Leiomyosarcoma is a malignant mesenchymal neoplasm of smooth muscle origin that is characterized by a wide spectrum of histopathologic features and clinical behavior. Leiomyosarcoma has increased in incidence from 10% to 31% of all sarcomas over the last 10 years. Because of the unpredictable clinical behavior and the lack of objective surrogate markers for its evaluation, progress in the management of these tumors has been minimal. Although traditional clinical and pathologic parameters remain the basis for evaluating these tumors, predicting the response and biologic outcome within the different histopathologic grades is difficult. Surrogate endpoint markers for stratifying patients within histopathologic categories therefore are needed.
Despite efforts to characterize these neoplasms, an unequivocal genetic fingerprint for leiomyosarcoma currently is lacking. General comparative genomic hybridization studies have been performed in an attempt to establish characteristic hallmarks for the differential diagnosis.1 Although these studies have provided global information regarding the alterations of the chromosomal regions, the information is insufficient for establishing precise signature profiles.
Similarly, the identification of mutated genes and fusion products in certain sarcomas has aided in the understanding of their development and diagnosis, but to our knowledge little has been achieved with regard to leiomyosarcoma because of its heterogeneous pathology and unpredictable clinical behavior. Some of the molecular abnormalities observed in sarcomas include mdm2 amplification, p53 gene mutations in adult soft tissue sarcomas,2 and mutations of c-kit in gastrointestinal stromal tumors.3 Establishing a molecular signature for the different sarcoma subtypes would help resolve these issues, but the feasibility of characterizing malignant tissues based on genetic molecular profiling rests on the critical assumption that intratumor heterogeneity is minimal.
An advantage of high-throughput genomic analysis is the power to classify neoplasms biologically on the basis of their genetic signatures. In that context, gene expression profiling of alveolar rhabdomyosarcoma already has been conducted using cDNA microarray technology,4 employing multidimensional scaling (MDS)5 for the analysis. We have used a similar genomic and computational approach to assess the degree of intratumor heterogeneity of leiomyosarcoma as a necessary prerequisite to identifying its molecular portrait.
In the current study, we used cDNA microarrays6, 7 to investigate gene expression profiles in tumor as well as normal tissue samples obtained from three patients. We also performed parallel gene expression profiling of peripheral and core sections from the tumor of one of the patients. The computational analysis of the data strongly suggests that the intratumor heterogeneity in leiomyosarcoma is minimal and that the tumor specific expression profile is regionally consistent in these tumors. In addition, we identified genes that were expressed differentially in tumor and normal tissue.
MATERIALS AND METHODS
Fresh tissue specimens from three patients with primary leiomyosarcoma were submitted to the Department of Pathology at the University of Texas M. D. Anderson Cancer Center and constitute the material for the current study. Tissues harvested for the analysis were snap-frozen shortly after surgical removal and stored at –80 °C. Frozen sections from all tissue specimens for cDNA array analysis were prepared and hematoxylin and eosin stained slides were evaluated by a pathologist (A.E.-N.). Macrodissection of normal and necrotic areas was performed for enrichment of tumor cells. Patients provided signed informed consent according to the institutional review board guidelines.
The microarray slides used in the current study were produced by the Cancer Genomics Core Laboratory at the M. D. Anderson Cancer Center. The slides contained 2304 sequence-verified cDNA clones8 printed in duplicate on the array for quality control purposes. Many of the clones have known functions in cell-cycle control, apoptosis regulation, DNA repair, metabolism, cell-cell adhesion, and communication.
Matched frozen tumor and normal specimens from Patients 1 and 2 were processed for RNA extraction. Four spatially different peripheral samples, one central sample, and a normal sample were obtained from Patient 3. A macroscopically and histologically normal specimen of gastric mucosa was used as a constitutional normal control. The tissues were ground to powder under frozen conditions and lysed in the lysis buffer TRI (Molecular Research Center, Inc., Cincinnati, OH). After this procedure, the tissue/TRI reagent mixture was homogenized using a power homogenizer. After the RNA was extracted, a 1% MOPS gel was run to confirm its quality.
