Identification of valid endogenous control genes for determining gene expression in human glioma

Abstract

In human glioma, quantitative real-time reverse-transcription PCR (qPCR) is a frequently used research tool. However, no systematic analysis of suitable reference genes for reliable gene expression analysis has been performed so far. In the current study, we tested 19 commonly used reference genes for their expression stability in human astrocytoma WHO Grade II, astrocytoma WHO Grade III, and glioblastoma (WHO Grade IV) both alone and compared with normal brain. First, equivalence tests for equal expression of candidate genes were applied, and those genes showing differential expression were ruled out from further analyses. Second, expression stability of the remaining candidate genes was determined by the NormFinder software. Generally, glioblastoma exhibited the highest expression levels and largest variability of candidate genes, whereas the opposite was true for normal brain. Even though Normfinder analyses revealed a large number of genes suitable for normalization in each of the tumor subgroups and across these groups, this number was drastically reduced after inclusion of normal brain into the analyses: Only GAPDH, IPO8, RPL13A, SDHA, and TBP were expected not to be differentially expressed; NormFinder analysis indicated favorable stability values for all of these genes, with TBP and IPO8 being the most stable ones. These 5 genes represent different physiological pathways and may be regarded as universal reference genes applicable for accurate normalization of gene expression in human astrocytomas of different grades (WHO Grades II–IV) alone and compared with normal brain, thereby enabling longitudinally designed studies (eg, in astrocytoma before and after malignant transformation).

Despite intensive scientific and medical efforts in the last decade, there still is a lack of definitive information regarding the etiology of malignant glioma, and identification of prognostic factors that influence survival remains only fragmentary.¹ Therefore, it is important to improve further the understanding of the complex biological interactions that regulate glioma development and growth control.

Gene expression analysis using quantitative real-time reverse-transcription PCR (qPCR) has been shown to be a promising approach to identify molecular markers associated with tumor grade, tumor progression, patient survival, and therapeutic accessibility and has increasingly gained importance due to its high sensitivity, specificity, reproducibility, and its potential for high throughput.^2–5 However, to obtain valid results by qPCR, it is essential to address the inherent issues associated with this technique arising from the varying type, quantity, and quality of the RNA, differences in reverse transcription, and PCR efficiencies: All data have therefore to be normalized to valid reference genes (“housekeeping genes”), and PCRs have to be efficiency corrected.^6–8

Ideal reference genes should be stably expressed in all samples under investigation, regardless of tissue type, developmental stages, disease state, and medical or experimental treatment.^9,10 As the “ideal” reference does not exist, convenient housekeeping genes have to be verified prior to qPCR data analysis for each tissue type under investigation and each experimental setting used.^11,12 So far, normalization of qPCR data in human glioma has mostly been performed with only 1 single reference gene (eg, ACTB,^4,13 GAPDH,^5,14,15 18srRNA,³ ALAS,¹⁶ or B2M¹⁷), and no study has sufficiently evaluated the appropriateness of the applied normalization strategy. Moreover, the necessity to use at least the combination of two validated reference genes for proper quantitative evaluation has not yet been considered.^18,19 In the present study, we tested 19 commonly used reference genes for their expression stability in human astrocytoma WHO Grade II, astrocytoma WHO Grade III, and glioblastoma (WHO Grade IV) both alone and compared with normal brain.²⁰ At a second step, the analysis was done across all of the 3 tumor entities and, additionally, compared with normal brain. The latter was done to identify a set of reference genes suitable as universal normalizers in astrocytoma of different grades, thereby enabling longitudinally designed quantitative expression studies, eg, in astrocytoma before and after malignant transformation.

