| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
Regular Articles |
1 Department of Biological Chemistry, Silberman Institute of Life Sciences’ Hebrew University, Jerusalem, Israel; 2 Bioinformatics Centre, University of Copenhagen, Copenhagen, Denmark; and 3 National Cancer Institute, NIH, Bethesda, Maryland
 |
ABSTRACT |
|---|
 Top ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
|
INTRODUCTION |
|---|
|
TopABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
What might be the basis of this tissue-of-origin determined resistance to chemotherapy? It has been argued that, to a large extent, it is an inherent property of the tissue from which the cancer derives. That so few exceptions to this tissue-dependence can be noted, such as the renal-derived Wilm’s tumor of childhood, supports this argument. The sensitive or resistance phenotype extends to metastatic tumors as well. For example, a testicular metastasis in the lung will respond to chemotherapy, whereas a pancreatic metastasis in the same site will not (6 , 7, 8, 9) . The range of tumor types from sensitive to resistant can be considered as though they were arrayed along an axis. We asked whether alterations in gene expression patterns could be identified based on this axis of sensitivity or resistance, termed here the "axis of intractability."
The advent of genome-wide mRNA and protein expression profiling techniques (10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20) , and the Serial Analysis of Gene Expression (SAGE) technique (10 , 21, 22, 23) , open a valuable new route to investigating drug resistance, or intractability, and drug sensitivity. Many cellular factors, such as the expression of drug-pump systems, apoptosis-regulating cascades, and growth factors and their receptors, impinge on drug sensitivity (24) .
In a similar approach, the genomic array survey of the 60 cancer cell lines of the National Cancer Institute drug screen (25) successfully assembled these cell lines into families of similar gene expression, and these families reflected the tissue of origin of the cancer. This implies that a set of tissue of origin-specific genes can be identified for each tumor type. The paradigm for the present study was, thus, to relate the gene expression profile for a given tumor type to its position on the axis of intractability, and to identify differential gene expression patterns based on the sensitivity or resistance of that tumor type, with the goal of establishing new insights into the molecular basis of tumor intractability to chemotherapy.
|
MATERIALS AND METHODS |
|---|
|
TopABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
lists the libraries that were used in this part of the present study. In each library, the occurrence of any tag can be normalized to a value of 200,000 total tags. In cases in which data on microdissected tumor was available, such libraries were chosen.
|
The second database used was the Surveillance, Epidemiology and End-Result (SEER) database, from which we used, in particular, the SEER Cancer Statistics Review,7 which contains the SEER Incidence, United States Mortality, and 5-Year Relative Survival Rates for the period 1973–1999 (26) . In particular, we used the 5-year survival data for patients presenting with tumors that were diagnosed as "distant." The SEER database defines distant as having distant metastases that "are tumor cells that have broken away from the primary tumor, have traveled to other parts of the body, and have begun to grow at the new location. Distant stage is also called remote, diffuse, disseminated, metastatic, or secondary disease."8
We entered the SEER database tissue by tissue and recorded the 5-year survival data listed as "Distant," "All Races," "Total." As we describe later, we use these data as a surrogate measure of the intractability of the tumors arising from these various tissues.
We used the SAGE database in two ways: (a) to explore the relationship between gene expression as a function of the state of the cells of a particular tissue (whether present as cells in culture or as a solid tumor); and (b) by correlating the SEER database with the SAGE databases for solid tumors, to explore any possible relationship between the gene expression of a tumor type and the survival of patients with distant tumors derived from that tissue. The correlation between SAGE tag numbers for the various tumor tissues and the SEER 5-year survival of patients with distant tumors of the same tissue type was performed using a download of part of the SAGE database9 as noted in the text, and then selecting for the 14 libraries of tumor tissues for which we had SEER data. The correlations were calculated on an Excel worksheet.10 We listed genes showing correlations having a probability of P < 0.01 (or, in some cases, P < 0.005) of being zero.
