Stem cell–ness: a "magic marker" for cancer
http://www.100md.com
《临床调查学报》
1Deartment of Molecular Theraeutics and
2Deartment of Biostatistics and Alied Mathematics, University of Texas M.D. Anderson Cancer Center, Houston, Texas, USA.
Abstract
Transcritional rofiling of atient tumors is a much-heralded advancement in cancer theray, as it rovides the oortunity to identify atients who would benefit from more or less aggressive theray and thus allows the develoment of individualized treatment. However, translation of this romise into atient benefit has roven challenging. In this issue of the JCI, Glinsky and colleagues used human and murine models to identify a otential stem cell mRNA signature, based on the hyothesis that tumors with stem cell–like characteristics are likely to have a oor rognosis. Remarkably, an 11-gene "exression signature" associated with "stem cell–ness" searated atients with different cancers into good- and oor-rognosis grous. Such a "magic marker" would, if validated, have a major imact on atient care. However, there remain challenges incumbent with creating and validating such signatures.
See the related article beginning on age 1503
rediction and cancer
The incetion of high-throughut analyses using oligonucleotide microarrays has given biologists the ability to globally assess RNA levels in a atient’s tumor samle. A tyical microarray study generates data on the exression of the aroximately 20,000 human genes and soon studies will be able to analyze the more than 150,000 slice variants of RNA that are likely to have functional roles. The inherent challenge is to convert this data into alicable knowledge. A otential strength of the technology lies in its ability to uncover comlex gene interaction atterns and correlate those atterns with clinically relevant outcomes. This "holy grail" could ultimately redict not only the theraeutic resonse of the tumor resent in each atient, but also the atient's survival, which would subsequently lead to the develoment of individualized theray for each atient based both on the genetic aberrations in the tumor and on the atient’s own genetic makeu. However, desite early enthusiasm, there have been considerable challenges in converting the romise of individualized molecular medicine into clinical ractice.
In this issue of the JCI, Glinsky and colleagues outline a ossible exression signature comrising 11 genes that has the ability, according to the authors’ analysis, to segregate tumor samles from multile tumor lineages into those that have good or oor rognoses (1). The authors have alied this gene set to multile tissue tyes from disarate data sets and have reeatedly observed its redictive ower. The alication of their 11-gene signature to these indeendent sets addresses an analytical limitation that is often overlooked when "redictive" gene exression signatures are found in microarray exeriments (2, 3). When thousands of measurements are taken on each atient, the number of ways to select some of those measurements as a attern classifying tumors or redicting outcomes is enormous. When selecting multile gene measurements, the robability of finding a combination with aarent clinical relevance just by chance is even higher.
This multile-measurements roblem can be addressed using a "training and test set" aroach, wherein redictive models are validated on searate, indeendent data sets of adequate size. However, maintaining consistency of the thresholds and cutoffs used across data sets is fundamental to this aradigm. If a certain threshold is used to indicate a secific change on the training set, then it is imortant that the same threshold be used with subsequent test sets. While Glinsky et al. (1) do searate each data set into training and test sets, they require a searate training and test set and different cutoffs for each cell lineage and otentially for each RNA analysis latform. Since the authors use different cutoffs to identify rognosis in different data sets, one must remain cautious about whether their aroach will be generally alicable to atient management. Furthermore, slitting a single data set into 2 smaller sets for training and test uroses also introduces statistical and analytical challenges, since the smaller data sets will have diminished ower (4).
Stem cells and cancer rogression
A key insight in the Glinsky et al. study (1) is the biological motivation driving their selection of the gene signature. The authors begin with the lausible hyothesis that transformed cells, in which self-renewal or stem cell–related athways are activated, may contribute to the survival of cancer cells in tumors and romote tumor rogression and oor rognosis for atients. They aly this idea across secies, combining a study of stem cells in a murine leukemia viral-1 (Bmi-1) knockout mouse model with a study of rimary and metastatic tumors in a model of transgenic adenocarcinoma of the mouse rostate (TRAM) in order to select genes that consistently dislay a stem cell self-renewal–like exression rofile in multile models. This aroach builds on the aradigm that cancer likely arises in a limited oulation of stem cells. These stem cells could otentially have a set of common characteristics, and thus gene exression atterns, across tumor lineages. Characterization of a common exression signature with sulemental tissue-secific gene changes could reflect the cell of origin of a cancer. This concet is comatible with the observation that most tumors are less differentiated than the utative recursor cell and that only a small number of normal cells have the otential for self-renewal (5).
Moreover, if this signature can delineate those atients whose tumors rely on stem cell–like gene exression, then targeting those genes within the tumors may result in a cell oulation with limited roliferative otential. There is strong suort for the hyothesis that the clones that initiated the cancer are different from the majority of cells within the tumor (6). Different athways may be activated within different clones, and thus theraeutic targeting of these initiating cells may lead to a better outcome. Traditional theraies are aimed at the raidly dividing cells within the tumor; while these may reduce tumor mass, they may not lead to cures.
The significance of the Bmi-1–based 11-gene signature
Most statisticians (and many biologists) are leery of studies that claim to find "gene signatures" or "atterns of gene exression" that can be used to redict clinical outcomes across tumor lineages. Much of the uneasiness arises because many studies neglect to recisely define the notion of a signature. By contrast, Glinsky and colleagues roduce a concrete definition of a gene signature (1). Their signature is defined quantitatively as an 11-dimensional vector of exression fold-change values in the base-10 logarithm of 11 genes in the eriheral nervous system (NS) neurosheres in the mouse Bmi-1 knockout model. Deviations from the average exression of these 11 genes in individual tumor samles are correlated with this 11-dimensional vector. While this signature had classification and redictive value when assessed on human tumors, it is imortant to note that other otential "stem cell–ness" signatures, such as the 14-gene grou discussed in the study, did not demonstrate redictive value.