Methods

Tissue Samples

Adult patients with a histologically proven diagnosis of a supratentorial so far untreated WHO Grade II astrocytoma (n = 9), WHO grade III astrocytoma (n = 9), or glioblastoma (WHO Grade IV, n = 9) were included. Simultaneous expression analysis of 19 commonly used reference genes (Table 1) was performed in this data set. To test for the influence of different normalization strategies on the obtained qPCR results, we analyzed gene expression levels of 2 selected genes in another data set of 15 primary untreated glioblastoma. All patients gave informed consent, and the study protocol was reviewed and approved by the institutional review board of the Ludwig-Maximilians-Universität, Klinikum Grosshadern, Munich, Germany. Glioma tissue samples were obtained from stereotactic biopsy procedures and every effort was made to ensure that exclusively solid viable tumor was used for qPCR: Based on multimodal planning following coregistration of computerized tomography (CT), magnetic resonance imaging (MRI; including T1- and T2-weighted sequences) and O-(2-[(18)F]fluoroethyl)-1-tyrosine positron emission tomography (FET-PET)^21,22 (i-plan stereotaxy^®, BrainLAB^®) for better visualization of the tumor, the tumor/vessel interface, as well as the intratumoral metabolic activity, a serial biopsy was taken along a trajectory representative of the solid tumor tissue (including the area with highest static and/or dynamic FET uptake kinetics (“hotspot”), if present). Using microforceps, the maximum amount of tissue per biopsy specimen was 1 mm³. The number of specimens taken in 1 mm steps along the chosen trajectory was in the range of 10–25 samples per tumor and exactly one of these samples was selected for qPCR analysis. The selection procedure was guided by intraoperative smear preparations, which were routinely done by the attending neuropathologist. Tissue samples taken for expression analysis were solely collected from tumor probes next (ie, 1 mm distance) to smear preparations exclusively showing solid vital tumor tissue. Additionally, a corresponding sample (level + 1 mm) was taken for paraffin embedding and state-of-the-art histopathologic examination. This procedure was used to verify the presence of solid viable tumor tissue next to the probe taken for expression analysis. For histopathologic evaluation, samples were fixed with 4% buffered formalin, paraffin embedded, and subjected to routine staining (Hematoxylin and Eosin, Elastica van Gieson, Periodic acid–Schiff) and immunohistochemistry with antibodies against human GFAP (monoclonal mouse, clone 6F2, Dako) and anti-MAP2 (clone HM-2, Sigma). The histologic diagnosis of all tissue specimens was made according to the WHO Classification of Tumors of the Central Nervous System, 4th edition, 2007.

Table 1.

Candidate normalization genes evaluated in this study

Gene	Accession numbera	Gene name
18S	emb|X03205	Human 18S ribosomal RNA
ACTB	ENST00000331789	Actin, cytoplasmic 1
ALAS	ENST00000394965	5-Aminolevulinate synthase
B2M	ENST00000349264	β-2-Microglobulin precursor
β-Globin	ENST00000335295	Hemoglobin subunit delta
G6PDH	ENST00000393564	Glucose-6-phosphate 1-dehydrogenase
GAPDH	ENST00000229239	Glyceraldehyde-3-phosphate dehydrogenase
GUSB	ENST00000304895	β-Glucuronidase precursor
HPRT1	ENST00000370796	Hypoxanthine-guanine phosphoribosyltransferase
IPO8	ENST00000256079	Importin-8
PBGD	ENST00000278715	Porphobilinogen deaminase
PGK1	ENST00000373316	Phosphoglycerate kinase 1
PPIA	ENST00000355968	Peptidyl-prolyl cis–trans isomerase A
RPL13A	ENST00000270634	60S ribosomal protein L13a
RPLP0	ENST00000228306	60S acidic ribosomal protein P0
SDHA	ENST00000264932	Succinate dehydrogenase
TBP	ENST00000392092	TATA-box-binding protein
TFRC	ENST00000392396	Transferrin receptor protein 1
YWHAZ	ENST00000353245	14-3-3 protein zeta/delta

^aPrimer design based on this sequence. The database source is the Ensembl database (http://www.ensembl.org).

Normal brain (from 8 patients) was obtained from epilepsy surgery. One additional normal brain sample mRNA was purchased from Ambion.

