ExploringtheDistinctiveBiologicalCharacteristicsofPilocyticandLow-GradeDiffuseAstrocytomasUsingMicroarrayGeneExpressionProfiles O A

Texte intégral

(1)

Copyright @ 2006 by the American Association of Neuropathologists, Inc. Unauthorized reproduction of this article is prohibited.

O

RIGINAL

A

RTICLE

Exploring the Distinctive Biological Characteristics of Pilocytic and Low-Grade Diffuse Astrocytomas Using Microarray

Gene Expression Profiles

Sandrine Rorive, MD, Calliope Maris, MD, Olivier Debeir, PhD, Flavienne Sandras, MSc, Michel Vidaud, MD, Ivan Bie`che, MD, Isabelle Salmon, MD, PhD, and Christine Decaestecker, PhD

Abstract

Although World Health Organization (WHO) grade I pilocytic astrocytomas and grade II diffuse astrocytomas have been classified for decades as different clinicopathologic entities, few, if any, data are available on the biologic features explaining these differences.

Although more than 50 microarray-related studies have been carried out to characterize the molecular profiles of astrocytic tumors, we have identified only 11 that provide sound data on low-grade astrocytomas. We have incorporated these data into a comparative analysis for the purpose of identifying the most relevant molecular markers characterizing grade I pilocytic and grade II diffuse astrocytomas. Our analysis has identified various interesting genes that are differentially expressed in either grade I or grade II astrocytomas when compared with normal tissue and/or high-grade (WHO grade III and IV) astrocytomas. A large majority of these genes encode adhesion, extracellular matrix, and invasion-related proteins. Interestingly, a group of 6 genes (TIMP4, C1NH, CHAD, THBS4, IGFBP2, and TLE2) constitute an expression profile characteristic of grade I astrocytomas as compared with all other categories of tissue (normal brain, grade II, and high-grade astrocytomas). The end products (proteins) of these genes act as antimigratory compounds, a fact that could explain why pilocytic astrocytomas behave as compact (well-circumscribed) tumors as opposed to all the other astrocytic tumor types that diffusely invade the brain parenchyma. Having validated these molecular markers by means of real-time reverse transcriptaseYpolymerase chain reaction, an integrated model was proposed illustrating how and

why pilocytic astrocytomas constitute a distinct biologic and pathologic entity when compared with diffuse astrocytomas.

Key Words: Cell migration, Diffuse astrocytoma, Molecular marker, Gene microarray, Pilocytic astrocytoma.

INTRODUCTION

The widely used term Blow-grade astrocytoma[ may lead to conceptual errors, to therapeutic uncertainty, or to misinterpretations of clinical data that result in the inadequate clinical management of patients (1). Indeed, this category includes mainly World Health Organization (WHO) grade I pilocytic and WHO grade II diffuse astrocytomasYY2 tumor groups with completely different behavior patterns. Pilocytic astrocytomas are generally indolent and noninfiltrative neo- plasms known for their high incidence in children and are associated with favorable prognoses (2, 3). Their neuroimaging is mostly characterized by a relatively well-circumscribed, solid, contrast-enhancing mass with a cystic component (2).

Seen under the microscope, pilocytic astrocytomas display a classic biphasic architectural pattern with an absence of parenchyma infiltration, thus allowing total surgical resection if the tumor site permits (2). In contrast, WHO grade II diffuse astrocytomas progress systematically to anaplasia and higher tumor grades (i.e. WHO grade III anaplastic astrocytomas and WHO grade IV glioblastomas). The neuroimaging of grade II astrocytomas most often shows poorly circumscribed and noncontrast-enhancing tumors (3). Seen under the microscope, these tumors are made up of well-differentiated neoplastic astrocytes that diffusely infiltrate the brain parenchyma and make total resection impossible. Although as summarized in Table 1, pilocytic and grade II diffuse astrocytomas are theoretically well defined by the WHO classification (4), drawing a distinction between them may be extremely difficult on histologic grounds alone. This is especially true when only small specimens are available, like in the case of stereotactic biopsies (2). Clinical and neuroradiologic features often facilitate diagnosis, but the data available are sometimes insufficient to distinguish between these 2 types of tumors. In addition, in the case of some pilocytic astrocytomas, the tumor cells infiltrate the surrounding normal parenchyma and, as J Neuropathol Exp Neurol Volume 65, Number 8, August 2006

794

J Neuropathol Exp Neurol

CopyrightÓ2006 by the American Association of Neuropathologists, Inc.

Vol. 65, No. 8 August 2006 pp. 794Y807

From Laboratory of Pathology (SR, CM, FS, IS), Erasmus University Hospital, Erasmus, The Netherlands; Department of Logical and Numerical Systems (OD), Faculty of Applied Sciences and the Laboratory of Toxicology (CD), Institute of Pharmacy; Universite´ Libre de Bruxelles, Brussels, Belgium;

and Laboratoire de Ge´ne´tique Mole´culaire-INSERM U745 (MV, IB), Faculte´ des Sciences Pharmaceutiques et Biologiques, Universite´ Paris V, Paris, France.

Send correspondence and reprint requests to: Isabelle Salmon, MD, PhD, Laboratory of Pathology, Erasmus University Hospital, 808 route de Lennik, B-1070 Brussels, Belgium; E-mail: isalmon@ulb.ac.be

CD is a Senior Research Associate with the BFonds National de la Recherche Scientifique,[Brussels, Belgium.

This work has been supported by grants awarded by the Fonds Yvonne Boe¨l (Brussels, Belgium).

Supplementary data is available online at http://jneuropath.com.

(2)

seen under the microscope, mimic the infiltrative pattern of diffuse astrocytomas (5).

In addition to the need for new diagnostic and prognostic markers to improve the pathologic evaluation of, and the distinction between grade I and grade II astrocytomas, an understanding of the biologic and genetic features leading to tumor carcinogenesis and progression is a major key to suitable patient management and treatment. In this context, the molecular profiling of astrocytomas constitutes an interesting line of investigation in the identification of specific gene products involved in tumor processes. A number of genetic alterations have already been identified in human astrocytic tumors, including certain ones associated with prognostic values in grade II tumors such as p53 (3Y8).

In the past few years, microarray technology has been extensively applied in the search for genetic alterations that differentially characterize the various histologic astrocytoma

types. However, a majority of the microarray studies under- taken have focused on the genetic profiles of high-grade tumors (especially glioblastomas), i.e. the ultimate stage in the development and aggressiveness of these cancers. Of more than 50 studies centered on microarray analyses of human astrocytic tumors, we identified only 11 that included data on low-grade astrocytomas (9Y19). The aim of the present study is to exploit these data as efficiently as possible to extract accurate information usable for the characterization of grade I as opposed to grade II astrocytomas. Unfortunately, the data provided by the studies on low-grade astrocytomas were not sufficiently detailed to carry out statistical meta-analyses (20, 21). Therefore, we used a comparative analysis (21) based on the identification of interstudy concordances and discor- dances. In the present study, we analyzed a compilation of 11 studies reporting altered gene expression profiles in grade I and grade II astrocytomas to identify the most promising molecular markers (oncogenes or tumor-suppressor genes) in- volved in the tumorigenesis and biology of grade I and grade II astrocytomas. We then used real-time reverse transcriptaseY polymerase chain reaction analyses on a series of normal and tumor tissue samples to confirm the molecular profiles of these markers.

