|Home | About | Journals | Submit | Contact Us | Français|
The purpose of this study was to identify new prognostic biomarkers with clinical impact in malignant peripheral nerve sheath tumor (MPNST), a highly aggressive malignancy for which no consensus therapy exists besides surgery. We have used tissue microarrays (TMAs) to assess in situ expression of 14 cell-cycle–regulating proteins in 64 well-characterized MPNST patients: 36 sporadic and 28 with neurofibromatosis type 1 (NF1). We developed a new software application for evaluation and logistics of the TMA images and performed a literature survey of cell cycle proteins in MPNST. For NF1-associated patients, there was a clear association between nuclear expression of p53 and poor survival (p = 0.004). Among the other proteins analyzed, we also found significant associations between survival and clinical variables, but none were as strong as that for p53. For the total series of MPNSTs, p53 was shown to be an independent predictor of survival, and patients without remission, with tumor size larger than 8 cm, and with positive p53 expression had a 60 times greater risk of dying within the first 5 years compared with the remaining patients (p = 0.000002). This is the most comprehensive study of in situ protein expression in MPNST so far, and expressed p53 was found to be a strong surrogate marker for outcome. Patients in complete remission with a primary p53-positive MPNST diagnosis may be considered in a high-risk subgroup and candidates for adjuvant treatment.
Malignant peripheral nerve sheath tumor (MPNST) is a rare tumor type with an incidence of 0.001% in the general population.1,2 For individuals with neurofibromatosis type 1 (NF1), which is caused by a germline mutation of the NF1 gene, the lifetime risk for obtaining MPNST is 6%–13%.3,4 The 5-year survival rate for this group of MPNST patients has been reported to be only half the survival rate for the sporadic MPNST cases.1,3,5–8 It is still unclear whether this is due to a difference in the molecular phenotype between sporadic and NF1-associated MPNST. Thus, there is a need for informative prognostic and predictive markers for this malignancy.
In general, MPNSTs display a complex and highly variable karyotype, but some recurrent genetic aberrations have been reported.9,10 Loss of proximal regions of chromosome arm 17q, which includes the NF1 gene, is found in most MPNSTs, as well as in plexiform neurofibromas, which are considered precursor lesions to MPNST in NF1-associated patients.11–15 Mutations of the NF1 gene are found in both NF1-associated and sporadic MPNSTs.16 We have previously reported copy number gain in the distal part of 17q, which harbors antiapoptotic and proliferative genes such as BIRC5 and TOP2A.12,17–19 Another frequently observed chromosomal aberration is deletion of 9p21, reflecting loss or large rearrangements of the cyclin-dependent kinase (CDK) inhibitor 2A gene (CDKN2A).20–23 This gene encodes the two nonhomologous isoforms, p14ARF and p16INK4a, both with important effects on cell cycle progression (Fig. 1).
Previous studies have addressed the possibility of using protein expression levels of some of the central cell cycle components as prognostic markers in MPNST,24–30 but little is known about how and to what extent these proteins contribute to the development of NF1-associated and sporadic MPNST. The clinical impact of different levels of protein expression remains mostly unknown. One reason for the limited knowledge is the low incidence rate of MPNST, making it difficult to obtain sufficiently large cohorts for proper evaluation of prognostic factors. Furthermore, the number of proteins analyzed has been limited, and larger numbers of proteins should be included in a single study to provide a more comprehensive picture of the complex regulation of the cell cycle. Finally, since there is no consensus treatment for MPNST, some patients have received radio- or chemotherapy prior to tumor resection, which complicates the interpretation of the impact of clinical and biomolecular features. In this study, we have used an immunohistochemical approach to analyze the expression of 14 cell-cycle–related proteins in a tissue microarray (TMA) that contains a joint series of Norwegian and Swedish MPNSTs (n = 64) for which long-term clinical follow-up data were available. A new software application was developed for visualization, scoring, and storage of the scanned TMA images.