RNA Extraction, Linear Amplification, and Reverse Transcription

To minimize the risk of RNA degradation, the tissue specimen was directly treated in RNA lysis buffer, and RNA isolation was performed immediately using the RNaequous Micro Kit (Ambion) with subsequent DNAse treatment (TURBO DNase, Ambion) following the manufacturer's protocol. The quantity and quality of the RNA was determined using a spectrophotometer; the quality of the RNA was satisfactory in all cases (260/280 nm absorbance ratio between 1.95 and 2.15). RNA recovery from each stereotactic biopsy sample was approximately 20–100 ng. To obtain RNA amounts suitable for gene expression analyses, RNA was amplified using the TargetAmp-Kit (Epicentre), as per manufacturer's recommendations. The used technique of RNA amplification by in vitro transcription was first described by Van Gelder et al.²³ and is now commonly referred to as the Eberwine method. It involves 2 general steps: reverse transcription using an oligo(dT) primer bearing a T7 promoter site and, after second-strand cDNA synthesis, antisense RNA (aRNA) is transcribed in vitro using T7 RNA polymerase. The resulting amplification factors were between 500 and 2500.

Hereafter, equal amounts of the different samples of amplified RNA (1000 ng) were transcribed into cDNA. The reverse transcription (RT) reaction was carried out using random primers and Superscript III reverse transcriptase (Invitrogen), as per manufacturer's instructions.

Reference Genes and Real-Time PCR

We used the RealTime ready Human Reference Gene Panel (Roche Diagnostics) allowing the simultaneous expression analysis of 19 commonly used reference genes (Tables 1 and 2). The panel contains prevalidated primer-probe sets for the quantification of 19 reference genes, 3 positive controls to check for degradation of the initial RNA, the quality of the RT reaction, and 2 negative controls to detect residual genomic DNA. For expression analyses of example genes, highly specific primer-probe sets were used:

Table 2.

Function of candidate normalization genes evaluated in this study

Gene	Function	Process
18S	Ribosomal RNA	Translation
RPL13A	Ribosomal protein
RPLP0	Ribosomal protein
PBGD	Heme biosynthetic pathway	Hemoglobin metabolism
β-Globin	Hemoglobin β-chain, oxygen transport
ALAS	Heme biosynthetic pathway
GAPDH	Glycolysis and gluconeogenesis	Carbohydrate transport and metabolism
PGK1	Glycolysis
G6PDH	Pentose phosphate pathway
HPRT1	Purine synthesis in salvage pathway	Nucleotide metabolism
IPO8	Intracellular protein transport into nucleus	Protein trafficking
GUSB	Lysosomal exoglycosidase	Glycosylation
ACTB	Cytoskeletal structural protein	Cell structure
PPIA	Isomerase, exerts chaperone activity	Protein folding
B2M	β-chain of MHC class I molecule	MHC-mediated immunity
SDHA	Cellular oxygen homeostasis	Oxidative phosphorylation
TBP	RNA polymerase II, transcription factor	Transcription
TFRC	Cellular iron ion homeostasis	Iron homeostasis
YWHAZ	Adaptor protein; regulation of signaling pathways	Signal transduction

RAD51: 5′-GGCGGTCAGAGATCATACAGA-3′ (forward primer), 5′-AGATCCAGTCTCAATTCCACCT-3′ (reverse primer); Probe: Universal ProbeLibrary probe: #36, cat.no.04687949001 (Roche Diagnostics Penzberg, Germany). Cyclin-dependent kinase inhibitor 2A (CDKN2A): 5′- GTGGACCTGGCTGAGGAG-3′ (forward primer), 5′-TCTTTCAATCGGGGATGTCT-3′ (reverse primer); Probe: Universal ProbeLibrary probe: #34, cat.no.04687671001 (Roche Diagnostics).

All assays were designed to be intron spanning, and amplicon sequences of the housekeeping gene assays are available upon request.

