The expression status of CD133 is associated with the pattern and timing of primary glioblastoma recurrence

Abstract

Background

Glioblastoma carries a poor prognosis primarily because of its high rate of recurrence. The ability to predict the recurrence pattern and timing would be highly useful for determining effective treatment strategies. We examined the correlation between prognostic factors and the pattern of recurrence in patients with primary glioblastoma. In particular, we examined whether there was a correlation between the expression of CD133 and glioblastoma recurrence.

Methods

We retrospectively analyzed 112 patients with primary glioblastoma. The timing and pattern (local or distant) of the initial recurrence were obtained from medical records. To identify factors predictive of recurrence, we examined CD133 expression by Western blots and immunohistochemistry, clinical (age, sex, KPS, Ki67 labeling index, surgery, ventricular entry) and genetic (IDH1, 7p, 9p, 10q, MGMT) factors.

Results

Of the 112 patients, 99 suffered recurrence. The first recurrence was local in 77 patients and distant in 22 patients. Among the factors to predict the pattern of recurrence, CD133 expression was significantly higher in distant than in local recurrence. Of the factors to predict the timing of recurrence, high CD133 expression was associated with shorter time to distant recurrence in both univariate and multivariate analyses (P = .0011 and P = .038, respectively).

Conclusions

The expression of CD133 may be a predictor of the pattern and timing of recurrence of primary glioblastoma.

Primary glioblastoma, classified as grade IV tumor by the World Health Organization (WHO), is the most malignant tumor of the human central nervous system.¹ Despite aggressive treatment, the median survival of patients with these tumors is ∼12.5–15 months.^2,3 During their clinical course, almost all patients suffer a posttreatment recurrence. Therefore, to improve the prognosis, the pattern and timing of tumor recurrence must be better understood.

Advances in surgical treatments such as 5-aminolevulinic acid fluorescence-guided surgery,⁴ intraoperative MRI,⁵ and functional mapping⁶ have enabled maximal resection of these tumors without eliciting severe morbidity. Postoperative treatment with temozolomide and radiation to achieve local tumor control also improves the prognosis slightly.⁷ Despite good control of the initial lesion, some patients suffer tumor recurrence at a distant location.^8–12 The reported rate of postoperative distant recurrence including dissemination ranges from 8% to 43.6%.^9,11–13 To predict and address the clinical course of patients with glioblastoma, we must acquire a better understanding of factors involved in the pattern and timing of not only local but also distant tumor recurrence. Detailed analysis of these factors may make prophylactic treatment for recurrences possible. However, there are few studies on such clinical and molecular features.

Here, we conducted a retrospective study to identify the factors associated with distant and local tumor recurrence. The factors analyzed were age, sex, KPS, extent of resection, Ki67 labeling index (LI), CD133 expression, isocitrate dehydrogenase 1 (IDH1) mutation, O⁶-methylguanine methyltransferase (MGMT) gene promoter methylation, and alterations of chromosome 7p (EGFR), 9p (CDKN2A), and 10q (PTEN), all of which are known prognostic factors associated with high-grade gliomas.^2,7,14–17 Ventricular entry during surgery was also analyzed because of the possible distribution of glioblastoma cells via the cerebrospinal fluid (CSF).¹² Because glioma stem cells are involved in the aggressive behavior of glioblastoma,^18,19 we hypothesized that they may play a role in glioblastoma recurrence. Therefore, we focused on the expression of CD133, a cell surface marker of glioma stem cells.^20–22

Materials and Methods

Patients and Samples

This retrospective study was conducted with the approval of the Ethics Committee of the Tohoku University School of Medicine, and written informed consent was obtained from all patients.

Between January 1997 and March 2010, we treated 224 consecutive glioblastoma patients in our department. Of these, 112 met our inclusion criteria: diagnosis of WHO grade IV glioblastoma²³ with no previous history of lower-grade tumors, availability of both genomic DNA and protein, events such as recurrence or death during follow-up, or no events for ≥12 months of follow-up. Patients who had undergone biopsy were excluded from the study. Tumor specimens were obtained from the lesion presenting enhancement in gadolinium (Gd)-enhanced MRI studies, immediately frozen in liquid nitrogen, and stored at −80°C until extraction of DNA and protein.