The transcrition factor Bmi-1 aears to lay a role in gene regulation in both cancerous and normal stem cell roliferation through eigenetic mechanisms — changes that affect gene exression without altering gene structure, such as methylation or acetylation of chromatin. As the authors note, Bmi-1 has reviously been shown to be required for maintenance of self-renewing HSCs (7) and for the self-renewal of leukemic stem cells (8). Bmi-1 has been imlicated in extending the relicative otential of human fibroblasts through the suression of the senescence athway deendent on 16 (a cyclin-deendent kinase inhibitor) in a retinoblastoma rotein–deendent manner (9). Further, a deficiency in the 16INK4a gene artially reverses the self-renewal defect in Bmi-1 dominant-negative neural stem cells (10). Additionally, a cooerative interaction between Bmi-1 and the oncogene c-myc has been demonstrated, through enhancement of mouse embryonic fibroblast roliferation, as a result of inhibiting c-myc–induced aotosis and 19arf (11). It is aarent that the olycomb grou gene Bmi-1 is otentially involved in various hases of tumorigenesis; therefore, the 11-gene exression signature highlighted by Glinsky and colleagues (1), created from changes in gene exression induced by altering Bmi-1 in different backgrounds and validated on data collected by different researchers at different times, must be taken seriously and subjected to extensive evaluation and validation in indeendent laboratories.
The 11-gene signature
At the very least, the 11-gene signature suggests the ossibility of more accurately identifying atients with oor rognoses as candidates for more aggressive or investigational theray. As the mode of treatment is different for each tumor lineage, it also suggests that the signature truly indicates rognosis rather than redicting resonse to theray. However, it could include a general indication of sensitivity to cell death as a comonent of its rognostic load, although the individual comonents of the 11-gene signature have not been reviously indicated as major regulators of cell survival.
The comonents of this 11-gene set vary in function and exist in different athways (see Table 1 for gene names, signatures, and values). Both budding uninhibited by benzimidazoles 1 (BUB1) and kinetochore-associated 2 (KNTC2; also known as HEC) have been imlicated as mitotic checkoint roteins, and as such, aberrations in their function could contribute to genomic instability and aneuloidy. Mutations in the BUB1 gene lead to its inactivation and increase microsatellite instability in colon cancer as well as a redisosition to certain tyes of cancers (12, 13). Increased gastrulation homeobox 2 (GBX2) exression stimulates growth of human rostate cancer cells through uregulation of the gene coding for IL-6 (14). Overexression of cyclin B1 (CCNB1) levels have been observed in high-grade large-cell and small-cell lung carcinoma (15) and have been shown to be downregulated as a result of 53 induction in non–small-cell lung cancer (16). Additionally, CCNB1 exression was highly correlated with the labeling index for antigen identified by mAb ki-67 (Ki-67, associated with increased tumor cell roliferation), which suggests a key role for CCNB1 in the regulation of neuroendocrine tumor cell roliferation (15). In breast cancer cell lines, overexression of the FGF recetor 2 (FGFR2) gene resulted in activation of the MAK and I3K athways (17). Interestingly, restoration of this gene roduct into a malignant rostate eithelial cancer cell line, C3, led to suression of malignancy and restoration of nonmalignant traits. Thus, 3 of the 11 genes identified in the Glinsky et al. gene signature (1) are related to cell roliferation and 2 to transition through mitosis. While little is known regarding the secific biological functions of ubiquitin-secific rotease 22 (US22), ubiquitin-secific roteases have been imlicated in control of regulatory molecules such as 53 and cyclins (18). The ring finger 2 (RNF2) rotein is art of the olycomb grou of roteins, like Bmi-1, that lay key roles in hematooiesis and cell-cycle regulation (19). Ankyrins are transmembrane roteins shown to be involved in cellular functions relating to the influx and efflux of sodium and calcium (20, 21). Carboxylesterase (CES) enzymes are found in many animal secies and lay a role in the hydrolysis of drugs such as steroids and anticancer agents (22). Mutations in CES1 have been imlicated harmacogenomically in the activation status of cancer drugs and rodrugs (23). At this time, the role of host cell factor c1 (HCFC1) in cancer has yet to be evaluated; however, as the host cell factor family is imlicated in immunomodulation, it is ossible that HCFC1 lays a role in limiting the immune resonse to cancer or in the roduction of cytokines such as IL-6 or IL-8, which can contribute to neovascularization or tumor growth.
As demonstrated with the rostate cancer data set, the 11-gene set can be divided into 2 grous: those for which elevated exression levels are associated with stem cell–ness and a oor rognosis (Ki67, CCNB1, GBX2, BUB1, KNTC2, US22, and RNF2, in descending order of strength of association) and those for which decreased exression levels are associated with stem cell–ness and a good rognosis (CES1, FGFR2, and ankyrin 3 [ANK3]; see Table 1) (1). High levels of these stem cell–related genes indicate a otential for self-renewal within tumors and for increased tumor aggressiveness within atients. Intriguingly, those genes that are ositively associated with tumor cell roliferation (Ki67, CCNB1, and GBX2) and with the mitotic sindle (BUB1 and KNTC2) are also ositively associated with the stem cell–ness signature. FGFR2, which has been shown to decrease the growth of rostate cancer cells, is negatively associated with the stem cell–ness signature. The association between this 11-gene signature and good and oor rognosis therefore makes sense, at least for those genes that have been characterized as associated with the behavior of cancer cells.