To discern the "themes" suggested by the sets of genes that we identified using the above tools, we used the EASE program11 found on the DAVID (Database for the Annotation, Visualization and Integrated Discovery) website. This program was developed by the Laboratory of Immunopathogenesis and Bioinformatics at SAIC-Frederick, Inc., for the National Institute of Allergy and Infectious Diseases of the NIH. Using this program, we entered the lists of genes that we had identified as being significantly over- or underexpressed in the comparison between solid tumors and cell lines, or that had been significantly correlated with 5-year survival of patients presenting with distant tumors. The program reports the sets of genes that share a common theme in terms of their biological role, as annotated by various gene ontology and gene mining databases. The statistical validity of the various themes is calculated by EASE by comparing two fractions: (a) the set of genes in the list that have the given theme compared with all of the genes on the given list of identified genes; and (b) all of the genes in the genome with this theme compared with all of the genes of the entire genome.
|
RESULTS |
|---|
|
TopABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
, depicting on a logarithmic scale the mean relative potency of 21 drugs used currently in cancer chemotherapy, when tested on the 60 cell lines of the National Cancer Institute Anticancer Drug Screen Panel (27)
. Excluding the highly sensitive leukemias, the sensitivity of all other cancer cells is seen to be very similar. The average value of the logarithm of the sensitivity for each tissue type differs from the overall mean of these averages by less than 1 SE for all except the leukemias, for which the deviation is more than 2 SEs. These data reflect the well-known phenomenon that cell lines appear to be far more sensitive to cytotoxin treatment, and far less variable in their response, than are cancers in patients. We wondered whether we could find some genetic basis for the difference between solid tumors and cell lines, and eventually between solid tumors in the clinic.
|
lists the libraries that were used in this part of the present study. Using the search tool DGED provided by the SAGE Genie database, as described in "Materials and Methods," the cutoff ratio for the comparisons was set as 2-fold overexpressed or 2-fold underexpressed for each gene, with a concomitant statistical likelihood of P < 0.05. Using these cutoff criteria, the total number of tag differences between various libraries were determined, all normalized to a total of 200,000 tags. For every tumor type (see supplementary material),10
the difference between the solid tumors and the cell lines was greater, often far greater, than that within the set of solid tumors or the set of cell lines. Comparing all of our chosen solid tumors, taken together, with all of the chosen corresponding cell lines, taken together, the number of tags that were 2-fold different (higher or lower) was 697, when the significance was set at P < 0.01.
We obtained, tissue-by-tissue, the ratio of the expression of each tag in the pool of solid tumors to the expression of that tag in the corresponding cell line libraries (Table 1A)
. A ratio of greater than 2 at a significance of P < 0.05 was defined as "overexpressed in solid tumors." Similarly, a ratio of expression greater than 2 (P < 0.05) in a group of cell lines as compared with the corresponding tumors was defined as "overexpressed in cell lines." Tags from the libraries were selected for inclusion in a comprehensive database based on the following arbitrary criteria: (a) the tag had to be expressed at a level of 10 tags per 200,000 tags or greater; and (b) the tag had to be overexpressed in at least 12 of the 16 solid tumor libraries or 11 of the 15 cell line libraries. Criterion 1 was set to reduce the possibility of error, because tags found at a small number of copies could arise from errors in sequencing a large copy number tag with a similar sequence. Criterion 2 was chosen to ensure the tissue generality of our findings, by excluding genes unique to particular tumor types. Sixty-two tags met the criteria for tags (genes) overexpressed in solid tumors, and these are included in Table 2
. The entries in the table are color-coded for ease of comparison. In a similar manner, Table 3
lists the 61 tags (genes) overexpressed in the cell lines, using the above criteria. In both Table 2
and Table 3
, the final column gives, for each gene (i.e., tag) the overall ratio for all of the solid tumors, taken together as a group, to all of the cell lines as a group, or the reverse ratio as indicated, at a significance of P < 0.01.