Real-time qPCR was performed in duplicate with the Light Cycler480 instrument 96-well plate (Roche Diagnostics) using Roche's Probes Mastermix and 10 ng cDNA per well. The thermal cycler conditions comprised 45 cycles of 95°C for 10 s, 60°C for 30 s, and 72°C for 1 s. Efficiencies of PCR for all housekeeping gene assays were determined as described by Pfaffl.²⁴

Data Analysis

NormFinder²⁵ was used to select the most stable candidate genes, whereby both the intra- and intergroup variation of candidate genes were modeled. Stability (M) values were calculated for each candidate gene under investigation. Lower M values, hereby referred to as higher expression stability. In the current report, stable gene expression was assumed for M < 0.5, intermediate stability for M values between 0.5 and 1.0, and unstable expression for M values >1. Valid NormFinder analysis requires a sample set of minimally 8 samples per group and the recommended number of candidate genes under investigation should lie in the range of 5–10²⁵; these requirements were fulfilled in the current study. For comparison of candidate gene transcription levels, the cycle threshold values (C_t) were plotted. The mean C_t values of the replicates for each sample were transformed into raw, nonnormalized quantities (Q) using the comparative ΔC_t method.²⁶ Reference gene mRNAs should be stably expressed, and their abundances should show strong correlation with the total amounts of mRNA present in the samples. Therefore, candidate genes with median C_t values in the range of 35 or higher were classified as low abundant genes not suitable for valid normalization and were excluded from further analysis. For analysis of normal distribution, the Kolmogorov and Smirnov test was applied. A prerequisite for valid Normfinder evaluation is that the average expression of candidate genes is not significantly different between subgroups. Therefore, as a preceding step, equivalence tests for equal expression of candidate genes were used and those genes showing differential expression were ruled out from further analysis. Intergroup comparisons were performed by t-test or Mann–Whitney test, if data were normal or not normally distributed, respectively. Multiple group comparison was done with one-way ANOVA (in case of normal distribution). Expression data were tested for the degree of variability (F-test), and the means and standard deviations were computed and compared. In the case of multiple testing, Bonferroni correction was carried out. Two-tailed levels of statistical significance are indicated by P < .05.

To substantiate the results, additionally the alternative software application geNorm²⁷ was used. Here, the average pairwise variation of a candidate genes with all other candidate reference genes under investigation is indicated by the reference gene stability value (M value). The M values above 1.5 indicate low expression stability. The software uses an algorithm to rank the candidate reference genes by a repeated process of stepwise exclusion of the worst scoring reference gene. For all analyses comparing gene expression in more than one group, NormFinder was exclusively used, because it allows performing both an intragroup and an intergroup comparison. The degree of association between expression levels of candidate genes and other parameters (such as patients' age) were analyzed with the t-test/Mann–Whitney test.

Results

Expression Levels of Candidate Reference Genes

Twenty-seven samples of glioma tissue (9 Astrocytoma WHO Grade II, 9 Astrocytoma WHO Grade III, 9 Glioblastoma) and 9 samples of normal brain tissue were evaluated; the mean age of patients was 57, 55, and 47 years, respectively. The difference was statistically significant (P < .05). The efficiency of all assays was confirmed to be ≥98% (data not shown). The C_t values were found to be normally distributed. The expression levels of the candidate genes for the different tumor entities and normal brain are demonstrated in Fig. 1. Median C_t values range from as low as 16.9 (B2M) to as high as 34.5 (G6PDH); G6PDH was therefore classified as a low-abundant gene and was excluded from further analyses. Expression levels were influenced by the tumor grade and patients' age, which were significantly intercorrelated with each other (patients with low-grade astrocytoma were significantly younger); sex had no influence. Generally, the lowest expression levels with the lowest variability of candidate genes were seen for normal brain when compared with tumor tissue; this difference was statistically significant in 6 candidate genes: RPLP0, ACTB, PBGD, B2M, TFRC, and GUSB. Four candidate genes, however, exhibited significantly higher expression levels in normal brain than in tumor tissue: YWHAZ, ALAS, 18S, and HPRT1. No significant differences were seen between WHO Grade II and Grade III astrocytomas. The glioblastomas exhibited the highest expression levels with the largest variability in all candidate genes when compared with Grade II and/or Grade III astrocytomas, which was statistically significant for 8 genes: RPL0, PPIA, YWHAZ, B2M, TFRC, PGK1, GUSB, and ALAS.

Fig. 1.

