Sage Journals: Discover world-class research

Abstract

The pathophysiologic similarities of many human and canine cancers support the role of the domestic dog as a model for brain tumor research. Here we report the construction of a custom canine brain-specific cDNA microarray and the analysis of gene expression patterns of several different types of canine brain tumor The microarray contained 4000 clones from a canine brain specific cDNA library including 2161 clones that matched known genes or expressed sequence tags (ESTs) and 25 cancer-related genes. Our study included 16 brain tumors (seven meningiomas, five glial tumors, two ependymomas, and two choroid plexus papillomas) from a variety of different dog breeds. We identified several genes previously found to be differentially expressed in human brain tumors. This suggests that human and canine brain tumors share a common pathogenesis. In addition, we also found differentially expressed genes unique to either meningiomas or the glial tumors. This report represents the first global gene expression analysis of different types of canine brain tumors by cDNA microarrays and might aid in the identification of potential candidate genes involved in tumor formation and progression.

Keywords

cDNA microarray canine choroid plexus tumor ependymoma glial tumor meningioma

Dogs share significant anatomical and physiological similarities with humans and are excellent models for human diseases. Unlike most experimental models, cancers in the canine population are spontaneous diseases, making them a valuable resource in the identification of genes associated with oncogenesis.^13, ¹⁸ Sequencing of the canine genome has generated great interest in characterizing similarities and differences between human and canine cancers.¹⁴ Thus, the canine genome project aids in the development and characterization of companion animal models for cancer research and in the design of new cancer therapeutic drugs.^6, ⁷

Primary tumors of the central nervous system are common in dogs and are histologically similar to those found in humans. Brain tumors occur in all dog breeds, although some breeds have a predisposition for certain tumor types. Most tumors occur in dogs over the age of 6 years.²⁸ Meningiomas are the most common brain tumors found in dogs, with the majority being intracranial and arising from arachnoid cells.^12, ²⁸ Histologically, these tumors are diverse due to their ability to undergo both mesenchymal and epithelial differentiation. In humans, 15 distinct meningeal variants have been classified by the World Health Organization (WHO) with three grades of malignancy.²² In domestic animals, nine histological patterns have been described with seven of them similar to the variants included in the human meningioma classification system.¹⁵ Of these, the meningothelial and transitional variants are the most commonly described in dogs.²³

A recent study showed that canine and human meningiomas have similar immunophenotypes and may therefore share similar genetic alterations leading to tumorigenesis.¹ Microarray analysis of human meningiomas has identified several genes previously known to have altered gene expressions (e.g., growth hormone receptor, insulin-like growth factor binding protein 7, endothelin receptor A, insulin-like growth factor 2) while also identifying novel differentially expressed genes.³¹ In another study, several genes involved in signaling pathways were found to be differentially expressed and could be linked to aberrant phenotypes exhibited by meningiomas.⁵

Glial tumors are the most common brain tumors in humans and the second most frequently reported brain tumor in dogs.²⁸ A recent study revealed that human and canine astrocytomas not only shared morphologic similarities but they both also exhibited overexpression of the genes encoding p53 and the epidermal growth factor receptor (EGFR).²⁶ Glioblastoma multiforme (GBM) is one of the most frequent types of astrocytomas seen in humans, and accounts for about 40% of all astrocytomas in dogs.²⁸ Canine GBMs share many microscopic and immunoreactivity features found in human GBMs.¹⁹ A recent study of human glial tumors identified a number of differentially expressed genes belonging either to growth factor-related genes or structural/extracellular matrix-related genes. In addition, EGFR and the α4 chain of laminin genes were found to be consistently overexpressed in low- and high-grade gliomas.²¹

In dogs, choroid plexus tumors (CPT) are diagnosed almost as frequently as gliomas, while ependymomas are more rare.²⁸ Choroid plexus papilloma (CPP) and choroid plexus carcinoma (CPC), the two variants of the tumor, are rare in humans and mostly occur in children.¹⁷ Pathologic and immunohistochemical studies revealed a complex antigenic profile in canine CPTs comparable with that of human CPTs.²

The majority of domestic dogs are pets that are exposed to the same environmental influences as humans. Naturally occurring cancers in dogs thus provide excellent animal models to further our understanding of tumorigenesis and epidemiology and expand the possibilities of developing new therapeutic treatments. While many studies have demonstrated morphological similarities between canine brain tumors and their human counterparts, very little is known about their comparability at the molecular level. The aim of this study was to apply cDNA microarrays to analyze gene-expression patterns in several different types of canine brain tumors. This analysis represents a first step toward gaining a greater insight into the genetic alterations present in canine brain tumors and identifying novel genes specifically involved in tumorigenesis.