|
|
(genes overexpressed in solid tumors relative to cell lines) in many cases, a gene was overexpressed more than 100-fold (red-shaded entries) in solid tumors compared with the cell lines established from histologically similar tumors. For many genes, the ratio of expression in all 16 solid tumors to that in all 15 cell lines, was greater than 5. A cursory examination of Table 2
suggests that three molecular groupings (shaded turquoise) encompass many of the genes listed: (a) a group encoding proteins that participate in cellular immune responses; (b) a group participating in the complement pathway; and (c) a group encoding extracellular matrix (ECM) proteins. This observation was confirmed using the EASE program,11
developed by Drs. R. A. Lampicki and D. A. Hosack of the Laboratory of Immunopathogenesis and Bioinformatics, NIH, as described in "Materials and Methods" (28)
. EASE identifies the biomolecular themes present in a list of genes and computes a statistical probability for the correctness of the identified theme. Table 4
lists the themes and relevant probabilities that were assigned by EASE to the genes listed in Tables 2
and 3
.
|
lists the 61 genes, overexpressed in cell lines that satisfied the inclusion criteria described above. In a few cases (red-shaded entries), a gene was more than 100 times overexpressed in cell lines compared with the solid tumors. In many cases, overexpression ratios of 5–100-fold (green and yellow shading) were found. For many genes, when all 16 solid tumors were taken together and compared with the 15 cell lines taken together, overexpression ratios of 2.5-fold or more were found. Inspection of Table 3
suggests many of the genes listed (shaded turquoise) encode proteins involved in protein synthesis. Table 4
lists the themes and relevant probabilities assigned for these genes by EASE. These results confirm that the majority of genes overexpressed in cell lines, as compared with tumors, are involved in protein biosynthesis, including genes that encode proteins involved in ribosome biogenesis.
Fig. 2A
depicts gene expression, represented as SAGE tags/200,000 tags, for 29 proteins involved in protein synthesis. Genes encoding proteins of the small subunit of the ribosome are represented by squares, those encoding large subunit proteins by circles and those involved in protein synthesis, initiation or elongation, by triangles. The average tag number in all of the cell lines in Table 1A
is plotted on the Y axis, against the average tag number in all of the solid tumors in Table 1A
on the X axis. All values are normalized to 200,000 tags. The regression line drawn shows a significant correlation of r = 0.97 and a slope of 2.8. The data suggest that many of the genes involved in protein synthesis are coordinately switched on as the cell lines are established from solid tumors, or during the time that they are maintained for many generations in cell culture. Fig. 2B
depicts, in a similar manner, the genes identified in the theme analysis of Table 4
as involved in the ECM. Here, the Y axis gives the normalized tag number for all of the cell lines taken together, and the X axis gives tag numbers for all of the tumors together. The regression line has a slope of the reciprocal of 8.3, meaning that these ECM genes are coordinately overexpressed in the tumors by an average ratio of some 8-fold. The correlation is significant at r = 0.92. For comparison, the genes identified in the theme analysis as immune-related, have a correlation of r = 0.57, and this is not significantly different from zero. Thus, the immune genes, although overexpressed in the solid tumors, appear not to be coordinately overexpressed (data not shown). Finally, Fig. 3
depicts these data manually rearranged so as to separate the genes overexpressed in solid tumors (upper half of the figure) from those overexpressed in cell lines (lower half). The rearrangement gives a clear dissection of these genes into a cluster of protein synthesis genes that are overexpressed in the cell lines, and a cluster of immune response, cell–cell interaction and ECM genes that are overexpressed in the solid tumors. (These immune response genes are overexpressed in the solid tumors, although this expression, as just discussed, does not seem to be coordinately regulated).
|
|
would be expressed at higher levels in the intact tumors (with their adjacent stromal elements) than in the microdissected samples. This was not the case. Because two microdissected breast tumors were available, these were compared with intact breast tumors. We found that 126 tags were more than 2-fold differentially expressed between the two microdissected breast cancers and two breast cancers (B 95–347 and B 95–259, both invasive ductal cancers), the latter presumably containing stromal elements. Of these 126, only 4 can, however, be found in Table 2
, as being high in solid tumors as compared with cell lines: IGFBP7, IGHG3, APOC1, and CD74. Of these four, two (IGFBP7 and IGHG3) are listed in Table 2
with the comment "may be stromal," derived from the gene ontology web site. Similarly, for prostate cancer, the SAGE libraries contain a single microdissected adenocarcinoma, and two comparable solid tumor samples that had not been dissected. Eighteen tags were differentially expressed between these samples using the criterion of 2-fold differential expression at a significance of P < 0.05, but only two of these are found in Table 2
: IGHG3 and CLU. One of these genes was the same IGHG3 gene found in the breast cancer comparison. The other was CLU, which might also originate from the stroma. Thus at most, 5 of the 62 genes listed in Table 2
(high in solid tumors as compared with cell lines) may be derived from stromal elements.