Expression of 19 reference genes in glioma and non-cancerous brain tissue specimen. Comparison of threshold cycles (C_t values) of the 19 candidate genes in glioblastoma (filled circles, n = 9), astrocytoma WHO Grade III (open circles, n = 9), astrocytoma WHO Grade II (filled inverted triangles, n = 9), and normal brain tissue (open triangles, n = 9). cDNA was synthesized from amplified RNA purified from tumor tissue obtained by stereotactic biopsy. Horizontal bars indicate medians.

Expression Stability of Candidate Reference Genes in Astrocytomas Grades II, III, and IV

Stability values for each of the three tumor entities are given in Table 3. β-Globin turned out to be the only unstable gene in each of the tumor entities investigated and was therefore excluded from further analysis. Highly stable genes for Grade III astrocytoma were also found to be highly stable in both the glioblastoma group, except for PBGD and for PGK1 (which were classified as intermediate stable in the glioblastoma group) and the Grade II astrocytoma group, except for GUSB (which was classified as intermediate stable in the astrocytoma Grade II group). The largest number of highly stable genes was seen in the glioblastoma group. The difference, however, was not statistically significant. The 5 most stable genes for each of the analyzed tumor entities were as follows: astrocytoma WHO Grade II: PPIA, YWHAZ, GAPDH, RPLP0, and TBP; astrocytoma: WHO Grade III. TBP, PPIA, PGK1, YWHAZ, and RPL13A; glioblastoma: YWHAZ, IPO8, 18S, B2M, and HPRT1.

Table 3.

Housekeeping genes in individual glioma entities ranked according to their expression stability estimated with NormFinder^a

	GBM	Mb	Astro III	Mb	Astro II	Mb
1	YWHAZ	0.211	TBP	0.305	PPIA	0.161
2	IPO8	0.219	PPIA	0.329	YWHAZ	0.180
3	18S	0.271	PGK1	0.330	GAPDH	0.276
4	B2M	0.314	YWHAZ	0.379	RPLP0	0.284
5	HPRT1	0.331	RPL13A	0.383	TBP	0.300
6	RPLP0	0.334	SDHA	0.386	ALAS	0.305
7	ACTB	0.344	GUSB	0.427	SDHA	0.343
8	TBP	0.346	PBGD	0.439	RPL13A	0.355
9	SDHA	0.383	GAPDH	0.465	HPRT1	0.403
10	GUSB	0.383	IPO8	0.473	IPO8	0.435
11	PPIA	0.392	HPRT1	0.548	PBGD	0.458
12	GAPDH	0.429	TFRC	0.564	ACTB	0.474
13	ALAS	0.431	B2M	0.567	PGK1	0.582
14	RPL13A	0.473	18S	0.577	TFRC	0.584
15	TFRC	0.561	ALAS	0.586	B2M	0.637
16	PBGD	0.570	ACTB	0.601	GUSB	0.737
17	PGK1	0.746	RPLP0	0.706	18S	0.771
18	β-Globin	1.043	β-Globin	1.221	β-Globin	1.697

^aThe candidates are listed with decreasing expression stability from top to bottom.

^bM, average expression stability.

Stability values with the geNorm algorithm revealed almost identical results: slight differences in the rank order of the reference genes occurred, even though the pattern of the previously identified most and least stable genes still remained the same (data not shown).

In a consecutive step, we intended to identify reference genes suitable across all 3 tumor entities. Prefiltering with one-way ANOVA analysis of 6 candidate genes revealed significant differences in gene expression (Fig. 2), which were excluded from further analyses (PPIA, YWHAZ, B2M, TFRC, PGK1, and GUSB). The results of the NormFinder analysis are summarized in Table 4. All of the remaining 11 reference genes exhibited low M–values (<0.2), and TBP and RPL13A were found to be the most stable genes in this series.

Fig. 2.

Identification of housekeeping genes that are valid for all 3 tumor entities together. Comparison of threshold cycles (C_t values) of 17 candidate genes in glioblastoma (filled circles, n = 9), astrocytoma WHO Grade III (shaded circles, n = 9), astrocytoma WHO Grade II (inverted triangles, n = 9). The 6 candidate genes on the right side of the vertical line exhibited significant differences in gene expression (P < .05, one-way ANOVA with Bonferroni correction).