Materials and Methods

Tissue samples

Brain tumor samples were obtained from the veterinary schools at North Carolina State University (NCSU; dogs 1, 2, 3, 4, 5, 8, 15, 16) and University of California at Davis (dogs 6, 7, 9, 10, 11, 12, 13, 14). Tissue samples were flash frozen immediately upon surgical removal or at necropsy. Diagnosis was confirmed for each case by histological examination. A total of 16 tumor samples were analyzed and included seven meningiomas, two choroid plexus, two ependymomas, one oligodendroglioma, one glioblastoma, one astrocytoma, and two oligoastrocytoma (Table 1). The reference RNA sample used in this study was generated from pooling RNA derived from whole-brain sections of two mixed breed dogs with no signs of any neurologic disorder. Because the reference sample did not contain meninges, we also included normal meninges samples taken from two individual dogs that did not show any clinical neurologic defects.

Table 1.

Summary of the 16 brain tumor samples used to generate cDNA array and Q-PCR data in this study.


Sample	Breed	Sex	Age∗	Tumor Type	Location

1	Golden Retriever	Female	13	Meningioma, low grade	Occipital lobe
2	Mixed breed	Male	7	Meningioma, high grade	Frontal lobe
3	Boxer	Female	12	Meningioma, low grade	Cerebellum
4	Shepherd mix	Female	10	Meningioma, low grade	Brainstem
5	Rhodesian Ridgeback	Male	6	Meningioma, low grade	Left frontoparietal lobe
6	Golden Retriever	Male	8	Transitional meningioma	Frontal lobe to olfactory bulb
7	West Highland Terrier	Female	9	Psammomatous type meningioma	Cerebrum
8	Mixed breed	Female	3	Papilloma choroid plexus, low grade	3rd ventricle
9	English Bulldog	Male	12	Glioblastoma multiforme (grade IV)	Cerebrum
10	Labrador Retriever	Female	10	Oligodendroglioma	Frontal lobe to olfactory bulb
11	Boxer	Female	9	Multifocal oligoastrocytoma	Cerebrum
12	Mixed breed	Female	12	Ependymoma	In 4th ventricle
13	Dalmatian	Female	12	Gemistocytic (diffuse)	Cerebellum
				Astrocytoma (grade II)
14	Mixed breed	Female	10	Mixed oligoastrocytoma	Cerebrum, frontal lobe to optic chiasm
15	Terrier mix	Female	10	Papilloma choroid plexus, low grade	3rd ventricle extending to thalamus
16	Mixed breed	Female	1	Ependymoma, high grade	Thalamus

∗ Age of dog at diagnosis.

Microarray preparation

Custom cDNA microarrays were constructed from a canine brain cDNA library that contained approximately 4,000 clones. Of those 4,000 clones, sequence analysis indicated that 2,161 matched to known genes or ESTs and included matches to the mitochondrial genome, unannotated human genomic DNA clones, and novel ESTs.²⁵ In addition, 25 cancer-related genes previously identified from canine bacterial artificial chromosome genomic clones were polymerase chain reaction (PCR) amplified from canine cDNA and added to the arrays.²⁹ Library clones were grown overnight and PCR amplification of plasmid inserts was performed with M13 forward and reverse primers. PCR products were precipitated and resuspended in water prior to the addition of an equal volume of di-methyl sulfoxide (DMSO). DNA solutions in 50% DMSO were spotted onto GAPII (Corning) slides by an Affymetrix 417 arrayer. Microarray slides were ultraviolet cross-linked and stored in a desiccator at room temperature until further use.

Total RNA preparations

Total RNA was extracted using Trizol as outlined by the manufacturer (Invitrogen) and analyzed for integrity by gel electrophoresis. Total RNA was DNase treated and repurified using the RNeasy Mini kit (Qiagen). Five hundred nanograms of total DNase-treated RNA was used for RNA amplification (aRNA) with a Low RNA Input Fluorescent Linear Amplification kit (Agilent Technologies).

Fluorescence labeling of probe

Three micrograms of amplified RNA was used for Cy5/Cy3 labeling with the SuperScript Indirect cDNA labeling system (Invitrogen). For each sample, dye swapping was performed such that each sample was labeled with both Cy5 and Cy3 and used to generate duplicate data sets. Each labeled tumor sample was combined with the normal, differentially labeled brain reference sample and lyophilized.

Hybridization

Each array slide was prehybridized in 5× sodium chloride sodium citrate (SSC), 0.1% sodium dodecyl sulfate (SDS), and 1% bovine serum albumin (BSA) for 45 minutes at 42°C. After prehybridization, the slides were rinsed briefly twice in water and then once in isopropanol, prior to being dried by centrifugation at 500 × g for 5 minutes. The lyophilized labeled probe was resuspended in hybridization buffer containing 5× SSC, 5× Denhardt's solution, 50% formamide, 0.5% SDS, 200 ng poly A, and 200 ng of calf thymus DNA. After denaturation at 95°C for 3 minutes, the probe was applied directly to the prehybridized array slide, covered with a HybriSlip (Schleicher & Schuell) cover slip, and placed in a hybridization cassette (Corning). Arrays were incubated for 16 hours at 42°C and then processed through three consecutive washes of decreasing ionic strength and dried by centrifugation for 5 minutes at 500 × g.

