2

2 2

Recurrence 15 yrs No follow-up

40 60

F

0 9

2 2

2 2

Recurrence 20 yrs Alive 1 yr

16 17

M

1 36

2 2

2 2

Recurrence 1 yr No follow-up

42 52

F

2 4

1 2

2 2

Recurrence 10 yrs Presented with hemorrhage; died postoperatively

17 18 20 21

M

3 7 7 36

2 2 2 1

2 2 2 1

Recurrence 1 yr Recurrence 1 yr Recurrence 2 yrs Died 8 mos later from intratumoral hemorrhage

17 24

M

1 4

2 1

2 2

Recurrence 6.7 yrs Alive with disease 3.1 yrs

52 53

M

7 7

2 2

2 2

Recurrence 0.9 yr Alive 1.4 yrs

8 9

M

5 5

2 2

2 2

Recurrence 1 yr Alive 1 mo follow-up

MI: mitotic index (number of mitoses per 10 high-power fields; EP: endothelial proliferation; P: primary tumor; R1, R2, R3: first, second, and third recurrence.

Histologic Findings at Recurrence We had the opportunity to review specimens obtained at multiple points in time in 9 of 12 patients who underwent surgery for recurrence on one (7 cases) or more occasions (2 cases). Data regarding histologic features and follow-up in these cases are summarized in Table 4. Of the 9 specimens studied, 5 showed histologic signs of anaplasia, as manifested by increased mitotic activity, and in 2 instances, by the acquisition of necrosis and/or endothelial proliferation. Four tumors, 2 of which had a low MI (,5 mitoses per 10 HPF) and 2 of which a high MI ($5 mitoses per 10 HPF), remained histologically unchanged at recurrence.

TABLE 5 Pleomorphic Xanthoastrocytoma: Therapy Literature review

Current series Therapy

No.

Extent of resection Gross total removal Subtotal removal Biopsy only Adjuvant therapy None Radiation Chemotherapy Radiation 1 chemotherapy

57

%

No.

%

97 68 32 —

65

68 30 2 89

57 29 1.5 12.5

64 34 1 1

No. indicates the number of patients for whom information was available when this is different from the total number.


2039

TABLE 6 Pleomorphic Xanthoastrocytoma: Significant Prognostic Factors in Recurrence Free Survival P value Factors and models

FIGURE 3. Recurrence free survival (Kaplan–Meier method). The curve for the present series (solid line) is compared with that based on previously reported cases (dashed line): The two are similar.

Univariate analyses Extent of resection 0, ,5, $5 mitosesa Atypical mitoses Necrosis Multivariate analyses Extent of resection 0, ,5, $5 mitosesa

Current series

Literature review

0.003 0.007 0.04 ns

nd na na 0.008

0.007 0.01

— —

ns: Not significant; nd: not done; na: not available. a Number of mitoses per 10 high-power fields.

TABLE 7 Pleomorphic Xanthoastrocytoma: Significant Prognostic Factors in Patient’s Overall Survival P value Factors and models

FIGURE 4. Overall survival curves (Kaplan–Meier method) for patients in the present series (solid line) and for previously reported cases (dashed line) are quite similar. Treatment Modalities Table 5 summarizes our data regarding initial treatment. Extent of surgical removal is based on estimates of the neurosurgeon. Information regarding adjuvant therapy is fragmentary. Twenty-eight patients received adjuvant therapy: 19 underwent external beam radiation therapy (EBRT), 1 received chemotherapy, and 8 underwent a combination of chemotherapy and EBRT. Because most of our patients received adjuvant therapy in their home community, there was no control over total dose of radiation employed, the ports used to administer the radiation, or the drug used.