Table 4.

Candidate genes not differentially expressed when comparing GBM, astrocytoma Grade III, and astrocytoma Grade II, ranked according to their expression stability estimated with NormFinder^a

Gene	Stability value
TBP	0.075
RPL13A	0.109
SDHA	0.117
GAPDH	0.123
IPO8	0.130
RPLP0	0.132
ALAS	0.153
PBGD	0.155
ACTB	0.155
18S	0.178
HPRT1	0.180

^aThe candidates are listed with decreasing expression stability from top to bottom.

Expression Stability of Candidate Reference Genes in Glioma Compared with Normal Brain

Intergroup comparison of each of the tumor subgroups vs normal brain revealed a larger number (10 genes) of differentially expressed candidate genes in glioblastoma than in Grade III (4 genes: YWHAZ, 18S, HPRT1, and PPIA) and Grade II astrocytoma (5 genes: YWHAZ, GUSB, ALAS, 18S, and HPRT1). The difference was statistically significant (P < .05). The results of the intergroup comparison of tumor subgroups vs normal brain are summarized in Fig. 3A–C. ANOVA analysis of all tumor entities plus normal brain revealed that only 5 genes were expected not to be differentially expressed in all three groups (ie, GAPDH, IPO8, RPL13A SDHA, and TBP). Subsequent Normfinder analysis resulted in low M–values in a comparable range (0.146–0.223) for all of the 5 genes, with TBP and IPO8 being the most stable candidates (results are given in Table 5; for visualization, see Fig. 4). These 5 genes were identical with the best ranked highly stably expressed reference genes evaluated for glioma WHO Grade II–IV without normal brain and represent different physiologic pathways (which makes co-regulation unlikely, see Table 2). They may therefore be regarded as universal normalizers within the context of the current data set.

Fig. 3.

Expression of candidate reference genes in astrocytoma compared with normal brain. The vertical lines separate differentially (right side) from nondifferentially (left side) expressed genes (P < .05, Student's t-test). (A) Comparison of threshold cycles (C_t values) of 17 candidate genes in glioblastoma (filled circles, n = 9) and normal brain tissue (open circles, n = 9): 10 genes were differentially expressed. (B) C_t values of astrocytoma WHO Grade III (filled circles, n = 9) vs normal brain (open circles, n = 9): 4 genes were differentially expressed. (C) C_t values of astrocytoma WHO Grade II (filled circles, n = 9) vs normal brain (open circles, n = 9): 4 genes were differentially expressed.

Table 5.

Suitable reference genes for GBM, astrocytoma Grade III, astrocytoma Grade II, and normal brain, ranked according to their expression stability-estimated with NormFinder^a

Gene	Stability value
TBP	0.146
IPO8	0.170
RPL13A	0.194
GAPDH	0.220
SDHA	0.223

^aThe candidates are listed with decreasing expression stability from top to bottom.

Fig. 4.

Expression of 5 reference genes evaluated as stably expressed in all glioma entities compared with noncancerous brain tissue. Comparison of threshold cycles (C_t values) in glioblastoma (filled circles, n = 9), astrocytoma WHO Grade III (open circles, n = 9), astrocytoma WHO Grade II (filled inverted triangles, n = 9), and normal brain tissue (open triangles, n = 9).

Effects of Using Different Normalizers on Gene Expression Analyses of Target Genes

To test for the influences of different normalization strategies on the accuracy of qPCR results, we selected 2 target genes of interest in the evaluation of prognosis and therapeutic response of glioblastoma (CDKN2A and RAD51)^26,28 and performed qPCR experiments to analyze their expression in human glioblastoma (n = 15) vs normal brain. Data were normalized to the geometric mean of (i) a combination of 2 reference genes being evaluated as suitable (eg, RPL13A and GAPDH) and (ii) a combination of B2M and HPRT1, which both have been shown to be unsuitable for normalization in the experimental setting here tested.