Target Preparation and Hybridization
RNA samples were subjected to an amplification protocol before hybridization on microarray slides. Approximately 4–10 μg of total RNA was annealed to an oligo dT25-T7 primer (5′- AAA CGA CGG CCA GTG AAT TGT AAT ACG ACT CAC TAT AGG GCG ATT-3′) and a up-primer (5′- AAG CAG TGG TAA CAA CGC AGG GAC CGGG-3′). Reverse transcription to cDNA then was performed at 42 °C for 2 hours using Superscript II enzyme (Gibco, Grand Island, NY). RNase H was added and the sample was incubated at 37 °C for 15 minutes to remove the RNA template. Three cycles of polymerase chain reaction (PCR) then were performed to obtain double-stranded cDNA. The PCR product was purified on a PCR purification column (Qiagen, Chatsworth, CA) and reduced to half using a Speed Vac (Savant Instrumetns, Holbrook, NY). An in vitro transcription reaction (T7 Megascript; Ambion, Austin, TX) then was performed at 37 °C for 2 hours to amplify the product. DNase I was added to remove the DNA template and an Rneasy minicolumn (Qiagen) then was used for RNA cleanup. The purified product was eluted in 50 μL of water and reduced to 9 μL in the Speed Vac. Next, the product was prepared for labeling with a fluorescent dye by annealing it with a random hexamer at 70 °C for 10 minutes. A labeling reverse transcription reaction was performed at 42 °C for 2 hours during which either Cy3 or Cy5-dCTP was added. The RNA template was removed by alkaline hydrolysis at 65 °C for 1 hour. The final target product was purified with a MicroSpin G-50 column (Amersham Pharmacia, Arlington Heights, IL) and then the volume was reduced to 10 μL with a Speed Vac. Cy3 and Cy5 targets thus prepared then were mixed together for cohybridization. Express Hyb (ClonTech, Palo Alto, CA) and blocking reagents containing 8 μg of polydA40-60 (Amersham), 2 μg of yeast tRNA (Gibco-BRL), and 10 μg of human Cot I DNA (Gibco-BRL) were added. The target mixture was warmed at 60 °C for 1 hour and spun down. It then was applied to the slide and covered with a glass coverslip. The slides were incubated at 60 °C overnight and then washed at 37 °C in 1X standard saline citrate (SSC) (3M sodium chloride and 0.3M sodium citrate) 0.01% sodium dodecyl sulfate (SDS) once, 0.2X SSC 0.01% SDS once, and then 0.1X SSC twice. They were spun at 400 × g for 1 minute to dry.
Imaging and Data Analysis
The hybridized slides were scanned with a GeneTAC LS IV laser scanner (Genomic Solutions, Ann Arbor, MI) and the obtained signal intensities were quantified with ArrayVision (Imaging Research Inc., St. Catherines, Ontario, Canada). The log-transformed background-subtracted signal intensities in the current study were standardized by subtracting the sample mean and then dividing the difference by the sample standard deviation of each array. For the data analysis, we employed multidimensional scaling (MDS) as well as hierarchical tree clustering. For identifying the differentially expressed genes, a method9 developed by Baggerly et al. was used. Briefly, the arrays used in the current study include duplicate spots for each gene of interest. Experimental evidence suggests that within each channel, the variability of the difference in log intensities is a decreasing function of the mean log intensity. This observation is exploited by fitting a nonparametric smooth curve to the data and using it as an estimate of the standard deviation. The estimates from both channels are pooled to produce an overall estimate of standard deviation, which is used to compute t statistics for each gene on the array. The genes then are ranked using the absolute value of the t statistic so that the most highly ranked genes are the ones that are most likely to exhibit differential expression.
RESULTS AND DISCUSSION
The main objective of the current study was to determine the degree to which tumor heterogeneity affects the gene expression profiles of leiomyosarcoma. Although certain single gene approaches (such as c-kit) may provide some useful information regarding the biology of these tumors, they are inadequate for assessing the entire spectrum of genetic events associated with highly heterogeneous solid tumors that are expected to exhibit complex genetic changes. This is due to the wide variability at the level of a single gene, rendering the information less reliable than profiling the gene expression level of groups of genes. This is especially relevant in the case of large leiomyosarcomas with the possibility of regional heterogeneity that may profoundly influence the gene expression profiling of these tumors. Thus, it was crucial to determine first whether dissimilarities between different parts of the same tumor have an impact on the determination of a tumor specific signature from a random single tumor specimen. To gain insight into this issue and to assess whether we should perform an expanded gene expression profiling project for leiomyosarcoma, we first collected three leiomyosarcoma specimens from three different patients. Second, we harvested tissues from four different peripheral regions and one central specimen from the tumor from one patient. In addition, fragments of normal gastric mucosa and spleen that were removed as part of the surgical procedure were collected and used as the normal controls. All the tissues were evaluated macroscopically and microscopically by a pathologist to confirm the nature and quality of the specimens.