Survival Analysis At termination of data collection, 59 patients (83%) were alive, a rate that is strikingly similar to the literature value of 83% of the 100 reported patients in whom follow-up data were available. Overall survival in our study was 81% at 5 years and 70% at 10 years. Recurrence free survival was 72% at 5 years and 61% at 10 years. Estimates of recurrence free survival and overall survival curves based on our data and the those

Univariate analyses 0, ,5, $5 mitosesa Extent of resection Atypical mitoses Necrosis Multivariate analyses 0, ,5, $5 mitosesa

Current series

Literature review

0.001 0.08 0.02 0.04

na nd na 0.01

0.005

—

ns: Not significant; nd: not done; na: not available. a Number of mitoses per 10 high-power fields.

generated from published data were comparable (Figs. 3, 4). Results of univariate and multivariate analyses for recurrence free and overall survivals are summarized in Tables 6 and 7. These show the P values for all prognostic factors (see above) that were found to be of significance in both the univariate and the stepwise multivariate models. Variables are listed in order of decreasing P values and are therefore of decreasing probability. Where they were available, correlative data based on our literature review also are provided.

Recurrence Free Survival Extent of resection was the single factor most strongly associated with recurrence free survival (Table 6, Fig. 5). Taken individually, MI and presence of atypical mitoses also showed a significant correlation (Fig. 6). In contrast to what was reported previously in the

2040


FIGURE 5. Recurrence free survival curves (Kaplan–Meier method) in the present series based on extent of primary tumor resection. This measure of survival was significantly longer in patients who underwent gross total resection rather than subtotal resection. In 14 cases, extent of resection could not be established. (log-rank test; P 5 0.003).

FIGURE 6.

Recurrence free survival curves (Kaplan–Meier method) in the present series based on mitotic index [0, ,5, and $5 mitoses per 10 high-power fields (HPF)]. There are significant differences in this measure of survival among patient groups (log-rank test; P 5 0.007). Two group comparisons demonstrate a significant difference in recurrence free survival between tumors with 0 and $5 mitoses per 10 HPF [P 5 0.002; relative risk (RR), 7.5], whereas the difference between tumors with ,5 and $5 mitoses per 10 HPF approached significance (P 5 0.06).

FIGURE 7.

Overall survival curves (Kaplan–Meier method) in the present series based on mitotic index [0, ,5, and $5 mitoses per 10 high-power fields (HPF)]. Significant differences in survival were noted among these patient groups (log-rank test; P 5 0.001). Two group comparisons demonstrate a significant difference in overall survival between tumors with ,5 versus $5 mitoses per 10 HPF [P 5 0.007; relative risk (RR), 6.4] and those with 0 versus $5 mitoses per 10 HPF (P 5 0.005; relative risk, 23). The difference between tumors with 0 versus ,5 mitoses per 10 HPF was not significant (ns).

literature and confirmed by our analysis of previously reported cases (Table 6), in our series necrosis was not significantly associated with recurrence free survival. In the multivariate model, extent of resection and MI were the only independent predictors.

Overall Survival The MI (0, ,5, and $5 mitoses per 10 HPF) was the single factor that was associated strongly with overall survival (Table 7, Fig. 7); taken individually, the presence of atypical mitoses and of necrosis also were significantly associated (Fig. 8). We noted no significant difference in overall survival based on extent of resection (Fig. 9). In this multivariate model, MI was the only independent predictor of overall survival.

FIGURE 8.

Overall survival curves (Kaplan–Meier method) in the present series based on the presence of necrosis. Survival was significantly longer in patients with tumors that did not show necrosis at first resection compared with those in which this parameter was present.(log-rank test; P 5 0.04).

Table 8 summarizes the data regarding extent of resection (GTR vs. STR), MI (0, ,5, and $5 mitoses per 10 HPF), presence of necrosis, and MIB-1 LI (,2 vs. $2) in patients who did not have tumor recurrence, in those who had tumor recurrence but did not die, and in patients who died. The course of disease in patients who died and the histologic features of their tumors were as follows: Four patients experienced continuous tumor progression and died 0.3–1.6 years after initial surgery. Of these, the tumors in 3 patients exhibited high mitotic activity, 1 with necrosis at primary resection; the tumor in the fourth patient showed low mitotic activity and no necrosis. Two patients whose tumor showed high mitotic activity and necrosis at initial surgery despite GTR followed by radiation and chemotherapy,


FIGURE 9.