Definition of Recurrence Patterns

After the initial surgery, radiation, and chemotherapy, all patients underwent Gd-enhanced MRI studies every 2–3 months. Any new enhanced lesions on Gd-enhanced MRI were recorded as the first day of recurrence. For patients with symptoms suggestive of CSF dissemination,^8,24 we performed spinal MRI or CSF studies. Definitions for tumor recurrence were modified from the previously reported criteria that categorized the recurrence as local, distant, diffuse, or multifocal.^25,26 In this study, we defined “local recurrence” as a combination of “local” and “diffuse,” a new enhanced lesion at or within 3 cm of the primary site resection cavity (Fig. 1A). We defined “distant recurrence” as a combination of “distant” and “multifocal,” a new enhanced lesion centered more than 3 cm away from the primary site resection cavity or margin of the primary residual tumor, or more than one lesion site with each lesion having a well-defined border with normal brain signal (Fig. 1B). For patients manifesting a distant recurrence after initial local recurrence, we recorded a local recurrence (Fig. 1C).

Clinical Parameters

Clinical patient profiles were obtained from medical records. All patients underwent radical surgery followed by chemotherapy (nimustine hydrochloride [ACNU] or temozolomide) and radiation therapy (total dose, 60 Gy). Total surgical resection was defined as the disappearance of enhanced tumor lesion on the basis of pre- and postoperative Gd-enhanced MRI studies. Ventricular entry during surgery was identified from patient records or pre- and postoperative MRI scans. On tumor recurrence, the patients underwent salvage surgery, second-line chemotherapy, or radiation therapy. Ki67 LI was determined by immunohistochemical staining of resected specimens with Ki67 antigen (Dako).

Expression of CD133

The expression of CD133 was analyzed by Western blots. To validate the results from Western blots, immunohistochemistry (IHC) was also performed.

Protein was extracted from frozen tissue specimens with a tissue protein extraction reagent (T-PER, Thermo Scientific). The primary antibodies for Western blots were anti-CD133 (1 : 250; Miltenyi Biotec) and anti–β-actin (1 : 500; Santa Cruz Technology). On 8% polyacrylamide gels (Invitrogen) the expression of CD133 and of β-actin was visualized at ∼130 kDa and 47 kDa, respectively. The intensity of bands on Western blots was analyzed using ImageJ software (National Institutes of Health) and the ratio of CD133 to β-actin was determined for each patient sample.

Paraffin-embedded 2-μm-thick sections from primary glioblastoma tissues were stained using antibody to CD133 (CD133/1 AC133, Miltenyi Biotec). Briefly, after deparaffinization, rehydration, and blocking, the sections were incubated with CD133 antibody (1 : 10) for 1 h at room temperature. IHC was carried out using the avidin–biotin method (Vectastain ABC kit, Vector Laboratories).²⁷ CD133 staining was scored by 2 neuropathologists (M.W., H.S.) unaware of the clinical information. The whole tissue section excluding necrotic and normal brain area was semiquantitatively reviewed, and the percentage of CD133-positive cells was calculated as reported previously.¹⁴

Based on the result from Western blots and IHC, CD133 expression was assessed for local and distant recurrence. To identify a correlation between CD133 expression on Western blots and IHC, we calculated the Pearson correlation coefficient. To set the optimal cutoff value for differentiating between high and low CD133 expression in univariate survival analysis, we modified the method previously reported.¹⁴ CD133/β-actin ratio from Western blots was graded as ≥1 or < 1, ≥2 or <2, and ≥3 or <3. Each cutoff was analyzed by logistic regression analyses, and their odds ratios were compared. Similar analysis was performed using data from IHC to verify CD133 expression on Western blots. Correlation between CD133 expression and other factors are listed in Supplementary Table S1.