Correlating Genes Differentially Expressed in the SAGE Analysis with 5-Year Survival As Tabulated in the SEER Database.
Having identified genes differentially expressed in cell lines and solid tumors using the SAGE database, we explored the possibility that some of the genes identified may also predict drug sensitivity in solid tumors. Although a prospective study or a retrospective analysis of banked tumors from a completed clinical trial with response to therapy and survival data would directly address this question, this would require an enormous effort, and we, thus, sought to extend the results using the SEER database. Fig. 4
depicts 5-year survival using information from the SEER database for patients presenting with a variety of tumors that were diagnosed as distant. Because these tumors are unlikely to be amenable to surgical resection, survival reflects the inherent proliferative rate or "biology of the tumor" and, at least in part, their sensitivity to systemic therapy, i.e., chemotherapy. The data on survival range from close to 80% for cancer of the testis to 1% for pancreatic tumors. The figure depicts what we term "the axis of intractability," with pancreatic cancer at one extreme and testicular cancer, the least intractable or most sensitive tumor type, at the other extreme. We wondered whether the SAGE database might allow us to identify genes that may possibly explain the differential survival of "distant tumors" and that could provide some insight into the presumed differences in chemosensitivity. To accomplish this, we evaluated the correlations between the expression frequency of a given tag in a particular tissue and the percentage survival of patients bearing tumors of that tissue. Table 1B
lists the 14 SAGE libraries used in this part of the study. A library was included in this analysis only if the tumor from which it was derived was included in the SEER database with a percentage distant survival value. Included in these 14 libraries were 13,648 individual genes (i.e., tags). We chose, for further analysis, only those tags that had an average expression across all of the libraries of at least 5 tags/200,000 tags (see "Materials and Methods"). Tables 5
and 6
list the tag, gene name, and function of genes with identified functions, that showed a positive (Table 5)
or a negative (Table 6)
correlation coefficient, significant at P < 0.01 for Table 5
or at P < 0.005 for Table 6
, against the SEER data for survival of patients presenting with distant disease. The tables also contain the mean tag number normalized to 200,000 tags. (The full data set for the 13,648 tags and the restricted data set are available as supplementary materials on the website.10
)
|
|
|
. The number of tags per 200,000 tags is plotted on the X axis against the percentage distant survivors on the Y axis. Similar plots can be made for genes that correlate positively with the SEER data for survival and, of course, for functionally identified genes whose expression shows a nonsignificant correlation with percentage survival (graphical data not shown).
|
whose expression in a solid tumor correlates positively with the 5-year survival of patients presenting with distant cancers, are associated with protein synthesis, being concerned in ribosome biogenesis (this theme appears with a value of P < 1 x 10–4 using the EASE program). This was, in general, the same class of genes that distinguished the cell lines from the solid tumors. Equally strikingly, several of the genes in Table 6
whose expression correlates negatively with survival and, hence, perhaps chemosensitivity, are associated with the ECM, the cytoskeleton, or with cell adhesion (this latter theme appears with a value of P < 0.001 using the EASE program), including one of the same class of genes that distinguished solid tumors from cell lines. Interestingly, the association with the immune system, a striking finding in the comparison between solid tumors and cell lines, did not appear when the SAGE data were correlated with the SEER data. Thus the complement components, C1QA, C1QB, C4B, and the MHC class II interacting factor CD74, correlate with nonsignificant r values of only –0.076, –0.242, 0.257, and 0.270 against percentage distant survivors, although they are overexpressed in solid tumors as compared with cell lines by >100-, >100-, 30-, and 20-fold, respectively. If, for the immune-related genes, a continuum of differential expression existed from resistant to sensitive cells, one might expect here, too, coordinately regulated correlations against percentage distant survivals. These were not found for the immune-related genes.
