In order to ensure the propagation of tumor epithelium without apparent contamination with fibroblasts and other cellular components, previously described methods including the selective release of nests of malignant cells from the connective tissue, were used [2
]. As illustrated in Figure , the characteristic epithelial morphology, and microscopic heterogeneity between tumor cultures from independent cases was observed. To confirm that these cultures represented pure populations of epithelial cells, indirect cytokeratin immunostaining was performed (Figure ). Clinical information and additional cell culture details are listed in Table .
Figure 1 A. Microscopic phenotype of primary tumor tissue and corresponding tumor-derived epithelial and stromal cell cultures. (1–3) – H & E-stained frozen sections showing histology of representative cases processed for RNA isolation (more ...)
Clinical characteristics of Primary Tumors used for Gene Expression Analysis
Concordant aspects of global gene expression in breast cancer tissue and tumor-derived cultures
Seventeen RNA samples were analyzed by cDNA microarrays. These were comprised of 4 cases of primary breast tumor tissue and 1–2 matched tumor cultures, 2 additional unmatched tumor cultures, and 4 immortal breast cell lines. Cell lines were selected to represent estrogen receptor (ER) positive (T47D) and ER negative (SKBR3, BT20) tumors, and non-cancerous breast epithelium (ENUt7, ref. [13
Unsupervised clustering analysis of 7362 clones (4743 unique genes/ESTs) grouped the samples into three separate clusters representing immortal cell lines, tumor tissue, and tumor cultures. Figure displays a cluster of genes, which were over expressed in immortal cell lines but under expressed in tumor tissue and tumor cultures, while Figure displays a 'mirror image' gene expression pattern (under expressed genes in immortal cell lines, which were over expressed in tumor tissue and tumor cultures). The clusters selected for illustration represent ~90% gene correlation. The entire data set displayed additional clusters [Figure S1 – see http://genome-www.stanford.edu/breast_cancer/PTCC/].
Comparisons of large data sets as described above frequently result in "significant" patterns of gene expression by chance alone. We employed the Significance Analysis of Microarrays (SAM) for independently verifying genes that are differentially expressed between classes or groups of samples. SAM analysis identified 930 clones, representing 681 unique genes/ESTs, whose expression was significantly (>2-fold) different between group 1 comprised of immortal cell lines, and group 2 comprised of tumor tissue and tumor cultures (0.05% false discovery rate). A full list of the differentially expressed genes is provided in [Table S2 – see http://genome-www.stanford.edu/breast_cancer/PTCC/]. As expected on the basis of the similarity in gene expression observed between tumor tissue and tumor cultures in the unsupervised array data (Figure ), these clusters were also present in the SAM profile (Figure ). As shown in Figure , genes upregulated in immortal cell lines (indicated by red vertical bar) reflecting significantly shorter doubling times (for example, RFC4
) were primarily those associated with the 'proliferation' cluster described by Ross et al [14
]. In contrast, upregulated transcripts in tumor tissue and tumor cultures (Figure , indicated by green vertical bar) included genes involved in epithelial differentiation (MAL, MAFB, RUNX1, KRT5
), in the induction of apoptosis (DAPK1
), and in tumor angiogenesis, and extravasation (SPARC
) (Gene Ontology – GO annotations, ref [15
Primary epithelial cell cultures, in contrast to fibroblasts and rapidly growing cell lines, undergo rapid growth arrest in 10% fetal calf serum (FCS). This is why we, and others have propagated cultures of primary tumor epithelium in 0–5% FCS (1–9). Immortal cell lines, however, grow optimally in 10% FCS; lower concentrations retard growth. Therefore, to optimize growth conditions for both, 2% FCS was chosen for primary tumor cultures and 10% FCS for cell lines. It is conceivable that an increased concentration of FCS may account for differences in gene expression between primary tumor cultures and immortalized cell lines. In this case, one would expect increased proliferation in the primary tumor cultures at 10% FCS, when in fact, growth is severely inhibited by this approach.
As summarized in Table , based on microarray expression data of 38,999 cDNA clones, the average correlation between matched tumor tissue and short-term tumor cultures (7 sets) was 0.41 (sd = 0.03) in contrast to 0.10 (sd = 0.09) between tumor tissue and immortal cell lines (16 pairs). These correlation coefficients differed significantly in a Mann-Whitney rank sum test (p = 0.007) demonstrating that the gene expression profile of tumor tissue was more consistent with that of the corresponding tumor culture than it was with any of the immortal cell lines tested.