Data analysis

The arrays were scanned at a 10-μm resolution by an Axon GenePix scanner and image analysis was performed using the GenePix Pro software (Axon Instruments, Inc). Raw intensity values minus the background for each array were scaled, log₂ transformed, and the average for each gene across all the arrays was calculated. Genes containing expression values at or below background were removed from the data set (this process eliminated 725 genes).

The data set was analyzed using a two-step mixed model analysis of variance in SAS version 7.1. To adjust for overall array and dye effects, a first ANOVA model was used to express the relative fluorescence intensity of each gene as the fold difference in expression level relative to the sample mean. Expression differences were then evaluated for each gene separately according to the model

{relative fluorescence intensity}_{i j k} = μ + T_{i} + D_{j} + A_{k} + \in_{i j k,}

where the fixed effects are represented by T_i for tissue class (reference, meningioma, other tumor, or normal meninges) and D_j for the jth dye (Cy3 or Cy5). μ is the grand mean and A_k and ∊_ijk are random effects of array and residual error assumed to be normally distributed with a mean of zero. The array effect controls for correlation between the two measurements on each array. The magnitude and significance of expression differences between particular tissue classes were computed with the DIFFS option in PROC MIXED. Supplementary data comprising the raw intensity values, results of the SAS analysis, tables of all significant differentially expressed genes for the tumor samples at the Bonferroni level P < 0.00001, and a MIAME-compliant description of the experiment are provided at http://www.cvm.ncsu.edu/mbs/breen_matthew.htm.

The false discovery rate was controlled using Storey's q-value procedure implemented in Bioconductor. For the meningioma versus normal meninges contrast, q < 0.05 corresponds to P < 0.008. For the tumor versus reference contrast, q < 0.05 corresponds to P < 0.016. Qualitatively similar results were obtained by comparing ratios of each sample to the reference but with a much lower significance rate due to the reduced number of degrees of freedom and less efficient control of residual error. Hierarchical clustering was based on the average of the relative fluorescence intensity values for each Cy5 and Cy3 sample and calculated using Ward's average linkage method.

Quantitative real-time PCR

Half a microgram of aRNA was reverse transcribed with random primers and SuperScript III reverse transcriptase (Invitrogen) according to manufacturer's instructions and synthesized cDNA was diluted 1 : 5. Quantitative real-time PCR was performed on an i-Cycler iQ system (BioRad). Samples were run in duplicates with 2 μl of template in a 25-μl reaction using the Platinum Quantitative PCR SuperMix-UDG (Invitrogen) under the following PCR parameters: 95°C for 2 minutes followed by 50 cycles of 95°C for 15 seconds, primer-specific annealing temperature for 30 seconds, 72°C for 30 seconds. Melt curve analysis was performed by increasing the temperature in 0.5°C increments starting at 55°C for 80 cycles of 10 seconds each. Sequences of primers, annealing temperature, and size of amplified product are shown in Table 2. From the amplification curve, a cycle threshold (C _t) number was calculated using the iCycler iQ software (BioRad) and the mean C _t value from the duplicate reactions was used for fold-change analysis. The two housekeeping genes beta2-microglobulin gene (B2M) and hypoxanthine-phosphoribosyl-transferase (HPRT) were used for normalization of samples by subtracting the C _t value of the housekeeping gene from the C _t value of the candidate gene. The 2^−ΔΔCt method was used to determine relative gene expression for each gene.²⁰

Table 2.

Primers used for quantitative PCR of selected canine genes, which were shown by cDNA array analysis to be either up- or down regulated in canine brain tumors.


Gene	Forward and Reverse Primers	Temp	Size (bp)