Overall survival curves (Kaplan–Meier method) in the present series based on extent of primary tumor resection. There was not a significant difference in survival between patients who underwent gross total removal and who underwent only subtotal resection (log-rank test; P 5 0.08). In 14 patients, data regarding the extent of resection were not available.

experienced recurrences at 0.6 years and 0.8 years and died at 2.5 years and 1.8 years, respectively. None of these 6 patients underwent further surgery or autopsy. Three patients whose tumors showed low mitotic activity and no necrosis at primary resection experienced one or multiple recurrences and eventually died between 2.1 years and 4.8 years after primary surgery. A recurrent tumor could be examined in only 1 patient and showed histologic progression (see case 14, Table 4). One patient whose tumor exhibited necrosis but a low MI experienced a recurrence after 10 years and died postoperatively (see case 57, Table 4). Yet another patient with a high MI but no necrosis at first surgery died 5.7 years after primary surgery of a contralateral, postirradiation anaplastic oligodendroglioma. Finally, 1 patient was well and free of seizures at 20.5 years, when he died in a motor vehicle accident.

DISCUSSION PXA is a new addition in the 1993 WHO classification of tumors of the central nervous system.1 Although it is rare and accounts for less than 1% of all astrocytic neoplasms,72 it is the subject of numerous articles, many reporting only single cases or small series. Since the original description by Kepes et al. of 12 cases in 1979,2 their clinical and pathologic description of a neoplasm with relatively favorable prognosis despite ominous histologic features generally has been confirmed.4 – 67 In many cases, its superficial location, relative circumscription, and cyst/mural nodule architecture facilitate a gross total removal. These architectural features, therefore, are considered major determinants of the often favorable outcome.62,68 The role of histologic features and of tumor grade in predicting biological behavior has received less at-

2041

tention, although increased mitotic activity and necrosis have been associated with increased aggressiveness.62 For that reason, we studied a large group of patients with PXA and undertook an analysis of previously published cases in which adequate information regarding morphology and outcome had been provided. From the analysis of the data, it became apparent that these two patient populations, one mainly representing the combination of two large North American consult practices and the other a compilation of case reports and small series from the international neuropathology community, were remarkably similar in their characteristics and closely mirrored the original experience of Kepes et al.2 Numerous similarities, including age distribution, with a high incidence in children and young adults, equal frequency of occurrence in both males and females, seizures as the most common presenting sign, predilection for the temporal lobe, and a high frequency of cystic lesions, all contribute to making these two groups of patients comparable. Furthermore, frequencies of GTR as well as of radiation therapy were quite similar. Thus, it was not surprising that both recurrence free and overall survival curves were essentially superimposable. Our study confirmed that PXA, with its 70% 10-year survival rate, behaves significantly better than those lesions with which it was and continues to be mistaken. Nonetheless, unlike other astrocytic tumors of favorable prognostic type e.g., pilocytic astrocytoma and subependymal giant cell astrocytoma, PXA is associated with a higher frequency of recurrence, anaplastic transformation, and death. The question remained: Could one predict which tumors would recur and/or behave more aggressively based on morphologic and clinical information available at time of diagnosis? We carefully analyzed our cases as well as published PXA cases in an effort to extract as much useful information as possible regarding prognostic parameters of proven significance in ordinary, diffuse astrocytic tumors1,73 and PXA,62,72 specifically, mitotic activity, endothelial proliferation, and necrosis. Table 3 shows that, although the presence of necrosis has been recorded frequently in the literature, information regarding mitotic activity is often barely mentioned. We found that, in univariate analysis, the MI appeared to be predictive of both recurrence free and overall survival. In contrast, the presence of necrosis correlated significantly with overall survival, but not with recurrence free survival. This last finding is at odds with what was reported previously. Our analysis of previously reported cases (Tables 6, 7), showed that necrosis was a significant predictor of both recurrence

2042


TABLE 8 Pleomorphic Xanthoastrocytoma: Summary Data by Group Extent of surgery (%)a

Mitoses (%)b

MIB-1 LI (%)

Group

No.

