Relationship between ornithine decarboxylase levels in anaplastic gliomas and progression-free survival in patients treated with DFMO–PCV chemotherapy

Abstract

The purpose of this study was to assess the relationship between progression-free survival (PFS) in patients treated with DFMO + PCV (procarbazine, CCNU, vincristine) chemotherapy for malignant gliomas with tumor cell ornithine decarboxylase (ODC) activity. Formalin-fixed slides were obtained for study patients with anaplastic gliomas (AGs) and glioblastoma treated on protocol DM92-035. ODC levels were measured using an antibody to ODC coupled to Alexa 647 dye (Ab-ODC-Alexa 647). Ab-ODC-Alexa 647 intensity in transgenic murine hearts of differing ODC activity was used to calculate ODC activity in tumor cell nucleoplasm. In total, tumor specimens for 31 of 114 (27%) patients treated on the AG strata and 10 patients from the GBM strata were obtained. We found that tumor ODC level heterogeneity increased with increasing tumor malignancy. In a Cox proportional hazards model, PFS was found to be inversely related to median tumor ODC activity, with an unadjusted hazard ratio for median ODC group (>3.3 vs. ≤3.3 nmol/30 min/μg protein) of 5.8 (p < 0.0001); a median PFS of 522 weeks for patients with AGs with median ODC activity ≤ 3.3 and 31 weeks for the 8 AG and 10 glioblastoma patients with ODC activity > 3.3 nmol/30 min/μg protein. Of AG tumors in which ODC activity was evaluated, 26% had ODC levels > 3.3 nmol/30 min/μg protein. This study shows that Ab-ODC-Alexa 647 fluorescence intensity can be used as a surrogate marker of ODC biochemical activity in AGs and can predict PFS to DFMO-based chemotherapy. © 2007 Wiley-Liss, Inc.

Of the many specific cellular targets of anticancer agents that have been developed, most of the older targets were precursors of DNA, alkylators or binders of DNA, or inhibitors of key DNA structural enzymes. It was usually not possible, however, to measure the levels of these actual targets in the tumors of patients, and thus to be able to confidently predict response to treatment and plan treatments that would be most effective in certain patients. For many of the newer signal transduction inhibitors, efforts have been made to measure both the levels of the presumed target and the relevant downstream and upstream targets affected. This has led to the recognition of Bcr-Abl in acute lymphoblastic leukemia,1 c-kit in gastrointestinal stromal tumors2 and a specific mutations of the EGF receptor in nonsmall cell lung cancer3, 4; in all cases drugs inhibiting tumor-site specific tyrosine kinases led to improved treatment outcomes.

In previous studies, we showed that α-difluoromethylornithine (DFMO; eflornithine), a specific irreversible inhibitor of ornithine decarboxylase (ODC), has antitumor activity against malignant gliomas in adults in the following regimens: as a single agent in patients with recurrent glioma,5, 6 in combination with BCNU (carmustine),7 and in combination with PCV (procarbazine, CCNU, vincristine).8, 9 To determine whether ODC levels differ among tumors, and thus whether the ODC level could be used to predict response to treatment, we developed an assay for measuring ODC levels in formalin-fixed tumor tissue. Our findings from the use of this assay showed that ODC levels were considerably greater in glioblastomas (GBMs; WHO IV) than in astrocytomas (WHO II) and anaplastic astrocytomas (AAs; WHO III) and that the level of ODC in anaplastic oligodendrogliomas (AOs; WHO III) was in between the two.10 We also showed in these studies that ODC activity, as measured using standard assay procedures, correlated directly with the fluorescence intensity emitted by a polyclonal ODC antibody conjugated to an Alexa fluorescent dye.10 This prompted us in the present study to determine whether the progression-free survival (PFS) of patients treated on our previously published DFMO–PCV studies8, 9 correlated with the pretreatment tumor ODC levels in the patients with anaplastic gliomas (AGs) or GBMs and thereby to support or refute our hypothesis that PFS is inversely related to tumor ODC levels. We report here our finding of a good inverse correlation between PFS and tumor levels of ODC in patients with AGs treated with the specific irreversible inhibitor of ODC.