MPNSTs from 64 patients (31 women and 33 men) were included in this study. The samples were collected at tumor orthopedic centers at Lund University Hospital (Lund, Sweden) and the Norwegian Radium Hospital (Oslo, Norway) during 1980–2002. Twenty-eight of these patients had NF1, and 36 were sporadic cases. The median age at diagnosis for these two groups was 24 and 54 years, respectively. Histopathologic classification and grading were reviewed by sarcoma reference pathologists following published guidelines.10,31,32 The tumors of nine patients were low grade and those of 52 were high grade; the tumor grade was unknown in the remaining three cases. At latest follow-up, 39 patients had died after 1–225 months (median 19), 23 were alive 7–369 months (median 122) after diagnosis, and two patients were lost to follow-up. The clinical data are summarized in Table 1.
The study was approved by the regional ethics committee for medical research of the South-Eastern Norway Regional Health Authority and by local ethics committees at Lund University.
The basic TMA technology was originally described in 1998.33 The TMA block was initially constructed by transferring 79 cylindrical tissue cores (0.6 mm diameter) from formalin-fixed and paraffin-embedded tumor samples into a recipient paraffin block.18 Later, 27 cores were added to the same TMA that we present here, including one core from each of two benign neurofibromas, as well as two cores from a single plexiform neurofibroma. The TMA block also contained 30 cores from normal tissues and other cancer types that were used as staining controls. At least one core from each of the 64 MPNSTs was included in the TMA, and for 21 tumors, two to five parallel cores were included. For the in situ protein expression analyses, 5-μm-thick sections of the TMA block were adhered onto Superfrost Plus microscope slides (Menzel GmbH & Co. KG, Braunschweig, Germany).
In situ protein expression was probed by incubating the TMA slides with selected primary antibodies listed in Supplementary Table 1. Sections of the TMA block were deparaffinized in xylene (Merck, Whitehouse Station, NJ, USA) and by rinsing twice in 100% ethanol, followed by 96% and 70%, and then in water. Antigen retrieval was performed by heating in a microwave oven at 850 W for 5 min and then 100 W for 15 min, and then immersion in one of the following buffers: 10 mM sodium citrate (pH 6.0), 1 mM EDTA (pH 8.0), or 10 mM Tris, 1 mM EDTA (pH 9.0), depending on the primary antibody used (see Supplementary Table 1). The buffers also contained 0.05% Tween 20 (all reagents were from Sigma-Aldrich, St. Louis, MO, USA). Staining was performed according to the protocol for DAKO EnVision+ using the reagents supplied with the K5007 kit (Dako, Glostrup, Denmark). Briefly, this included blocking of endogenous peroxidase activity for 5 min before incubation with the primary antibody of choice for 30 min at room temperature. A negative control experiment was provided by omitting the primary antibody for one slide. Then, the secondary antibody, conjugated with horseradish-peroxidase–labeled polymer, was applied and incubated for 30 min. Staining was completed by incubation for 5 min with 3,3′-diaminobenzidine (DAB), which results in a brown precipitate at the antigen site. Excess DAB was rinsed off before the slides were counterstained with hematoxylin and dehydrated in increasing concentrations of ethanol and xylene. Finally, glass cover slips were mounted with Depex glue (Chemiteknikk, Oslo, Norway).
The TMA sections were scanned at ×400 magnification into digital high-resolution images using the Nano-Zoomer Digital Pathology (Hamamatsu Photonics K.K., Hamamatsu, Japan).
A new software application, TMA-ImageAnalyzer, was developed for processing the scanned TMA slides and as infrastructure for the images and scoring results from the individual tissue cores and antibodies. A representation of the graphical user interface of the software is shown in Fig. 2. The software was a beta release of a product developed in the Department of Medical Informatics, Rikshospitalet University Hospital, Oslo, Norway. A commercial version will be made available through Room4 Group Ltd. (East Sussex, UK).
The expression of the 14 cell-cycle–related proteins was scored independently by two researchers by visual inspection of the immunohistochemical staining of the TMAs. For nuclear protein expression, we categorized the samples into five groups according to the percentage of nuclei with positive staining: 0%, 1%–5%, 6%–10%, 11%–50%, and >51%. For the statistical analyses, we grouped the samples with staining less than 5% as negative and more than 5% as positive. For cytoplasmic protein expression, moderate and strong staining was considered positive. If a number of samples were taken from one tumor, the scoring was considered positive if at least one of these samples was stained positively on the microarray. The interobserver variability was 13% on average. Before reevaluation of these cases, a pathologist was consulted to establish the boundaries between positive and negative samples.