Of the 17 genes in Table 5
that correlate positively with survival, 1 is found in Table 3
as a gene overexpressed in cell lines, but another 7 have counterparts from the same gene ontology class. Among the 14 genes in Table 6
correlating negatively with survival, 6 have ontology class matches in Table 2
. To extend these observations, we asked whether, in general, the genes differentially expressed between solid tumors and cell lines were differentially expressed in tumors with different "percentage survival." To this end, we combined the prostate and ovary tumor samples from Table 1B
into one pool, defining this as a pool of "tumors associated with better survival" and compared gene expression in these two cancers with expression in a pool of "tumors associated with poor survival," including the stomach and pancreas samples of Table 1B
. As we had done in Fig. 2
, we plotted, in Fig. 6
, gene expression as SAGE tags per 200,000 tags. Fig. 6A
examines expression of genes found in Tables 3
and 5
that are involved in protein biosynthesis. The number of tags in the ovary + prostate pool are plotted on the Y axis and those in the stomach + pancreas pool on the X axis. The circles (Fig. 6A)
are the tags (genes) from Table 3
that are involved in protein biosynthesis and were found to be overexpressed in cell lines, whereas the squares (Fig. 6A)
are the genes from Table 5
that are likewise involved in protein synthesis and correlate significantly with better survival. The regression through all of the points shows a highly significant correlation of 0.96, and a slope of 2.84. This suggests that the protein biosynthesis genes are coordinately regulated and are almost three times as highly expressed in the tumors associated with better survival. Similarly in Fig. 6B
, tags from Tables 2
(solid circles, Fig. 6B
) and 6
(open squares, Fig. 6B
) identified as "ECM" or cell adhesion genes are plotted with, again, the tags for stomach + pancreas on the X axis and those for ovary + prostate on the Y axis. The regression has a significant correlation of r = 0.80 and a slope of 0.44. Thus these genes involved in the ECM or cell–cell interaction appear to be coordinately regulated and, as a group, are more than twice as heavily expressed in tumors associated with a poor survival than in those with a better prognosis.
|
|
DISCUSSION |
|---|
|
TopABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
This study is based in large part on the existing SAGE libraries. To what extent are conclusions based on the SAGE approach likely to be valid? Two main problems are associated with the current SAGE libraries. First, there is some measure of ambiguity in assigning the 10-nucleotide tag sequences to particular genes. Although the SAGE database uses a sophisticated algorithm to make such identification as error free as possible, incorrect identifications may still arise. The 22-nucleotide "long SAGE" protocol (29)
will largely remove this source of ambiguity, but most of the existing SAGE libraries are still based on the 10-nucleotide tags, and "short SAGE" tags will remain the major source for SAGE analysis for the immediate future. Second, genes that are expressed at low levels will give rise to only a small number of tags. A tag that is represented by only a few examples in a tissue might, however, itself arise as an artifact from erroneous sequencing of a tag present at high levels. DNA sequencing errors can be high, hence a 10-nucleotide tag present at 1000 copies might appear with dozens of "neighbors" that differ from it by a single nucleotide in 10. Any such artifactual tag would appear as a low copy tag and be incorrectly assigned. Thus, low copy number tags are inherently suspect, so that a gene that is indeed represented by a low number of tags cannot be unambiguously reported. The question of the ambiguity of tag interpretation has been avoided as much as possible in the present study by considering large groups of tags, present in many SAGE libraries (as recorded in Tables 2
, 3
, 5
and 6
) so that ambiguity in one or two identifications should not affect the broad conclusions drawn. All conclusions are based on tags present at an average of more than five copies per library, normalized to 200,000 tags. This does mean, unfortunately, that genes that may be very important in their effects, being present as tags at only low copy number, will have been missed. Finally, our analysis is based on only a limited data set. We could find only seven tissue types in the SAGE database for which matched pairs of solid tumor and counterpart cell lines were available. We could find only nine tissue types that are listed in both the SAGE and the SEER 5-year distant survival databases. Extending both these data sets might lead to some surprises. With these caveats, however, the analysis as presented does lead to some interesting conclusions and hypotheses.
