| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
* Simons Center for Systems Biology, Institute for Advanced Study, Princeton, New Jersey, USA; and the
Helen Diller Family Comprehensive Cancer Center and Cancer Research Institute, University of California, San Francisco, California, USA
1Correspondence: Institute for Advanced Study, 1 Einstein Drive, Princeton, NJ 08540, USA. E-mail: chyeang{at}ias.edu
 |
ABSTRACT |
|---|
Key Words: cell cycle control • Ras pathway • Wnt pathway • P53 pathway • Igf • Akt pathway • TGFβ pathway
|
INTRODUCTION |
|---|
; the Cancer Genome Atlas project launched by the U.S. National Cancer Institute, ref. 2
). However, despite the fact that sequencing technologies have become cheaper and faster, large amounts of resources are still required to sequence many samples of all the common tumors, and the completion of these ambitious projects may still be years ahead. While the completion of these large-scale sequencing/resequencing projects will provide unparalleled quantities of information about genetic and molecular alterations of cancer, the data accumulated from decades of published work can already reveal the recurrent mutational patterns and crosstalk of major cancer pathways. In this review, we present a useful and unique analysis of the somatic mutation data from thousands of previous studies and identify the recurrent combinatorial mutational patterns in 45 different tissue types. The results demonstrate the heterogeneity of combinatorial patterns in different tissues, confirm the functional relations of genes in the pathways, indicate the differential couplings between pathways in different tissues, and reveal the temporal orders of mutational events. Furthermore, using the information of the combinatorial patterns we suggest candidates for new sequencing projects that can either reveal novel patterns or validate predictions deduced from existing patterns.
|
DATABASE OF SOMATIC GENE MUTATIONS IN CANCER |
|---|
. It extracts from a large number of publications the mutational records of 
3000 genes in various types of primary tumor tissues or cancer cell lines. The version of May 2007 contains 218,323 cancer samples and 514,020 records extracted from 4138 studies. Currently, COSMIC collects only small mutations in protein-coding regions, including point mutations, small insertions and deletions, frame shifts, and unspecified mutations in the sources.
As most studies were undertaken before the postgenomic era, when large-scale sequencing was either unattenable or nonaffordable, the majority of the 218,323 samples probe only one (89.71%) or a few genes. As expected, most studies target only a small set of well-known oncogenes and tumor suppressors, such as TP53, KRAS, and APC. An exception is the 785 cell line samples where large-scale screenings of 3303 genes were performed by the Sanger Institute (Cancer Cell Line Project; ref. 4
).
The sparseness of genes probed in most samples explains the low frequency of mutations: only 19% (41,475 of 218,323) of the samples contain at least one detected mutation, and only 9.04% (46,491 of 514,020) of the records are mutations. However, even common oncogenes or tumor suppressors are not mutated in more than half of the samples in which they are probed. TP53 mutation, for instance, comprises only 46% (419 of 902) of the samples probed for this gene.
|
MUTATIONAL SIGNATURES ARE HETEROGENEOUS IN DIFFERENT TISSUES |
|---|
lists the 44 genes and their mutational frequencies in 17 selected tissue types. (The mutational frequency table of all 45 tissue types is reported in the Supplemental Material.) Nearly all genes are known oncogenes or tumor suppressors in major signaling or regulatory pathways of cancer. Many of them are involved in cell growth and proliferation (Ras pathway: KRAS, NRAS, HRAS, BRAF; IGF-AKT pathway: PTEN, PIK3CA; Wnt pathway: APC, β-catenin; TGF-β pathway: SMAD4), cell cycle control (TP16, RB1), and stress response and apoptosis (TP53).
|
The mutational signatures of those genes are heterogeneous in different tissues. Some genes, such as TP53 and TP16, are frequently mutated across most tissues. Other common oncogenes and tumor suppressors, by contrast, exhibit clear tissue specificity. For instance, APC mutations are found primarily in gastrointestinal tissues such as pancreas (24 of 152 samples), stomach (113 of 606 samples), large intestine (1192 of 3602 samples), and small intestine (31 of 211 samples). They are rare in lung adenocarcinoma (2 of 145 samples) and lung small cell carcinoma (1 of 144 samples). By contrast, RB1 mutations appear primarily in lung small cell carcinoma (52 of 116 samples), central nervous system (17 of 249 samples), urinary tract (18 of 54 samples), and eye tissue (76 of 164 samples). They are rare in large intestine tissue (1 of 78 samples).