Western blot analyses34 of the antibodies were performed as a quality control to confirm that they bind to proteins with the anticipated molecular masses. This was done using 12 cell line extracts as protein sources, as previously described.23
The statistical significance of the differences in protein expression observed between groups of patients was calculated using Fisher’s exact test. The bivariate correlation for coexpression of proteins was described using the Spearman correlation coefficient. Disease-specific and disease-free survival curves were analyzed using the Kaplan-Meier method, and the Breslow test was used to compare the equality of the survival functions. Finally, we used Cox regression for multivariate analyses to determine the parameters with the greatest impact on the survival. All statistical analyses were performed using SPSS software, version 15.0 (SPSS Inc., Chicago, IL, USA). For our hypothesis testing, we report p-values lower than 0.05, but the true significance of these findings should be viewed in light of the number of tests performed in each analysis. For the 14 proteins, the Bonferroni corrected significance level should be 0.0036.
The scoring results from the in situ protein expression analysis of the MPNSTs are illustrated in Fig. 3. The detailed results for nuclear and cytoplasmic staining for each antibody are shown in Table 2, and our staining remarks are presented in Supplementary Table 1. Results for each patient group (with and without NF1) are shown in detail in Table 3. The largest difference between NF1-associated and sporadic MPNSTs was found for staining of nuclear p27Kip1 (n-p27Kip1) and cytoplasmic p27Kip1 (c-p27Kip1) (p = 0.03).
The Western blot quality controls for each antibody confirmed binding to protein bands of the expected size in at least one of the cell lines. For a few antibodies, we also observed additional bands that we have not analyzed further (Table 4). We expected to see a variety of protein isoforms in different cancer tissues depending on posttranslational modifications, alternative splicing events, partial degradation, and variable degrees of oligomerization, and in the following immunohistochemical analyses, we have assumed that all antibodies are specific for one protein.
We calculated the bivariate correlation for all the in situ expression results and found that c-CDK4 and c-p14ARF had the highest Spearman correlation factor (r = 0.56, p = 0.00002). Among the sporadic MPNST samples, n-p21Cip1 and n-cyclin D1 correlated best (r = 0.61, p = 0.0001), and for the NF1-associated samples, n-MDM2 and n-p14ARF had the highest correlation (r = 0.69, p = 0.00006). All significant correlations are presented in Table 5.
For 21 patients, the TMA contained between two and five parallel samples taken from different parts of the same tumor. Comparisons of the immunohistochemical scoring results for these samples provide information about the heterogeneity of the protein expression within one particular tumor. The proteins CDK2, p53, MDM2, and Ki67 displayed the largest variation within individual tumors, with opposite results (positive in one core, negative in another) found in 7–9 of the 21 tumors in question. In contrast, RB1, p14ARF, p21Cip1, cyclin D1, cyclin D3, and CDK4 showed variation in two or fewer of the 21 tumors. For the following calculations, a tumor was considered positive if at least one of the parallels was scored as positive.
Table 6 lists the significance values for all clinical parameters and selected proteins in relation to 5-year disease-specific survival. Table 7 lists the test statistics for NF1-associated and sporadic tumors separately. Kaplan-Meier plots comparing the survival curves related to clinical parameters and the most significant immunohistochemical results are shown in Fig. 4. Our data show that the 5-year survival is close to 50% for both NF1-associated patients and sporadic MPNST patients (Fig. 4A). As expected, complete remission after initial surgery, tumor size, and tumor grade were significantly associated with survival (Fig. 4B–D). Among the proteins, patients with p53-positive tumors had a strikingly poorer 5-year survival compared with the group with p53-negative tumors (p = 0.008; Fig. 4E; Table 6), an association that was particularly strong for the NF1-associated patients (p = 0.004; Table 7). Furthermore, if the nuclear expression of p53 is stratified into five levels according to the percentage of positive nuclei, the survival rate decreases with increasing amounts of p53 (p = 0.002 for the trend; Fig. 4F). The association between p53 and survival holds even if patients that were not in complete remission are excluded (p = 0.03, n = 40; Fig. 4G), and if patients that had metastatic cancer at the time of the initial diagnosis are excluded (p = 0.03; n = 54; data not shown). A positive c-p14ARF is also correlated with poor survival (p = 0.04; Table 6). For n-cyclin D1, we found a correlation with good prognosis in the group of sporadic MPNST patients (p = 0.011), but not among the NF1 patients (Table 7).