MATERIALS AND METHODS Selection of the Studies Analyzed

To identify which studies were eligible for inclusion in our comparative analysis, we surveyed the MEDLINE cita- tions by means of the National Library of Medicine_s PubMed online search engine. We used different combinations of the following key words: Bhuman,[ Bglioma,[ Bastrocytoma,[

Barray[ (or Bmicroarray[), and Bgene.[ Our last query was updated on November 3, 2005. From the approximately 50 studies identified, we selected those which included low-grade astrocytomas in their series but excluded those in which grade II and grade III astrocytomas were grouped together and compared with glioblastomas. This led to in the selection of 11 studies published by 10 independent laboratories between December 2000 and October 2005 (9Y19).

Description of the Studies Analyzed

Table 2 describes the main characteristics of the 11 studies selected. The biologic materials used are mentioned in terms of the series of human astrocytoma samples analyzed (pooled specimens are specified) and the normal brain samples taken as control (with the mention of their various origins). The astrocytomas were classified according to the WHO criteria as follows: grade I = pilocytic astrocytomas, grade II = diffuse astrocytomas, grade III = anaplastic astro- cytomas, and grade IV = glioblastomas (4). The patients_

ages were specified when they were detailed by the authors.

It should be noted that although only 2 of the 11 studies reported separate data on both grade I and grade II astrocytomas (13, 19); another, which only looked at child- hood astrocytomas, pooled grade I and II tumors into a low- grade group (15). These latter data were therefore treated with caution in the comparative analysis. The microarray technologies used in each study are summarized in Table 2 TABLE 1. Major Clinicopathologic Features Distinguishing

Grade I Pilocytic from Grade II Diffuse Astrocytomas Pilocytic

Astrocytomas

Diffuse Astrocytomas (WHO grade I) (WHO grade II)

Clinical Features

Age Children9adults* Adults9children

Site Infratentorial9 supratentorial cerebellum, midline structures

Supratentorial9 infratentorial cerebral hemispheres, brainstem

Development Mostly indolent tumors, anaplastic transformation:

11%

leptomeningeal dissemination:

4Y12%

Tendency toward anaplasia

Morphologic features Neuroimaging Well-circumscribed

tumors

Ill-circumscribed tumors Frequently cystic Mostly solid

Contrast enhancement No contrast enhancement Under the

microscope

No tumor cell infiltration (mostly)

Diffuse tumor cell infiltration Biphasic pattern:

piloid/microcystic

Monophasic pattern Nuclear atypia Nuclear atypia

Mitosis† No mitosis‡

Vascular proliferation†

No vascular proliferation‡

Necrosis† No necrosis‡

Clinical management Complete tumor

resection (whenever possible)

Gross tumor resection followed by aBwait and see[approach or adjuvant therapy

*, X9Y means more frequent in the case of X than of Y.

†, Controversial prognostic significance (2, 3).

‡, Hallmarks of high-grade astrocytomas (4).

WHO, World Health Organization.

(3)

Copyright @ 2006 by the American Association of Neuropathologists, Inc. Unauthorized reproduction of this article is prohibited.

TABLE 2. Description of the Microarray Studies Including Low-Grade Astrocytomas in Their Series

Reference Authors

Number of Human Samples (normal brain and astrocytomas) Origins of Normal Brain

Control Samples Array type

Signal Revelation (measurement)

Selection of Differentially

Expressed Genes

Available Data for Selected

Genes Normal Grade I Grade II Grade III Grade IV

9 Huang et al,

2000

3A 0 11A 0 0 One hippocampus

(epilepsy), 1 postmortem cortex, 1 postmortem medulla

cDNA clone Radioactivity (absolute)

Intensity ratio 92 (orG0.5)

Ratios of mean intensity values

10 Sallinen et al, 2000

1U 0 2U pooled 2U pooled 3U pooled Total RNA

from a normal human brain

cDNA clone Radioactivity (absolute)

Intensity ratio 91.8 (orG0.56)

Semiquantitative intensity values

11 Godard et al,

2003

2U 0 2C + 22A 0 27A One lobectomy

(after brain edema), 1 total RNA from a normal human brain

cDNA clone Radioactivity (absolute)

Intensity ratio 92 (orG0.5) and statistical test

p values between pairs of groups

12 Hunter et al,

2002

0 3U pooled 0 4U pooled 0 V cDNA clone Fluorescence

(ratio)

Intensity ratio 91.8 (orG0.56)

Mean intensity values per group (and their ratios)

13 Rickman et al,

2001

6A 19U 5U 0 21U Normal temporal

cortex

Oligonucleotide Fluorescence (absolute)

Intensity ratio 91.5 (orG0.67) and statistical test

Ratios of mean intensity values and p values

14 Gutmann

et al, 2002

3A 7C + 1A 0 0 0 Normal white

matter

Oligonucleotide Fluorescence (absolute)

Statistical test Intensity values per case and p values

15 Khatua et al,

2003

0 @Y7C @Y6C V Oligonucleotide Fluorescence

(absolute)

Statistical test Mean intensity values per group and p values

16 Ljubimova

et al, 2001

70U pooled 0 2A 0 5A Pooled RNA

from normal corpus callosum 70 trauma patients)

cDNA clone Fluorescence (ratio)

Intensity ratio 92 (orG0.5)

Ratios of mean intensity values

17 van den Boom

et al, 2003

0 0 8A @Y8A V Oligonucleotide Fluorescence

(absolute)

Intensity ratioQ2 (orG0.5) and statistical test

Ratios of mean intensity values and p values

18 Wong et al,

2005

2C + 1A 21C 0 0 Cerebellar tissue

adjacent to tumor

Oligonucleotide Fluorescence (absolute)

Intensity ratioQ3 (orG0.33) and statistical test

Ratios of mean intensity values

19 Huang et al,

2005

4U pooled 8C pooled 15A 0 Postmortem

normal cerebellum

cDNA clone Radioactivity (absolute)

Intensity ratio 92 (orG0.5)

Ratios of mean intensity values

C, children; A, adults; U, unmentioned.

RoriveetalJNeuropatholExpNeurolVolume65,Number8,August2006

Ó2006AmericanAssociationofNeuropathologists,Inc.

796

(4)

under the following headings: types of arrays used in each experiment (cDNA clones or oligonucleotides), signal revelation techniques (radioactivity or fluorescence), and types of measurements produced (absolute expression levels or expression ratios). Table 2 also lists the criteria used by the authors to identify the genes differentially expressed across the sample categories as well as the available data on the selected genes. The selection criteria were usually based on signal intensity ratios (between the mean intensity values) computed for each pair of compared tissue categories. These ratios were compared with a priori fixed thresholds com- plemented or otherwise by statistical tests to identify statis- tically significant differences. In some cases, the selection was based on statistical tests only. The available data on gene expression generally concern genes selected as differentially expressed only and are often expressed in terms of mean intensity value ratios with or without any associated statistical p values (Table 2). In the supplementary materials, Supple- mentary Table 1 completes this description by reporting various details mentioned in the studies analyzed (acquisition of the sample materials, quality controls, and validations).