Patients and methods

Study patients and formalin-fixed tumor samples

Letters were sent directly to and/or through the Community Cancer Oncology Program (CCOP) coordinators at The University of Texas M. D. Anderson Cancer Center to all pathology departments that provided original tumor tissue to the study referee pathologist, Janet Bruner, M.D., requesting tumor samples for all 114 patients with AGs (i.e., anaplastic astrocytoma [AA], anaplastic oligoastrocytoma [AOA], and anaplastic oligodendroglioma [AO]) with evaluable outcomes who had been treated postirradiation on protocol DM92-035 with the DFMO–PCV combination.9 In the letter, we asked for the formalin-fixed tissue block or, if that was not available, 3–4 unstained slides of the tumor tissue. Enclosed with the letter was a copy of the IRB approval for the study granted by The University of Texas M. D. Anderson Cancer Center. In addition, we called pathologist offices if tissue blocks/slides were not received. Post hoc we also obtained tissue blocks on 17 patients treated with PCV on protocol DM92-035 from The University of Texas M. D. Anderson Cancer Center.

Study PFS

Survival data for these patients were updated from the original published reports8, 9 by the CCOP Data Management Office through March 13, 2006. PFS was measured from the time of study registration.

ODC antibody

The rabbit polyclonal antibody used was a previously developed polyclonal antibody to ODC (Ab-ODC).11 It was purified on an amino-link column (Pierce Chemical; Rockford, IL) to which purified 6× His-ODC had been cross-linked. Molecular Probes (Eugene, OR) performed a custom conjugation of Ab-ODC to Alexa 647 dye. Prior to conjugation, bovine serum albumin and azide were removed. Binding efficiency was determined to be 2.9 mol dye/mol protein.

Preparation of tumor tissue microarrays

The tissue microarrays were prepared in the Department of Pathology at M. D. Anderson. Tumor blocks to be used for the array were first interrogated by neuropathologists12 to mark the region to obtain cores indicative of the tumor. Two arrays were created with 57 (55 patients as 2 cases were in duplicate) validated human brain tumors. On each array were also 4 heart samples previously extracted from transgenic mice with cardiac ODC overexpression. The cardiac tissue on the array was from formalin-fixed tissue blocks taken from mice sacrificed between day 2 and week 4 after birth. Portions of the original heart tissue had been studied by biochemical techniques previously to determine ODC activity (nmol/30 min/μg protein).13 Arrays were constructed with 0.6-mm diameter cores. Serial array sections were cut at a nominal thickness of 4 μm. In addition to the 55 patient tumors on the microarrays, 8 individual microscope slide cases were used.

Immunohistochemical staining protocol for Ab-ODC-Alexa 647

The experimental methodology used in these studies was similar to that previously used.10 The technique we used for antibody staining was one developed for bright-field in situ hybridization and modified for Ab-ODC-Alexa 647. First, slides with sections of heart or tumor were deparaffinized in xylene and dehydrated with decreasing concentrations of ethanol. Slides were then placed in 1× Target Retrieval Solution (#S169984; DakoCytomation, Copenhagen, Denmark) and steamed for 30 min for antigen unmasking. After cooling, samples were permeabilized with 0.2% Triton X-100 and then rinsed with pyrogen-free distilled water and Ca²⁺- and Mg²⁺-free phosphate-buffered saline (PBS). Samples were blocked with serum-free protein blocking solution (#X090930; DakoCytomation) to quench nonspecific binding. Samples were then incubated with 20-μl stock concentrations (0.8 μg/μl Ab-ODC) of Ab-ODC-Alexa 647 in a humidity chamber at 4°C for 48 hr. Slides were rinsed 3 times in PBS and mounted with VECTASHIELD Mounting Medium (#H-1000; Vector Labs, Burlingame, CA). In all cases, binding of the ODC antibody to human brain tumors was performed simultaneously with transgenic heart tissues on the microarray. Heart tissue was checked for consistency in the ODC values between consecutive staining sessions.