While the relevance to cancer rogression and malignancy has been studied and even established in certain members of this 11-gene set, others have yet to be studied in detail. This underscores another strength of high-throughut analysis: the use of such a global aroach may identify new and unexected targets for study, some of which may rove to be otential theraeutic targets. The inclusion of genes involved in cancer cell roliferation as well as unexlored genes within this redictive grou suggests that further analysis into their mechanisms and otential involvement in signaling athways germane to cancer is warranted.
Can their analysis be relicated?
In many microarray studies, how the data are analyzed may be more imortant than how they were generated. It is therefore incumbent uon researchers to describe their analytical methods in enough detail to allow indeendent researchers to relicate their comutations on the same data. Ideally, these details take the form of recise equations or algorithms or the actual code used to analyze the data (24-26). Although the Glinsky et al. study (1) reorts many of the critical values needed to relicate their results, their reliance on urely verbal descritions leaves room for some ambiguity.
We attemted to relicate art of their analysis and turned our attention to the same lung cancer study used by the authors (27), which included survival data on 125 atients and microarray data from Affymetrix U95Av2 GeneChis. Our estimate of the standard gene exression signature is listed in Table 1. Because of redundancy, a total of 18 robe sets reresent these 11 genes on the U95Av2 array (Table 2). We used all 18 robe sets, which exanded the fixed gene signature vector from Table 1 into an 18-dimensional vector. For each of the 18 robe sets, we comuted the average exression over all 125 samles. We divided the exression vectors in individual tumor samles by the average exression value and then transformed the ratios by comuting the base-10 logarithm, roducing 18-dimensional vectors. We then comuted the earson correlation coefficient between the fixed gene exression signature vector and the individual vector of log ratios. This rocedure yielded 1 number, xi, for each tumor samle i = 1, . . . , 125, which reresented our best estimate of the stem cell–like henotye association index (SAI) described in the Glinsky et al. study (1).
We used SAI as a continuous covariate in a Cox roortional hazards model to redict survival in the full data set and found that SAI was significant (likelihood ratio test, = 0.0179). Because the coefficient xi in the model was ositive (0.798), higher values of SAI were associated with shorter survival. Using trial and error, we determined a threshold on the full data set that, when used to slit the samles into 2 grous, yielded the best results in a survival analysis. The otimum threshold occurred at 0.32 and searated the data into 40 samles with a stem cell–like rofile (xi > 0.32) and 85 samles without a stem cell–like rofile. To this extent, we were able to confirm the analysis reorted by Glinsky et al. (Figure 1).
Based on this reliminary attemt to relicate the authors’ analyses, we believe that their results should be greeted with cautious otimism. However, we were unable to validate their method when slitting the data multile times into training and test sets. Our slits were more challenging than the ones used by the authors, since we did not use outcome information to balance the slits. It is also not clear whether we followed the same rocedure for comuting the SAI that was used by Glinsky et al. (1).
Conclusions
Glinsky and colleagues (1) have roduced a stimulating analysis of a collection of microarray data exeriments. Using ideas that were well motivated by the underlying biology and combining data across secies from 2 different microarray studies, they identified an 11-gene signature that might be related to tumor behavior, and thus atient survival, in cancer. In order to validate this signature, they tested it retrosectively in a broad sectrum of microarray studies of different kinds of cancer.
In our hands, unsuervised analyses of exression levels across some of the same data sets used by Glinsky and colleagues (1) did not reveal any clinically relevant or statistically significant correlations with outcome. By introducing a weighting coefficient into their redictor model, they are in fact adding an element of suervision into the analysis. Further, as these coefficients are re-derived on each subsequent data set, the statistical and analytical issues incumbent with high-throughut technologies are insufficiently addressed (2).
Glinsky et al. roosed several models, but only the 11-gene metastatic TRAM tumor samle/NS set consistently differentiated samles in a clinically relevant manner (1). An imortant asect of this gene set and the corresonding coefficients used in determining the weighted survival redictor is that those genes that have been reviously identified as otential indicators of oor rognosis are given the greatest coefficients. That is, negative weighting correlates with longer survival. For examle, in the rostate cancer SAI, the authors assign the largest coefficient value to Ki-67, which is associated with increased roliferation in cancer and could therefore lead to greater aggressiveness and oorer rognosis.
In the search for a owerfully redictive set of cancer genes, there have been various signatures roosed that contain anywhere from 10–100 genes. However, recent studies have cast doubt on the ower of those sets that were validated within a single data set or even a single tumor tye (2-4). As such, the need for validation of a roosed gene exression signature across indeendent data sets is warranted. To their credit, Glinsky and colleagues have gone to great lengths to validate their 11-gene signature, albeit in a suervised manner, across multile tumor samles and multile tissue tyes. As researchers begin to question the results reorted from microarray studies (2-4, 28), the need for relicable analyses and indeendent corroboration grows more acute. Regardless of whether it was said by Niels Bohr or Yogi Berra, it is still the case that "rediction is very difficult, esecially about the future."
Footnotes
Nonstandard abbreviations used: ANK3, ankyrin 3; Bmi-1, murine leukemia viral-1; BUB1, budding uninhibited by benzimidazoles 1; CCNB1, cyclin B1; CES, carboxylesterase; FGFR2, FGF recetor 2; GBX2, gastrulation homeobox 2; HCFC1, host cell factor c1; Ki-67, antigen identified by mAb ki-67; KNTC2, kinetochore-associated 2; NS, eriheral nervous system; RNF2, ring finger 2; SAI, stem cell–like henotye association index; TRAM, transgenic adenocarcinoma of the mouse rostate; US22, ubiquitin-secific rotease 22.