Comparison of the Microarray Targets

To evaluate the number of genes commonly analyzed in the different studies (2 by 2), we compared the sets of transcripts targeted by the different microarray platforms used by the authors on the basis of the GenBank identifiers present in publicly available array descriptions (electronic files provided by the manufacturers). Having suppressed the identifier repetitions, we then counted the number of common (and unique) identifiers found (case-insensitive text comparison) per pair of arrays. However, because the same gene could be labeled with different GenBank identifiers, the results provided subsequently are pessimistic evaluations of the number of targets common to pairs of arrays. To provide more realistic information on the data actually available for our analysis, we also report the number of gene expression values explicitly provided by the authors and thus actually exploitable by our method of analysis, which is detailed in the next section.

Method of Analysis

The data reported in the 11 publications considered in our analysis were carefully reviewed by 2 members of our team (SR and CD). To characterize grade I and grade II astrocytomas, 5 comparisons were made: 1) grade I versus grade II astrocytomas; 2) grade I astrocytomas versus normal brain tissue; 3) grade II astrocytomas versus normal brain tissue; 4) grade I versus high-grade (grade III and/or IV) astrocytomas; and 5) grade II versus high-grade astrocytomas.

It should be kept in mind that in comparisons 4 and 5, the data provided by Khatua et al did not enable any distinction to be drawn between grade I and grade II and could thus be a source of discordance in comparison with other data (15).

Because the available data did not enable any statis- tical meta-analysis to be carried out on the different studies, we adopted the following methodology, based, as far as possible, on interstudy validations, with a view to identify- ing the genes that (most likely) exhibit differential levels of

expression in the context of the 5 comparisons mentioned previously. First, to permit an interstudy concordance and discordance analysis, we selected the genes whose expres- sion data had been reported in at least 2 different studies.

This selection was made across 2, 4, and 3 studies in the case of grade I-related comparisons 1, 2, and 4, respectively (taking the Khatua et al study [15] into consideration for comparison 4), and across 6 studies in the case of grade II- related comparisons 3 and 5 (taking the Khatua et al study [15] into consideration for comparison 5). In this first selection, we identified the genes for which a differential expression was concordantly reported in the studies in- volved. Because of the small number of studies on grade I astrocytomas, in the case of grade I-related comparisons, we additionally selected the genes reported as having heavily and significantly modified levels of expression; if the ex- pression ratios between 2 groups of tissue samples exceeded 8 or was below 0.125 (i.e. at least 4 times above or below the threshold value usually adopted in the various studies;

Table 2) and was statistically significant. This drastic selection was motivated by the fact that no interstudy validation was possible and that the number of tumors analyzed was often low. Finally, to provide a complete range of information, when a gene was selected as being differentially expressed across 2 tissue categories, the data available for the other comparisons are also mentioned.

Real-Time Reverse TranscriptaseYPolymerase Chain Reaction Analyses

Real-time reverse transcriptaseYpolymerase chain reaction (RT-PCR) analyses were carried out on a series of tissue samples to validate a number of the conclusions result- ing from our comparative analysis. To parallel the hetero- geneity of the normal samples encountered in the microarray studies analyzed (Table 2), 9 normal postmortem brain tissue samples of different origins (3 from the white matter, 3 from the frontal cortex, and 3 from the cerebellum) were obtained within 24 hours after death and stored atY80-C until RNA isolation. Frozen tumor samples from 16 primary tumors (6 grade I, 5 grade II, and 5 grade IV) were added to this series. RNA extraction, cDNA synthesis, and PCR reaction conditions have been described previously (22).

Quantitative values were obtained from the threshold cycle number (Ct value) at which the increase in the fluorescent signal associated with an exponential growth of PCR products begins to be detected by the laser detector of the ABI Prism 7700 Sequence Detection System (Perkin- Elmer Applied Biosystems, Foster City, CA) using the PE Biosystems analysis software according to the manufac- turer_s manuals. The precise amount of total RNA added to each reaction mix (based on optical density) and its level of quality are both difficult to assess. We therefore also quantified the transcripts of RPLP0, an endogenous RNA control gene (22).

As previously detailed (22), the results were expressed as N-fold differences in target gene expression relative to the RPLP0gene and termedBNtarget.[The Ntargetvalues of the samples were subsequently normalized so that the median of Ntargetvalues of the 9 normal brain tissue samples was one.

(5)

Copyright @ 2006 by the American Association of Neuropathologists, Inc. Unauthorized reproduction of this article is prohibited.

A cutoff threshold of 3-fold differences in terms of the Ntarget median values was applied to detect gene up-/

downregulation in one tissue category compared with another.

Primers for the RPLP0 and the 10 target genes were chosen with the assistance of the Oligo 5.0 computer program (National Biosciences, Plymouth, MN). We con- ducted searches in the dbEST and nr databases to confirm the absence of single nucleotide polymorphisms and the total gene specificity of the nucleotide sequences chosen as primers. To avoid the amplification of contaminating genomic DNA, one of the 2 primers was placed at the junction between 2 exons.

Statistical Analysis of Real-Time

Reverse Transcriptase

Y

Polymerase Chain Reaction Results

Statistical comparisons of the 4 different groups of tissue samples analyzed by means of real-time RT-PCR were carried out using the nonparametric Kruskal-Wallis (analysis of variance) test because it is the most powerful in comparing more than 2 groups of data with reduced sizes (23, 24). We also compared the grade I tumor group with another group (3 comparisons) by means of the Dunn

procedure adapted for multiple comparisons with respect to one fixed group (23, 24). However, this test is very conser- vative in that it produces a large type II error. Consequently, the Mann-Whitney test between pairs of groups was also carried out to indicate which differences contribute the most to the significance level evidenced by means of the Kruskal- Wallis test. All the statistical analyses were carried out using Statistica (Statsoft, Tulsa, OK).

RESULTS

Table 3 summarizes the results obtained by comparing the lists of genes targeted by the different microarrays, as well as the manufacturer_s reference, the number of GenBank identifiers before and after the suppression of the replicates present on Affymetrix arrays and the number of common identifiers per pair of arrays. However, the actual amount of expression data that we were able to exploit in our comparative analysis was strongly reduced in some cases, as detailed in Table 4. Nonetheless, interesting data were found by comparing the different studies, as detailed subsequently.