GTR

STR

0

5

Necrosis

2

No recurrence Recurrence Deaths

46 13 12

32 (70) 3 (23) 4 (33)

9 (20) 3 (23) 6 (50)

20 (43) 4 (31) 2 (17)

21 (46) 7 (54) 4 (33)

5 (11) 2 (15) 6 (50)

4 (9) 0 (0) 4 (33)

14 (82) 5 (100) 4 (57)

3 (18) — 3 (43)

GTR: gross total resection; STR: subtotal resection; LI: labeling index. a Information was available in only 57 of 71 cases. b Number of mitoses per 10 high-power fields.

and survival. Given the paucity of published data, the significance of the MI could not be evaluated formally in cases from the literature. Furthermore, in a multivariate analysis of our series, MI emerged as the only histologic finding that, together with extent of surgical resection, could independently predict tumor recurrence and overall survival. Such a comparison of the relative weight of MI versus necrosis could not be performed in previously reported cases. We propose the designation “PXA with anaplastic features” to denote PXA, featuring high mitotic activity ($5 mitoses per 10 HPF) with or without accompanying necrosis. At present, it is unclear whether such tumors should be termed anaplastic or Grade 3 (of 4). Use of the term “anaplastic” might be found objectionable when applied to a tumor that behaves less aggressively than ordinary anaplastic astrocytoma and in which resectability more often affects the outcome. It is even less clear just when, if ever, a PXA should be considered Grade 4 or be equated with glioblastoma, as was suggested in the 1993 WHO blue book.1 We fully concur with Pahapill and coauthors, who stress that necrotic PXAs are not and should not be termed glioblastoma.62 Indeed, PXA appears to be yet another example of an astrocytic tumor distinct from ordinary diffuse astrocytomas and to which criteria used in grading such astrocytomas1,73 should not be applied. Five of the 9 cases in which we reviewed multiple resection specimens acquired histologic anaplastic features, as manifested by increased mitotic activity and, in 2 cases, necrosis and/or endothelial proliferation (Table 4). It should be pointed out, however, that histologic malignancy in PXA does not correlate with prognosis as reliably as in patients with ordinary, diffuse astrocytomas. Indeed, the clinical course of patients with histologically malignant PXAs was often more favorable and less precipitous than that of patients with fibrillary astrocytomas whose tumors showed the same features.42,62 That these tumors are

inherently different is also supported by molecular genetic studies, which suggest that the genetic events underlying PXA formation and progression differ significantly from those involved in diffuse astrocytoma tumorigenesis.64 These provide further support for maintaining a separation between anaplastic PXA and diffuse astrocytomas of Grade 3 (anaplastic) and Grade 4 (glioblastoma). Our study also confirmed the importance of extent of primary resection in the prevention of tumor recurrence. However, it is noteworthy that the extent of surgical resection did not appear to significantly affect long term survival. Because 1) the extent of resection was assessed by review of operative reports, a method known to be imperfect compared to immediate postoperative magnetic resonance imaging evaluation, and 2) even this limited information was unavailable in nearly 20% of cases, caution must be used in evaluating these data. A possible benefit to complete surgical resection, which did not reach statistical significance in our series, was suggested in the literature review by Macauley et al.42 The more recent review of Pahapill clearly demonstrated a beneficial effect of complete resection on survival in patients with PXAs that lacked necrosis,62 although survival was not influenced by the extent of resection in cases featuring necrosis. We conclude that, whenever possible, GTR remains the goal of primary treatment. Furthermore, in cases in which a GTR is achieved, we recommend a “wait and see” approach followed, if necessary, by additional surgery for residual and/or recurrent tumor.68 At present, the role of adjuvant radiotherapy and/or chemotherapy remains uncertain. Our own data could not provide precise guidelines, since only 19 patients underwent adjuvant EBRT, and 1 patient received chemotherapy alone. One might extrapolate from current management recommendations to employ EBRT in those patients in whom there is postop-


erative residual tumor74 and who are thought to have “anaplastic” features.75 We recognize that, in a slowly growing tumor such as PXA, one with an inherently good prognosis, more cases and longer periods of follow-up will be essential to confirm or refute our findings. Nonetheless, given the infrequency of these tumors72 and the difficulties inherent in obtaining homogeneous data, particularly long, uniform follow-up on large patient numbers, we believe that our data provide some guidelines. Performing a prospective, randomized clinical trial of treatment modalities is practically impossible in a single center. Therefore, it becomes of importance, as suggested by Pahapill et al., to make the best of the available, albeit imperfect data and to follow as many patients as possible for long periods of time.62 In fact, a “registry ” approach to such uncommon tumors may be the only way for the knowledge base to expand. The neurooncology community may need to assume the responsibility of creating such databases. The establishment of some mechanism to announce, facilitate data transfer, maintain, and analyze this information will become a challenge for the future. It is our goal to follow our patient cohort and to prospectively add each new case coming to our attention. The formation of such a PXA “registry,” to which cases would be gratefully received, will, hopefully, shed light on the natural history of this unusual tumor and on factors influencing its prognosis.