Molecular Analysis

Genomic DNA was extracted with the QIAamp DNA Mini Kit (Qiagen) using the manufacturer's protocol. IDH1 gene was amplified by PCR; sequencing was conducted as previously described.^28,29 For MGMT promoter methylation analysis, we performed methylation-specific PCR.^23,29,30 The chromosome copy numbers of 7p (EGFR), 9p (CDKN2A), and 10q (PTEN) were analyzed by multiplex ligation-dependent probe amplification (MLPA) using the probe mix from the SALSA MLPA Kit P105 Glioma-2 (MRC-Holland), as previously reported.^29,31

Prognosis

The time to distant recurrence (TTD) or time to local recurrence (TTL) was the interval between the day of first surgery and the day of recurrence detection on MRI scans. Overall survival (OS) was recorded as the interval between the day of first surgery and the day of death or the last follow-up examination.

Statistical Analysis

The relationship between the recurrence pattern and expression of CD133 was evaluated using Mann–Whitney U and Fisher's exact tests. For analyses of correlation between CD133 expression in Western blots and IHC, Western blot data were log-transformed to reduce right skewness. To obtain the best cutoff value for CD133 expression, odds ratios from each cutoff value were compared, and the largest was applied in subsequent studies. The probabilities of TTD, TTL, and OS were calculated using the Kaplan–Meier method and compared with the log-rank test. For multivariate analysis, factors achieving P < .10 in univariate analysis were introduced in a backward stepwise Cox regression analysis for estimating the hazard ratios (HRs) and their 95% confidence intervals (CIs). In analyses of competing risks, Fine–Gray proportional hazard models were used, and distant recurrence and local recurrence were considered as the competing events. All statistical analyses were performed using the SPSS program, Prism (GraphPad Software), and R2 version 15.0. Differences of P < .05 were considered statistically significant.

Results

Population Characteristics

The 112 glioblastoma patients who fulfilled our inclusion criteria consisted of 64 males and 48 females, with median age 57 years (range, 7–77) and a median preoperative KPS score of 70 (range, 20–90). Genomic DNA and protein were obtained from all patients, and paraffin-embedded samples from 95 patients. Median follow-up was 25.7 months (range, 3–152); 94 patients (83.9%) died. Of the 112 patients, 99 manifested recurrence; the recurrence was local in 77 patients and distant in 22. Of the remaining 13 patients, 8 manifested neither local nor distant recurrence, and for 5 patients, we were unable to identify the day of recurrence. Postoperative treatment consisted of radiation alone for 11 patients, and the remaining 101 patients received combined radiation and chemotherapy with temozolomide (n = 27), ACNU (n = 55), or other agents (n = 19). There was no significant difference in OS, TTD, and TTL between patients treated with ACNU and temozolomide (data not shown).

Expression of CD133

Results of CD133 expression assessed by Western blots are shown in Fig. 2A. To validate CD133 expression analyzed by Western blots, IHC was additionally performed (Fig. 2B). Pearson correlation coefficient analysis of CD133 expression analyzed using Western blots and IHC showed a significant correlation (P = .0003; Fig. 2C).

Correlation Analyses to Predict the Pattern of Recurrence

First, we analyzed several factors to see whether they could predict a recurrence pattern. In particular, CD133 expression determined using Western blots was significantly higher in distant recurrence than in local recurrence (P = .0002; Table 1, Fig. 2D). CD133 expression demonstrated using IHC was also higher in distant recurrence than in local recurrence (P = .0043; Fig. 2E). These results showed that high CD133 expression was significantly associated with the pattern of distant recurrence. Homozygous deletion of 9p was also correlated with distant recurrence (P = .045), but other factors did not show significant correlation (Table 1).

Univariate Analysis to Predict the Timing of Recurrence

Next, we investigated whether CD133 expression or other factors are associated with the timing of recurrence. High CD133 expression irrespective of the cutoff value was associated with shorter TTD (Supplementary Fig. S1A). Thus, high CD133 expression may be a predictor of shorter TTD. To determine the optimal cutoff value, odds ratios were examined. The odds ratio of the CD133/β-actin ratio ≥1 or <1 was 9.9 (95% CI 3.5–28.2, P = .000018), that of the CD133/β-actin ratio ≥2 or <2 was 5.5 (95% CI 1.7–18.1, P = .0046), and that of the CD133/β-actin ratio ≥3 or <3 was 2.7 (95% CI 0.6–12.2, P = .20). Therefore, the value 1, indicating the largest odds ratio, was used in this study to distinguish high CD133 expression (CD133/β-actin ratio ≥1) from low CD133 expression (CD133/β-actin ratio <1). To validate this result, CD133 expression determined using IHC was analyzed to predict TTD; we found that higher expression was associated with shorter TTD (Supplementary Fig. S1B).