Tables 5 and 6 list those genes whose expression data reported in at least 2 different studies agreed on differential TABLE 3R Comparative Analysis of the Set of Genes Analyzed and the Data Provided in the Different Studies

Number of Common Targets Between Each Pair of Arrays

Reference

Microarray (manufacturer:

reference) Sequences*

Unique ID†

Clontech:

7851-1huCa12

Clontech:

7742-1huCa

Affymetrix:

Hu6800

Affymetrix:

U95Av2

Affymetrix:

U133A

9, 11, 19 Clontech: 7851-1huCa12 1,185 1,185 594 765 822 155

10 Clontech: 7742-1huCa 597 597 594 352 372 83

13, 17 Affymetrix: Hu6800 7,129 6,948 765 352 4,890 823

14, 15 Affymetrix: U95Av2 12,625 11,302 822 372 4,890 1,941

18 Affymetrix: U133A 22,283 21,196 155 83 823 1,941

12, 16 Incyte Genomics: UniGemV‡ 11,004

*, Number of targets labeled with GenBank identifiers, a (relatively reduced) number of additional consensus sequences present on Affymetrix arrays (without Genbank identifiers) were not taken into account.

†, Number of unique targets after suppression of replicated identifiers (ID).

‡, No data provided by the manufacturer.

TABLE 4R Comparative Analysis of the Set of Genes Analyzed and the Data Provided in the Different Studies Number of Genes Whose Expression Was Mentioned in Each Study and Was Exploitable in Our Analysis

Reference Authors

Subgroup Comparisons Grade I/

Normal

Grade I/

Grade II

Grade I/High Grade Grade II/

Normal

Grade II/

High Grade II/IV II/III + IV I/III I/IV I/III + IV

9 Huang et al 24

10 Sallinen et al 50 50 50

11 Godard et al 8 36

12 Hunter et al 31

13 Rickman et al 102 102 102 102 102

14 Gutmann et al 47

15 Khatua et al 144 144

16 Ljubimova et al 39 17

17 van den Boom et al 70

18 Wong et al 84

19 Huang et al 129 129 129

Rorive et al J Neuropathol Exp Neurol Volume 65, Number 8, August 2006

Ó2006 American Association of Neuropathologists, Inc.

798

(6)

expression levels between grade I or grade II astrocytomas and normal tissue or high-grade tumors. In addition, Table 5 includes genes with a marked difference in expression between grade I tumors and one of the other categories of samples (normal tissue, grade II, or high-grade astrocy- tomas). Although the results for high-grade astrocytomas are detailed in terms of grade III, grade IV, or both (III + IV), depending on the data available per study, we considered the concordance of data in terms of the high-grade category taken as a whole (given the lack of a satisfactory amount of data in each category). Tables 5 and 6 list the genes grouped by class according to their putative functions and detail the ratios (and/or the associated p values when available) of

the average levels of gene expression measured in the 2 tissue groups concerned. These ratios are either directly mentioned in the studies referenced or were computed from the data provided. When a gene was identified as being differentially expressed across 2 tissue categories, the ratios (and/or p values) available for the other pairs of categories are also mentioned.

Information in Supplementary Table 2 shows genes that, although belonging to the first selection, do not appear in Tables 5 and 6 because 2 studies either disagreed on their expression levels or agreed on the absence of any variation in expression between the 2 tissue categories under comparison.

TABLE 5R Genes Differentially Expressed in Grade I Astrocytomas

Gene

Putative Function

Grade I/ Normal 4 Studies (13, 14, 18, 19)

Grade I/

Grade II 2 Studies

(13, 19) I/III

Grade I/High-Grade 3 Studies (12, 13, 15)

Reference I/IV I/III + IV

TIMP4 Tissue inhibitor of metalloproteinases 4

Invasion 12.4 (pG0.001)* 3.4 (p = 0.02) 2.3 (pG0.001) 13

39.0 14

4.7 (p = 0.004) 15 C1NH First component of

complement inhibitor

Protein degradation regulation (invasion)

7.7 (pG0.001) 3.6 (pG0.001) 2.2 (p = 0.001) 13

CHAD Chondroadherin ECM 8.0 (p = 0.001) 8.0 (p = 0.01) 8.0 (pG0.001) 13

THBS4 Thrombospondin 4 ECM 11.4 (pG0.001) 8.0 (p = 0.007) 13

CSPG2 Chondroitin sulfate proteoglycan 2

ECM 20.5 (pG0.001) 13

COL9A1 Collagen type IX,

>1 chain

ECM 9.2 (p = 0.007) 5.6 (pG0.001) 13

9.4 (pG0.05) 18

SPARC Secreted protein ECM 3.0 12

Acidic, rich in cysteine

1.8 (pG0.05) 2.3 (p = 0.04) 4.4 (pG0.001) 13

SEMA5A Semaphorin 5A Adhesion 6.4 (p = 0.000) 13

8.2 18

AGC1 Aggrecan 1 Adhesion 2.7 (p = 0.001) 13

5.3 18

IGFBP2 Insulin-like growth factor-binding protein 2

Growth 4.0 4.0 0.31 (p = 0.002) 13

6.0 14

0.07 (p = 0.001) 15 EGFR Epidermal growth

factor receptor

Growth 0.28 12

0.02 (p = 0.003) 15 LPL Lipoprotein lipase Lipid

metabolism

16.2 (pG0.001) 4.3 (pG0.001) 13

APOD Apolipoprotein D Lipid metabolism

8.5/6.5† 12

p90.05 p90.05 2.8 (pG0.001) 13

1.4 14

TLE2 Transducin-like enhancer of split 2

Signal transduction

2.1 (p = 0.008) 2.1 (pG0.001) 2.1 (p = 0.003) 13

*, Ratios of the average gene expression levels measured in the 2 tissue groups indicated in the column headings and the associated statistical p values (when available).

†, Results obtained with 2 separate probes.

(7)

Copyright @ 2006 by the American Association of Neuropathologists, Inc. Unauthorized reproduction of this article is prohibited.

TABLE 6R Genes Differentially Expressed in Grade II Astrocytomas

Gene

Putative Function

Grade II/

Normal 6 Studies (9Y11, 13, 16, 19)

Grade II/High-Grade 6 Studies (10, 11,13, 15, 16, 17)