As shown in Fig. 5A and B, gene expression levels vary considerably depending on the applied normalization strategy. When normalized to the geometric mean of 2 stably expressed genes, RAD51 expression (Fig. 5A) increased 10.1-fold in glioblastoma vs normal brain, which should be regarded as the value of highest correctness. Normalization to B2M and HPRT1 resulted in a false-high (37-fold) increase. Expression analyses of CDKN2A (Fig. 5B) confirmed these results. When using the optimal normalization strategy with 2 stably expressed genes, an almost unchanged CDKN2A expression was found (1.2-fold), whereas, when normalizing to B2M and HPRT1 was done, a 20-fold upregulation was obtained. As an additional control, normalization was also performed to the geometric mean of other combinations out of the 5 universal normalizers (TBP/RPL13A and TBP/GAPDH), which led to almost identical results.

Fig. 5.

Normalization strategy strongly influences results of qPCR analyses. Expression of RAD51 and CDKN2A in GBM relative to normal brain. Data were normalized to the geometric mean of the 2 suitable housekeeping genes (RPL13A and GAPDH), and of the 2 not stably expressed genes (B2M and HPRT1). RNA samples derived from glioblastoma tissues of 15 patients and 4 normal brain tissue samples were used as real-time PCR templates. Expression levels were calculated using different normalization strategies as indicated. Results are expressed as the mean mRNA expression ratio (±SEM) of (A) RAD51-expression and (B) CDKN2A expression in tumor tissue relative to normal brain (*P < .05).

Discussion

Real-time PCR is becoming the method of choice for high-throughput gene expression analysis because of its wide dynamic range of quantification combined with high sensitivity and precision. However, careful evaluation of reference gene stability is necessary to generate reliable results.^29–31 The current study is the first systematic analysis of a wide set of reference genes in human astrocytoma of different grades (WHO Grades II, III, and IV), leading to the identification of reliable normalizing strategies for the different tumor entities alone, in comparison with each other, and compared with normal noncancerous brain tissue. As “tumor up-grading” is a frequent but poorly understood phenomenon in astrocytoma Grade II, the identification of “universal” reference genes allowing longitudinal analyses in cases of tumor progression is an important issue that has been addressed in the current study.

Nineteen commonly used candidate reference genes were analyzed. Molecular stereotactic biopsy technique was used for highly controlled sampling of solid viable tumor tissue, and immediate processing of tissue samples to RNA ensured that analyses were performed with high-quality RNA recovered from tissue of unambiguous tumor classification.³²

Out of the 19 reference genes investigated, G6PDH and β-globin revealed unsuitable for normalization purposes in human glioma: The first because of low expression levels and the latter because of its high expression instability.

The modeling of the intragroup variation of candidate genes with NormFinder demonstrated that, at first glance, a remarkable number of the investigated genes exhibited sufficient expression stability (ie, 13, 11, and 10 genes have been classified as suitable normalizers in GBM, astrocytoma Grade III, and astrocytoma Grade II, respectively). This suggestion, however, should be regarded with caution, as a general upregulation of the candidate genes' expression (particularly observed for GBM in this series) may diminish the effects of differential expression, which might become apparent when further medical or experimental treatment occurs. Thus, false-positive classification of some candidate genes could not be excluded, and only a combination of the most stable reference genes should be used for normalization.

It is important to note in this context that the number of suitable candidate genes decreased to 11 and 5, respectively, when stability evaluation was done across all 3 tumor entities and, additionally, in tumor tissue compared with normal brain. Interestingly, although highly ranked in the single-tumor entities, the genes YWHAZ and PPIA (Cyclophilin A) turned out to be differentially expressed across the tumor subgroups and not suitable for proper intergroup comparative expression analysis. This finding may be in line with recently published studies, which report that (i) YWHAZ mRNA is upregulated in head and neck squamous cell carcinoma and may serve as a candidate proto-oncogene³³ and (ii) PPIA mRNA is upregulated in endometrial carcinoma and associated with decreased survival.³⁴ The detected differential expression of these genes across different astrocytoma grades might be not surprising and deserves further evaluation.