Among the three normal tissue specimens, the one obtained from Patient 3 (N3) was a smooth muscle specimen from the gastric wall, which we considered the most appropriate control for leiomyosarcomas. The other two normal tissue specimens were mostly normal epithelial and lymphoid cells, and therefore were not considered as the ideal control in a search for genes that were expressed differentially between normal and tumor tissues.
Our strategy to test regional heterogeneity from different peripheral and central regions was based on the possibility that peripheral clonal differences at the leading edges and the inefficient vascularity leading to a hypoxic state may affect gene expression. If hypoxia and clonal heterogeneity did indeed result in marked differences between core and peripheral regions, a strategy for sampling these tumors for gene expression analysis that accounted for these factors would have to be developed.
After extracting total RNA and confirming its quality, we proceeded with the microarray experiments. For each microarray experiment, matched samples for comparison were labeled with Cy5 or Cy3, and cohybridized to the microarray slides. To determine gene expression ratios, we cohybridized the following fluorescently labeled pairs: P1/N3, P2/N3, P3/N3, P4/N3, C/N3, T1/N1, T2/N2, T1/N3, and T2/N3 (P1-P4 were peripheral samples from Patient 3, C was a core sample from Patient 3, N1-N3 were normal tissues from Patients 1–3, and T1-T2 were tumor tissues from Patients 1 and 2).
Currently, data from microarray experiments are presented as ratios, or as log-transformed ratios, of test samples against a common reference sample that functions as an internal standard to avoid gross errors caused by uneven hybridization. In addition, using ratios accomplishes a sort of self-normalization, because extraneous factors not related to gene expression levels may affect both channels equally. Signal intensity ratios, which are used to quantify the relative expression of one sample with respect to the other, are widely preferred for the gene expression profiling of differentially expressed genes.10
Figure 1 shows the two-dimensional MDS solution containing nine tissues, each one cohybridized with a common reference (N3), in which the Euclidean distance was calculated for each pair of standardized log-ratios. It can be seen readily from Figure 1 that the peripheral and core sections from Patient 3 exhibited a substantially higher degree of similarity to each other than to the other tissues. As expected, we observed a substantial difference between normal control and tumor tissues.
An alternative way to visualize clusters is to construct a dendrogram in which links between objects correspond to the distances between them, with objects being joined into successively larger clusters. Figure 2 shows a dendrogram of the hierarchical cluster tree corresponding to the nine tissues displayed in Figure 1. The metric used for this analysis was the incremental sum of squares (i.e., the increase in the total within-group sum of squares as a result of joining two groups). The within-group sum of squares of a cluster is defined as the sum of the squares of the distances between all objects in the cluster and the centroid of the cluster. In keeping with the MDS findings, the peripheral and core sections of Patient 3 formed a rather tight cluster compared with the other four tissues.
It also was important to compare the variability of gene expression in peripheral and core tissue specimens from Patient 3 with the typical variability observed for replicates of single tissues. This allows for the assessment of other nonexperimental factors that may have an impact on the analysis. That is to say, if the variability exhibited between the different tumor sections proved to be considerably higher than that between replicates of a single tissue, this would imply that factors other than experimental conditions led to the observed variation.
To test the reproducibility of our gene expression measurements, we performed several experiments in which a tumor sample from each patient was cohybridized with a matched normal tissue sample from the same patient. This was repeated three times using the same RNA samples, for two different patients. For example, from Patient 1, T1 was cohybridized with N1, the latter being the reference channel, for a total of three replicates (similarly, for T2 and N2).
We once again used the Euclidean distance to compute the proximities between standardized samples corresponding to the three replicates from Patient 1, Patient 2, and the peripheral and core sections from Patient 3 and, as before, MDS was performed to visualize proximity (Fig. 3). An informal inspection of this figure suggests that the scatter of the cluster containing different tumor sections from Patient 3 is comparable to the scatter of the clusters containing replicate samples from Patients 1 and 2. Specifically, the scatter of a cluster, which in our case is produced by the two-dimensional MDS solution, was defined as the sum of the eigenvalues of the computed covariance matrix for that cluster,11 which is equivalent to the trace of the covariance matrix. The computed scatter was 0.26 for the cluster of T1 samples, 0.05 for the T2 samples, and 0.17 for the peripheral/core samples. From this, we can conclude that the variability exhibited by the different tumor sections was well within the observed experimental variability of replicate experiments.