Reference

II/III II/IV II/III + IV

TYRO3 Tyrosine protein kinase SKY Adhesion 0.25* 9

G0.56 10

0.55 13

0.13 19

CD9 CD9 antigen Adhesion 2.3 11

2.3 19

NCAM1 Neural cell adhesion Adhesion up (pG0.05) 11

molecule 1 1.8 13

TIMP3 Tissue inhibitor of metalloproteinases 3 Invasion >20,000† 9

2.3 (pG0.05) 11

MMP16 Matrix metalloproteinase 16 Invasion Up (pG0.05) 11

3.00 (p = 0.006) 15

CSPG2 Chondroitin sulfate proteoglycan 2 ECM 10.2 13

4.7 16

FN Fibronectin ECM Off/off Off/on 10

Down (pG0.05) 11

0.44 (p = 0.03) 15

0.07 16

COL4A1 Collagen IV>1 chain ECM 0.18 16

0.13 (p = 0.004) 17

COL4A2 Collagen IV>2 chain ECM 0.5 (p = 0.04) 15

0.09 (p = 0.003) 17

COL5A2 Collagen V>2 chain ECM 0.39 (p > 0.05) 15

0.28 (p = 0.01) 17 KCNN3 Potassium small conductance

calcium-activated channel, subfamily N, member 3

Neuronal 4.2 9

Excitability 20.8 19

EGFR Epidermal growth factor receptor Growth 10,000Y20,000† 9

3.6 0.46 16

6.8 19

Down (pG0.05) 11

0.02 (p = 0.003) 15 PDGFRA Platelet-derived growth

factor receptor>subunit

Growth 2.1 9

6.7 19

NTF3 Neurotrophin 3 Growth 2.1 11

2.0 19

VEGF Vascular endothelial growth factor Growth Off/off Off/on 10

0.47 (p > 0.05) 15

0.30 16

IGFBP2 Insulin-like growth factor binding protein 2

Growth On/off G0.56 10

0.08 (pG0.001) 13

0.07 (p = 0.001) 15 IGFBP3 Insulin-like growth factor

binding protein 3

Growth Off/off Off/on 10

Down (pG0.05) 11

G0.01 16

IGFBP5 Insulin-like growth factor binding protein 5

Growth similar G0.56 10

0.02 16

*, Ratios (and their associated statistical p values, when available) of the average gene expression levels measured in the 2 tissue groups indicated in the column headings.

†, Absolute expression in grade II because undetected in normal tissue (100% of cases with a detectable mRNA level).

Off, undetected; on, detected; up, upregulated; down, downregulated.

Rorive et al J Neuropathol Exp Neurol Volume 65, Number 8, August 2006

Ó2006 American Association of Neuropathologists, Inc.

800

(8)

Identification of the Most Differentially Expressed Genes in Grade I and

Grade II Astrocytomas

The 2 studies that permitted a comparison of grade I and grade II astrocytomas (13, 19) reported 29 and 129 genes, respectively, as being differentially expressed in these 2 astrocytoma subgroups. Unfortunately, only one gene (PIK3R1) was mentioned in both studies, although with discordant results (Supplementary Table 2), and thus no concordant variations are reported in Table 5. In contrast to the study by Rickman et al, which provides the level of statistical significance of each expression variation (13), Huang et al do not provide this information or any other data concerning the heterogeneity levels of the individual expression values (19). Table 5 thus mentions significant results from the study by Rickman et al only (seeBMaterials and Methods[), and we subsequently validated these by means of real-time RT-PCR, as detailed subsequently. The 2 most differentially expressed genes reported by these authors code for ECM components chondroadherin (CHAD) and thrombospondin-4 (THBS4), both of which show a high level of overexpression in grade I as compared with grade II astrocytomas (the expression ratios are equal to 8). Table 5 also mentions 2 other genes characterized by lower expression ratios but which yield the most significant results (p G 0.001). These data concern the first component of complement inhibitor (C1NH) and the transducin-like enhancer of split-2 (TLE2), both of which are overexpressed in grade I astrocytomas. Although the latter codes for a mammalian homolog to the Drosophila Groucho gene that represses the transcriptional response to several extracellular signals, the former (also known as SERPING1, a class G, member 1, serine proteinase inhibitor) regulates protein degradation in the usual activation pathway of the comple- ment system and is also associated with antiinvasive properties, as detailed in theBDiscussion.[

The other expression ratios mentioned in Table 5 for the grade I/II comparison in fact offer complementary information on the genes identified as differentially expressed in grade I astrocytomas as compared with normal brain and/or high-grade astrocytomas, as detailed in the following section.

In summary, CHAD, THBS4, C1NH, and TLE2 are the genes that exhibit the most modified levels of expres- sion between grade I pilocytic and grade II diffuse astro- cytomas (in terms of either the expression ratios or the associated p values). In that it provides all the data available on these 4 genes, Table 5 also shows other interesting data on their expression in the other categories of tissue, as de- tailed subsequently.

Identification of Genes With an Altered Expression in Grade I Astrocytomas as Compared With Normal Brain Tissue and High-Grade Astrocytomas

We focused on the identification of the genes potentially involved in the tumorigenesis of grade I astro- cytomas, i.e. those that exhibit differential expression when

these tumors are compared with normal brain tissue. In this context, the strong overexpression of the tissue inhibitor of metalloproteinases 4 (TIMP4, coding for a multifunctional protein with antiinvasive properties) occurring in grade I astrocytomas was evidenced in 2 studies, both of which reported high expression ratios (12.4 and 39; Table 5). Our analysis also pointed to the fact that 4 genes encoding different ECM components are strongly upregulated in grade I astrocytomas as compared with normal brain tissue (signif- icant ratios = 8). These genes are chondroadherin (CHAD), thrombospondin 4 (THBS4), chondroitin sulfate proteo- glycan 2 (CSPG2)Valso known as versicanVand the collagen type IX>1 chain (COL9A1) (each reported in a single study with the exception of COL9A1, which is evidenced in 2;

Table 5). In addition, semaphorin 5A (SEMA5A) and aggrecan 1 (AGC1), 2 genes encoding adhesion molecules, are reported in 2 studies as being upregulated in grade I astrocytomas. This is also the case for the insulin-like growth factor binding protein-2 (IGFBP2). Finally, one study reports the very strong (and significant) upregulation (ratio >16) of lipoprotein lipase (LPL, involved in lipid metabolism) in grade I astrocytomas. Supplementary Table 2 mentions 4 additional genes reported in at least 2 different studies that either disagreed on their expression levels (2 genes) or agreed on similar expression levels (2 genes) when grade I tumors were compared with normal tissue.

It should be noted that all 4 genes identified as being overexpressed in grade I as compared with grade II astrocytomas (CHAD, THBS4, TLE2, and C1NH) also remain overexpressed when compared with normal brain tissue and thus appear to be very characteristic of grade I astrocytomas. Two additional genes (TIMP4 and IGFBP2) identified as being overexpressed in grade I tumors as compared with normal brain tissue also show overexpression when compared with grade II astrocytomas (Table 5).

To complete the characterization of grade I astrocy- tomas, we identified the genes differentially expressed in grade I as compared with high-grade (grade III or IV) astro- cytomas. It would be useful to compare these identified genes with those having a modified level of expression in grade II as opposed to high-grade astrocytomas (analyzed in the next section), both categories belonging to the diffuse astrocytoma continuum.

Of the 10 genes reported in Table 5 as being differ- entially expressed between grade I and high-grade astrocy- tomas, only 3 had not already been identified in the earlier comparisons reported here (i.e. grade I versus grade II and grade I versus normal). These 3Bnew[genes areSPARC(an ECM-related gene coding for osteonectin), APOD (coding for apolipoprotein D, a member of the lipocalin family), and EGFR(the epidermal growth factor receptor). In each case, 2 different studies agreed to the extent that they evidenced either upregulation (SPARC andAPOD) or downregulation (EGFR) in grade I as compared with high-grade astrocy- tomas. Additional data on APOD report that this gene is expressed similarly in normal tissue and in grade I and grade II astrocytomas (Table 5). In the case of SPARC gene expression, it should be mentioned that Khatua et al report similar levels of expression in high-grade and low-grade

(9)

Copyright @ 2006 by the American Association of Neuropathologists, Inc. Unauthorized reproduction of this article is prohibited.

astrocytomas (15); this may result from the inclusion of grade II tumor samples in the group of low-grade astro- cytomas analyzed by these authors. Indeed, another study reports fairly similar levels of SPARC gene expression in grade II and high-grade astrocytomas (13). No other discordant data were evidenced in the grade I/high-grade tumor comparison.