Fluorescence microscopy

Images were acquired using an Olympus BX61 fluorescence microscope containing a mercury arc discharge for fluorometric excitation. This microscope has a trinocular observation head coupled to a Hamamatsu ORCA-ER digital charged coupled device camera system. The microscope, camera and data analysis are operational using a Dell Optiplex GX260 PC and ImagePro 5.1 software developed by Media Cybernetics (Carlsbad, CA). For the measurement of Ab-ODC-Alexa 647 intensity, our approach was to measure optical density in the Cy5 window (Chroma Technology; Rockingham, VT) using 12-bit grayscale images obtained at 40× magnification for 100 msec for the heart samples and 1,000 msec for the tumor samples. Core samples without antibody were also measured in the Cy5 window for 100 msec for the heart samples and 1,000 msec for the tumor samples. Lag time between subsequent exposures was 2–3 min.

Image analysis

Twelve-bit grayscale images were used for all measurements. For heart samples, a pixilated bitmap was created as a single pixel value obtained every 20 pixels using Image Pro 5.1 software. This yielded ∼3,100 pixilated intensity values (i.e., measurements) per heart field. We repeated this analysis on 3 microscopic fields per heart core. We used 4 transgenic heart samples used in a previous study10 that represented the expected range of ODC activity.10

For each tumor sample, we studied 3 microscopic fields per tumor core in 3 tumor core/pathology samples (equal to 9 microscopic fields per patient). The fields chosen were contiguous across the widest horizontal plane. Using Image Pro 5.1 software, a manual tagging approach was applied for each 40× field. Using the grayscale image, presumed tumor nuclei were tagged for analysis. Between 20 and 100 nuclei were tagged per field, depending on the cell density of the histologic specimen. Each tag point was set to measure the average intensity of an area of three pixels (0.5 μm²).

Pathology verification

Using a feature of Image Pro v.5.1 (Snap Measurements feature), the tag points were “burned” onto their corresponding field such that tagged locations could later be compared to hematoxylin and eosin (H&E)-stained tumor core slides. The H&E-stained slides and tagged grayscale template were independently compared by Kenneth Aldape, a neuropathologist at M. D. Anderson, to determine the accuracy of our tumor nucleus tagging method.

Statistical considerations

For both heart and tumor core samples, mean grayscale intensity values were transferred to Microsoft Excel and statistically evaluated using SPSS for Windows (version 12) and/or Prism 4 software (GraphPad; San Diego, CA). In all cases, mean grayscale intensity values were corrected for blank mean grayscale intensity for data analyses. Linear least squares regression analysis was then used to relate the mean grayscale intensity to ODC activity. Mann–Whitney U-tests were used to compare the mean ODC levels between histology groups. Cox proportional hazards regression analysis was used to relate ODC levels and clinical covariates.

Results

Formalin-fixed tumor samples

Of the 114 patients entered into the DFMO–PCV AG arm of protocol DM92-035, we received formalin-fixed specimens for 31 (27%). Written and telephone contact was made with all pathology offices involved with study patients, and we generally had good cooperation from the various pathology departments. The two main reasons given for not being able to provide tumor tissue was (a) insufficient tissue remaining and (b) loss of the material at the Baylor College of Medicine pathology facility during the 2001 flood in Houston. Of the 31 AG tumors, 22 were AAs, 4 AOs and 5 AOAs or malignant gliomas not otherwise specified. Because of the low number of tissue samples obtained and to improve the robustness of the statistical comparison, we obtained 10 GBM tissue blocks from patients treated with DFMO–PCV on the GBM strata of DM92-035 after surgery performed at M. D. Anderson. These 41 tumors from patients in the DM92-035 protocol constituted the evaluable outcomes group. To increase the database of ODC levels in AGs and to determine if there was an ODC effect on PFS independent of treatment (PCV instead of DFMO–PCV), we also obtained tissue blocks from 17 patients (11 AAs, 3 AOs, 1 AOAs and 2 GBMs) treated only with PCV on protocol DM92-035 and 5 (3 AAs, 2 AOs) other patients treated with PCV off protocol bringing the total number of tumors studied to 63 (36 AAs, 9 AOs, 6 AOAs/AGs and 12 GBMs).