No essential differences were seen when we used alternative clinical end points, such as disease-free survival or time to first event (data not shown).
A multivariate Cox regression was performed with the three proteins and the three clinical variables that were shown to affect the disease-specific survival in the Kaplan-Meier plots (Table 6). The parameter “metastasis at time of diagnosis” was omitted since all eight cases with metastases were included in the group without complete remission. These analyses revealed complete remission as the strongest predictor of overall survival (p = 0.000012), followed by n-p53 (p = 0.014) and tumor size (p = 0.021), whereas tumor grade, c-p53, and c-p14ARF did not provide additional information. For patients with all three characteristics combined—no remission, large tumor size, and positive p53 expression—the relative risk of dying within 5 years was 60 times higher (95% confidence interval, 11–329; p = 0.000002; n = 58).
To our knowledge, the present study is the most comprehensive immunohistochemical investigation of cell-cycle–related proteins in MPNST. Previous studies have often been limited by the number of antibodies considered, the number of tumor samples included, or both (Table 8). We also provided long and reliable follow-up data for patients still alive and included several quality control steps to ensure optimal immunohistochemical staining for each antibody. The histopathology of all the MPNST tissue cores on the TMA was confirmed by an expert soft tissue tumor pathologist, and at least two persons independently scored each antibody. All the staining results were scanned into digital high-resolution images, making any discrepancies easy to evaluate on a computer screen using the software TMA-Image-Analyzer.
In contrast to earlier reports,26,28,35 we found no significant difference in the distribution of p53 or Ki67 when we compared the high- and low-grade MPNSTs. This could be related to the absence of a universally accepted grading system for sarcomas.36 The Scandinavian grading system is primarily based on Broder’s grading of malignant tumors.32 None of our three benign neurofibroma samples expressed Ki67, whereas more than half of the MPNSTs had positive expression (Tables 2 and and3).3). Although expression of p53 is expected to be lower in neurofibromas, we were not able to distinguish them from MPNSTs in our data based on the small number of neurofibromas (Table 3).30
A large variation is reported for p53 expression in MPNST (Table 8). Our nuclear scoring results are within the range of these previous studies, and we also observed cytoplasmic expression. The large variation among different studies could be explained by the use of different antibodies in various concentrations, as well as different cutoff levels for scoring.
A common reason for accumulation of p53 in cancer cells is mutation in the TP53 gene. For MPNST, there are conflicting data regarding the influence of TP53 mutations. A handful of studies, each with only four to seven cases, report up to 75% mutations or deletions in TP53,37–39 and a higher frequency is generally reported for sporadic cases compared with NF1-associated MPNST. In contrast, in a series of 16 MPNSTs, including 11 from NF1-associated patients and 5 sporadic MPNST patients, we did not detect any mutations in the coding exons 2–11 of TP53.40 Our findings are supported by other studies (range, 5–32 cases) reporting infrequent mutations.27,30,41–43 The stability of p53 is also linked to the negative feedback loop involving MDM2.44 Among the 18 of our samples that were negative for p53, only one was positive for MDM2. In tumors with both p53 and MDM2 expression, p14ARF is a candidate for suppression of MDM2 activity.45 We found p14ARF expressed in all samples that also expressed MDM2.
The large intratumor variation of protein expression for CDK2, p53, MDM2, and Ki67 suggests that events leading to altered expression of these genes, such as mutations, deletions, amplifications, or epigenetic modifications, occur late in tumor development.
The most homogeneous expression patterns were seen for RB1, p14ARF, p21Cip1, cyclin D1, cyclin D3, and CDK4. Events affecting the expression of these proteins therefore most likely occur prior to, or early in, the development of MPNST.