Six other genes (TIMP4, CHAD, COL9A1, LPL, TLE2, andC1NH) are reported as being overexpressed in grade I as compared with high-grade astrocytomas, similar to their overexpression when compared with normal tissue and/or grade II astrocytomas. Finally, 2 studies agree on the down- regulation of IGFBP2 in grade I as compared with high- grade astrocytomas. This contrasts with the upregulation reported when grade I astrocytomas are compared with either normal brain tissue or grade II astrocytomas.

All these results indicate that these different genes could constitute a potential tool for the characterization of pilocytic astrocytomas (grade I). In particular, 5 genes (TIMP4, C1NH, CHAD, TLE2, and IGFBP2) exhibit differential levels of expression that enable grade I astrocytomas to be distin- guished from all the other categories of samples analyzed.

Identification of Genes With an Altered Expression in Grade II Astrocytomas As Compared With Normal Brain Tissue or High-Grade Astrocytomas

We first compared the levels of gene expression in grade II astrocytomas and normal brain tissue to identify potential carcinogenesis agents. In this case, different pairs (or even a larger number) of studies agree to the extent that they report changes in expression in 4 different adhesion, invasion or ECM-related genes involved in cell migration, and interactions with the surrounding environment. In contrast to TYRO3(coding for protein tyrosine kinase SKY involved in cellYcell and cellYECM adhesion), which is downregulated in grade II astrocytomas, the 3 other genes are overexpressed in these tumors (as compared with normal brain tissue). These genes areCD9(coding for a cell surface glycoprotein that is a member of the tetraspanin family and is able to modulate cell adhesion and migration),TIMP3,the tissue inhibitor of metalloproteinases 3, and CSPG2. Dif- ferent studies also agree on the overexpression in grade II astrocytomas as compared with normal brain tissue of 3 genes involved in cell growth, namelyEGFR andPDGFRA (2 well-known growth factor receptors), and NTF3, which encodes a member (neurotrophin 3) of a family of proteins that controls neuronal survival and differentiation. Finally, 2 studies from the same group agree (but with very different ratios; Table 6) on the overexpression ofKCNN3, a neuronal potassium channel gene regulating neuronal excitability. It should be noted that our comparative analysis also evi- denced a relatively large number of genes reported with discordant results when grade II tumors were compared with normal tissue. As detailed in supplementary Table 2, the majority of these results concern the most recent study included in our analysis (19). This study provides very dif- ferent data as compared with the others, including a previous one from the same group (9).

An analysis of the gene expression in grade II as compared with high-grade tumors evidenced the largest number of differentially expressed genes (11 of the 18 genes listed in Table 6) split into cell migration (6 genes) and cell growth (5 genes) actors. Only 2 of these actors are over- expressed in grade II tumors. This is the case for the neural cell adhesion molecule 1 (NCAM1) and the matrix metal- loproteinase 16 (MMP16), adhesion- and invasion-related genes, respectively. The other genes (all downregulated in grade II astrocytomas) included 4 coding for ECM compo- nents such as fibronectin and 3 collagen subtypes (collagen IV>1 and>2 chain and collagen V>2 chain) and 5 growth factor-related genes, namely the EGFR, the vascular endo- thelial growth factor (VEGF), and the insulin-like growth factor-binding protein 2, 3, and 5 (IGFBP2, IGFBP3, and IGFBP5). It should be noted that all these changes in gene expression involving grade II and high-grade astrocytomas seem to concern grade II and grade IV astrocytomas more specifically (Table 6). In addition, our comparative analysis evidenced 5 genes (related to invasion or ECM) reported with discordant results when grade II tumors were compared with high-grade tumors (Supplementary Table 2).

Identification and Validation of the Most Characteristic Variations in Gene Expression Level in the Different Categories Analyzed

Figure 1 summarizes the most characteristic variations in gene expression extracted from Tables 5 and 6 and lists the 5 genes (TIMP4, C1NH, CHAD, TLE2, and IGFBP2) whose differential levels of expression enable grade I astrocytomas to be distinguished from all other categories of samples analyzed. AlthoughTIMP4, C1NH, CHAD,and TLE2 show their highest levels of expression in grade I astrocytomas,IGFBP2exhibits a more complex expression pattern. Although this latter gene is overexpressed in grade I astrocytomas as compared with normal tissue and grade II astrocytomas, it is downregulated in grade I as compared with high-grade astrocytomas. In addition,THBS4is over- expressed in grade I astrocytomas as compared with normal brain tissue and grade II astrocytomas (no comparison of the levels of expression of grade I and high-grade astrocy- tomas is available either in the 11 microarray studies or in the rest of the literature). To complete this characterization, we selected from Tables 5 and 6 the 2 most important variations in gene expression observed when normal tissue is compared with either grade I or grade II astrocytomas and when each of these low-grade categories is compared with high-grade astrocytomas. This selection evidenced other genes as being differentially expressed in pilocytic astrocytomas as compared with either normal brain tissue (i.e. CSPG2 and LPL) or high-grade astrocytomas (i.e.

APOD and EGFR). Two of these genes were similarly identified as being differentially expressed in grade II as- trocytomas as compared with normal brain tissue (CSPG2 andEGFR).EGFRoverexpression also paralleled the tran- sition from grade II to high-grade astrocytomas. Finally, IGFBP2overexpression was evidenced as one of the largest variations occurring in high-grade astrocytomas as com- pared with grade II astrocytomas.

Rorive et al J Neuropathol Exp Neurol Volume 65, Number 8, August 2006

Ó2006 American Association of Neuropathologists, Inc.

802

(10)

Each of the 10 genes shown in Figure 1 was submitted to real-time RT-PCR analyses carried out on normal tissue, grade I, grade II, and high-grade astrocytoma samples. The results shown in Table 7 agree with the majority of the variations in expression schematized in Figure 1 by exhibit- ing clear trends in terms of the median values (ratio Q3 or

e0.33), whereas certain statistical confirmations require large series of cases (such as in the case of TIMP4). These latter difficulties seem to be the result of the broad ranges of expression particularly observed in normal and grade I samples in accordance with the individual data provided by Gutmann et al (14). For 3 genes (TIMP4, CHAD, and TLE2), these variations could be related to differences in tissue location encountered in these 2 tissue categories, in particular between infratentorial and supratentorial areas (data not shown). It should be also noted that the different EGFR variations mentioned in Figure 1 were not all confirmed. Although we evidenced clear upregulation of this gene in high-grade astrocytomas when compared with grade I tumors, this upregulation remained true for only 2 of 5 grade IV cases when they were compared with grade II tumors (data not shown). Similarly, when compared with normal tissue, theEGFRupregulation in grade II tumors was not confirmed in the samples analyzed and neither was the upregulation of CSPG2 in grade I and grade II tumors as compared with normal tissue. The differentIGFBP2 varia- tions and the upregulation of TLE2 in grade I tumors as compared with all the other tissue categories (Fig. 1) were confirmed except in the comparison of grade I tumors and normal tissue. Finally, the upregulation ofC1NHexhibited a 4-fold difference (in terms of the median values) when compared with grade II tumors and was characterized by only a 2-fold difference in comparisons with normal tissue.