First, compared with cell lines, solid tumors (a) overexpress genes concerned with cell–cell communication and with the ECM (Table 2)
; (b) overexpress genes involved in the immune response (Table 2)
; and (c) underexpress genes involved in protein synthesis (Table 3)
. Second, the 5-year survival of a patient with a given tumor is correlated with the first and third of the above factors. Using the SEER 5-year survival data as a surrogate for chemosensitivity (and recognizing that factors other than chemosensitivity also influence the survival data), we observed that improved survival was associated with (a) relative overexpression of genes coding for protein synthesis components (Table 5)
; and (b) underexpression of cell adhesion and cytoskeletal genes (Table 6)
. What might be the significance of these findings?
At the outset, the question arises as to whether normal tissue contamination of the solid tumors could give rise to the molecular results described here. Infiltrating normal cells, i.e., stromal cells, could perhaps account for both the tumor versus cell line results, and for the differences between solid tumor types. Indeed, normal immune cells in the tumor tissue could explain the presence of complement genes and immune response genes. Cells such as tumor infiltrating lymphocytes (TIL cells) have been isolated from melanomas and renal cell cancers in trials aimed at enhancing host immunity to cancer (30 , 31) . Normal blood elements can be seen in any solid tumor biopsy, and it is argued that angiogenesis is critical to tumor progression. The fact that the complement and immune response genes were not found to be significant when the SEER database was examined supports the notion that this set of genes may indeed be attributable to normal tissue contamination. Arguing, however, that this pattern is not due to normal tissue contamination is that three of the libraries were derived from microdissected tumors. A direct comparison of the microdissected tumors with their solid tissue counterparts that were not microdissected showed that the vast majority of the genes identified in the present study as differentially expressed between solid tumors and cell lines, or significantly correlated with 5-year survivals, did not appear to arise from stromal elements. The ECM gene expression persists in the SEER correlations.
Accepting that the gene expression pattern, higher cell–cell interaction and ECM genes in solid tumors that confer a poor 5 year survival, and higher protein synthesis genes in cell lines and tumors associated with a better 5-year survival, is derived from the malignant cell itself, the results appear to be eminently reasonable. It is obvious that solid tumors are held within a tissue and will be communicating with their neighbors (32) . They will be synthesizing an ECM. The proteins involved in these activities have little function in the case of cells maintained in culture, and their expression is presumably repressed during the transition from solid tumor to cell line. Also, the cell lines are continuously being selected for rapid growth, so that their ability to rapidly synthesize proteins will be enhanced during generations in culture. These relations become less obvious when seen in the context of the continuum suggested by the SEER data, which stretches from solid tumors that respond poorly to chemotherapy (as measured by our surrogate, the 5-year survivals of distant tumors), through those tumors that respond better, to cells in culture. In some ways it is counter intuitive to find that the tumors associated with a reduced 5-year survival, which one would expect to be dividing faster, appear to have lower potential for synthesis of ribosomal proteins. Such tumors might also be expected to be more prone to metastasize and to have lower expression of genes related to adherence. Yet the reverse seems to be the case, from the present data, for both of these properties. (Rapid growth may indeed confer drug sensitivity and, although it could potentially affect survival measured in months, is less likely to impact 5-year survival.)
It has, however, long been a paradox of clinical oncology that certain rapidly growing tumors are more susceptible to chemotherapy when effective therapy exists, whereas rapidly growing tumors lead to earlier death when chemotherapy is ineffective. There is much data indicating that a high level of cell division, measured by rapid incorporation of labeled DNA precursors (33) or the activation of cell cycle proteins (34) , is correlated within a particular tumor type with a poor survival. In addition, studies of the expression of genes involved in protein synthesis suggest that a high level of protein synthesis is also associated with poor prognosis (35 , 36) . However, it is also true that some studies in tumors for which effective chemotherapy is available have shown that the reverse is true, that cancers undergoing more rapid cell division have a greater potential for long-term survival after treatment. This is well accepted in hematological malignancies. Consider the example of a Burkitt’s lymphoma or large cell lymphoma versus an indolent lymphoma or the example of an acute leukemia versus a chronic leukemia (37) .