B2M	5′-CGG AAA GGA GAT GAA AGC AG-3′	60°C	125
	5′-CAT GCT TTA CAC GGC AGC TA-3′
CREG	5′-GCG TGC CCT ACT TCT ACC TG-3′	55°C	124
	5′-GGG TCA AAT CCC TTC TTC CT-3′
LAMP1	5′-CTT TCC AGG TGC AAG GTG AC-3′	60°C	129
	5′-GGT AGG CGA TGA GGA CGA TA-3′
HPRT	5′-CAC TGG GAA AAC AAT GCA GA-3′	60°C	102
	5′-ACA CTT CGA GGG GTC CTT TT-3′
YWHAE	5′-CAG CAG CAT TGA ACA GAA AGA-3′	60°C	117
	5′-TCC AGA ATG TCA CAA CAG ATT AAC-3′
PRDX1	5′-TAT CAG ACC CCA AGC GTA CC-3′	60°C	118
	5′-ACA GTG ATC TGC CGA AGG AT-3′
TCEB1L	5′-TGC AAG ACT GTT GCC AAT ATG-3′	60°C	120
	5′-CCA CTG GTT CTC TTT GCG TA-3′
NCAM	5′-CAG CGT TGG AGA ATC CAA AT-3′	60°C	118
	5′-GGA GAT CCG TTG TTG GTT TG-3′
RAB9	5′-CGA TTC AGA AGC CTG AGG AC-3′	60°C	146
	5′-CTT TCG GGC TCT TTC ACA TC-3′
OLFM1	5′-CTC CTC AGC CTC CTC TTC CT-3′	60°C	134
	5′-GCG ACC ACT GTG CAG ATA CA-3′
NCOA4	5′-GGT GTG GCT GTA TGA ACA GGT-3′	60°C	133
	5′-GTT TTG GGT GCA CTC CAG TT-3′
SNAP25	5′-TCA TGG AGA AGG CTG ATT CC-3′	60°C	126
	5′-GCT ATC TTG CCC AAC AGC AT-3′
STMN2	5′-AGC TTC CCC GAA GAA GAA AG-3′	60°	112
	5′-CTC TGC CAA TTG TTT CAG CA-3′
PLP	5′-GTG GCA GCA ATC ATG AAG GT-3′	60°C	87
	5′-CCA AAT GAC GTT CCA CCT GT-3′
NDRG4	5′-GCT TCA ACA GGA CAC CTC CA-3′	60°C	114
	5′-ACC GCG TCA CTC ACC AGT-3′
GNB2L1	5′-CTG AGT GTG GCC TTC TCC TC-3′	60°C	126
	5′-CCA CTC TGA ATG GCT CTC GT-3′
APOD	5′-CCC AGT GAG CTT TGA GAA GG-3′	60°C	156
	5′-TTG GCA GGC TCT GTA AGG TT-3′

Results

The experimental design consisted of 18 duplicated experimental samples (seven meningiomas, nine other brain tumors, and two normal meninges, each from a different dog) against a whole-brain reference sample. Rather than simply comparing ratios, we used the statistically more powerful procedure of mixed model analysis of variance of relative fluorescence intensities to search for genes that are differentially expressed between each tissue class (see supplementary data for details).^10, ³² The results are presented graphically in Fig. 1 in the form of volcano plots of significance against fold change. Significance on the y-axis is on a negative log P scale such that increasing significance is toward the top, while fold change on the x-axis is shown on a log base 2 scale such that ±1 represents 2-fold increase and decrease, respectively. Experiment-wide significance was established by converting P-values into q-values and adopting a 5% false-discovery-rate cutoff.²⁷

Fig. 1.

Volcano plots of differentially expressed genes in canine brain tumors. Calculated differences of gene expression are presented by plotting the estimate values on the x-axis over the significance, negative log 10 of P-value (nlp). Dotted line indicates cutoff value of significance (Bonferroni P < 0.00001). Fig. 1A. Meningiomas minus normal meninges. Fig. 1B. Neuroepithelial tumors minus normal brain reference samples.

To validate our cDNA array, we conducted a self-self hybridization (reference versus reference sample and normal meninges versus normal meninges) experiment and, as expected, we did not detect any significant differentially expressed genes (data not shown). Overall, 327 genes were found to be differentially expressed between meningiomas and normal meninges, while 564 genes differentiate all other tumors as a class from the reference sample. Among the 50 genes significant at the Bonferroni level (P < 0.00001) for the comparison of meningiomas against normal meninges, approximately 30% represent the mitochondrial genome, 10% ribosomal proteins, 20% unknown sequences, and 40% correspond to predicted genes (see Table 3 and Fig. 1A—see supplementary data for comprehensive details). Among the 404 most highly significant genes for the other tumor samples, 30% correspond to unknown genes and only 20% to known genes (see Table 4 and Fig. 1B—see supplementary data for comprehensive details). No attempt was made to distinguish between different neuroepithelial tumors (astrocytoma, oligodendroglioma, oligoastrocytoma, ependymoma, and choroids plexus tumor), as there were too few samples of each class. Some of the more notable gene expression differences are discussed below.

Table 3.

List of differentially expressed genes with a fold change of three or greater in canine meningioma samples (see SOM Table 4 for a more complete list of the data).