REFERENCES 1.

2.

3.

4.

5.

6.

7. 8.

9.

Kleihues P, Burger PC, Scheithauer BW. Histological typing of tumours of the central nervous system. World Health Organization. 2nd ed. New York: Springer-Verlag, 1993: 11– 4. Kepes JJ, Rubinstein LJ, Eng LF. Pleomorphic xanthoastrocytoma: a distinctive meningocerebral glioma of young subjects with relatively favorable prognosis. A study of 12 cases. Cancer 1979;44:1839 –52. Kepes JJ, Kepes M, Slowik F. Fibrous xanthomas and xanthosarcomas of the meninges and the brain. Acta Neuropathol 1973;23:187–99. Kuhajda FP, Mendelsohn G, Taxy JB, Long DM. Pleomorphic xanthoastrocytoma: report of a case with light and electron microscopy. Ultrastruct Pathol 1981;2:25–32. Sakai H, Kawano N, Okada K, Tanabe T, Yagishita S. A case of pleomorphic xanthoastrocyoma. No Shinkei Geka 1981;9: 1519 –24. Heyerdahl Strom E, Skullerud K. Pleomorphic xanthoastrocytoma: report of 5 cases. Clin Neuropathol 1983;2: 188 –91. Jones MC, Drut R, Raglia G, Pleomorphic xanthoastrocytoma: a report of two cases. Pediatr Pathol 1983;1:459 – 67. Maleki M, Robitaille Y, Bertrand G. Atypical xanthoastrocytoma presenting as a meningioma. Surg Neurol 1983;20: 235– 8. Kawano N. Pleomorphic xanthoastrocytoma: pathological

10.

11. 12.

13.

14.

15.

16.

17.

18. 19. 20.

21. 22. 23.

24.

25. 26.

27.

28.