Other genes are mutated only in a specific set of cancers. For example, FLT3 encodes a receptor tyrosine kinase that regulates some events in hematopoiesis (5
6
7)
and is frequently mutated in acute myeloid leukemia (AML) (3108 of 13,468 samples). WT1 encodes a zinc-finger transcription factor involved in the development of the urogenital system (8)
and is mutated primarily in kidney (Wilms tumor, 86 of 975 samples). EGFR encodes an epidermal growth factor receptor (9
10
11)
and is mutated commonly in lung adenocarcinoma (713 of 3783 samples) and non small cell carcinoma (1303 of 3679 samples). Note that mutations in hematopoietic/lymphoid (e.g., FLT3) and lung (e.g., EGFR) cancers are often specific for certain subtypes of a tumor (e.g., FLT3 in acute myeloid leukemia and EGFR in adenocarcinoma/non-small cell carcinoma).
|
SOME DIFFERENCES BETWEEN TUMOR AND CELL LINE MUTATIONAL FREQUENCIES ARE STATISTICALLY SIGNIFICANT |
|---|
shows the mutational frequencies in cell lines and tumors. We employed the gene-tissue entries where the genes were probed in more than 50 samples in both cell lines and tumors. For each gene-tissue combination, we evaluate the P value of the null hypothesis that the mutational frequencies of the cell line and tumor data are identical, and report the significant patterns in Table 3
(also see the Supplemental Material). Six mutational signatures have significant differences (P<0.001). The disparate differences between cell lines and tumors are often confounded by the unbalanced sample sizes between the two subsets. More samples of cell lines or tumors are needed in order to understand whether these differences are truly meaningful.
|
|
|
TYPES OF SOMATIC MUTATIONS ARE CONSISTENT WITH THE FUNCTIONS OF GENES |
|---|
shows the occurrences of each type of mutation in each gene among all tissues. The refined version of Table 4
subdivided by tissue type and cell line/non-cell line data is reported in the Supplemental Material. Frame shifts occur almost exclusively in tumor suppressors, as they disrupt the proteins and their tumor-suppressing functions. Point mutations are common in both oncogenes and tumor suppressors, since they may either enhance or degrade the function of a protein. Insertions that result from translocations occur primarily in oncogenes (FLT3, 2915; KIT, 173; EGFR, 46; ERBB2, 32), except tumor-suppressor CEBPA (23 insertions). Many insertions are tandem duplications of protein domains (gene amplifications) (e.g., in FLT3, ref. 12
, and EGFR, ref. 13
), which may enhance the functions of oncogenes. By contrast, deletions have been observed in both oncogenes (e.g., EGFR, 728; KIT, 406; β-catenin, 182; PDGFRA, 46) and tumor suppressors (e.g., TP16, 964; PTEN, 48; CEBPA, 11). Most deletions are detrimental to protein functions and thus should be observed in tumor suppressors. However, deletion of the extracellular domains of receptors (e.g., EGFR; ref. 14
) may cause a ligand-independent firing and promote cell growth. Moreover, deletions of the amino acid residues near the NH2-terminal of β-catenin stabilize the protein against degradation and thus enhance its function (15)
.
|
|
COMBINATORIAL MUTATIONAL PATTERNS REVEAL THE FUNCTIONAL RELATIONS OF GENES IN MAJOR CANCER PATHWAYS |
|---|
, 17)
. Tumorigenesis is a consequence of a series of mutations accumulated over years. Mutations of two genes participating in the same pathway or process rarely confer a significant selective advantage compared to the single mutation, since the functional consequences of single and double mutations are similar. By contrast, functional consequences of mutations of multiple genes that participate in different pathways or functions may be additive or even synergistic in conferring an advantage to the tumor. Therefore, we would expect to observe a tendency of mutually exclusive mutations of genes in the same pathway and the tendency of co-occurring mutations of genes that populated distinct pathways.
