Abstract

Comprehensive expression profiling of tumors using DNA microarrays has been used recently for molecular classification and biomarker discovery, as well as a tool to identify and investigate genes involved in tumorigenesis. Application of this approach to a cohort of benign and malignant adrenocortical tissues would be potentially informative in all of these aspects. In this study, we generated transcriptional profiles of 11 adrenocortical carcinomas (ACCs), 4 adrenocortical adenomas (ACAs), 3 normal adrenal cortices (NCs), and 1 macronodular hyperplasia (MNH) using Affymetrix HG_U95Av2 oligonucleotide arrays representing ∼10,500 unique genes. The expression data set was used for unsupervised hierarchical cluster analysis as well as principal component analysis to visually represent the expression data. An analysis of variance on the three classes (NC, ACA plus MNH, and ACC) revealed 91 genes that displayed at least threefold differential expression between the ACC cohort and both the NC and ACA cohorts at a significance level of P 8.0UNHighMixedYNNNoneACC 71° CarcinomaLAFUN172300HighLipid-poorYNYLiverACC 81° CarcinomaLAFUN9180HighMixedYYYSkullACC 11Metastatic carcinomaLAFLiver metastasisUNUNHighMixedYNANALiverACC 131° CarcinomaLAFUN16460LowMixedYNNLung, liver, boneACC 141° CarcinomaRAMR leg edema222000HighLipid-poorYYYLiverACC 15Metastatic carcinomaRAMLung metastasisUNUNHighMixedYNANALung and chest wallACC 171° CarcinomaRAFRight abdominal pain262300HighLipid-poorYYYLung and liverACC 181° CarcinomaLAFCushing's syndrome18900HighMixedYYYNoneACC 191° CarcinomaRAFR Flank pain9UNHighLipid-poorYYNLiverACC 331° CarcinomaLPMIncreased testosterone12450HighLipid-poorYYNLiver, lymph nodesACA 21AdenomaRAFCushing's syndrome4.565NALipid-richNNANANAACA 22AdenomaLAFCushing's syndrome2.512NAMixedNNANANAACA 28AdenomaLAFCushing's syndrome3.525.2NAMixedNNANANAACA 30AdenomaRAFCushing's syndrome2.516NAMixedNNANANAMNH 29Macronodular hyperplasiaRAFCushing's syndrome5.550NALipid-richNNANANANC 6Normal adrenal cortexLAFAdrenalectomy for metastatic lung carcinomaNANANANANANANANANC 9Normal adrenal cortexUNAUNAdrenalectomy for metastatic renal cell carcinomaNANANANANANANANANC 10Normal adrenal cortexUNAUNAdrenalectomy for metastatic gastrinomaNANANANANANANANANA, not applicable; UN, unknown.* A, Adult; P, child.† Tumors graded according to Weiss et al3Weiss LM Medeiros LJ Vickery Jr, AL Pathologic features of prognostic significance in adrenocortical carcinoma.Am J Surg Pathol. 1989; 13: 202-206Crossref PubMed Scopus (663) Google Scholar. Open table in a new tab NA, not applicable; UN, unknown. Single isolates of tissue samples were homogenized in the presence of Trizol reagent (Life Technologies, Inc., Gaithersburg, MD) and total cellular RNA was purified according to manufacturer's procedures. RNA samples were further purified using acid phenol extraction and RNeasy spin columns (Qiagen, Valencia, CA) and used to prepare cRNA probes. RNA quality was assessed by 1% agarose gel electrophoresis in the presence of ethidium bromide. Samples that did not reveal intact 18S and 28S ribosomal bands were excluded from further study. This study used commercially available high-density oligonucleotide microarrays (HG_U95Av2; Affymetrix, Santa Clara, CA). Preparation of cRNA, hybridization, scanning, and image analysis of the arrays were performed according to manufacturer's protocols and as previously described.25Giordano TJ Shedden KA Schwartz DR Kuick R Taylor JM Lee N Misek DE Greenson JK Kardia SL Beer DG Rennert G Cho KR Gruber SB Fearon ER Hanash S Organ-specific molecular classification of primary lung, colon, and ovarian adenocarcinomas using gene expression profiles.Am J Pathol. 2001; 159: 1231-1238Abstract Full Text Full Text PDF PubMed Scopus (183) Google Scholar, 33Rickman DS Bobek MP Misek DE Kuick R Blaivas M Kurnit DM Taylor J Hanash SM Distinctive molecular profiles of high-grade and low-grade gliomas based on oligonucleotide microarray analysis.Cancer Res. 2001; 61: 6885-6891PubMed Google Scholar The U95A arrays consist of 12,625 probe sets, each representing a transcript. Each probe set typically consists of 16 perfectly complementary 25 base long probes (PMs) as well as 16 mismatch probes (MMs) that are identical except for an altered central base. A normal adrenal cortex sample was selected as the standard and probe pairs for which PM-MM <−100 on the standard were excluded from further analysis. The average of the middle 