2043

viewpoints on pleomorphic xanthoastrocytoma. Noshuyo Byori 1983;1:85–90. Weldon-Linne CM, Victor TA, Groothuis DR, Vick NA. Pleomorphic xanthoastrocytoma: ultrastructural and immunohistochemical study of a case with a rapidly fatal outcome following surgery. Cancer 1983;52:2055– 63. Bukeo T, Nakakura S, Nishimoto A, Tabucki K. Pleomorphic xanthoastrocytoma (Kepes). No Shinkei Geka 1985;13:773–7. Goldring S, Rich KM, Picker S. Experience with gliomas in patients presenting with a chronic seizure disorder. Clin Neurosurg 1985;3:15– 42. Gomez JG, Garcia JH, Colon LE. A variant of cerebral glioma called pleomorphic xanthoastrocytoma: case report. Neurosurgery 1985;16:703– 6. Palma L, Maleci A, Di Lorenzo N, Lauro GM. Pleomorphic xanthoastrocytoma with 18-year survival: case report. J Neurosurg 1985;63:808 –10. Pasquier B, Kojder I, Labat F, Keddari E, Pasquier D, Stoebner P, et al. Le xanthoastrocytome du sujet jeune: revue de la litte´rature a` propos de deux observations d’e´volution discordante. Ann Pathol 1985;5:29 – 43. Grant JW, Gallager PJ. Pleomorphic xanthoastrocytoma: immunohistochemical methods for differentiation from fibrous histiocytomas with similar morphology. Am J Surg Pathol 1986;10:336 – 41. Iwaki T, Fukui M, Kondo A, Matsushima T, Takeshita I. Epithelial properties of pleomorphic xanthoastrocytomas determined in ultrastructural and immunohistochemical studies. Acta Neuropathol 1987;74:142–50. MacKenzie JM. Pleomorphic xanthoastrocytoma in a 62year-old male. Neuropathol Appl Neurobiol 1987;13:481–7. Severi P, Gambini C, Andrioli GC, Boccardo M. Pleomorphic xantoastrocytoma. Pathologica 1987;79:507–12. Gaskill SJ, Marlin AE, Saldivar V. Glioblastoma multiforme masquerading as a pleomorphic xanthoastrocytoma. Child Nerv Syst 1988;4:237– 40. Paulus W, Peiffer J. Does the pleomorphic xanthoastrocytoma exist? Acta Neuropathol 1988;76:245–52. Stuart G, Appleton DB, Cooke R. Pleomorphic xanthoastrocytoma: report of two cases. Neurosurgery 1988;22:422. Ametani T, Gi H, Yamamoto S, Tsuda T, Kako M, Fukumitsu T, et al. Pleomorphic xanthoastrocytoma: report of a case and review of the literature. Nisseki Igaku Jpn Red Cross Med J 1989;41:199 –206 Kepes JJ, Rubinstein LJ, Ansbacher L, Schreiber DJ. Histopathological features of recurrent pleomorphic xanthoastrocytomas: further corroboration of the glial nature of this neoplasm. Acta Neuropathol 1989;78:585–93. Whittle IR, Gordon A, Misra BK, Shaw JF, Steers AJW. Pleomorphic xanthoastrocytoma. J Neurosurg 1989;70:463– 8. Sugita Y, Kepes JJ, Shigemori M, Kuramoto S, Reifenberger G, Kiwit JCW, Wechsler W. Pleomorphic xanthoastrocytoma with desmoplastic reaction: angiomatous variant. Report of two cases. Clin Neuropathol 1990;9:271– 8. Allegranza A, Ferraresi S, Bruzzone M, Giombini S. Cerebromeningeal pleomorphic xanthoastrocytoma. Report on four cases: clinical, radiologic and pathological features (including a case with malignant evolution). Neurosurg Rev 1991; 14:43–9. Hosokawa Y, Tsuchihashi Y, Okabe H, Toyama M, Namura K, Kuga M, et al. Pleomorphic xanthoastrocytoma: ultrastructural, immunohistochemical, and DNA cytofluorometic study of a case. Cancer 1991;68:853–9.