More importantly, perhaps, many studies have demonstrated a role of cell–cell communication in reducing sensitivity to chemotherapy. Garrido et al. (38) have shown that growing acute myeloid leukemia cells that are in contact with bone marrow stromal cells, as opposed to freely in suspension, greatly reduce the apoptosis induced by exposing such cells to the cytotoxins ara-C or daunomycin. Overexpression of ß-catenin, a protein with a role in cell–cell adhesion (39) , rendered epithelial cells resistant to anoikis (the apoptosis that is induced when the interaction between normal cells and their ECM is disrupted (40) and, concomitantly, to cytotoxin- and to radiation-induced damage, and enabled them to continue cycling when grown to confluence (41) . Furthermore, methods to grow and study cells in solid, three-dimensional cultures are available (42, 43, 44, 45) . In most such studies, a very significant increase in resistance to cytotoxins is found in cells grown as solid three-dimensional cultures, as opposed to the same cells grown as monolayers (46) . Cells grown as multicellular tumor spheroids are similarly rendered resistant to drugs and demonstrate a diminished capacity for apoptosis (47 , 48) . Disruption of cell–cell adhesion (by enzymes or anti-adhesion antibodies) can be shown to enhance drug cytotoxicity in cells grown as such spheroids (49 , 50) . Interactions with the ECM and its components have been investigated by Dalton (51) and Nefedova et al. (52) . In particular, myeloma cells are rendered resistant to cytotoxins (mitoxantrone is an example) when they find themselves in contact with extracellular components such as fibronectin or with bone marrow cells or soluble factors derived from them. Indeed, this phenomenon was designated cell adhesion-mediated drug resistance. In all of these studies, a clear relationship has been established between a high level of cell–cell interaction and resistance to chemotherapeutic agents. The correlation that we have described between the insensitivity of a tumor type to chemotherapy and the overexpression of genes involving cell–cell interactions is fully in accord with these in vitro findings.
It might be thought that genes controlling cell–cell interactions might be down-regulated in metastatic cells, having enabled such cells to escape from their neighbors and to metastasize. However, no such gene pattern is observed in the molecular signature of metastatic tumors recently proposed by Ramasawamy et al. (13)
. Down-regulated genes in that study included actin, myosin, and MHC components, perhaps derived from nonepithelial cells in the tumors. In addition, a propensity to metastasis was associated with an overexpression of components of the protein translation apparatus, of securin (a gene involved in chromosome separation), of lamin, and also of genes derived from nonepithelial components of the tumor, such as the type I collagens. It should be emphasized that our study does not necessarily reveal any of the factors that might be associated with metastasis, because we have compared, within the different tissue types, tumors that are already distant and hence already metastatic. Thus, we would not have expected to find factors that are important in producing metastasis per se. Finally, we would note that a number of genes that appear in Table 6
have previously been linked to cancer. These include CD44 (53
, 54)
, LGALS3 (galectin-3, involved in cell adhesion; Refs. 55
, 56
), and clathrin (57
, as well as the catenin that interacts with it (58, 59, 60)
.
What are the implications of these findings for chemotherapy of tumors? The analysis suggests that solid tumors, as opposed to cell lines, are adherent and relatively inert, i.e., low in protein synthesis potential. Perhaps they are relatively inert because they are adherent. This accords with the findings that cells induced in vitro to grow in close association with each other have a lower cell proliferation potential. The analysis suggests, furthermore, that the tumors that are more intractable, and presumably more chemotherapy-resistant, are more adherent, and more inert as far as protein synthesis is concerned. The SAGE analysis has provided a molecular window into tumor biology that in some settings could mean the difference between drug sensitivity and drug resistance, and may provide insight that could lead to the circumvention of this type of drug resistance.
|
ACKNOWLEDGMENTS |
|---|
; to Drs. Doug Hosack and R. Lempicki of the Laboratory of Immunopathogenesis and Bioinformatics, NIAID/NIH, for making the EASE program available and helping us to identify the themes and relevant probabilities that were assigned by EASE; to Dr. Greg J. Riggins of the Department of Pathology, Duke University Medical Center, Durham, North Carolina, and Dr. Carl Schaefer of the National Cancer Institute Center for Bioinformatics for help with SAGE, and Dr. Schaefer for sending us the SAGE files; and to Drs. Aryeh Stein and Shimrit Ginott for help with statistics. W. D. Stein was a Fogarty Senior Research Fellow of the NIH during the initial phases of this work, and thanks Dr. Michael Gottesman of the National Cancer Institute/NIH for his hospitality during that period. |
FOOTNOTES |
|---|
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
Requests for reprints: Wilfred Stein, Room 208, Bldg. 10, 9000 Rockville Pike, Bethesda, MD 20892. E-mail: wdstein{at}vms.huji.ac.il