We found that especially the NF1-associated patients with accumulation of nuclear p53 had poor survival (p = 0.004; Table 7). The Cox regression analysis identified p53 as the strongest predictor of poor survival for the entire patient group, including the sporadic cases (Table 6). Positive c-p53 expression was also associated with reduced survival (p = 0.01), and there were no disease-specific deaths within the group with negative cytoplasmic staining. However, since only nine samples were negative for c-p53, this parameter reached a significance of only p = 0.13 in the multivariate analysis. The association between p53 expression and poor patient survival in MPNST has to our knowledge been reported only in one previous study of 28 cases.24 In addition, a difference between the NF1-associated patients and the sporadic MPNST patients regarding p53 expression and time of recurrence has also been reported.26 For p16INK4a, there seems to be a tendency for lower expression with increasingly aggressive phenotypes; one study reported positive immunoreactivity for p16INK4A in virtually all neurofibromas and in almost as many plexiform neurofibromas,28 whereas the frequency of MPNST with positive staining for p16INK4a is severely reduced (Table 8). This can be explained in part by deletions of the CDKN2A gene, which have been reported by us and others.20–22 Furthermore, p16INK4A immunoreactivity is typically observed in those that do not have homozygous deletions of the CDKN2A gene.23,41 There was only one disease-related death among the eight patients with negative c-p14ARF expression (p = 0.04; Fig. 4G), an alternative transcript of the CDKN2A gene. We found that expression of both n-p27KIP1 and c-p27KIP1 was significantly lower in NF1-associated MPNSTs than in sporadic cases (Table 3); in fact, no tumors with cytoplasmic staining could be found among the NF1-associated MPNSTs. A correlation between c-p27 expression and poor survival, as observed by Kourea et al.,25 could not be verified in our data. Positive cyclin E1 and CDK4 were found expressed in almost all samples, which is in strong contrast to two earlier reports.25,41 This could be due to differences in the immunohistochemistry protocols, for example, use of other dilution levels or reagents.
Many studies report that the 5-year survival from diagnosis of NF1-associated patients with MPNST is only half of that of sporadic MPNST cases (~30% for NF1 vs. 55% for sporadic).1,3,5–8 However, this is not a consistent finding, as others report no significant difference.46,47 In the present series, no significant difference was observed in the survival rate when we compared NF1-associated MPNSTs with sporadic cases; the 5-year survival among the patients in this cohort was about 50% for both groups (Fig. 4A). We can disregard diagnostic uncertainty for the present series, since all cases were reevaluated by reference pathologists, and furthermore, six NF1-associated and eight sporadic cases were analyzed for germline mutations, confirming the clinical status (I. Bottillo et al., unpublished data). The largest difference between the two groups is observed after 36 months, where the disease-specific survival for the NF1-associated and sporadic MPNST patients is 52% (±10) and 65% (±8), respectively. However, this difference is not significant (p = 0.46). One might speculate that the relatively high survival rate for the NF1-associated patients in our study compared with the previous studies could be correlated to a good monitoring protocol of the NF1-associated patients, allowing for early detection of malignancies.
We have here presented the most comprehensive in situ protein profiling of sporadic and NF1-associated MPNST so far, which has allowed us to delineate the expression levels of a number of proteins affecting the regulation of the cell cycle in this cancer type.
Notably, tumors with high expression of n-p53, c-p53, or c-p14ARF were correlated with poor survival for both patient groups. Nuclear cyclin D1 was correlated with prolonged survival for the sporadic MPNST patients. Overall, for both groups of MPNST patients, we showed that complete remission was the most decisive parameter to predict disease-specific survival, followed by nuclear expression of p53 and tumor size. Patients without complete remission, with tumor size larger than 8 cm, and with positive p53 expression had a 60 times greater risk of dying within the first 5 years compared to the remaining patients (p = 0.000002). These data suggest that assessment of p53 expression in situ should be performed routinely on these tumors. Patients with curative intent who are in complete remission with a known positive p53 tumor status form a high-risk subgroup for whom adjuvant treatment should be considered.
This study was financed by grants from the Norwegian Cancer Society (R.A.L.: A95068), from which H.R.B. is supported with a Ph.D. grant, and from the Norwegian Research Council (R.A.L.: 163962).