Gene Name	Fold Change∗	P-Value	Gene Acc No.	Gene ID

PODXL	−3	8.96E−05	BQ091456	5420
NCAM	−3	4.07E−06	BQ091464	4684
CREG	−3	3.09E−05	BQ091879	8804
LAMP1	+18	7.92E−06	BI395313	3916
RAB9	+15	1.09E−07	CFRAB9	403947
UBA52	+13	3.69E−05	CFA388512	7311
MT1	+13	4.84E−05	D84397	403800
RPUB1	+11	1.18E−08	BQ839561	281370
TCEB1L	+8	3.95E−05	BI395330	6500
RPL15	+8	4.90E−05	NM_002948	6138
PRDX1	+7	1.85E−05	NM_002574	5052
RPL9	+7	6.54E−06	BQ091592	6133
YWHAE	+6	3.58E−06	BQ091844	7531
RPL9	+6	3.42E−05	BQ091592	6133
RPL19	+5	5.47E−05	CFA388522	403682
MPEG1	+5	2.26E−05	BI395335	219972
EEF1A1	+5	1.00E−05	AY195837	403506
RPS25	+4	1.13E−05	BI817150	6230

∗ − = downregulated genes in meningiomas compared to normal meninges; + = upregulated genes in meningiomas compared to normal meninges.

Table 4.

List of differentially expressed genes with a fold change of three or greater in nonmeningioma canine brain tumor samples.


Gene Name	Fold Change∗	P-Value	Gene Acc No.	Gene ID

Golli-MBP	−18	2.30E−11	BI794829	4155
PLP1	−11	2.65E−13	BI395987	5354
NCOA4	−8	3.42E−14	BI395858	8031
OLFM1	−7	2.76E−06	BQ091496	10439
STMN2	−6	1.30E−05	BQ091692	11075
KNS2	−5	7.23E−05	BI397412	3831
NCOA4	−5	4.97E−05	BI395858	8031
SNAP25	−5	6.30E−08	BQ091870	6616
NDRG4	−4	9.59E−08	BI395331	65009
MBP	−4	2.71E−13	BI397414	4155
APOD	−4	3.76E−06	BI395384	347
SNAP25	−4	8.97E−05	BQ091870	6616
S100B	−4	9.29E−06	BI405287	6285
PTGDS	−3	4.28E−09	AB026988	5730
PPP3CA	−3	7.46E−12	BI395264	5530
GNB2L1	+3	9.86E−09	BI817105	10399
TRHDE	+3	3.09E−08	BQ091424	29953
RPL18a	+3	6.82E−05	BQ091679	6142
RPL10	+3	3.21E−07	BM735680	6134
RPS14	+3	2.61E−06	AJ388514	403690
PRL7a	+3	1.97E−07	AJ388527	403678

∗ − = downregulated genes in brain tumors compared to normal brain reference; + = upregulated genes in brain tumors compared to normal brain reference.

Hierarchical cluster analysis was used to look for possible differentiation between the meningioma samples. Among the significantly differentially expressed genes, a number appear to define two distinct types of meningiomas, as shown in Fig. 2. Group one contains samples 1, 5, 6, and 7 while group two includes samples 2, 3, and 4. In addition, this analysis highlighted a tumor sample that was first believed to be a meningioma but clustered more closely with the normal meninges. Upon histological examination, it was found to be a granular cell tumor (data not shown).

Fig. 2.

Hierarchical cluster analysis of meningioma samples (M1–7), granular cell tumor (GCT), and normal meninges (N1–2). The red and green bars indicate genes demonstrating increased and decreased expression, respectively, based on the average relative fluorescence intensity values for each Cy5 and Cy3 sample. Only the top 100 genes based on nlp values are shown in the cluster.

To verify the results of our array analysis, we used quantitative real-time PCR on several differentially expressed genes. To compensate for the initial aRNA and cDNA concentrations, we normalized each sample to two housekeeping genes, B2M and HPRT. B2M is a medium-expressed gene and HPRT is considered a low-expressed gene and, because the expression levels of many of our genes in normal meninges are not known, we used both housekeeping genes for our analysis. While the number of fold change differed depending on which housekeeping gene was used, the overall expression patterns of each gene were similar in both. Expression levels of the neural cell adhesion molecule (NCAM), cellular repressor of E1A (CREG), lysosomal-associated membrane protein 1 (LAMP1), rab9 ras-related GTP-binding protein (RAB9), transcription elongation factor B (TCEB1L), peroxiredoxin 1 (PRDX1), and tyrosine 3-monooxygenase (YWHAE) were analyzed in all seven meningioma samples. The cDNAs of both normal meninges samples were pooled together into one sample for the gene expression comparison with the meningioma samples. For the downregulated genes, NCAM expression was underexpressed in five of the meningioma samples while CREG was down in all of the samples. LAMP1, PRDX1, TCEB1L, and YWHAE were upregulated in all meningioma samples and RAB9 in six of the samples.