Conflict of interest: The authors have declared that no conflict of interest exists.
References
Glinsky, G.V., Berezovska, O., and Glinskii, A.B. 2005. . Microarray analysis identifies a death-from-cancer signature redicting theray failure in atients with multile tyes of cancer. J. Clin. Invest. 115::1503-1521. doi:10.1172/JCI23412.
Michiels, S., Koscielny, S., and Hill, C. 2005. . rediction of cancer outcome with microarrays: a multile random validation strategy. Lancet. 365::488-492.
Ransohoff, D.F. 2004. . Rules of evidence for cancer molecular-marker discovery and validation. Nat. Rev. Cancer. 4::309-314.
Ransohoff, D.F. 2005. . Lessons from controversy: ovarian cancer screening and serum roteomics. J. Natl. Cancer. Inst. 97::315-319.
Cheng, W., Liu, J., Yoshida, H., Rosen, D., and Naora, H. 2005. . Lineage infedility of eithelial ovarian cancers is controlled by HOX genes that secify regional identity in the reroductive tract. Nat. Med. 11::531-537.
Dick, J.E. 2003. . Breast cancer stem cells revealed. roc. Natl. Acad. Sci. U. S. A. 100::3547-3549.
ark, I.K. et al. 2003. . Bmi-1 is required for maintenance of adult self-renewing haematooietic stem cells. Nature. 423::302-305.
Lessard, J., and Sauvageau, G. 2003. . Bmi-1 determines the roliferative caacity of normal and leukaemic stem cells. Nature. 423::255-260.
Itahana, K. et al. 2003. . Control of the relicative life san of human fibroblasts by 16 and the olycomb rotein Bmi-1. Mol. Cell. Biol. 23::389-401.
Molofsky, A.V. et al. 2003. . Bmi-1 deendence distinguishes neural stem cell self-renewal from rogenitor roliferation. Nature. 425::962-967.
Jacobs, J.J. et al. 1999. . Bmi-1 collaborates with c-Myc in tumorigenesis by inhibiting c-Myc-induced aotosis via INK4a/ARF. Genes Dev. 13::2678-2690.
Cahill, D.. et al. 1998. . Mutations of mitotic checkoint genes in human cancers. Nature. 392::300-303.
Hanks, S. et al. 2004. . Constitutional aneuloidy and cancer redisosition caused by biallelic mutations in BUB1B. Nat. Genet. 36::1159-1161.
Gao, A.C., Lou, W., and Isaacs, J.T. 2000. . Enhanced GBX2 exression stimulates growth of human rostate cancer cells via transcritional u-regulation of the interleukin 6 gene. Clin. Cancer Res. 6::493-497.
Igarashi, T. et al. 2004. . Divergent cyclin B1 exression and Rb/16/cyclin D1 athway aberrations among ulmonary neuroendocrine tumors. Mod. athol. 17::1259-1267.
Dubrez, L. et al. 2001. . Cell cycle arrest is sufficient for 53-mediated tumor regression. Gene Ther. 8::1705-1712.
Moffa, A.B., Tannheimer, S.L., and Ethier, S.. 2004. . Transforming otential of alternatively sliced variants of fibroblast growth factor recetor 2 in human mammary eithelial cells. Mol. Cancer Res. 2::643-652.
Chen, C., et al. 2005. Ubiquitin-roteasome degradation of KLF5 transcrition factor in cancer and untransformed eithelial cells. Oncogene. doi:10.1038/sj.onc.1208497..
Sanchez-Beato, M. et al. 2004. . Abnormal cG rotein exression in Hodgkin’s lymhoma. Relation with E2F6 and NFkaaB transcrition factors. J. athol. 204::528-537.
Hayashi, T., and Su, T.. 2001. . Regulating ankyrin dynamics: Roles of sigma-1 recetors. roc. Natl. Acad. Sci. U. S. A. 98::491-496.
Kretschmer, T. et al. 2002. . ainful neuromas: a otential role for a structural transmembrane rotein, ankyrin G. J. Neurosurg. 97::1424-1431.
Redinbo, M.R., Bencharit, S., and otter, .M. 2003. . Human carboxylesterase 1: from drug metabolism to drug discovery. Biochem. Soc. Trans. 31::620-624.
Marsh, S. et al. 2004. . harmacogenomic assessment of carboxylesterases 1 and 2. Genomics. 84::661-668.
Gentleman, R. 2005. . Reroducible research: a bioinformatics case study. Statistical Alications in Genetics and Molecular Biology [serial online]. 4::Article 2. htt://www.beress.com/sagmb/vol4/iss1/art2.
Gentleman, R., and Lang, D.T. 2004. Statistical Analyses and Reroducible Research. Bioconductor roject Working aers [serial online]. Working aer 2. htt://www.beress.com/bioconductor/aer2..
Leisch, F., and Rossini, A. 2003. . Reroducible statistical research. Chance. 16::46-50.
Bhattacharjee, A. et al. 2001. . Classification of human lung carcinomas by mRNA exression rofiling reveals distinct adenocarcinoma subclasses. roc. Natl. Acad. Sci. U. S. A. 98::13790-13795.
