as a potential mediator of tumorigensis by Cyclin D1 overexpression(10).
- pathway finding by gene expression profiling
Huan Liu
Genetics 144, Oncogenomics (Winter 2005, Dr. Charles Brenner)
I. Expression profiling helps cancer research
Human cancers are very complex and heterogenous diseases which progress from multiple genetic hits and involve defects in various biological pathways and factors at different levels. For studying the layered complexity of cancer, gene expression profiling using cDNA microarray and data mining offers a powerful tool as it allows analysis of patterns of gene expression in genome-wide scale(1).
II. Application of expression profiling in gene function and pathway discovery
Gene deletion or inactivation, for example, by knock-out or mutation, can be combined with microarray data to investigate gene function and novel interactions (2).
Many of the early studies that have proved the power of this approach are done in lower organism such as Saccharomyces cerevisiae because of the relative small transcriptome size and convenience of genetic manipulation. Hughes et al.(3) constructed 300 different single-gene disruption mutants in yeast, and then the transcription profiles of the mutants are compared to the wild type strain.
In the resulting patterns (fig1B), horizontal axis represents groups of co-regulated transcripts. These genes respond similarly to the disruption of other genes. This kind of similarity is often the result of uninteresting downstream convergence of pathways, such as global stress signature. However, proximity on the vertical axis means, two genes, when disrupted, produce similar molecular phenotype. And these mutants that display similar expression profiles are very likely to be involved in the same cellular process and share cellular function (2).
So, this provided a way of major pathway groupings. And by comparing expression profiles of uncharacterized ORF mutants to profiles of known mutants, they also successfully assigned new genes to cellular pathways.
Changes in gene expression patterns by introducing normal or mutated genes can also be used with microarray to determine gene function.
Typical study of this kind is overexpression of certain protein or its mutant type in cell lines and examine its effect on global gene expression profiling. Co-regulation of genes across these conditions reveals functional gene groups. This can lead to the identification of potential targets of the protein of interest in cancers and contribute to the understanding of its possible oncogenic or tumor suppressing mechanism. Proteins that have been investigated this way include some oncoproteins, such as PAX-FKHR(4), CMYC(5,6),cyclinD1(9), and tumor suppressors, for example, P53(7,8).
An elegant example of the power of this approach is provided by Lamb et al
at Dana-Farber Cancer Institute (9). By using a combination of molecular genetics, cDNA
microarray and data mining of human tumor expression databases, they successfully identified transcription factor C/EBP as a potential mediator of tumorigensis by Cyclin D1 overexpression(10).
III. Mechanisms of cyclin D1 in tumorigenesis: cdk-dependent vs cdk-independent activities?
Protooncogene cyclin D1 is overexpressed in 45% of breast cancers and a broad range of other human tumors(10), in many cases, as a consequence of chromosomal translocations, amplifications(11), etc.
Physiological role of Cyclin D1 is positive regulator of G1 phase progression in cell cyle(12), which is achieved in a cdk-dependent manner. It functions through activating cdk4/6 to catalyze initial phosphorylation of Rb(13), releases E2F transcription factors from pRb-E2F complexes and derepresses E2F target genes(14) which are genes required for progression from G1 to S phase including cyclin E(fig1A). Then, complete inactivation of pRb is mediated by cyclinE-cdk2 complex, and this is facilitated by a non-catalytic funciton of cyclinD-cdk4/6 complex by sequestering cdk inhibitors, p21 and p27(15, fig1B). In addition, cyclin D1 has also been shown to affect the activity of several transcription factors in a cdk-independent manner(fig1C), including oestrogen receptor, androgen receptor, STAT3 and BETA/NeuroD(16).
It has long been assumed that cyclin D1, as activator of cdk4/6, also underlies its pathological activity in tumorigenesis by promoting proliferation. And cdks became drug development targets, with drugs such as flavopirdol and CYC202(17). However, experimental evidence failed to support this model(18,19), suggesting that the pathological function of cyclinD1 in tumors is not simply to promote proliferation by activating cdks.
IV. Identification a cdk-independent mechanism of cyclin D1 in human cancer
- gene expression profiling study by Lamb et al(9)
1. Schematic overview
Beauty of this work by Lamb and colleagues lies in that they generated a mutant
cyclin D1 protein which was incapable of activating cdk to compare it's
overexpression microarray data with that of wild type protein, and then by
integrating the gene profiling data from both cell line and hundreds of human
tumors and by using data mining technology they developed a mechanistic
understanding of cyclin D1 function in human cancer, which was further confirmed
by functional analysis.
Pestell at DDT 2003 (17) has nicely reviewed the schematic methodology used by Lamb and colleagues to identify the cdk-independent oncogenic mechanism of cyclin D1.