Heart ODC activity versus ODC intensity

Figure 1 shows the plots of heart tissue ODC activity (nmol/30 min/μg protein) vs. ODC-Ab-Alexa 647 intensity for the 2 tissue microarrays measured over a period of 2 months (the time that elapsed between the beginning and end of the laboratory portion of these studies). The linear least squares fits are nearly the same with respect to slope and intercept. The greatest variance was seen for the third heart sample and may reflect a real difference in regional heart values for ODC activity between these 2 arrays and the original dataset from the study in which these same heart samples were used.10 The equations for the two least-squares fits are

• 

• 

When these intensities are corrected for the 10-fold difference in exposure time between heart (100 msec) and tumor (1000 msec), and converted to yield ODC activity, they become

• 

• 

Figure 1. Linear least-squared plots of transgenic heart activity (nmol/30 min/μg protein) vs. ODC-Ab-Alexa 647 intensity. Each plot represents four heart samples. The data in Set A (—) and Set B (– – –) were obtained 2 months apart. The bit-mapped intensity for each of the three 40× fields per heart sample (∼3100 total measurements per heart) was used to compute a mean ± SEM. The mean values overlap and the SEM are not visualized since, in all cases, it is less than 1.4% of the mean.

Tumor ODC activity

The calculated values for each tumor represent the total individual measurement of nuclear ODC intensity. If fluorescence intensity was below background, we, for the purposes of this analysis, assigned to this intensity the ODC activity at the lower end of the sensitivity in our assay, 0.05 nmol/30 min/μg protein. The distribution (heterogeneity) of ODC activity for the 63 gliomas from the current study and 18 gliomas from a previously published study10 are shown for all 81 patients in Figure 2. Tumor ODC activities collated in Table I consists of the means of the medians of individual types of tumors and of the 75th percentile ODC activity levels. A Mann–Whitney U-test comparison of the results for the four tumor histology groups demonstrated that ODC activity levels in AOA and AA were similar, in GBMs were 4-fold higher (95% CI 2.7, 20.7) than those in AOAs/AGs, greater than 3-fold higher (95% CI 3.0, 4.3) than those in AAs (p < 0.0001), and 2-fold higher (95% CI 1.7, 2.5) than those in AOs (p < 0.0001). ODC levels in AOs were also nearly 2-fold higher (95% CI 1.7, 1.7) than those in AAs (p < 0.003).

Figure 2. Plot of mean (±95% CI) ODC activity values for 81 tumors. The black diamonds represent the median ODC value. The lower limit of activity was set at zero as the lower measure of ODC activity was computed to be approximately 0.05 nmol/30 min/μg protein. The numbers on the x-axis represents number of patients: Orange bar, anaplastic astrocytoma; blue bar, anaplastic oligodendroglioma; green bar, anaplastic oligoastrocytoma; dark gray bar, GBM multiforme.

Table I. Comparison of ODC Activity (nmol/30 min/μg Protein) Computed Based on Ab-ODC-Alexa 647 Intensity and Heart ODC Activity Standards from Current and Previous7 Tumor Measurements

Pathology	N	Means (SE) calculated from median or 75^th percentiles		Mann–Whitney U-test of medians
		Medians	75th percentiles	AOA/MG	AA	AO	GBM
AO, anaplastic oligodendroglioma; AA, anaplastic astrocytoma; AOA, anaplastic oligoastrocytoma; GBM, glioblastoma.
AOA	6	1.6 (0.5)	2.4 (0.6)		0.59	0.024	0.0003
AA	42	2.0 (0.3)	2.7 (0.3	0.59		0.003	<0.0001
AO	15	3.4 (0.4)	4.1 (0.5)	0.024	0.003		<0.0001
GBM	18	6.9 (0.3)	7.9 (0.3)	0.003	<0.0001	<0.0001