Ioannidis, J.. 2005. . Microarrays and molecular research: noise discovery? Lancet. 365::454-455.(John . Lahad, Gordon B. M)
2Deartment of Biostatistics and Alied Mathematics, University of Texas M.D. Anderson Cancer Center, Houston, Texas, USA.
Abstract
Transcritional rofiling of atient tumors is a much-heralded advancement in cancer theray, as it rovides the oortunity to identify atients who would benefit from more or less aggressive theray and thus allows the develoment of individualized treatment. However, translation of this romise into atient benefit has roven challenging. In this issue of the JCI, Glinsky and colleagues used human and murine models to identify a otential stem cell mRNA signature, based on the hyothesis that tumors with stem cell–like characteristics are likely to have a oor rognosis. Remarkably, an 11-gene "exression signature" associated with "stem cell–ness" searated atients with different cancers into good- and oor-rognosis grous. Such a "magic marker" would, if validated, have a major imact on atient care. However, there remain challenges incumbent with creating and validating such signatures.
See the related article beginning on age 1503
rediction and cancer
The incetion of high-throughut analyses using oligonucleotide microarrays has given biologists the ability to globally assess RNA levels in a atient’s tumor samle. A tyical microarray study generates data on the exression of the aroximately 20,000 human genes and soon studies will be able to analyze the more than 150,000 slice variants of RNA that are likely to have functional roles. The inherent challenge is to convert this data into alicable knowledge. A otential strength of the technology lies in its ability to uncover comlex gene interaction atterns and correlate those atterns with clinically relevant outcomes. This "holy grail" could ultimately redict not only the theraeutic resonse of the tumor resent in each atient, but also the atient's survival, which would subsequently lead to the develoment of individualized theray for each atient based both on the genetic aberrations in the tumor and on the atient’s own genetic makeu. However, desite early enthusiasm, there have been considerable challenges in converting the romise of individualized molecular medicine into clinical ractice.
In this issue of the JCI, Glinsky and colleagues outline a ossible exression signature comrising 11 genes that has the ability, according to the authors’ analysis, to segregate tumor samles from multile tumor lineages into those that have good or oor rognoses (1). The authors have alied this gene set to multile tissue tyes from disarate data sets and have reeatedly observed its redictive ower. The alication of their 11-gene signature to these indeendent sets addresses an analytical limitation that is often overlooked when "redictive" gene exression signatures are found in microarray exeriments (2, 3). When thousands of measurements are taken on each atient, the number of ways to select some of those measurements as a attern classifying tumors or redicting outcomes is enormous. When selecting multile gene measurements, the robability of finding a combination with aarent clinical relevance just by chance is even higher.
This multile-measurements roblem can be addressed using a "training and test set" aroach, wherein redictive models are validated on searate, indeendent data sets of adequate size. However, maintaining consistency of the thresholds and cutoffs used across data sets is fundamental to this aradigm. If a certain threshold is used to indicate a secific change on the training set, then it is imortant that the same threshold be used with subsequent test sets. While Glinsky et al. (1) do searate each data set into training and test sets, they require a searate training and test set and different cutoffs for each cell lineage and otentially for each RNA analysis latform. Since the authors use different cutoffs to identify rognosis in different data sets, one must remain cautious about whether their aroach will be generally alicable to atient management. Furthermore, slitting a single data set into 2 smaller sets for training and test uroses also introduces statistical and analytical challenges, since the smaller data sets will have diminished ower (4).
Stem cells and cancer rogression
A key insight in the Glinsky et al. study (1) is the biological motivation driving their selection of the gene signature. The authors begin with the lausible hyothesis that transformed cells, in which self-renewal or stem cell–related athways are activated, may contribute to the survival of cancer cells in tumors and romote tumor rogression and oor rognosis for atients. They aly this idea across secies, combining a study of stem cells in a murine leukemia viral-1 (Bmi-1) knockout mouse model with a study of rimary and metastatic tumors in a model of transgenic adenocarcinoma of the mouse rostate (TRAM) in order to select genes that consistently dislay a stem cell self-renewal–like exression rofile in multile models. This aroach builds on the aradigm that cancer likely arises in a limited oulation of stem cells. These stem cells could otentially have a set of common characteristics, and thus gene exression atterns, across tumor lineages. Characterization of a common exression signature with sulemental tissue-secific gene changes could reflect the cell of origin of a cancer. This concet is comatible with the observation that most tumors are less differentiated than the utative recursor cell and that only a small number of normal cells have the otential for self-renewal (5).
Moreover, if this signature can delineate those atients whose tumors rely on stem cell–like gene exression, then targeting those genes within the tumors may result in a cell oulation with limited roliferative otential. There is strong suort for the hyothesis that the clones that initiated the cancer are different from the majority of cells within the tumor (6). Different athways may be activated within different clones, and thus theraeutic targeting of these initiating cells may lead to a better outcome. Traditional theraies are aimed at the raidly dividing cells within the tumor; while these may reduce tumor mass, they may not lead to cures.
The significance of the Bmi-1–based 11-gene signature
Most statisticians (and many biologists) are leery of studies that claim to find "gene signatures" or "atterns of gene exression" that can be used to redict clinical outcomes across tumor lineages. Much of the uneasiness arises because many studies neglect to recisely define the notion of a signature. By contrast, Glinsky and colleagues roduce a concrete definition of a gene signature (1). Their signature is defined quantitatively as an 11-dimensional vector of exression fold-change values in the base-10 logarithm of 11 genes in the eriheral nervous system (NS) neurosheres in the mouse Bmi-1 knockout model. Deviations from the average exression of these 11 genes in individual tumor samles are correlated with this 11-dimensional vector. While this signature had classification and redictive value when assessed on human tumors, it is imortant to note that other otential "stem cell–ness" signatures, such as the 14-gene grou discussed in the study, did not demonstrate redictive value.