50% of the PM-MM differences was used as the expression measure for each probe set. A quantile normalization procedure was used to adjust for differences in the probe intensity distribution across different chips. We applied a monotone linear spline to each chip that mapped quantiles 0.01 up to 0.99 (in increments of 0.01) exactly to the corresponding quantiles of the standard. Then, the transform log[100 + max(X + 100; 0)] was applied to the data from each chip. Code to perform these computations is freely available at http://dot.ped.umich.edu:2000/ourimage/pub/index.html. A tissue microarray34Kononen J Bubendorf L Kallioniemi A Barlund M Schraml P Leighton S Torhorst J Mihatsch MJ Sauter G Kallioniemi OP Tissue microarrays for high-throughput molecular profiling of tumor specimens.Nat Med. 1998; 4: 844-847Crossref PubMed Scopus (3558) Google Scholar containing 4 normal adrenal cortex (NC) samples, 24 adrenocortical adenoma (ACA) samples, 12 adrenocortical hyperplasias (including both diffuse and macronodular types), 62 ACC samples, along with 3 pheochromocytomas and several various normal tissues, was constructed for immunohistochemical validation studies using the Beecher manual tissue arrayer and 0.6-mm-diameter cylindrical cores. Donor blocks were retrieved from the archives of the University of Michigan Department of Pathology and the corresponding H&E slides were reviewed and representative viable areas were chosen for sampling. Given the relatively homogeneous nature of adrenocortical tumors in general, each case was arrayed in duplicate. Immunohistochemistry was performed using formalin-fixed, paraffin-embedded sections from routine paraffin blocks (mib-1) or the adrenal tissue microarray (TOP2A) using the avidin-biotin complex method.35Sheibani K Tubbs RR Enzyme immunohistochemistry: technical aspects.Semin Diagn Pathol. 1984; 1: 235-250PubMed Google Scholar The following antibodies, dilutions, and pretreatment conditions were used: anti-human Ki-67, mib-1 antibody (DAKO, Carpinteria, CA), 1:100 dilution, Tris-ethylenediaminetetraacetic acid, microwave, pH 9.0; anti-human TOP2A, clone 3F6 (Novocastra Laboratories, Newcastle, UK), 1:40 dilution, citric acid, microwave, pH 6.0; and anti-human α-inhibin (Serotec, Raleigh, NC), prediluted antibody, Tris-ethylenediaminetetraacetic acid, microwave, pH 9.0. The mib-1 and TOP2A immunostains were evaluated by counting the number of mib-1 or TOP2A immunoreactive tumor nuclei and the total number of tumor nuclei. The results for both immunostains were recorded as the percentage of immunoreactive tumor nuclei/total tumor nuclei. Tonsil with germinal centers was used as a positive control and negative controls were performed with no primary antibody. cDNA was synthesized from 1 μg of total RNA using a first strand synthesis kit for RT-PCR (Retroscript; Ambion, Austin, TX) and poly(dT) primers. The relative abundance of IGF2 transcripts was assessed using the 5′ fluorogenic nuclease assay to perform real-time Q-PCR.36Heid CA Stevens J Livak KJ Williams PM Real time quantitative PCR.Genome Res. 1996; 6: 986-994Crossref PubMed Scopus (5040) Google Scholar IGF2 primers (forward 5″-CCGTGCTTCCGGACAACT-3, reverse 5′-GGACTGCTTCCAGGTGTCATATT-3′) and a fluorogenic probe (5′-CCCCAGATACCCCGTGGGCAA-3′) were designed using the Primer Express software package (Applied Biosystems, Foster City, CA) and obtained from Applied Biosystems. The sequence of the IGF2 amplicon was compared to reported genomic sequences using the BLAST program to assure the amplicon was unique. The primers and probe set were optimized with respect to MgCl2 concentration and time and temperature of the hybridization step. Multiplex Q-PCR using a SmartCycler (Cepheid, Sunnyvale, CA) was performed in 30-μl reactions consisting of 1× Q-PCR Supermix-UDG reaction mix (Life Technologies, Inc., Gaithersburg, MD) supplemented with the appropriate MgCl2 concentration. Relative expression of mRNA for IGF2 was calculated using the comparative CT method as described37Collins C Rommens JM Kowbel D Godfrey T Tanner M Hwang SI Polikoff D Nonet G Cochran J Myambo K Jay KE Froula J Cloutier T Kuo WL Yaswen P Dairkee S Giovanola J Hutchinson JB Isola J Kallioniemi OP Palazzolo M Martin C Ericsson C Pinkel D Albertson DL Li WB Gray JW Positional cloning of ZNF217 and NABC1: genes amplified at 20q13.2 and overexpressed in breast carcinoma.Proc Natl Acad Sci USA. 1998; 95: 8703-8708Crossref PubMed Scopus (275) Google Scholar