2. Molecular signature of cdk-independent effects of cyclin D1 overexpression in breast cancer cell line
To differentiate between cdk-dependent and cdk-independent cyclin D1 activities, Lamb and colleagues generated a mutant cyclinD1 which is incapable of activating cdk4. They ectopically expressed either this mutant or wild type cyclin D1 or GFP (as control) in breast cancer cell line MCF7 by recombinant adenoviral vector(fig2A), collected RNA at intervals up to 24 hours post-infection, then used cDNA microarrays to compare the gene expression profile by overexpressing this mutant with the profile by overexpressing wild type cyclinD1.
16 genes that consistently had a greater than 3 fold change in expression following cycin D1 overexpression were identified as wild type cyclin D1 expression signature(fig2B). Surprisingly, this wild type cyclinD1 signature was reproducibly induced by cyclin D1 K112E mutant although the mutant is incapable of activating cdk4(fig2C), indicating cdk-independent effects of cyclinD1 overexpression. Same selection methods identified 13 genes following cyclin D1 K112E mutant overexpression, 8 of which are already among the targets of wild type cyclin D1, and the other five are also largely affected by wild type protein(fig2D). These 21 genes were identified as induced by both wild type and mutant cyclin D1 and used as cyclin D1 expression signature for the putative cdk-independent function of cyclinD1 in the rest of the study.
3. Validation of the molecular signature of cyclin D1 in human tumors
This signature was established by overexpressing in experimental cell culture system. To figure out whether what was found in cell line really represents situation in human tumors, they further analyzed the Global Cancer Map (GCM) database(20), which contains gene expression profiles from 190 primary human tumors of 14 different histological types. By reasoning that biological relevant cyclin D1 target genes would be frequently co-expressed with endogenous cyclin D1 in these tumor samples, they ordered 6471 genes in GCM according to how closely their expression pattern across all 190 tumors approximated that of cyclin D1, then KS nonparametric statistic was employed to capture the position of the 21 signature genes identified previously in the cell culture system within this ordered list. So, if these 21 genes are really biological relevant cyclin D1 target genes, they should be frequently co-expressed with cyclin D1 in tumor samples, and should appear early in this ordered list, and the statistic S (KS score) should be large (Analysis methods).
And it was indeed that this set of target genes was significantly correlated with the level of cyclin D1 in human tumors (high KS score, and p=0.048). Most of the signature genes appear early in the list(fig3A). This correlation can be more clearly seen if comparing the pattern of cyclin D1 expression across the entire GCM with the expression of those individual targets(fig3B). So, the expression signature identified in vitro recapitulated cyclin D1 function in vivo, and supported the cdk-independent activity of cyclin D1 in human tumors.
Further evidence of this cdk-independent activity of cyclin D1 in human tumors was apparent when using the same KS analysis to examine the relationship between cyclin D1 levels and E2F target genes in human tumors. A set of 22 E2F target genes had been identified in another microarray experiment published before, however, this set of genes was not correlated with cyclin D1 in GCM , low KS score, p=0.668(fig3C). Instead, these E2F target genes were highly correlated with cyclin D3 expression across GCM, P=0.002, indicating a striking difference between D-type cyclins which are previously considered to perform similar biochemical activities, and supporting the idea that cyclin D1 may have unique mechanisms in tumorigenesis.
4. Discovery by data mining C/EBP
as participant in cyclin D1 pathway in human cancers
Next, Lamb et al. used KS scanning as data mining tool to identify genes among over 16,000 genes in GCM that are highly co-expressed with the 21 genes of the cyclin D1 signature. To achieve this goal(fig 4A), each individual gene X in the 16000 gene database was used as a prototype. The remaining genes are ordered according to degree of co-expression to that gene X, just as how they produced the ordered list of genes according to their co-expression with cyclin D1. Then, a KS score for gene X with the cyclin D1 target gene set was calculated, just as the KS score for cyclin D1 with the cyclin D1 target gene was calculated before. So, the larger the KS score, the earlier the cyclin D1 target genes appear in the ordered list, also indicating expression pattern of this gene X matches the expression pattern of most of the cyclin D1 target genes to a larger degree.
Genes ranked in the top 50 by this method are
listed in (fig4B).
The expression pattern of these genes across tumor samples are highly correlated
with expression of the 21 cyclinD1 target genes(fig4C).
Together with independent screens with several other tumor gene expression
databases(table1), a transcription factor CCAAT enhancer binding protein (C/EBP)
was consistenetly observed to be co-expressed with cyclin D1 expression signature
across 500 huamn tumor specimens, strongly indicating that it is very likely
involved in regulating genes affected by cyclinD1 overexpression in human
tumors. And the rapidity of some of the changes in gene expression followed by
overexpression of cyclin D1 in cell culture(fig2B)
did suggested involvement of transcription factor.
5. Confirmation of C/EBP
involvement by Functional analysis
To confirm the involvement of C/EBP
in the oncogenic mechanism of cyclin D1 which was identified in silico, they
conducted functional analysis of cyclin D1 target gene promoters by traditional
reporter assay.
First, an independent cyclin D1 target gene set
was selected for this validation studies(fig5A,5B),
according to similarity in temporal expression pattern to HSP70-2, a known
cyclin D1 target.