The transcrition factor Bmi-1 aears to lay a role in gene regulation in both cancerous and normal stem cell roliferation through eigenetic mechanisms — changes that affect gene exression without altering gene structure, such as methylation or acetylation of chromatin. As the authors note, Bmi-1 has reviously been shown to be required for maintenance of self-renewing HSCs (7) and for the self-renewal of leukemic stem cells (8). Bmi-1 has been imlicated in extending the relicative otential of human fibroblasts through the suression of the senescence athway deendent on 16 (a cyclin-deendent kinase inhibitor) in a retinoblastoma rotein–deendent manner (9). Further, a deficiency in the 16INK4a gene artially reverses the self-renewal defect in Bmi-1 dominant-negative neural stem cells (10). Additionally, a cooerative interaction between Bmi-1 and the oncogene c-myc has been demonstrated, through enhancement of mouse embryonic fibroblast roliferation, as a result of inhibiting c-myc–induced aotosis and 19arf (11). It is aarent that the olycomb grou gene Bmi-1 is otentially involved in various hases of tumorigenesis; therefore, the 11-gene exression signature highlighted by Glinsky and colleagues (1), created from changes in gene exression induced by altering Bmi-1 in different backgrounds and validated on data collected by different researchers at different times, must be taken seriously and subjected to extensive evaluation and validation in indeendent laboratories.
The 11-gene signature
At the very least, the 11-gene signature suggests the ossibility of more accurately identifying atients with oor rognoses as candidates for more aggressive or investigational theray. As the mode of treatment is different for each tumor lineage, it also suggests that the signature truly indicates rognosis rather than redicting resonse to theray. However, it could include a general indication of sensitivity to cell death as a comonent of its rognostic load, although the individual comonents of the 11-gene signature have not been reviously indicated as major regulators of cell survival.
The comonents of this 11-gene set vary in function and exist in different athways (see Table 1 for gene names, signatures, and values). Both budding uninhibited by benzimidazoles 1 (BUB1) and kinetochore-associated 2 (KNTC2; also known as HEC) have been imlicated as mitotic checkoint roteins, and as such, aberrations in their function could contribute to genomic instability and aneuloidy. Mutations in the BUB1 gene lead to its inactivation and increase microsatellite instability in colon cancer as well as a redisosition to certain tyes of cancers (12, 13). Increased gastrulation homeobox 2 (GBX2) exression stimulates growth of human rostate cancer cells through uregulation of the gene coding for IL-6 (14). Overexression of cyclin B1 (CCNB1) levels have been observed in high-grade large-cell and small-cell lung carcinoma (15) and have been shown to be downregulated as a result of 53 induction in non–small-cell lung cancer (16). Additionally, CCNB1 exression was highly correlated with the labeling index for antigen identified by mAb ki-67 (Ki-67, associated with increased tumor cell roliferation), which suggests a key role for CCNB1 in the regulation of neuroendocrine tumor cell roliferation (15). In breast cancer cell lines, overexression of the FGF recetor 2 (FGFR2) gene resulted in activation of the MAK and I3K athways (17). Interestingly, restoration of this gene roduct into a malignant rostate eithelial cancer cell line, C3, led to suression of malignancy and restoration of nonmalignant traits. Thus, 3 of the 11 genes identified in the Glinsky et al. gene signature (1) are related to cell roliferation and 2 to transition through mitosis. While little is known regarding the secific biological functions of ubiquitin-secific rotease 22 (US22), ubiquitin-secific roteases have been imlicated in control of regulatory molecules such as 53 and cyclins (18). The ring finger 2 (RNF2) rotein is art of the olycomb grou of roteins, like Bmi-1, that lay key roles in hematooiesis and cell-cycle regulation (19). Ankyrins are transmembrane roteins shown to be involved in cellular functions relating to the influx and efflux of sodium and calcium (20, 21). Carboxylesterase (CES) enzymes are found in many animal secies and lay a role in the hydrolysis of drugs such as steroids and anticancer agents (22). Mutations in CES1 have been imlicated harmacogenomically in the activation status of cancer drugs and rodrugs (23). At this time, the role of host cell factor c1 (HCFC1) in cancer has yet to be evaluated; however, as the host cell factor family is imlicated in immunomodulation, it is ossible that HCFC1 lays a role in limiting the immune resonse to cancer or in the roduction of cytokines such as IL-6 or IL-8, which can contribute to neovascularization or tumor growth.
As demonstrated with the rostate cancer data set, the 11-gene set can be divided into 2 grous: those for which elevated exression levels are associated with stem cell–ness and a oor rognosis (Ki67, CCNB1, GBX2, BUB1, KNTC2, US22, and RNF2, in descending order of strength of association) and those for which decreased exression levels are associated with stem cell–ness and a good rognosis (CES1, FGFR2, and ankyrin 3 [ANK3]; see Table 1) (1). High levels of these stem cell–related genes indicate a otential for self-renewal within tumors and for increased tumor aggressiveness within atients. Intriguingly, those genes that are ositively associated with tumor cell roliferation (Ki67, CCNB1, and GBX2) and with the mitotic sindle (BUB1 and KNTC2) are also ositively associated with the stem cell–ness signature. FGFR2, which has been shown to decrease the growth of rostate cancer cells, is negatively associated with the stem cell–ness signature. The association between this 11-gene signature and good and oor rognosis therefore makes sense, at least for those genes that have been characterized as associated with the behavior of cancer cells.