using the CT of GAPDH as the reference. Transcriptional profiles of 11 ACCs (nine primary and two metastatic), 4 ACAs, 3 NCs, and 1 macronodular hyperplasia (MNH) were generated using oligonucleotide arrays with 12,625 probe sets interrogating ∼10,500 genes. To provide visual representations of the tissues and tumors based on gene expression, we used principal component analysis (PCA) to locate the two-dimensional views that capture the greatest amount of variability in the data, using several thousand of the most variable probe sets. The resulting PCA view showed a clear separation between the ACC and benign cohorts (Figure 1A). The greatest variability was seen within the ACC cohort, with two tumors, ACC19 and ACC14, located far from the remaining ACCs. One of these tumors (ACC19) was a high-grade ACC with little morphological (Figure 2) or transcriptional evidence of adrenocortical differentiation, although the tumor was focally and weakly positive for α-inhibin (not shown). The other outlying tumor, ACC14, was a myxoid variant of ACC, a rare but recognized variant38Brown FM Gaffey TA Wold LE Lloyd RV Myxoid neoplasms of the adrenal cortex: a rare histologic variant.Am J Surg Pathol. 2000; 24: 396-401Crossref PubMed Scopus (66) Google Scholar (Figure 2). Interestingly, the ACC closest to the NC/ACA cohort, ACC13, was a low-grade ACC2Weiss LM Comparative histologic study of 43 metastasizing and nonmetastasizing adrenocortical tumors.Am J Surg Pathol. 1984; 8: 163-169Crossref PubMed Scopus (779) Google Scholar that shared some histological features with the benign cohort (Figure 2). A typical high-grade ACC (ACC17) is also shown for comparison in Figure 2. The PCA view showed little differences between the NC and the ACA/MNH cohorts (Figure 1A), a not entirely unexpected finding based on their similar morphologies.Figure 2Histology of typical ACC and the three outlying tumors identified by PCA. ACC13 is a well-differentiated or low-grade ACC. ACC14 is a myxoid variant of ACC. ACC19 is a poorly differentiated ACC. ACC17 is a typical high-grade ACC. H&E; original magnification, ×200.View Large Image Figure ViewerDownload Hi-res image Download (PPT) In addition to PCA, we used hierarchical clustering to generate another visual relationship between the adrenocortical tissues and tumors. The resulting dendrogram (Figure 1B) placed 10 of 11 ACCs on the same branch, with the low-grade tumor ACC13 on the branch with the NCs, ACAs, and MNH, which likely reflects its well-differentiated nature. However, the dendrogram clearly showed a difference between ACC13 and the NC/ACA cohort. With respect to this cohort, the hierarchical clustering was in agreement with the PCA analysis (Figure 1A), as the dendrogram (Figure 1B) showed minimal differences within the NC/ACA cohort. One of the main interests in this study was to identify transcripts of greater abundance in ACC samples than in ACA/MNH and NC samples. A simple one-way analysis of variance using the entire data set of 12,625 probe sets on these three groups of samples obtained 677 probe sets that gave P < 0.01 when comparing ACCs with ACAs, and 473 when ACCs were compared with NCs. These numbers of probe sets are several times larger than the number expected by chance alone. To highlight those differences of potentially greatest biological interest, as well as to reduce the fraction of false-positive findings, we further required that expression values in the ACCs be at least twofold different from the average value in the other group, and that the probe sets meet these requirements in both comparisons (ACC versus NC, ACC versus ACA). One hundred fifty-eight probe sets met these criteria, and for 91 probe sets the fold changes were greater than three for both comparisons. Randomly permuting the sample labels 10,000 times, on average less than one probe set met the final criterion of P < 0.01 and fold change larger than three for both comparisons, and no permutations gave more than the observed number of 91 probe sets. This permutation result shows that very few of the selected 91 genes have been identified because of chance variation alone. A PCA view of the data for these 91 probe sets (Figure 1C), similar to the previous PCA view based on thousands of variable probe sets (Figure 1A), showed a greater relative distance between the ACC and benign cohorts, as expected because of the method of probe set selection. Interestingly, there was persistent intragroup variability