For the neuroepithelial brain tumor samples, we analyzed the expression levels of proteolipid protein (PLP1), neuroblastoma protein (OLFM1), stathmin-like 2 (STMN2), nuclear receptor coactivator 4 (NCOA4), synaptosomal-associated protein (SNAP25), NDRG family member 4 (NDRG4), apo-lipoprotein D (APOD), and guanine nucleotide binding protein beta polypeptide 2-like 1 (GNB2L1). All samples were compared with the normal brain reference and each sample showed downregulation for PLP1, OLFM1, STMN2, SNAP25, and APOD and upregulation for GNB2L1. Five samples were underexpressed for NDRG4 (oligodendroglioma, oligoastrocytoma, CPT, ependymoma, and GBM) and NCOA4 (oligoastrocytoma, two CPT, ependymoma, and GBM).

Discussion

Gene expression analyses of human brain tumors are beginning to identify genes and pathways that may be involved in tumorigenesis. However, human tumor samples are often heterogeneous and identifying genes that are responsible for tumor initiation versus tumor growth and progression can be troublesome. The domestic dog offers an excellent model system for human cancer because the same types of spontaneous cancers affect this species, and through extensive inbreeding, selected dog breeds represent a more genetically homogeneous population. The restricted level of genetic heterogeneity in a variety of dog breeds should allow us to more easily identify genes directly involved in tumorigenesis as well as to identify genes previously not associated with cancer.

In addition to being pathophysiologically similar, brain tumors in both human and dog have been shown to share similar immunohistologic phenotypes.^1, ^19, ²⁶ However, these studies only included a few protein markers and very little is known about the molecular changes that occur in the canine tumors. To address this issue, we constructed a canine cDNA microarray that contained genes specific to the brain. Our samples included a variety of different types of brain tumor across a number of dog breeds. In a recent study, we compared a number of different brain tissues across several dog breeds and showed that differences in gene expression levels were not affected by the breed but rather by the type of brain tissue.¹¹

In our gene expression analysis of canine meningiomas, we found several genes that had similar expression patterns to genes identified previously in human meningiomas. For example, we detected upregulation of ribosomal proteins in meningiomas versus normal meninges. This is consistent with data previously reported for human meningiomas.⁵ We also identified several downregulated genes that have not previously been reported in human meningiomas. The cellular repressor of E1A gene, CREG, is a secreted protein and plays a role in cell proliferation, differentiation, and development.³⁰ A recent study showed that CREG is able to bind to the tumor suppressor mannose-6-phosphate/insulin-like growth factor II receptor and inhibit cell growth.⁴ In our canine meningioma samples, CREG was downregulated by threefold in the tumors compared with normal meninges.

These results were confirmed by quantitative real-time PCR and showed that CREG was downregulated in all samples. The level of differences in expression was quit variable among the samples and overall greater than in our array analysis. However, this was expected because quantitative real-time PCR is a more sensitive method for detecting differences in gene expression and our array data were based on a mean value of expression across all samples. In addition, we confirmed the expression patterns of a number of other differentially expressed genes identified by our array analysis.

Granular cell tumors (GCT) in the central nervous system are common in rats but are rare in dogs. In the rat, histological evidence suggests a meningeal origin for these tumors, while in the dog, the histogenesis remains unknown.^16, ³³ The most common site for human GCTs is the neurohypophysis, which are postulated to arise from pituicytes, while in dogs, they are usually intradural. A recent study of two cases of GCTs found that the origin of the granular cells might have been from the meninges.⁸ Our cluster analysis shows that the GCT contains many similarities with normal meninges but also revealed distinct differences when compared with meningioma samples. In addition, we also found two subgroups within our meningioma samples indicating that cDNA expression analysis could be useful for classification of meningioma variants.

In the neuroepithelial tumor samples, the total number of up- or downregulated genes was comparable. However, downregulated genes showed higher fold changes for differentially expressed genes. Upregulated genes included ribosomal proteins, cell signaling proteins, and extracellular proteins. Apolipoprotein D (ApoD) has been found highly expressed in human pilocytic astrocytomas, but not in infiltrating astrocytomas, and has been associated with a decrease in cell proliferation in some cell types.⁹ This previous study suggests that upregulation of ApoD might be responsible for the slow proliferation characteristics found in pilocytic astrocytomas. Interestingly, ApoD showed a fourfold downregulation in our tumor samples and these results were confirmed by quantitative real-time PCR. This might suggest that canine neuroepithelial tumors can be more proliferative than human tumors.

NDRG2 and NDRG4 belong to the N-myc down-regulated gene family and are highly expressed in the brain.²⁴ NDRG2 has been found to inhibit cell proliferation and has been shown to be downregulated in high-grade glioblastomas but not in low-grade gliomas.³ In our study, NDRG2 showed a twofold decrease in expression, but the differences in expression were not found to be statistically significant. On the other hand, NDRG4 expression was statistically significant and was found downregulated in the tumor samples. Quantitative real-time PCR confirmed these results in five of the brain tumors, which might indicate that NDRG4 could play a role in a number of different types of brain tumors while NDRG2 is specific to high-grade gliomas.