2044


29. Kawano N. Pleomorphic xanthoastrocytomas (PXA) in Japan: its clinicopathologic features and diagnostic clues. Brain Tumor Pathol 1991;8:5–10. 30. Kros JM, Vecht CJ, Stefanko SZ. The pleomorphic xanthoastrocytoma and its differential diagnosis: a study of five cases. Hum Pathol 1991;22:1128 –35. 31. Matsuhisa T, Maeshiro T, Kunishio K, Shigematsu H, Tsuno K, Mishima N, et al. Proliferative potential of pleomorphic xanthoastrocytoma: determined with nucleolar organizer regions and monoclonal antibodies of Ki-67 and anti-DNA polymerase a. Brain Tumor Pathol 1991;8:55– 8. 32. Sawyer JR, Roloson GJ, Chadduck WM, Boop FA. Cytogenetic findings in a pleomorphic xanthoastrocytoma. Cancer Genet Cytogenet 1991;55:225–30. 33. Furuta A, Takahashi H, Ikuta F, Onda K, Takeda N, Tanaka R. Temporal lobe tumor demonstrating ganglioglioma and pleomorphic xanthoastrocytoma components. J Neurosurg 1992;77:143–7. 34. Kawano N. Pleomorphic xanthoastrocytoma: some new observations. Clin Neuropathol 1992;11:323– 8. 35. Lindboe CF, Cappelen J, Kepes JJ. Pleomorphic xanthoastrocytoma as a component of a cerebellar ganglioglioma: case report. Neurosurgery 1992;31:353–5. 36. Rippe DJ, Boyko OB, Radi M, Worth R., Fuller GN. MRI of temporal lobe pleomorphic xanthoastrocytoma. J Comput Assist Tomogr 1992;16:856 –9. 37. Sawyer JR, Thomas EL, Roloson GJ, Chadduck WM, Boop FA. Telomeric associations evolving to ring chromosomes in a recurrent pleomorphic xanthoastrocytoma. Cancer Genet Cytogenet 1992;60:188 –9. 38. Tien RD, Cardenas CA, Rajagopalan S. Pleomorphic xanthoastrocytoma of the brain: MR findings in six patients. AJR 1992;159:1287–90. 39. Yoshino MT, Lucio R. Pleomorphic xanthoastrocytoma. AJNR 1992; 13:1330 –2. 40. Zorzi F, Facchetti F, Baronchelli C, Cani E. Pleomorphic xanthoastrocytoma: an immunohistochemical study of three cases. Histopathology 1992;20:267–9. 41. Lipper MH, Eberhard DA, Phillips CD, Vezina LG, Cail WS. Pleomorphic xanthoastrocytoma, a distinctive astroglial tumor: neuroradiologic and pathologic features. AJNR 1993; 14:1397– 404. 42. Macaulay RJB, Jay V, Hoffman HJ, Becker LE. Increased mitotic activity as a negative prognostic indicator in pleomorphic xanthoastrocytoma. J Neurosurg 1993;79:761– 8. 43. Ozek MM, Sav A, Pamir MN, Ozer AF, Ozek E, Erzen C. Pleomorphic xanthoastrocytoma associated with von Recklinghausen neurofibromatosis. Child Nerv Syst 1993;9:39–42. 44. Thomas C, Golden B. Pleomorphic xanthoastrocytoma: report of two cases and brief review of the literature. Clin Neuropathol 1993;12:97–101. 45. Herpers MJHM, Freling G, Beuls EAM. Pleomorphic xanthoastrocytoma in the spinal cord: case report. J Neurosurg 1994;80:564 –9. 46. Wasdahl DA, Scheithauer BW, Andrews BT, Jeffrey RA Jr. Cerebellar pleomorphic xanthoastrocytoma: case report. Neurosurgery 1994;35:947–51. 47. Bicik I, Raman R, Knightly JJ, Di Chiro G, Fulham MJ. PETFDG of pleomorphic xanthoastrocytoma. J Nucl Med 1995; 36:97–9. 48. De Jesu ´ s O, Rifkinson N. Pleomorphic xanthoastrocytoma with invasion of the tentorium and falx. Surg Neurol 1995; 43:77–9.