We categorize the mutations of multiple genes into two types of combinatorial patterns: co-occurrence of mutations in multiple (
2) genes and mutual exclusion of mutations in two genes. Mutual exclusion of more than two genes can be derived from pairwise relations because mutual exclusion relations are transitive. We define a test statistic for combinatorial patterns as the likelihood ratio (LR) between the empirical frequency of co-occurrence and the expected frequency according to the best simpler model that fits the data (see Supplemental Material). For two genes g1 and g2, this score is reduced to P(g1,g2 mutated)/P(g1 mutated)P(g2 mutated). High scores indicate co-occurrence and low scores suggest mutual exclusion. The cutoff values of the score separating co-occurrence and mutual exclusion are determined by the background distribution of the scores (see Supplemental Material). To eliminate the effect of a small sample size we also evaluate the P value of the likelihood ratio (see Supplemental Material) and report only the combinatorial patterns with values of P 
0.05.
We obtain 105 significant combinatorial mutational patterns. The false discovery rate of permutation tests on the data is <0.01. Table 5
shows the frequencies of significant combinatorial mutational patterns in different tissues. These combinatorial patterns cover genes in 6 major pathways relevant to cancer: cell cycle control, stress response, Ras, insulin growth factor (IGF-AKT), Wnt, and TGF-β signaling pathways. The majority of the combinatorial patterns conform with the simple hypotheses of pathways. Table 6
counts the combinatorial patterns covering genes in each pair of pathways. Most co-occurring patterns contain genes in different pathways (off-diagonal entries in the top section of the table), whereas most mutually exclusive pairs are the genes in the same pathways (diagonal entries in the bottom section of the table). Furthermore, many combinatorial patterns appear in multiple tissues and multiple studies, suggesting they are not the artifacts from the contamination of tissue samples or special cases of certain tumor types. We present a summary of the combinatorial patterns observed within and between the 6 pathways examined.
|
|
|
A BRIEF OVERVIEW OF CANCER PATHWAYS |
|---|
shows the genes in the combinatorial patterns and the simplified pathways in which they are involved. The cell cycle is regulated by several pairs of cyclins/cyclin dependent kinases. The decision to enter cell division or remain in a stationary phase is controlled by cyclin D/CDK4/6 and cyclin E/CDK2 pairs at the G1 phase. CDK4 is inhibited by various proteins, such as TP15, TP16 (18
, 19)
, and TP21. Both cyclin D/CDK4/6 and cyclin E/CDK2 complexes activate E2F, a master transcription factor at the G1 phase (20)
. E2F is also inhibited by retinoblastoma (RB1) protein. The synthesis of cyclin D is regulated by several signal transduction pathways including Ras, Wnt, TGF-β, and IGF-AKT.
|
The Ras pathway is a MAP-kinase signaling pathway constituting families of growth factors, receptors, e.g., EGFR and ERBB in lung (9
10
11)
, PDGFRA and KIT in gastrointestinal soft tissues (21)
, and KIT in hematopoietic tissues (22)
, G-proteins (KRAS, ref. 23
; NRAS, ref. 24
; and HRAS, ref. 25
), receptor tyrosine kinases (e.g., FLT3; refs. 5
, 6
), protein kinases (e.g., BRAF; refs. 26
, 27
), protein phosphatases (e.g., PTPN11; ref. 28
), and transcription factors (e.g., CEBPA; ref. 29
). The Wnt pathway is important in development and tissue regeneration and is often altered in gastrointestinal tumors (30)
. Two genes in the Wnt pathway undergo frequent mutations: β-catenin and APC. APC, GSK-3β, and other proteins bind and phosphorylate β-catenin, targeting its degradation in the absence of the WNT ligand (30)
.