The upregulation of THBS4 in grade I tumors was characterized by at least a 3-fold difference in terms of the median values (but requires statistical confirmation on more data) as compared with each of the other categories of tissue, including the high-grade tumors (a comparison missing in the microarray studies). A similar profile was also observed in the case of APOD (i.e. an upregulation in grade I astrocytomas as compared with each of the other categories).

This contrasts with the clear downregulation of EGFR in these tumors as compared with any other tissue category.

Finally, when compared with normal tissue, upregulation of LPLwas observed in all the tumor categories (exhibiting at least a 3-fold difference in terms of the median values but requiring statistical confirmation on more data).

FIGURE 1. Characteristic variations in gene expression levels evidenced across the different tissue categories analyzed. This figure presents a group of 6 genes (TIMP4, CHAD, IGFBP2, TLE2, C1NH, andTHBS4) whose differential levels of expres- sion enable grade I astrocytomas to be distinguished from all the other categories of samples analyzed (except in the case of THBS4, in which data coming between grade I and high-grade astrocytomas are lacking). In addition, this figure also men- tions the 2 most important variations in gene expression observed when normal tissue is compared with either grade I or grade II astrocytomas, and when each of these low-grade categories is compared with high-grade astrocytomas. Next to each label, there is a white arrow pointing either upward or downward, which indicates the direction of the variation in expression (down-/upregulation) observed from one category of tissue to another (in the direction indicated by the black arrows). *, Apart from the upregulation evidenced when grade I astrocytomas are compared with high-grade astro- cytomas, downregulation ofIGFBP2when grade I astrocytomas are compared with all the other categories.

TABLE 7R Real-time Reverse TranscriptaseYPolymerase Chain Reaction Analysis on the 10 Genes in Figure 1

Gene K-W Normal (n = 9) Grade I (n = 6) Grade II (n = 5) Grade IV (n = 5)

TIMP4 NS 1.0 (0.3Y14.9) 7.1 (1.7Y18.6) 2.2 (0.7Y2.3)* 2.1 (1.7Y2.8)

C1NH * 1.0 (0.2Y2.1)* 2.0 (0.9Y5.1) 0.5 (0.2Y2.0)* 1.6 (0.7Y4.7)

CHAD 1.0 (0.0Y10.2) 3.3 (0.0Y11.7) 0.0 (0.0Y1.0) 0.0 (0.0Y0.1)*

THBS4 NS 1.0 (0.2Y2.4) 3.7 (0.6Y10.7) 0.3 (0.1Y1.9)* 1.1 (0.1Y4.2)

CSPG2 NS 1.0 (0.6Y3.7) 1.8 (0.3Y5.1) 1.7 (1.3Y5.0) 1.6 (0.6Y2.4)

IGFBP2 1.0 (0.2Y3.4) 2.3 (1.2Y17.4) 0.8 (0.3Y2.3) 16.0 (14.6Y40.8)*

EGFR 1.0 (0.1Y2.1)* 0.2 (0.2Y1.1) 1.3 (1.2Y1.9)† 1.7 (1.2Y27.1)†

LPL NS 1.0 (0.2Y11.2) 4.1 (1.6Y16.3) 3.2 (0.3Y8.8) 4.3 (3.1Y55.7)

APOD 1.0 (0.4Y4.5)* 2.9 (1.8Y5.0) 0.5 (0.3Y1.6)† 0.4 (0.0Y2.0)†

TLE2 1.0 (0.1Y38.1) 0.7 (0.3Y3.3) 0.1 (0.0Y0.4)* 0.2 (0.0Y0.4)*

The data are expressed as the median (minimumYmaximum range) of normalized mRNA levels.

*, pG0.05;†, pG0.01. In the table, the significant p values indicated were computed by means of the Mann-Whitney test between the grade I group and each other group with confirmation by means of the Dunn procedure (underlined stars), seeBMaterials and Methods.[

K-W, multiple-group Kruskal-Wallis test; NS, not significant (p90.05).

(11)

Copyright @ 2006 by the American Association of Neuropathologists, Inc. Unauthorized reproduction of this article is prohibited.

DISCUSSION

Pilocytic astrocytomas constitute a well-defined clinico- pathologic entity that should be considered as being sepa- rate from WHO grade II diffuse astrocytomas. Nevertheless, in certain cases, the difference between these 2 astrocytoma subtypes is complicated by the absence of any diagnostic markers and may appear to constitute a real challenge to pathologists. A more detailed knowledge of the molecular biology of these tumors is therefore fundamental for both the better characterization of the distinctive features of grade I and grade II astrocytomas and the improvement of man- agement and treatment of patients. This knowledge can be provided by means of microarray analyses of gene- expression profiling.

However, comparisons of data sets generated by different array approaches may be difficult. In particular, different sources of variation may affect the results of the gene-expression analysis. These variations may be the result of biases in the biologic samples, specific technical aspects (e.g. in the preparation and hybridization of probes from biologic samples), the chips used in the analysis, and the image acquisition and data analysis methodology (21, 25).

Moreover, the application of microarray technology to neuropathology is confronted by additional difficulties, as recently reviewed by Buesa et al (25). In particular, the choice of samples is of necessity limited by 1) the relative infrequency of certain astrocytoma subtypes (such as pilocytic astrocytomas as compared with widespread glio- blastomas), and 2) access to normal brain tissue, which led to a degree of heterogeneity in the normal reference samples observed across the 11 studies that we analyzed (Table 2).

As mentioned in Table 1, patients_ages and tumor locations, factors rarely mentioned in the different studies, would be controlled for an efficient comparison of gene-expression levels between astrocytic tumors (in particular, low-grade ones). These different factors, combined with the lack of exhaustive gene expression data sets, may explain the rel- atively small number of concordant results evidenced in our comparative analysis. In all events, we were able to extract concordant information for 29 different genes reported as differentially expressed in grade I and/or grade II astrocy- tomas in comparison with normal tissue and/or high-grade astrocytomas. Interestingly, a majority of these genes (17 of 29) encode proteins that are involved in interactions with the surrounding environment (ECM) and relate to processes involved in cell migration (i.e. adhesion and invasion). As detailed subsequently, these results may at least partly explain the behavioral differences between grade I astrocy- tomas, which are noninfiltrative, well-delineated tumors, and grade IIYIV astrocytomas, which are prone to diffuse paren- chyma infiltration.