Correlation of tumor ODC activity with PFS

We performed Cox regression analyses to evaluate the relationships of PFS to age, Karnofsky performance score and different percentiles of individual tumor ODC activity (mean, median, and 75th percentile). Since there was no basis to determine which ODC value to use as the cut point, and the data was not normally distributed, we chose to use a log transformed mean (mean/LN(2)) ODC activity value for AG tumors, 3.3 nmol/30 min/μg protein, as the break point. While all GBM were above the 3.3 nmol/30 min/μg protein, 26% of AG tumors were also in that group. Figure 3 is a Kaplan–Meier plot of PFS for the 41 DFMO–PCV patients stratified into 2 groups by ODC activity. The median PFS was 522 (95% CI 424, 620) and 31 (95% CI 0, 68) weeks in these 2 groups, respectively. The differences were statistically significant at p < 0.0001 by log rank test.

Figure 3. Kaplan–Meier representations of the median PFS for each of the 41 evaluable patients (31 anaplastic gliomas and 10 GBM multiforme) stratified by ODC ≤ 3.3 (—) and ODC > 3.3 (···) nmol/30 min/μg protein are 522 weeks (events =12/23) and 31 weeks (events = 17/18) with a p < 0.0001 by log rank test.

We also performed a Cox regression analysis of PFS using as covariates, age, all AG vs. GBM, median ODC activity and ODC median category (either < 3.3 or ≥ 3.3). For ODC median activity, the hazard ratio was 1.5 (95% CI 1.2, 1.7; p-value < 0.0001), and for ODC median category, the hazard ratio was 5.8 (95% CI 3.6, 13.2; p-value < 0.0001). Table II summarizes the Cox regression analysis for the covariates age and median ODC activity that, in addition to the expected significance of age, ODC median activity was significant at p = 0.002. Since histology and median ODC activity are coupled (Table I), a second Cox regression analysis was computed to evaluate the covariates of age, AG or GBM, and ODC median group (< 3.3 or ≥ 3.3). ODC median group defined by ODC activity at 3.3 nmol/30 min/μg protein was used since 26% of AG patients had ODC activity levels seen in the GBM patient tumors. AG tumor grouping was used since there were insufficient numbers of the three histologies that make up the AG group to analyze separately. In this analysis, also summarized in Table II, found that all three covariates were statistically significant. When we did the same Cox regression analyses with the 17 PCV patients, none of the analyses were significant. In the 17-patient PCV dataset, the same Cox regression analyses failed to show significance for the 2- and 3-covariate analyses; this may have reflected the smaller dataset.

Table II. Summary of Cox Regression Model Analysis of PFS and the Following Covariates: (A) Median ODC Activity and Age and (B) ODC Group (Categorical 0 = Less than OR Equal to 3.3 and 1 = Greater than 3.3 nmol/30 min/μg Protein), Age and Glioma Strata (AG = 0 OR GBM =1)

	Hazard ratio (95% CI)	p-value
A. Two covariates
Median ODC activity	1.34 (1.11, 1.61)	0.002
Age	1.07 (1.03, 1.11)	<0.0001
B. Three covariates
ODC activity group	2.6 (1.1, 6.3)	0.033
Glioma strata	0.25 (.09. 0.68)	0.007
Age	1.07 (1.03, 1.11)	0.001

Discussion

Our hypothesis that the intensity of nuclear fluorescence or the pattern of its distribution would be inversely correlated with PFS in patients treated with DFMO–PCV was proven correct in this study, as ODC activity were inversely correlated with higher PFS based on the results in the 41 patients in our study treated with DFMO–PCV.