The IGF-AKT pathway transduces the survival signal in response to growth factor stimulation (31)
. The pathway is activated when PIP3 is phosphorylated by PI3 kinase PIK3CA and inhibited when it is dephosphorylated by PI3 phosphatase PTEN (32)
. In addition, PI3 kinase is also regulated by the Ras proteins (33
, 34)
. The TGF-β signaling pathway imposes growth inhibition under stressful conditions. SMAD4 encodes a protein in the TGF-β pathway (35
, 36)
. It activates the synthesis of TP15, which inhibits CDK4 (37)
.
The TP53 pathway responds to a variety of stress signals and controls major stress response processes including apoptosis, cell cycle arrest, and senescence (38
, 39)
. To coordinate the stress response TP53 transcriptionally regulates genes in other pathways such as TP21 in the cell cycle control (40)
and PTEN in the IGF-AKT pathway (41)
.
|
CELL CYCLE CONTROL, STRESS RESPONSES, IGF-AKT, AND TGF-β PATHWAYS |
|---|
). For instance, in 44 lung adenocarcinoma samples where TP16 and TP53 are cosequenced, TP16 and TP53 are mutated in 18 and 29 samples, respectively, and both genes are mutated together in 13 samples (LR 1.1; P8.25 x 10–4). RB1 and TP53 have a significant co-occurring pattern in lung small cell carcinoma (LR 1.11; P8.51x10–6), central nervous system (LR 1.66; P4.61x10–3), and urinary tract (LR 1.29; P6.14x10–3) tumors. Comutations between stress response and IGF-AKT pathways and between cell cycle control and IGF-AKT pathways are also often observed. PTEN and TP53 have a co-occurring mutational pattern in 6 different tissue types. PIK3CA and TP53 have a co-occurring mutational pattern in large intestine (LR 0.67; P3.45x10–2), breast (LR 0.82; P1.43x10–2), and endometrium (LR 0.94; P4.88x10–2). TP16 and PTEN have a co-occurring mutational pattern in acute lymphoblastic leukemia (ALL) (LR 1.6; P8.39x10–3), skin (LR 1.05, P7.62x10–4), and central nervous system (LR 1.17; P1.06x10–3). Furthermore, three genes belonging to each pathway—TP16, TP53, PTEN—also demonstrate a significant (yet weaker) co-occurring pattern in ALL (LR 0.98; P1.13x10–2), central nervous system (LR 0.85; P1.38x10–2), and urinary tract (LR 1; P4.87x10–2) tumors.
Genes in the TGF-β pathway are also comutated in the same sample with genes in the cell cycle control and stress response pathways. SMAD4 and TP53 have a co-occurring mutational pattern in the pancreas (LR 0.94; P3.04x10–2), large intestine (LR 1.33, P3.45x10–2), and upper aerodigestive tract (LR 1.15, P3.23x10–2), and TP16 and SMAD4 have a co-occurring mutational pattern in the pancreas (LR 1.15, P6.11x10–5). The co-occurring mutations of the three genes are statistically insignificant due to the insufficient number of samples probing all the three genes. No significant co-occurrence or mutually exclusive pattern was found between the IGF-AKT (PTEN, PIK3CA) and TGF-β (SMAD4) genes, since SMAD4 mutations occur mainly in the pancreas and large intestine, where PTEN and PIK3CA mutations are rare.