Again, CyclinD1 and CyclinD1K112E have similar effects on transcription of
target gene promoters. Through transfection-reporter assays and deletion or
point mutation(fig6A),
cyclin D1-responsive elements were mapped in the promoter of 6 out of 7 selected
target genes, and showed strong similarity to the consensus C/EBP
binding motif(fig6B).
Gel shift assay revealed formation of DNA-protein complexes containing C/EBP
with cyclin D1 response element(fig6C). Mutation
of these elements disrupted the complex formation(fig6C) and resulted in
refractory to cyclinD1 induction(fig6A)
, but showed higher basal activity. In addition, all of the cyclin D1 responsive
promoters tested were also activated by dominant-negative mutant C/EBP(C/EBPSpl)
to comparable degrees(fig6A),
and wild type C/EBP
negated transcription activation by cyclin D1 (fig6D),
suggesting that C/EBP
is a constitutive repressor of cyclin D1 targets and cyclin D1 overexpression
antagonizes this repression.
However, functional and biochemical analysis also revealed functional
interdependency between C/EBP
and cyclinD1 for gene transcription. The effects of cyclin D1 and C/EBP
are exerted through same sequence elements. Mutations on promoters that render
refractory to cyclin D1 were also unresponsive to C/EBPSpl(fig6A).
Ablation of C/EBP
by transfecting its dominant negative form C/EBPSpl
would abolish the ability of cyclin D1 to activate responsive promoters in a
dose-dependent manner(fig6E). In
C/EBP
null cells, no responsiveness of the promoter to cyclin D1 could be
detected(fig6F). Co-IP pulled down cyclin D1 with C/EBP
when co-transfected in MCF-7(fig6H),
indicating the function maybe mediated by physical contact of the two proteins.
V. Summary
may mediate tumorigenesis
VII. Reference
(1) Liotta, L., Petricoin, E.,(2000). Nature Review Genetics 1, 48-56.
(2) Cooper,C.S.,(2001) Breast Cancer Res. 3, 158-75.
(3) Hughes, T.R., Marton, M.J., Jones, A.R., Roberts, C.J., Stoughton, R., Armour, C.D., Bennett, H.A., Coffey, E., Dai, H., He, Y.D. et al., (2000). Cell 102, 109-126.
(4) Khan J, Bittner M, Saal LH, Teichmann U, Azorsa DO, Gooden GC, Pavan WJ, Trent JM, Meltzer PS,(1999). Proc Natl Acad Sci USA, 96, 13264-13269.
(5) Coller HA, Grandori C, Tamayo P, Colbert T, Lander ES, Eisenman RN, Golub TD,(2000), Proc Natl Acad Sci USA, 97, 3260-3265.
(6) Molecular Pattern Recognition: Whitehead/MIT Center for Genome Research [http://www.genome.wi.mit.edu/MPR]
(7) Zhao R, Gish K, Murphy M, Yuxin Y, Nottermann D, Hoffman WH, Tom E, Mack DH, Levine AJ,(2000). Genes Dev, 14, 981-993.
(8) http://www.rockefeller.edu/labheads/levine/levine.html
(9) Lamb, J., Ramaswamy, S., Ford, H.L., Contreras, B., Martinez, R.V., Kittrell, F.S., Zahnow, C.A., Patterson, N., Golub, T.R. and Ewen, M.E.,(2003). Cell 114, 323-334.
(10) Rosen JM (2003). Cell, 114,271-272.
(11) Hall, M., Peters, G.,(1996). Adv. Cancer Res. 68, 67-108.
(12) Sherr, CJ (1994). Cell 79, 551-555.
(13) Weinberg, RA (1995). Cell 81, 323-330.
(14) Dyson, N (1998). Genes Dev. 12, 2245-2262.
(15) Sherr, CJ, Roberts, JM (1999). Genes Dev. 13, 1501-1512.
(16) Coqueret, O (2002), Gene 299, 35-55.
(17) Pestell K (2003). DDT 8, 911-912.
(18) Weinstat-Saslow, D., Merino, M.J., Manrow, R.E., Lawrence, J.A., Bluth, R.F., Wittenbel, K.D., Simpson, J.F., Page, D.L. and Steeg, P.S.,(1995). Nat. Med. 1, 1257-1260.
(19) Oyama, T., Kashiwabara, K., Yoshimoto, K., Arnold, A. and Koerner, F.,(1998). Cancer Res. 58, 2876-2880.
(20) Ramaswamy, S., Tamayo, P., Rifkin, R., Mukherjee, S.C.-H., Yeang, C.-H., Angelo, M., Ladd, C., Reich, M., Latulippe, E., Mesirov, J.P. et al.,(2001). Proc. Natl. Acad. Sci. USA 98, 15149-15154.
(21) Hanash, S., (2004). Nat. Rev. Cancer 4, 638-643.
(22) Ewen, ME, Lamb, J (2004).
Trends in Mol.Med. 10, 158-162.