49. Petropoulou K, Whiteman MLH, Altman NR, Bruce J, Morrison G. CT and MRI of pleomorphic xanthoastrocytoma: unusual biologic behavior. J Comput Assist Tomogr 1995;19: 860 –5. 50. Sakamoto T, Sakakibara Y, Hayashi T, Yamashita K. Recurrence of pleomorphic xanthoastrocytoma six years after. No Shinkei Gika 1995;23:941–5. 51. Cervoni L, Salvati M, Santoro A, Celli P. Pleomorphic xanthoastrocytoma: some observations. Neurosurg Rev 1996;19: 13– 6. 52. Daita G, Yonemasu Y, Muraoka S, Nakai H, Maeda T. A case of anaplastic astrocytoma transformed from pleomorphic xanthoastrocytoma. Brain Tumor Pathol 1991;8:63– 6. 53. Glasser RS, Rojiani AM, Mickle JP, Eskin TA. Delayed occurrence of cerebellar pleomorphic xanthoastrocytoma after supratentorial pleomorphic xanthoastrocytoma removal. J Neurosurg 1995;82:116 – 8. 54. Katoh M, Aida T, Sugimoto S, Suwamura Y, Abe H, Isu T, et al. Immunohistochemical analysis of giant cell glioblastoma. Pathol Int 1995;45:275– 82. 55. Kordek R, Biernat W, Sapieja W, Alwasiak J, Liberski PP. Pleomorphic xanthoastrocytoma with a glangliomatous component: an immunohistochemical and ultrastructural study. Acta Neuropathol 1995;89:194 –7. 56. Li YS, Ramsay DA, Fan YS, Armstrong RF, Del Maestro RF. Cytogenic evidence that a tumor suppressor gene in the long arm of chromosome 1 contributes to glioma growth. Cancer Genet Cytogenet 1995;84:46 –50. 57. Haga S, Morioka T, Nishio S, Fukui M. Multicentric pleomorphic xanthoastrocytomas: case report. Neurosurgery 1996;38:1242–5. 58. Kubo O, Sasahara A, Tajika Y, Kawamura H, Kawabatake H, Takakura K. Pleomorphic xanthoastrocytoma with neurofibromatosis type 1: case report. Brain Tumor Pathol 1996;13: 79 – 83. 59. Lach B, Duggal N, DaSilva VF, Benoit BG. Association of pleomorphic xanthoastrocytoma with cortical dysplasia and neuronal tumors. Cancer 1996;78:2551– 63. 60. Lee TT, Landy HJ, Bruce JH. Arteriovenous malformation associated with pleomorphic xanthoastrocytoma. Acta Neurochir 1996;138:590 –1. 61. Levy RA, Allen R, McKeever P. Pleomorphic xanthoastrocytoma presenting with massive intracranial hemorrhage. AJNR 1996;17:154 – 6. 62. Pahapill PA, Ramsay DA, Del Maestro RF. Pleomorphic xanthoastrocytoma: case report and analysis of the literature concerning the efficacy of resection and the significance of necrosis. Neurosurgery 1996;38:822–9. 63. Pasquier B, Bost F, Peoc’h M, Barnoud R, Pasquier D. Donne´es neuropathologiques dans I’e´pilepsie partielle pharmacore´sistante. Ann Pathol 1996;16:174 – 81. 64. Paulus W, Lisle DK, Tonn JC, Wolf HK, Roggendorf W, Reeves SA, Louis DN. Molecular genetic alterations in pleomorphic xanthoastrocytoma. Acta Neuropathol 1996;91:293–7. 65. Powell SZ, Yachnis AT, Rorke LB, Rojiani AM, Eskin TA. Divergent differentiation in pleomorphic xanthoastrocytoma. Am J Surg Pathol 1996;20:80 –5. 66. Van Roost D, Kristof R, Zentner J, Wolf HK, Schramm J. Clinical, radiological, and therapeutic features of pleomorphic xanthoastrocytoma: report of three patients and review of the literature. J Neurol 1996;60:690 –2.

Natural History and Prognosis of PXA/Giannini et al. 67. Tonn JC, Paulus W, Warmuth-Metz M, Schachenmayr W, So¨renson N, Roosen K. Pleomorphic xanthoastrocytoma: report of six cases with special consideration of diagnostic and therapeutic pitfalls. Surg Neurol 1997;47:162–9. 68. Kepes JJ. Pleomorphic xanthostrocytoma: the birth of a diagnosis and a concept. Brain Pathol 1993;3:269 –74. 69. Giannini C, Scheithauer BW. Classification and grading of low grade astrocytic tumors in children. Brain Pathol 1997; 7:785–98. 70. Giannini C, Lopes MBS, Scheithauer BW, Hirose T, Vandenberg SR, Kros M. Immunophenotype of pleomorphic xanthoastrocytoma. J Neuropathol Exp Neurol 1998;57:501. 71. Giannini C, Scheithauer BW, Burger PC, Christensen MR, Wollan PC, Sebo TJ, et al. Cellular proliferation in pilocytic and diffuse astrocytomas. J Neuropathol Exp Neurol (in press).

2045

72. Kepes JJ, Louis DN, Paulus W. Pleomorphic xanthoastrocytoma. In: Kleihues P, editor. Pathology and genetics of tumours of the nervous system. Lyon: W.K. Cavenee International Agency for Research on Cancer, 1997:34 – 6. 73. Daumas-Duport C, Scheithauer BW, O’Fallon J, Kelly P. Grading of astrocytomas. A simple and reproducible method. Cancer 1988;62:2152– 65. 74. Berger MS, Deliganis AV, Dobbins J, Keles GE. The effect of extent of resection on recurrence in patients with low grade cerebral hemisphere gliomas. Cancer 1995;75:2785–7. 75. Dinapoli RP, Brown LD, Arusell RM, Barle JD, O’Fallon JR, Buckner JC, Scheithauer BW, Krook JE, Tsechetter LK, Maier JA. Phase III comparative evaluation of pCNU and carmustine combined with radiation therapy in high grade glioma. J Clin Oncol 1993;11:1316 –21.