4 Internet address for the SAGE program: http://cgap.nci.nih.gov/SAGE. 
5 Internet address for the DGED tool of the SAGE database: http://cgap.nci.nih.gov/SAGE/SDGED.
6 Significant differences in gene expression profiles are distinguished at: http://cgap.nci.nih.gov/SAGE/Significance.
7 Internet address for the SEER database with Cancer Statistics Review: http://seer.cancer.gov/csr/1973_1999.
8 Internet address for the 5-year survival data for patients presenting with tumors that were diagnosed as distant: http://training.seer.cancer.gov/module_ss2k/ss_cate_distant.html.
9 Internet address: ftp://ftp1.nci.nih.gov/pub/SAGE/.
10 The full data set and calculated correlations can be found in supplementary material at the following web site: http://www.binf.ku.dk/users/tlitman/SAGE.html.
11 Internet address for the EASE program: http://david.niaid.nih.gov/david/ease.htm.
Received 10/28/03. Revised 1/ 9/04. Accepted 2/ 5/04.
|
REFERENCES |
|---|
|
TopABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
This article has been cited by other articles:
|
|
 |
|
  D. A. Solomon, J.-S. Kim, H. W. Ressom, Z. Sibenaller, T. Ryken, W. Jean, D. Bigner, H. Yan, and T. Waldman Sample Type Bias in the Analysis of Cancer Genomes Cancer Res., July 15, 2009; 69(14): 5630 - 5633. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 L. Guo, Y. Ma, R. Ward, V. Castranova, X. Shi, and Y. Qian Constructing Molecular Classifiers for the Accurate Prognosis of Lung Adenocarcinoma. Clin. Cancer Res., June 1, 2006; 12(11): 3344 - 3354. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
C. R. Estrada, M. Salanga, D. R. Bielenberg, W. B. Harrell, D. Zurakowski, X. Zhu, M. R. Palmer, M. R. Freeman, and R. M. Adam Behavioral Profiling of Human Transitional Cell Carcinoma Ex vivo. Cancer Res., March 15, 2006; 66(6): 3078 - 3086. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
J. Terry, J. M. Lubieniecka, W. Kwan, S. Liu, and T. O. Nielsen Hsp90 Inhibitor 17-Allylamino-17-Demethoxygeldanamycin Prevents Synovial Sarcoma Proliferation via Apoptosis in In vitro Models Clin. Cancer Res., August 1, 2005; 11(15): 5631 - 5638. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 D. G. Peters, D. M. Kudla, J. A. DeLoia, T. J. Chu, L. Fairfull, R. P. Edwards, and R. E. Ferrell Comparative Gene Expression Analysis of Ovarian Carcinoma and Normal Ovarian Epithelium by Serial Analysis of Gene Expression Cancer Epidemiol. Biomarkers Prev., July 1, 2005; 14(7): 1717 - 1723. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 R. Sandberg and I. Ernberg Assessment of tumor characteristic gene expression in cell lines using a tissue similarity index (TSI) PNAS, February 8, 2005; 102(6): 2052 - 2057. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 G. Szakacs and M. M. Gottesman Comparing Solid Tumors with Cell Lines: Implications for Identifying Drug Resistance Genes in Cancer Mol. Interv., December 1, 2004; 4(6): 323 - 325. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| Cancer Research | Clinical Cancer Research |
| Cancer Epidemiology Biomarkers & Prevention | Molecular Cancer Therapeutics |
| Molecular Cancer Research | Cancer Prevention Research |
| Cancer Prevention Journals Portal | Cancer Reviews Online |
| Annual Meeting Education Book | Meeting Abstracts Online |
, genes concerned in protein synthesis initiation or elongation; 
, genes coding proteins of the small subunit of the ribosome; 
, the large subunit of the ribosome.