Of the known genes identified in this study, many are involved in signaling pathways, transcription activity, cytoskeleton rearrangements, and cell transformation. We did not find any of the 25 cancer-related genes to be differentially expressed in our analysis. However, many tumor-suppressor genes are expressed at low levels in normal tissue and small changes in expression might not be discovered when applying a high stringency on the data analysis. Oncogenes may contain mutations that lead to abnormalities in the cell, but their expression pattern might not be altered. Therefore, to detect alterations in these cancer-related genes in canine brain tumors, a more direct screening method is needed.

In summary, we found differentially expressed genes in canine brain tumors that have similar expression patterns to those described in human brain tumors. We also identified genes that have not been found in human brain tumor studies. Some of the cDNAs that indicated differential expression on our custom array contained ESTs and hypothetical proteins. Identification of those genes could lead to the discovery of yet unknown cancer-related genes. With this analysis, we have shown that canine brain tumors undergo similar genetic changes to those of human brain tumors on a molecular level. A larger, more in-depth comparative study on gene expression patterns between human and canine brain tumors is needed to filter out genes that play a role in tumorigenesis. By extending our study, we hope to be able to identify genes directly involved in tumor formation and progression, ultimately leading to new drug targets and therapies.

Footnotes

Acknowledgements

This work was supported by the American Kennel Club Canine Health Foundation (Award 249 to MB, NO, GG), a NCSU CVM First Award (MB, NO, GG), and by the Paul C. and Borghild T. Petersen Foundation (PJD).

References

1 Barnhart

Wojcieszyn

Storts

: Immunohistochemical staining patterns of canine meningiomas and correlation with published immunophenotypes. Vet Pathol 39:311–321, 2002

2 Cantile

Campani

Menicagli

Arispici

: Pathological and immunohistochemical studies of choroid plexus carcinoma of the dog. J Comp Pathol 126:183–193, 2002

3 Deng

Yao

Chau

Peng

Liu

Wang

Han

Nie

Kung

Leung

Lin

: N-Myc downstream-regulated gene 2 (NDRG2) inhibits glioblastoma cell proliferation. Int J Cancer 106:342–347, 2003

4 Di Bacco

Gill

: The secreted glycoprotein CREG inhibits cell growth dependent on the mannose-6-phos-phate/insulin-like growth factor II receptor. Oncogene 22:5436–5445, 2003

5 Fathallah-Shaykh

Zhao

Engelhard

Cerullo

Lichtor

Byrne

Munoz

Von Roenn

Rosseau

Glick

Sherman

Farooq

: Genomic expression discovery predicts pathways and opposing functions behind phenotypes. J Biol Chem 278:23830–23833, 2003

6 Guyon

Lorentzen

Hitte

Kim

Cadieu

Parker

Quignon

Lowe

Renier

Gelfenbeyn

Vignaux

DeFrance

Gloux

Mahairas

Andre

Galibert

Ostrander

: A 1-Mb resolution radiation hybrid map of the canine genome. Proc Natl Acad Sci USA 100:5296–5301, 2003

7 Hansen

Khanna

: Spontaneous and genetically engineered animal models; use in preclinical cancer drug development. Eur J Cancer 40:858–880, 2004

8 Higgins

LeCouteur

Vernau

Sturges

Obradovich

Bollen

: Granular cell tumor of the canine central nervous system: two cases. Vet Pathol 38:620–627, 2001

9 Hunter

Young

Olson

Brat

Bowers

Wilcox

Jaye

Mendrinos

Neish

: Differential expression between pilocytic and anaplastic astrocytomas: identification of apolipoprotein D as a marker for low-grade, non-infiltrating primary CNS neoplasms. J Neuropathol Exp Neurol 61:275–281, 2002

10.

10 Jin

Riley

Wolfinger

White

Passador-Gurgel

Gibson

: The contributions of sex, genotype and age to transcriptional variance in Drosophila melanogaster . Nat Genet 29:389–395, 2001

11.

11 Kennerly

Thomson

Olby

Breen

Gibson

: Comparison of regional gene expression differences in the brain of the domestic dog and human. Human Genomics, 2004 (in press)

12.

12 Kepes

: Presidential address: the histopathology of meningiomas. A reflection of origins and expected behavior? J Neuropathol Exp Neurol 45:95–107, 1986

13.

13 Khanna

Wan

Bose

Cassaday

Olomu

Mendoza

Yeung

Gorlick

Hewitt

Helman

: The membrane-cytoskeleton linker ezrin is necessary for osteosarcoma metastasis. Nat Med 10:182–186, 2004

14.