The median TTD, TTL, and OS for all patients were 13, 9, and 19 months, respectively (Table 2, Fig. 3). Increase in TTD correlated with low CD133 expression (P = .0011, Fig. 3A), low Ki67 LI (<35%; P = .015), and 9p homozygous deletion (−) (P = .021; Table 2). Increased TTL correlated with total resection (P < .0001) and high CD133 expression (P = .012; Fig. 3B, Table 2). Factors associated with prolonged OS were young age (<60 y; P = .012), total resection (P = .0013), low Ki67 LI (<35%; P = .0009), 10q loss (–) (P = .0050), and MGMT gene promoter methylation (P = .049) (Table 2). As shown in Supplementary Table S1, among 31 patients with high CD133 expression, 15 (48%) patients presented distant recurrence and 15 (48%) presented local recurrence, whereas among 81 patients with low CD133 expression, 7 (8.6%) patients showed distant recurrence and 62 (77%) showed local recurrence.

Multivariate Analysis of Prognostic Factors

In multivariate analysis for TTD, the factors introduced were CD133 expression, Ki67 LI (≥35%), 9p homozygous deletion, and 10q loss. We found that only high CD133 expression was an independent poor prognostic factor for TTD (HR 2.9, 95% CI 1.1–7.8, P = .038; Table 3). In analysis of competing risks, only high CD133 expression (HR 6.0, 95% CI 2.6–13.9, P = .000033) remained statistically significant. In multivariate analysis for TTL, the factors introduced were CD133 expression, extent of surgical resection, Ki67 LI (≥35%), and MGMT gene promoter methylation; high CD133 expression was an independent favorable prognostic factor (HR 0.44, 95% CI 0.3–0.8, P = .0056; Table 3). In addition, the absence of total resection and high Ki67 LI (≥35%) were independent poor prognostic factors for TTL (Table 3). In analysis of competing risks, high CD133 expression (HR 0.34, 95% CI 0.20–0.60, P = .00017) and the absence of total resection (HR 1.9, 95% CI 1.1–3.2, P = .017) remained statistically significant.

In multivariate analysis for OS, the factors introduced were age (≥60 y), extent of surgical resection, Ki67 LI (≥35%), IDH1 mutation, 10q loss, and MGMT gene promoter methylation; age ≥60 years (HR 1.7, 95% CI 1.1–2.7, P = .010), the absence of total resection (HR 2.0, 95% CI 1.3–3.2, P = .0023), and 10q loss (HR 1.7, 95% CI 1.0–2.7, P = .037) were independent poor prognostic factors.

Discussion

The evaluation of clinical and molecular factors involved in glioblastoma should advance the understanding of the clinical course of these tumors and help to predict future recurrences. To date, factors associated with recurrence pattern were PTEN mutation,⁹ Ki67 LI (≥35%),⁹ and 1p36 gain¹⁰ with dissemination, MGMT promoter status with the site of recurrence either inside or outside the radiotherapy field,³² and glioblastoma location contacting the subventricular zone with distant recurrence.³³ However, this is the first and largest report analyzing the recurrence pattern and timing of glioblastoma, taking into account multiple clinical and molecular factors. The present study reports several important findings. As expected, the main pattern of initial recurrence in our patients was “local recurrence” (77 of 99 patients, 78%). However, “distant recurrence” was observed in 22 patients (22%), indicating that this type of recurrence is also an important clinical feature of glioblastoma progression. These results are in good agreement with the data of Wick et al,¹³ given that the overall incidence of distant recurrence analyzed by new software MRIcro was also 20%.