Overexpression of Genes Contributing to Protease Pathway Inhibition in Pilocytic Astrocytomas

Figure 1 summarizes the most characteristic variations in gene expression extracted from Tables 5 and 6 and con- cern 10 genes. A majority of these variations were validated

a posteriori by means of real-time RT-PCR carried out on a completely independent series of cases. For the first time (at least to our knowledge), this figure evidences a characteristic gene profile for pilocytic astrocytomas in relation to all other (normal or diffuse astrocytoma) categories. Of the 6 genes in this profile, which were first reported by Rickman et al (13), 4 (TIMP4, C1NH, CHAD, and THBS4) are overexpressed in pilocytic astrocytomas and relate to the inhibitory processes of cell migration/invasion. These gene overexpressions were confirmed by real-time RT-PCR, at least in comparisons with grade II diffuse astrocytomas (in terms of the median values).

Full statistical validation requires an increased series of cases, in particular in the case ofTIMP4and CHADfor which our RT-PCR analysis revealed expression heterogeneity that seemed linked to tissue location (infratentorial versus supra- tentorial, data not shown). As illustrated in Figure 2,TIMP4 and C1NHcode for proteins that act as inhibitors of proteo- lytic enzymes such as matrix metalloproteinases (MMPs) and the serine proteases of the plasminogen activator/plasmin system. These enzymes are well known for playing a number of critical roles in the invasive properties of diffuse grade II to IV astrocytomas (26Y28). Although TIMP4 is able to directly inhibit MMP2, MMP9, and MT1-MMP (the membrane-type 1 matrix metalloproteinase, also referred as MMP14) and, indi- rectly, block the MT1-MMP-mediated activation of MMP2 and the MMP2-mediated activation of MMP9 (29Y31), C1NH recognizes and inhibits certain noncomplementary serine proteases such as plasmin (32), which is particularly able to activate certain MMPs, including MMP2 and MMP9 (Fig. 2).

Together with plasmin, these different MMPs are among the main agents involved in tumor astrocyte migration and invasion of healthy brain tissue (28, 33). As compared with diffuse astrocytomas (grades IIYIV), this hypothesized simul- taneous inhibition of both protease pathways might make the tumoral microenvironment less permissive to the migration of pilocytic astrocytoma cells into the brain parenchyma.

Overexpression of Genes Contributing to ECM Reinforcements in Pilocytic Astrocytomas

As illustrated in Figure 1,CHADandTHBS4are 2 other genes characteristically overexpressed in pilocytic astrocy- tomas. They code for ECM proteins (chondroadherin and thrombospondin-4) with adhesive properties and the ability to bind to collagens and other ECM proteins. As illustrated in Figure 2, these ECM partners are targeted by the protease pathways mentioned here (34Y36). More particularly, chon- droadherin is known to bind to collagen subtypes collagen II and VI (37Y39). Thrombospondin-4 appears to be a prefer- ential adhesive substrate for neurons and interacts strongly with both collagenous (i.e. collagen IYIV) and noncollagenous ECM proteins such as laminins, fibronectin, and tenascins (40, 41). Figure 2 shows that in addition to the inhibition of the protease pathways resulting from the upregulation of TIMP4 and C1NH, CHAD and THBS4 9 upregulation in pilocytic astrocytomas might cause the reinforcement of the ECM lattice in the tumoral microenvironment, with both events contributing to inhibition of tumor cell migration.

Involvement of IGFBP2 and TLE2 in the Biology of Pilocytic AstrocytomasIGFBP2is another gene participating

Rorive et al J Neuropathol Exp Neurol Volume 65, Number 8, August 2006

Ó2006 American Association of Neuropathologists, Inc.

804

(12)

in the characteristic gene profile of pilocytic astrocytomas (Fig. 1). It is one of the members of theIGFBPgene family, which codes for multifunctional proteins (i.e. the insulin-like growth factor binding proteins) (42). IGFBP2 is produced in various regions of the central nervous system and constitutes one of the major IGFBPs in the cerebrospinal fluid (43). By sequestering IGFs (Fig. 2), IGFBP2 inhibits IGF-stimulated events such as mitogenesis and cell migration (42); these events can also be blocked in the presence of serine protease inhibitors (44). As illustrated in Figure 2, the joint over- expression of IGFBP2 and C1NH (coding for a serine protease inhibitor) observed in pilocytic astrocytomas might suggest the inhibition of the plasmin-induced IGFBP2 proteolysis and the resulting preservation of the control of IGF-induced cell migration and proliferation. Although IGFBP2 seems able to affect cell adhesion and migration directly (and negatively), its functional properties need further investigation (42). However, all the inhibitory effects of IGFBP2 on cell proliferation and migration may be frustrated by its positive action on both MMP2 gene transcription (Fig. 2) and on cell growth (42, 45). A parallel can be drawn between these data and the overexpression of the IGFBP2 gene evidenced in high-grade as opposed to both grade I and grade II astrocytomas (Table 7). The positive correlation between IGFBP2 (gene and protein) expression and tumor grade was previously shown in the case of diffuse (grades IIYIV) astrocytomas (46).

Relatively little is known about TLE2, the sixth gene characteristic of pilocytic astrocytomas (Fig. 1). This tran- scriptional corepressor, which is a member of the Notch signaling pathway, is coexpressed and interacts with mam- malian HES (Hairy/Enhancer of Split) proteins in the developing mammalian nervous system and may perform

transcriptional repression functions during cell differentia- tion (47). A recent study suggests thatTLE2is also involved in the genesis of aggressive meningiomas (48).

Overexpression of Genes Involved in Lipid Metabolism in Pilocytic Astrocytomas

Our analysis identified 2 genes encoding proteins involved in lipid metabolism as being overexpressed in pilocytic astrocytomas (LPL and APOD). LPL is known to hydrolyze triglyceride molecules found in lipoprotein par- ticles. Recent studies suggest LPL involvement in Alzheimer disease, neuronal differentiation, and the pathophysiological response of the brain to ischemia (48Y51). APOD is a lipo- calin present in some scattered neurons, oligodendrocytes, and astrocytes and, in particular, in reactive astrocytes, in which an increased synthesis of APOD has been observed (52, 53). Hunter et al validated this marker (at RNA and protein levels) as the best potential marker for distinguishing between pilocytic and anaplastic astrocytomas, and they suggest that APOD might play a role in either decreased proliferation or cyst formation in pilocytic astrocytomas (12). More recently, the same research group extended these results by showing that this APOD upregulation in pilocytic astrocytomas is maintained when this category of tumor is compared with normal brain tissues and glioblastomas (54), which we also showed by means of real-time RT-PCR an- alyses (Table 7).

The Growth- and Migratory-Related

Status of Epidermal Growth Factor Receptor in Astrocytomas

EGFR is a well-known transmembrane receptor that transduces a mitotic signal after binding to extracellular ligands FIGURE 2. Illustration of the inter-

actions between the proteins encoded by TIMP4, C1NH, and IGFBP2 and the different protease pathways that target some of the ECM proteins (collagens, fibronectin, laminins, and tenascins) acting as the partners of thrombospondin-4 (THBS4) and chondroadherin (CHAD) in the ECM lattice. The standard arrows repre- sent positive actions (i.e. production or stimulation), the links terminating in vertical lines the inhibiting ac- tions, and the hatched arrows the proteolytic actions. See the text for more detail.

Figure

Updating...

Références

Updating...

Sujets connexes :