As shown in Table I, this study demonstrated a range of Ab-ODC-Alexa 647 nuclear fluorescence intensity calculated ODC activity among the four histologies studied. In particular, Table I demonstrates a general difference in the ODC activity between GBM and AOA, AA or AO and that AOA and AA are nearly the same. In the current study, ODC activity appeared to increase in the following manner: AOA = AA → AO → GBM. This pattern is qualitatively similar to that seen in our previous study.10

For each tumor studied, we also saw a range of ODC activities that reflects the heterogeneity of nuclear ODC activity within each tumor. Since ODC is a rate-limiting step in the synthesis of polyamines and polyamines are important to DNA synthesis and transcription, the observed variability of nuclear ODC likely reflects the tumor phenotype, its temporal commitment to cell division, and distance of a tumor cell from capillary nutrients that are critical to cell viability. The consequence of these possibilities is reflected in the observation that heterogeneity (larger range for the 95% CI) increases with higher median ODC activity and tumor malignancy. This is logical since the ODC activity levels in AA and AOA tumors are quite close to nontumor brain levels10 and have low levels of mitosis and, therefore, would be expected to be more homogeneous than mitotically more active gliomas such as AO and GBM.

The increasingly wider range of ODC activity values with increasing mitotic activity posed a question about which single measure for each tumor best predicts whether the tumor will be sensitive to DFMO: will it be the mean or median value, a value from the low end of the ODC activity percentile distribution, or a value from the higher end of ODC activity? We took a number of approaches and found that most approaches established the partial dependence of PFS on ODC activity from the 10th through 75th percentile of ODC activity. We concluded from this that the best way to gauge the impact of ODC activity on PFS was to stratify patients by median ODC activity above or below the log transformed mean value of 3.3 nmol/30 min/μg protein.

The ODC activity value was ≤3.3 nmol/30 min/μg protein in 74% of AG tumors (n = 23) and >3.3 nmol/30 min/μg protein in 26% of AG (n = 8), whereas none of the GBMs had ODC values ≤3.3 nmol/30 min/μg protein (Table I and Fig. 2). Using the 3.3 value as the break point, we found for the DFMO–PCV patient cohort that the median PFS was 522 weeks (10 years) in the 23 AG patients whose median tumor ODC activity was ≤3.3 nmol/30 min/μg protein and only 31 weeks for the 8 AG and 10 GBM patients whose median tumor ODC activity was >3.3 nmol/30 min/μg protein.

Since ODC activity generally corresponds to histology for AG in the astrocytic lineage, we considered that patients receiving PCV without DFMO might show that PFS was inversely related to ODC activity. When we stratified for ODC activity in the same manner for the 17 patients treated with PCV on protocol DM92-035 whose tumor samples were also included on the tissue microarray, we found that 41% (5 AGs and 2 GBMs) of the tumors had an ODC activity >3.3 nmol/30 min/μg protein. A Cox regression analysis showed no statistical significance for covariates important in the DFMO–PCV dataset. In addition, there was no statistical difference in the PFS at 5 years between the two ODC activity strata, with disease not progressing in 50% and 51% of patients. The similarity of the findings in the 2 strata may be explained by the small number of patients examined in the PCV arm and not enough GBMs in the PCV-only treatment group to yield a robust ODC range for the computations. Nonetheless, we can conclude that ODC activity did not affect PFS for those treated with PCV to the extent that it impacted PFS in those treated with DFMO–PCV.

In conclusion, our study suggests that patients with high-grade gliomas (AGs and GBMs) who have a median tumor ODC activity of less than 3.3 nmol/30 min/μg protein prior to the start of a DFMO-based therapy should be expected to have a longer PFS than those with higher ODC activity levels. Whether ODC levels can predict the efficacy of other anticancer agents awaits further study. One would anticipate, however, that this information would be helpful, especially for therapies directed against cell division, such as alkylating agents, and drugs directed against chromatin function, since ODC is induced in response to a variety of mitogenic stimuli and is increased in dividing cells.

Acknowledgements

The authors thank Sandra Ictech for updating the patient data, Kathleen Lamborn for statistical advice, and Joann Aaron and Betty Notzon for editorial assistance.