The common genes that were expressed differentially in the four peripheral and normal samples were identified and some of them are listed in Table 1. For each of the four peripheral samples, we considered the top 100 differentially expressed genes. Among those, 65 genes were common in all four samples. Figure 4 shows a segment of the microarray images corresponding to the cohybridization of the four peripheral (green) and normal (red) samples.
Table 1. Common Differentially Expressed Genes in the Four Peripheral Tumor Sections and Normal Tissue
Differentially expressed genes
Bold type indicates high expression in peripheral tumor compared with normal tissue.
Italic type indicates low expression in peripheral tumor compared with normal tissue.
Platelet-derived growth factor receptor-β
Immunoglobulin superfamily Leucine-rich repeat
Collagen-binding protein 1
Protein gene product 9.5
HREV 107-like protein
Rearranged immunoglobulin lamda light chain
As discussed previously, the MDS plot in Figure 1 demonstrates that the peripheral and the core sections from Patient 3 exhibited a high degree of concordance with regard to the global gene expression. We also analyzed the differential gene expression of the peripheral and core tumor specimens from Patient 3 and the resulting differential gene expression profile is given in Table 2. Among the genes that were expressed differentially between tumor and matched normal tissues, in both the peripheral and core materials, those encoding the metallothioneins (MT) were predominant. MTs are cysteine-rich proteins that have metalloregulatory functions, and hence they play an important role in the detoxification of heavy metals.12, 13 Studies have shown that overexpression of MT has been associated with aggressive biologic behavior and also may be involved in blocking apoptosis in some cell systems.14 It is interesting to note that MT expression was suppressed in leiomyosarcomas (Table 1), suggesting the existence of a tissue-dependent phenomenon. Similarly, the level of MT in renal carcinoma tissues was found to be consistently lower compared with that in surrounding tissues.15 We also observed that the Tob gene was suppressed in the fast-proliferating peripheral section of the tumor (Table 1). Tob is a member of the gene family with antiproliferative function and its overexpression results in the suppression of cell proliferation.16
Table 2. Genes Expressed Differentially in the Peripheral and Core Tissue Specimens
Differentially expressed genes
Bold type indicates high expression in peripheral tumor tissue.
Italic type indicates high expression in core tumor tissue.
Platelet-derived growth factor receptor-β
Immunoglobulin superfamily Leucine-rich repeat
Regulator of G-protein signaling 1
Nerve growth factor 1
M-phase inducer phosphatase
HREV 107-like protein
Similarly, overexpression of the platelet-derived growth factor (PDGF) receptor, a tyrosine kinase, also was more pronounced in the peripheral samples. It is interesting to note that the PDGF receptor displays extensive structural homology with c-kit. Both are tyrosine kinases that are targets for the drug STI-571, which currently is being tested in clinical trials for the management of smooth muscle tumors. Overexpression of PDGF receptor also has been identified in cervical carcinoma tissues17 and is an indicator of angiogenic potential in human gliomas.18
Genes encoding lysozyme and lymphotoxin-α that demonstrated moderate suppression in the peripheral samples also are known to have antitumor functions. Cathepsin E, an intracellular aspartic proteinase, consistently was found to be suppressed in the peripheral sections of the tumor (Table 1). The implications of this finding are not clear, because cathepsin E has been found to be present at the advancing margin of carcinoma tissues and may play an important role in tumor invasion and subsequent metastasis.
The results of the current study examining the regional differences in gene expression showed that the intratumor heterogeneity in leiomyosarcoma is insignificant from the standpoint of genetically profiling these tumors. This observation, in turn, permits one to establish an objective molecular model for leiomyosarcomas regardless of tumor size. When a large collection of leiomyosarcomas has been profiled, we hope to be able to identify profiles that are related clinically and, in the process, identify strong feature genes that are of diagnostic or prognostic importance for leiomyosarcomas. In addition, these expression profiles may identify therapeutic targets that are significant in the treatment of soft tissue sarcomas.