|
RAS, CELL CYCLE CONTROL, STRESS RESPONSE, IGF-AKT, AND TGF-β PATHWAYS |
|---|
2.08x10–2) and skin (LR 0.87, P2.10x10–3). KRAS and TP53 have a co-occurring pattern in lung adenocarcinoma (LR 0.85, P4.15x10–3), pancreas (LR 0.97, P4.44x10–4), and large intestine (LR 1.06, P3.35x10–3). NRAS and TP53 have a co-occurring pattern in both ALL (LR 1.11, P3.37x10–2) and AML (LR 2.07, P3.37x10–2). BRAF and TP53 have a co-occurring pattern in large intestine (LR 1.75, P3.59x10–2), thyroid (LR 0.96, P2.73x10–2) and skin (LR 0.73, P6.35x10–3). Other significant co-occurring pairs or triplets include KRAS and SMAD4 (pancreas); KRAS and PTEN (endometrium); NRAS and PTEN (ALL); BRAF and TP16 (skin); BRAF and PTEN (skin); KRAS and PIK3CA (large intestine); TP16, KRAS and SMAD4 (pancreas); TP16, KRAS, and TP53 (pancreas); KRAS, SMAD4, and TP53 (pancreas); BRAF, TP16, and PTEN (skin); TP16, NRAS, and PTEN (ALL); and TP16, NRAS, and TP53 (ALL). |
WNT, RAS, STRESS RESPONSE, IGF-AKT, AND TGF-β PATHWAYS |
|---|
. Co-occurring patterns of the Wnt genes with genes in Ras, stress response, IGF-AKT, and TGF-β pathways are observed. In large intestine, APC and KRAS demonstrate a strong co-occurring pattern (LR 1.23, P3.41x10–7). APC-BRAF (LR 0.95, P2.27x10–2), and β-catenin-KRAS (LR 1.13, P5.78x10–3) pairs also have co-occurring patterns. Large intestine samples also carry co-occurring patterns in the following pairs and triplets: APC and TP53; APC and SMAD4; APC and PIK3CA; APC, SMAD4, and TP53; APC, KRAS, and PIK3CA; and APC, KRAS, and TP53. No significant patterns between Wnt genes and cell cycle control genes were detected. This is mainly because the cell cycle genes (TP16, RB1) have rarely mutated or been cosequenced with APC or β-catenin in gastrointestinal tissues. |
INTRAPATHWAY PATTERNS |
|---|
1.05x10–31), NRAS and PTPN11 in ALL (LR 0, P3.63x10–2), KIT and PDGFRA in gastrointestinal soft tissues (LR 0.02, P8.22x10–25) (21)
, FLT3 and CEBPA in AML (LR 0.36, P1.54x10–2) (42)
. Curiously, these pairs include both genes that have clear upstream-downstream relations (KRAS and EGFR, RAS genes and BRAF, and FLT3 and CEBPA) and genes that serve complementary functions in the Ras pathway (different RAS genes, KIT, and PDGFRA). The alteration of any of those genes seems sufficient to promote abnormal cell growth.
Two genes involved in cell cycle control—TP16 and RB1—are mutually exclusive in the tissues of the central nervous system (LR 0.2, P1.21x10–2) and urinary tract (LR 0.25, P1.19x10–2). TP16 and RB1 reside along a linear pathway of cell cycle control (i.e., are epistatic). The mutation of either gene suffices to release the control of cell cycle progression and cause abnormal cell growth.
Two key genes of the Wnt pathway—APC and β-catenin—are mutually exclusive in the large intestine (LR 0.21, P1.53x10–2). Similar to TP16 and RB1, APC is upstream of β-catenin, regulating its degradation.