While the relevance to cancer rogression and malignancy has been studied and even established in certain members of this 11-gene set, others have yet to be studied in detail. This underscores another strength of high-throughut analysis: the use of such a global aroach may identify new and unexected targets for study, some of which may rove to be otential theraeutic targets. The inclusion of genes involved in cancer cell roliferation as well as unexlored genes within this redictive grou suggests that further analysis into their mechanisms and otential involvement in signaling athways germane to cancer is warranted.
Can their analysis be relicated?
In many microarray studies, how the data are analyzed may be more imortant than how they were generated. It is therefore incumbent uon researchers to describe their analytical methods in enough detail to allow indeendent researchers to relicate their comutations on the same data. Ideally, these details take the form of recise equations or algorithms or the actual code used to analyze the data (24-26). Although the Glinsky et al. study (1) reorts many of the critical values needed to relicate their results, their reliance on urely verbal descritions leaves room for some ambiguity.
We attemted to relicate art of their analysis and turned our attention to the same lung cancer study used by the authors (27), which included survival data on 125 atients and microarray data from Affymetrix U95Av2 GeneChis. Our estimate of the standard gene exression signature is listed in Table 1. Because of redundancy, a total of 18 robe sets reresent these 11 genes on the U95Av2 array (Table 2). We used all 18 robe sets, which exanded the fixed gene signature vector from Table 1 into an 18-dimensional vector. For each of the 18 robe sets, we comuted the average exression over all 125 samles. We divided the exression vectors in individual tumor samles by the average exression value and then transformed the ratios by comuting the base-10 logarithm, roducing 18-dimensional vectors. We then comuted the earson correlation coefficient between the fixed gene exression signature vector and the individual vector of log ratios. This rocedure yielded 1 number, xi, for each tumor samle i = 1, . . . , 125, which reresented our best estimate of the stem cell–like henotye association index (SAI) described in the Glinsky et al. study (1).
We used SAI as a continuous covariate in a Cox roortional hazards model to redict survival in the full data set and found that SAI was significant (likelihood ratio test, = 0.0179). Because the coefficient xi in the model was ositive (0.798), higher values of SAI were associated with shorter survival. Using trial and error, we determined a threshold on the full data set that, when used to slit the samles into 2 grous, yielded the best results in a survival analysis. The otimum threshold occurred at 0.32 and searated the data into 40 samles with a stem cell–like rofile (xi > 0.32) and 85 samles without a stem cell–like rofile. To this extent, we were able to confirm the analysis reorted by Glinsky et al. (Figure 1).
Based on this reliminary attemt to relicate the authors’ analyses, we believe that their results should be greeted with cautious otimism. However, we were unable to validate their method when slitting the data multile times into training and test sets. Our slits were more challenging than the ones used by the authors, since we did not use outcome information to balance the slits. It is also not clear whether we followed the same rocedure for comuting the SAI that was used by Glinsky et al. (1).
Conclusions
Glinsky and colleagues (1) have roduced a stimulating analysis of a collection of microarray data exeriments. Using ideas that were well motivated by the underlying biology and combining data across secies from 2 different microarray studies, they identified an 11-gene signature that might be related to tumor behavior, and thus atient survival, in cancer. In order to validate this signature, they tested it retrosectively in a broad sectrum of microarray studies of different kinds of cancer.
In our hands, unsuervised analyses of exression levels across some of the same data sets used by Glinsky and colleagues (1) did not reveal any clinically relevant or statistically significant correlations with outcome. By introducing a weighting coefficient into their redictor model, they are in fact adding an element of suervision into the analysis. Further, as these coefficients are re-derived on each subsequent data set, the statistical and analytical issues incumbent with high-throughut technologies are insufficiently addressed (2).
Glinsky et al. roosed several models, but only the 11-gene metastatic TRAM tumor samle/NS set consistently differentiated samles in a clinically relevant manner (1). An imortant asect of this gene set and the corresonding coefficients used in determining the weighted survival redictor is that those genes that have been reviously identified as otential indicators of oor rognosis are given the greatest coefficients. That is, negative weighting correlates with longer survival. For examle, in the rostate cancer SAI, the authors assign the largest coefficient value to Ki-67, which is associated with increased roliferation in cancer and could therefore lead to greater aggressiveness and oorer rognosis.
In the search for a owerfully redictive set of cancer genes, there have been various signatures roosed that contain anywhere from 10–100 genes. However, recent studies have cast doubt on the ower of those sets that were validated within a single data set or even a single tumor tye (2-4). As such, the need for validation of a roosed gene exression signature across indeendent data sets is warranted. To their credit, Glinsky and colleagues have gone to great lengths to validate their 11-gene signature, albeit in a suervised manner, across multile tumor samles and multile tissue tyes. As researchers begin to question the results reorted from microarray studies (2-4, 28), the need for relicable analyses and indeendent corroboration grows more acute. Regardless of whether it was said by Niels Bohr or Yogi Berra, it is still the case that "rediction is very difficult, esecially about the future."
Footnotes
Nonstandard abbreviations used: ANK3, ankyrin 3; Bmi-1, murine leukemia viral-1; BUB1, budding uninhibited by benzimidazoles 1; CCNB1, cyclin B1; CES, carboxylesterase; FGFR2, FGF recetor 2; GBX2, gastrulation homeobox 2; HCFC1, host cell factor c1; Ki-67, antigen identified by mAb ki-67; KNTC2, kinetochore-associated 2; NS, eriheral nervous system; RNF2, ring finger 2; SAI, stem cell–like henotye association index; TRAM, transgenic adenocarcinoma of the mouse rostate; US22, ubiquitin-secific rotease 22.