14 Kirkness

Bafna

Halpern

Levy

Remington

Rusch

Delcher

Pop

Wang

Fraser

Venter

: The dog genome: survey sequencing and comparative analysis. Science 301:1898–1903, 2003

15.

15 Koestner

Bilzer

Fatzer

Schulman

Summers

Van Winkle

: Histological classification of tumors of the nervous system of domestic animals, pp. 27–29. Armed Forces Institute of Pathology, World Health Organization, Washington, DC, 1999

16.

16 Krinke

Naylor

Schmid

Frohlich

Schnider

: The incidence of naturally-occurring primary brain tumours in the laboratory rat. J Comp Pathol 95:175–192, 1985

17.

17 Lantos

Vandenberg

Kleihues

: Greenfield's Neuropathology, 6th ed., vol. 2, pp. 647–653. Edward Arnold, London, 1997

18.

18 Lingaas

Comstock

Kirkness

Sorensen

Aarskaug

Hitte

Nickerson

Moe

Schmidt

Thomas

Breen

Galibert

Zbar

Ostrander

: A mutation in the canine BHD gene is associated with hereditary multifocal renal cystadenocarcinoma and nodular dermatofibrosis in the German Shepherd dog. Hum Mol Genet 12:3043–3053, 2003

19.

19 Lipsitz

Higgins

Kortz

Dickinson

Bollen

Naydan

LeCouteur

: Glioblastoma multiforme: clinical findings, magnetic resonance imaging, and pathology in five dogs. Vet Pathol 40:659–669, 2003

20.

20 Livak

Schmittgen

: Analysis of relative gene expression data using real-time quantitative PCR and the 2(-Delta Delta C(T)) method. Methods 25:402–408, 2001

21.

21 Ljubimova

Lakhter

Loksh

Yong

Riedinger

Miner

Sorokin

Ljubimov

Black

: Overexpression of alpha4 chain-containing laminins in human glial tumors identified by gene microarray analysis. Cancer Res 61:5601–5610, 2001

22.

22 Louis

Scheithauer

Budka

von Deimling

Kepes

: Pathology & Genetics: Tumours of the Nervous System, 2nd ed., pp. 176–184. International Agency for Research on Cancer, Lyon, France, 2000

23.

23 Patnaik

Kay

Hurvitz

: Intracranial meningioma: a comparative pathologic study of 28 dogs. Vet Pathol 23:369–373, 1986

24.

24 Qu

Zhai

Wei

Zhang

Xing

: Characterization and expression of three novel differentiation-related genes belong to the human NDRG gene family. Mol Cell Biochem 229:35–44, 2002

25.

25 Roberts

Hitte

Hendrickson

Hoffmann

Flickinger

Rutherford

Guyon

Galibert

Mickelson

: Characterization and radiation hybrid mapping of expressed sequence tags from the canine brain. Mamm Genome 14:203–213, 2003

26.

26 Stoica

Kim

Hall

Coates

: Morphology, immunohistochemistry, and genetic alterations in dog astrocytomas. Vet Pathol 41:10–19, 2004

27.

27 Storey

Tibshirani

: Statistical significance for genomewide studies. Proc Natl Acad Sci USA 100:9440–9445, 2003

28.

28 Summers

Cummings

de Lahunta

: Veterinary Neuropathology, pp. 362–373. Mosby, St. Louis, MO, 1995

29.

29 Thomas

Bridge

Benke

Breen

: Isolation and chromosomal assignment of canine genomic BAC clones representing 25 cancer-related genes. Cytogenet Genome Res 102:249–253, 2003

30.

30 Veal

Groisman

Eisenstein

Gill

: The secreted glycoprotein CREG enhances differentiation of NTERA-2 human embryonal carcinoma cells. Oncogene 19:2120–2128, 2000

31.

31 Watson

Gutmann

Peterson

Chicoine

Kleinschmidt-DeMasters

Brown

Perry

: Molecular characterization of human meningiomas by gene expression profiling using high-density oligonucleotide microarrays. Am J Pathol 161:665–672, 2002

32.

32 Wolfinger

Kass

: Nonconjugate Bayesian analysis of variance component models. Biometrics 56:768–774, 2000

33.

33 Yoshida

Mitsumori

Harada

Maita

: Morphological and ultrastructural study of the histogenesis of meningeal granular cell tumors in rats. Toxicol Pathol 25:211–216, 1997

Microarray Analysis of Differentially Expressed Genes of Primary Tumors in the Canine Central Nervous System

Abstract

Keywords

Materials and Methods

Tissue samples

Microarray preparation

Total RNA preparations

Fluorescence labeling of probe

Hybridization

Data analysis

Quantitative real-time PCR

Results

Discussion

Footnotes

Acknowledgements

References