|
MUTUALLY EXCLUSIVE PATTERNS BETWEEN PATHWAYS AND CO-OCCURRING PATTERNS WITHIN PATHWAYS |
|---|
and the off-diagonal entries in the bottom section of Table 6
). A remarkable example is the interaction between Ras and Wnt pathways in gastrointestinal tissues. In large intestines, several pairs of genes in the two pathways comutate (APC and KRAS, β-catenin and KRAS, and APC and BRAF). For example, in 389 samples, 121 and 118 carry APC and KRAS mutations, respectively, and 45 carry both mutations (LR 1.23; P3.47x10–7). By contrast, in pancreatic mutations, the genes on these two pathways are mutually exclusive (β-catenin and KRAS, APC and KRAS). In 78 pancreatic cancers, 23 and 33 samples carry β-catenin and KRAS mutations, and only 2 carry double mutations (LR 0.21; P6.18x10–5). The distinct mutational patterns of the same set of genes in different tissues suggest the functional interactions between pathways are tissue dependent. In the large intestine, both Ras and Wnt pathways appear to be active, and mutations of genes on both pathways are frequent. In the pancreas, however, each pathway seems to be active only in specific subtypes of tissues or tumors. For instance, APC mutations are observed in the familial adenomatous polypopsis (FAP) that can give rise to pancreatic carcinomas (43)
, and β-catenin mutations occur in solid pseudopapillary tumors of the pancreas (44)
. By contrast, KRAS mutations are rarely observed in these tissues but are frequently detected in ductal carcinoma of the pancreas (45)
, where APC and β-catenin are not frequently mutated.
Several gene pairs in the Ras and IGF-AKT pathways demonstrate co-occurring mutational patterns: KRAS and PTEN (endometrium, likelihood 1.43, P9.97x10–3), BRAF and PTEN (skin; LR 1.81; P3.56x10–3), and NRAS and PTEN (ALL; LR 1.94; P3.37x10–2). However, NRAS and PTEN mutations are mutually exclusive in skin (LR 0.27; P6.44x10–3). It is argued that RAS and PTEN are functionally equivalent since both regulate the PI3-kinase activity (34)
. This coupling, however, cannot explain the comutations of RAS and PTEN in ALL and endometrium. The difference may be attributed to the tissue-specific different roles of RAS genes in a signal transduction pathway.
Most gene pairs in the Ras pathway have mutually exclusive mutational patterns. Yet several pairs of RAS genes demonstrate co-occurring patterns: HRAS and KRAS in soft tissue sarcomas (LR 2.49; P1.13x10–2), KRAS and NRAS in AML (LR 0.63; P4.3x10–3), and KRAS and NRAS in skin (LR 0.66; P3.74x10–2). The coexisting mutations, identical to other gain-of-function mutations of RAS genes, are observed at positions 12, 13, and 61. These anomalies may be explained by several possible causes. Even though different RAS proteins have very similar structures and functions, they may act in different pathways, exhibit dosage-specific effects, or possess subtle genetic interactions between wild-type and mutated RAS (46
, 47)
. Alternatively, they may be artifacts from the tumor tissue mixing of single RAS mutants in the experiments.
|
COMBINATORIAL MUTATIONAL PATTERNS ARE HETEROGENEOUS IN DIFFERENT TISSUES |
|---|
and summarize them as follows. The most prominent combinatorial patterns in ALL are co-occurrences among the genes in cell cycle control (TP16), stress response (TP53), IGF-AKT (PTEN), and Ras (NRAS) pathways. By contrast, the prominent combinatorial patterns in AML often contain genes involved in hematopoiesis, such as the pairs of FLT3 and CEBPA (mutually exclusive), FLT3 and NPM1 (co-occurred), and FLT3 and KIT (mutually exclusive).
TP53 and cell cycle control genes form co-occurring patterns in three types of lung cancer: adenocarcinoma, small cell carcinoma, and squamous cell carcinoma. In adenocarcinoma and squamous cell carcinoma, TP53 comutates with TP16, whereas in small cell carcinoma it comutates with RB1. Mutually exclusive patterns within the Ras pathway and co-occurring patterns among Ras, TP53, or TP16 are observed in adenocarcinoma but not in small cell carcinoma. By contrast, PTEN mutations co-occur with mutations of TP53 and RB1 in small cell carcinoma. LKB1, a serine/threonine kinase in the mTOR pathway (31)
, is mutually exclusive with TP53 in adenocarcinoma. This is because both LKB1 and TP53 regulate the same protein kinase (AMP-kinase) downstream of each of these tumor suppressors (31
, 41)
.