The identified 5 genes suitable for both normalization in tumor tissue of different grades and normal brain (ie, GAPDH, IPO8, RPL13A, SDHA, and TBP) all exhibited high expression stability, represent different physiologic pathways (which makes coregulation unlikely) and, what is equally important, were identical to the top-ranked reference genes suitable for normalization across the different tumor entities, thereby supporting the validity of the results obtained from different modeling strategies. These genes may serve as “universal” normalizers allowing all qPCR experiments with combinations and comparison of GBM, astrocytoma Grade III, astrocytoma Grade II, and normal brain tissue. As recent literature clearly points out that normalization to one single gene is not sufficient,¹⁸ the optimal normalization strategy when evaluating gene expression in human glioma should at least (taking practicability and costs into consideration) include a combination of 2 or more out of the 5 reference genes evaluated here.

Out of the different software applications available for the selection of the most stable reference gene among a panel tested, we have not only used NormFinder in this study but also, for confirmation, the program geNorm. Both programs perform a ranking of the reference genes with respect to their expression stability. When comparing the gene stability calculations obtained by both programs for the single-tumor entities investigated, M values were almost identical. Even though the rank order of stability values of genes becomes slightly different after the stepwise exclusion procedure and recalculation by geNorm when compared with the initial data and those identified by NormFinder, the pattern of the most and least stable genes still remains the same, indicating well-matched estimations of intragroup variation of candidate genes by both software applications.

We illustrated the impact of different normalization strategies on the accuracy of qPCR results by analyzing the expression of RAD51 and CDKN2A which are genes of interest in the evaluation of prognosis and therapeutic response of glioblastoma. RAD51 is a central protein in mediating homologous recombination repair, and high RAD51 levels partially account for resistance to chemotherapy and radiotherapy in glioma. Furthermore, RAD51 protein reducing agents such as imatinib mesylate (Gleevec) are used to achieve radiosensitization.^26,28 Both deletion of the CDKN2A gene and hypermethylation of its promoter leading to decreased gene expression have been shown to be associated with malignant progression and shorter survival in human astrocytoma.^28,35,36 Therefore, accurate quantification of their mRNA expression may be a valuable tool to detect changes in the expression of those 2 genes, which may be of clinical relevance. Our results demonstrate the major influence of the normalization strategy on the expression profiles obtained by qPCR experiments with human glioma tissues. According to our previous analyses, normalization to a combination of the 2 stably expressed reference genes RPL13A and GAPDH should result in correct gene expression ratios. When normalizing to 2 reference genes with low expression stability (B2M and HPRT1), gene expression values clearly deviate from this “optimal standard” which leads to a wide overestimation of the RAD51 and CDKN2A expression.

A recently published report has identified TBP and HPRT1 (among 12 target genes evaluated) as adequate references for glioblastoma expression analysis alone and compared with normal brain.³⁷ These results were partially contradictory to the findings of the current analysis. HPRT1 indeed has been evaluated as a possible reference gene for glioblastoma; however, in a combination with normal brain it has been classified as an unsuitable candidate housekeeping gene. Suboptimal assays with significantly varying mRNA amplification efficiencies (87%–110%), highly varying expression levels even for the most suitable genes and an imbalanced sample size (30 glioblastoma samples were compared with 9 non-neoplastic brain samples), might have seriously biased the results of the study of Valente et al. and explain that their findings were not in accordance with our results. In the current series, amplification efficiency was always better than 98% and the range of the expression levels of the most suitable genes were generally extremely low. The distorting impact of comparing divergent amplification efficiencies on quantitative RT–PCR analysis has been underscored by others. Even with a 0.1 range in PCR efficiencies (ie, from 87% to 97%) already a more than 20-fold error in fold-difference occurs.^6,38

In conclusion, our study clearly shows that the accurate selection of reliable reference genes is an absolute prerequisite for measurement of correct gene expression in human glioma. We provide a set of 5 reference genes applicable for accurate normalization of gene expression studies in human glioma WHO Grades II, III, and IV as well as in comparison of glioma tissue to normal brain tissue.

Funding

The study was supported by the institutional funds of the Department of Anaesthesiology, LMU Munich, Germany.