Among the factors analyzed, high CD133 expression was a factor predicting the pattern of distant recurrence (Fig. 2D and E), which was also suggested in the report by Sato et al.³⁴ According to previous reports, glioma stem cells with high CD133 expression are resistant to chemoradiotherapy^18,35 and grow as floating cells.³⁶ Thus, these cells may survive the initial treatment and may be able to propagate through CSF, resulting in distant recurrence. Our results showed that the frequent homozygous deletion of 9p is also a predictive factor for distant recurrence. Some studies have reported that 9p homozygous deletion is associated with poorer survival of glioblastoma patients,³⁷ but the association with distant recurrence needs to be elucidated. Ventricular entry during surgery was observed in 73% of our patients. Although ventricular entry may create routes for glioma cell dissemination, it did not correlate with distant recurrence in our study (Table 1). This result supports the report by Elliott et al,¹² who concluded that neither radical surgery nor ventricular entry increases the risk of dissemination.

Another novel feature of our study was analysis of the recurrence timing. In particular, high CD133 expression presented sooner TTD in univariate analysis, and its median was only 13 months. Multivariate analysis of competing risks showed that high CD133 expression is an independent poor prognostic factor for TTD. Therefore, if the patients present with high CD133 expression, we may be able to predict that they will progress to distant recurrence within a short time. In multivariate analyses for TTL, the absence of total resection and high Ki67 LI were, predictably, the independent poor prognostic factors (Table 3). Low CD133 expression was also an independent poor prognostic factor for TTL (Table 3). This is probably because distant recurrence was rare and late in glioblastoma with low CD133 expression (Figs. 2D and E and 3A). In fact, only 8.6% of patients with low CD133 expression presented with distant recurrence (Supplementary Table S1), and it was in late periods. In addition, there is growing evidence that CD133 is not an absolute marker for cancer stem cells.^36,38,39 There may be some unidentified stem cells within the cell population with low CD133 expression that are specifically related to local recurrence. We investigated the expression of CD133, although other stem cell markers^27,40 may also regulate the recurrence pattern; thus, further study is warranted. CD133 positivity has been reported as a poor prognostic factor in patients with glioblastoma.^14,27 In the present study, patients with high CD133 expression tended to have poorer OS (Table 2 and Fig. 3C), but survival after recurrence was significantly shorter in patients with high CD133 expression than in those with low CD133 expression (P = .022; Supplementary Fig. S2A). Furthermore, the pattern of recurrence did not affect OS (data not shown), but survival after recurrence was significantly shorter in patients with distant recurrence than those with local recurrence (P = .0093; Supplementary Fig. S2B). Therefore, CD133 expression and distant recurrence are important factors for understanding the clinical course of patients with glioblastoma.

There are some limitations to the study. Firstly, because this was a retrospective study, not all patients had been treated in the same manner. There is a possibility that the treatments may have affected the pattern of recurrence²⁵; therefore, the interpretation of the results needs some caution, and studies with patients treated in the same manner are needed. Secondly, setting a cutoff value in molecular analysis is always debatable. Previous reports analyzing CD133 expression in glioma set cutoff values to differentiate high CD133 expression from low CD133 expression, but they used different values,^14,27,39,41 indicating that there is no absolute cutoff value. Therefore, the cutoff value may have affected our univariate analysis of the recurrence timing. In terms of assessing the pattern of recurrence, we examined the expression of CD133, finding a positive correlation between high CD133 expression and distant recurrence (Fig. 2D and E). Lastly, sample collection methods may affect the results. Tumor bulk is a mixture of CD133-positive neural precursor cells or endothelial precursor cells⁴² with CD133-positive and -negative glioblastoma cells. In our study, normal tissue and necrotic areas were excluded from IHC. Because the IHC dataset could validate the dataset from western blots, sampling error in Western blots would be limited. However, studies with tumor sampling from multiple lesions or enhanced lesion at the center or periphery of the tumor are needed.

To summarize, we retrospectively analyzed the factors associated with the pattern and timing of glioblastoma recurrence. Despite some inevitable limitations of the study, we suggest that high CD133 expression may predict distant recurrence. Therefore, these observations may be helpful in conducting future clinical studies.

Funding

This work was supported in part by Grants-in-Aid for Cancer Research from the Ministry of Health, Labour and Welfare of Japan to T.T.

Conflict of interest statement. None declared.

Acknowledgments

We thank T. Matsuki and M. Fue for their assistance in extracting genomic DNA and protein and in western blot analysis. The authors thank Ursula Petralia and Enago (www.enago.jp) for the English language review.

References