Conflict of interest: The authors have declared that no conflict of interest exists.
References
Glinsky, G.V., Berezovska, O., and Glinskii, A.B. 2005. . Microarray analysis identifies a death-from-cancer signature redicting theray failure in atients with multile tyes of cancer. J. Clin. Invest. 115::1503-1521. doi:10.1172/JCI23412.
Michiels, S., Koscielny, S., and Hill, C. 2005. . rediction of cancer outcome with microarrays: a multile random validation strategy. Lancet. 365::488-492.
Ransohoff, D.F. 2004. . Rules of evidence for cancer molecular-marker discovery and validation. Nat. Rev. Cancer. 4::309-314.
Ransohoff, D.F. 2005. . Lessons from controversy: ovarian cancer screening and serum roteomics. J. Natl. Cancer. Inst. 97::315-319.
Cheng, W., Liu, J., Yoshida, H., Rosen, D., and Naora, H. 2005. . Lineage infedility of eithelial ovarian cancers is controlled by HOX genes that secify regional identity in the reroductive tract. Nat. Med. 11::531-537.
Dick, J.E. 2003. . Breast cancer stem cells revealed. roc. Natl. Acad. Sci. U. S. A. 100::3547-3549.
ark, I.K. et al. 2003. . Bmi-1 is required for maintenance of adult self-renewing haematooietic stem cells. Nature. 423::302-305.
Lessard, J., and Sauvageau, G. 2003. . Bmi-1 determines the roliferative caacity of normal and leukaemic stem cells. Nature. 423::255-260.
Itahana, K. et al. 2003. . Control of the relicative life san of human fibroblasts by 16 and the olycomb rotein Bmi-1. Mol. Cell. Biol. 23::389-401.
Molofsky, A.V. et al. 2003. . Bmi-1 deendence distinguishes neural stem cell self-renewal from rogenitor roliferation. Nature. 425::962-967.
Jacobs, J.J. et al. 1999. . Bmi-1 collaborates with c-Myc in tumorigenesis by inhibiting c-Myc-induced aotosis via INK4a/ARF. Genes Dev. 13::2678-2690.
Cahill, D.. et al. 1998. . Mutations of mitotic checkoint genes in human cancers. Nature. 392::300-303.
Hanks, S. et al. 2004. . Constitutional aneuloidy and cancer redisosition caused by biallelic mutations in BUB1B. Nat. Genet. 36::1159-1161.
Gao, A.C., Lou, W., and Isaacs, J.T. 2000. . Enhanced GBX2 exression stimulates growth of human rostate cancer cells via transcritional u-regulation of the interleukin 6 gene. Clin. Cancer Res. 6::493-497.
Igarashi, T. et al. 2004. . Divergent cyclin B1 exression and Rb/16/cyclin D1 athway aberrations among ulmonary neuroendocrine tumors. Mod. athol. 17::1259-1267.
Dubrez, L. et al. 2001. . Cell cycle arrest is sufficient for 53-mediated tumor regression. Gene Ther. 8::1705-1712.
Moffa, A.B., Tannheimer, S.L., and Ethier, S.. 2004. . Transforming otential of alternatively sliced variants of fibroblast growth factor recetor 2 in human mammary eithelial cells. Mol. Cancer Res. 2::643-652.
Chen, C., et al. 2005. Ubiquitin-roteasome degradation of KLF5 transcrition factor in cancer and untransformed eithelial cells. Oncogene. doi:10.1038/sj.onc.1208497..
Sanchez-Beato, M. et al. 2004. . Abnormal cG rotein exression in Hodgkin’s lymhoma. Relation with E2F6 and NFkaaB transcrition factors. J. athol. 204::528-537.
Hayashi, T., and Su, T.. 2001. . Regulating ankyrin dynamics: Roles of sigma-1 recetors. roc. Natl. Acad. Sci. U. S. A. 98::491-496.
Kretschmer, T. et al. 2002. . ainful neuromas: a otential role for a structural transmembrane rotein, ankyrin G. J. Neurosurg. 97::1424-1431.
Redinbo, M.R., Bencharit, S., and otter, .M. 2003. . Human carboxylesterase 1: from drug metabolism to drug discovery. Biochem. Soc. Trans. 31::620-624.
Marsh, S. et al. 2004. . harmacogenomic assessment of carboxylesterases 1 and 2. Genomics. 84::661-668.
Gentleman, R. 2005. . Reroducible research: a bioinformatics case study. Statistical Alications in Genetics and Molecular Biology [serial online]. 4::Article 2. htt://www.beress.com/sagmb/vol4/iss1/art2.
Gentleman, R., and Lang, D.T. 2004. Statistical Analyses and Reroducible Research. Bioconductor roject Working aers [serial online]. Working aer 2. htt://www.beress.com/bioconductor/aer2..
Leisch, F., and Rossini, A. 2003. . Reroducible statistical research. Chance. 16::46-50.
Bhattacharjee, A. et al. 2001. . Classification of human lung carcinomas by mRNA exression rofiling reveals distinct adenocarcinoma subclasses. roc. Natl. Acad. Sci. U. S. A. 98::13790-13795.
Ioannidis, J.. 2005. . Microarrays and molecular research: noise discovery? Lancet. 365::454-455.(John . Lahad, Gordon B. M)