Pair wise correlations between primary tumors, matched epithelial cultures, and immortal breast epithelial cell lines
Determinants of replicative arrest in primary breast tumor-derived cultures
Towards the determination of specific molecular changes underlying the finite proliferative lifespan in tumor cultures, gene expression analysis, by QRT-PCR, was conducted on an expanded set of 39 samples comprised of multiple early passage epithelial cultures obtained from 16 primary breast cancers, 8 cases of normal breast epithelial organoids and, 12 immortal breast epithelial cell lines. As noted above, since genes in the proliferation cluster displayed minimal expression in primary tumor cultures, first we considered the possibility of growth arrest due to a lack of telomerase activity and subsequent telomeric attrition. The relative expression of hTERC and hTERT subunits of telomerase, encoding the structural RNA component, and the component with reverse transcriptase activity, respectively, was measured. While commonly used immortal cell lines (T47D, MDA231) displayed several-fold higher transcript levels for hTERT, undetectable to minimal levels were observed in 6/8 independent primary tumor cultures (1599T, 713T, 1569T, 1570T, 1617T, 1620T). The primary tumor culture with the highest relative expression of hTERT (257T) has developed into an immortal cell line (Figure ).
Figure 2 A. QRT-PCR analysis of hTERT and hTERC in primary tumor cultures compared to levels in normal breast organoids, matched fibroblasts, and immortal breast epithelial cell lines (T47D, MDA231, and ENUt7). The Y-axis is minimized to display the range of relative (more ...)
To confirm the functional impact of hTERC and hTERT down regulation, telomerase activity was measured directly by the TRAP assay. As expected, primary cultures with detectable transcript levels, and immortal cell lines of cancerous and non-cancerous origin displayed significant telomerase activity, while those tumor cultures that did not display gene expression showed no activity. In the primary tumor sample, 257T, robust telomerase activity was found as early as passage 8. Similarly, telomerase activity was detectable in early passage epithelial cultures propagated from cells isolated by mechanical dissociation (SP – spillage), or enzymatic digestion (DIG) of the tumor sample 054T (Figure ).
In the next step towards identifying the determinants of replicative arrest, primary tumor cultures were compared with immortal cell lines for relative expression of genes associated with the negative regulation of the cell cycle in general, and with epithelial cell proliferation in particular. This analysis included 9 candidate genes in 3 signaling pathways: (1) members of the CIP/KIP family of cyclin-dependent kinase inhibitors (CDKIs), p21CIP1/WAF1, p27KIP1, and p57KIP2 (2) members of the INK family of CDKIs, p15INK4B, and p16INK4A (3) members of the TGF-β family, TGFβI, TGFβII and the signaling receptors, TβRI, and TβRII. As illustrated in Figure , we observed that in 10/12 breast cancer cell lines, p21CIP1/WAF1 levels were 2 to138 fold lower than steady state levels in normal breast epithelium. In contrast, 4/20 primary tumor culture samples showed this range of p21CIP1/WAF1 expression (p = 0.0003). For the CDKIs, p27KIP1, and p57KIP2, several fold decrease in gene expression was observed in all primary tumor cultures and most immortal cell lines as well (p = 0.04 and 0.69, respectively). Similarly, no significant differences were apparent for p15INK4B, and p16INK4A gene expression in the two groups (p = 0.07 and 0.39, respectively). For members of the TGF-β family, while significant differences were not observed in the expression of TGFβI and TβRI, a median increase of 2 fold and 4 fold were found in the expression of TGFβII and TβRII respectvely. In contrast, immortal cell lines, showed a median decrease of 7-fold for TGFβII and 4-fold for TβRII (p = 0.0035 and 0.0011, respectively). All p values were derived by the Mann-Whitney test.
Figure 3 Comparative QRT-PCR analysis of genes encoding negative regulators of cell proliferation in primary tumor cultures and immortal cell lines. A – Expression levels of individual genes represented as fold increase or decrease over gene expression (more ...)
We used MANOVA to compare expression of the 9 above-mentioned genes in immortal cell lines, primary tumor cultures, and normal breast epithelium. It was apparent that the multivariate means for the 9 genes differed for each of the 3 groups above and that both immortal lines and primary tumor cultures differed from normal breast epithelium (p = 0.0001) as also depicted in the hierarchical clustering dendrogram of these samples (Figure ). A 3-D plot of the first three principal components, which together account for 83% of the total variation in expression in the 9-dimensional gene space, is shown in Figure . This display format demonstrates that the three types of cell samples cluster in different parts of the three dimensional space and that primary tumor cultures and immortal cell lines are significantly different from normal breast and from each other in the expression of the negative growth regulators evaluated here (p = 0.0002). The relatively large Wilks' lamda (0.271) for immortal cell lines vs. primary tumor cultures is most likely due to the greater spread of these groups in the 9-dimensional gene space, while normal samples are tightly grouped together. This finding may be related to the heterogeneity between individual tumors.
Overall, the genes in the array-based clusters, shown in Figure and , could be categorized as positive or negative regulators of proliferation respectively according to GO annotations. Genes such as, PCNA, CKS1B, TPX2, UBE2C, CDC6, confirmed by the statistically significant analysis of microarray data, SAM, were among the positive proliferation genes differentially expressed by immortal cell lines, and matched tumor tissue/cell culture samples (Figure , also see Table S2 – http://genome-www.stanford.edu/breast_cancer/PTCC/, for full list). Expression levels of negative proliferation genes identified in the SAM data, such as, TβRII, and CDKN1A (p21CIP1/WAF1) were confirmed by QRT-PCR (Figure )