Combinatorial patterns of TP53, TP16, and genes in Ras and Wnt pathways appear in gastrointestinal tissues such as pancreas, large intestine, and biliary tract. Most interpathway combinations are co-occurring patterns, except that the Ras-Wnt interactions in pancreas are mutually exclusive. Only one significant combinatorial pattern involved in TP16 (TP16 and KRAS are comutated) appears in the large intestine since TP16 is infrequently mutated in the large intestine (8 in 273 samples). This finding is consistent with the previous observation that TP16 is silenced by hypermethylation in colon cancer (48)
. Furthermore, SMAD4 (TGF-β pathway) comutates with TP16 and KRAS in pancreas and with APC and TP53 in large intestine.
In skin, intrapathway combinations of the Ras pathway genes and interpathway combinations occur with TP16, TP53, and PTEN. Most Ras pathway genes are mutually exclusive except for KRAS and NRAS mutations, which can co-occur in soft tissue tumors (sarcoma). In addition, NRAS and PTEN mutations demonstrate a mutually exclusive pattern.
In the central nervous system, both TP16 and RB1 mutations occur, but they are mutually exclusive in other tissue types. They can form co-occurring patterns with TP53 and PTEN.
|
THE ORDER OF MUTATIONAL EVENTS CAN BE INFERRED FROM THEIR SUBSET RELATIONSHIPS |
|---|
). However, the order of mutations of certain genes may not be random, since the fitness of the cancer cell population may be path dependent. Following the clonal expansion model, an early mutation should be more prevalent in the clonal population than a late mutation. Therefore, if the mutation of gene A precedes the mutation of gene B, then the samples carrying B mutations should be substantially subsumed to the samples carrying A mutations. For each pair of genes, we quantify the subsumed relationships using the product of P(A mutated, B mutated) and P(B not mutated, A not mutated) and calculate the P value according to random permutations (see the Supplemental Material). Table 7
shows 7 gene pairs that yield significant subsumed relationships (P<0.08). In skin, PTEN mutations (10 samples) are completely subsumed to BRAF mutations (48 samples). It is thus likely that the BRAF mutation precedes the PTEN mutation. In pancreas, TP16 mutations (72 samples) are substantially contained in KRAS mutations (103 KRAS mutation samples, 55 double mutation samples). In ALL, 7 of the 8 PTEN mutation samples are contained in TP16 (17 samples) and TP53 (14 samples) mutations, suggesting that TP16 and TP53 mutations precede PTEN mutations. In some tissues (liver, fat, muscle), TP53 can increase the level of PTEN (41)
. As such, TP53 inactivation would not have led to selection against PTEN. However, such selection observed in this dataset occurs in different tissue types. In the large intestine, all SMAD4 mutations (6 samples) are contained in APC mutations (27 samples), and PIK3CA mutations (14 samples) are largely contained in KRAS mutations (35 samples, 10 double mutation samples). Both results are consistent with the previous study in colon cancer (16)
, where APC and KRAS mutations occur at early stages of cancer progression. Finally, in the central nervous system, RB1 mutations (10 samples) are largely contained in TP53 mutations (32 samples, 9 samples of double mutations).
|
|
OBSERVED MUTATIONAL PATTERNS SUGGEST NEW COSEQUENCING TARGETS |
|---|
; the Cancer Genome Project, ref. 1
). To guide the analysis of those efforts, it is possible to identify and prioritize the cosequencing targets based on the information extracted from this dataset. This "experimental design" approach can save resources and has been shown to be useful in systems biology, e.g. (50
, 51)
.
We adopt two criteria to select cosequencing targets. The first criterion is to identify the gene pairs that could possibly reveal significant co-occurring or mutually exclusive patterns that are rarely sequenced together. For each tissue, we identified the gene pairs where each gene is mutated in more than 100, or 25%, of the probed samples, and both genes are cosequenced in less than 50 samples. Table 8
lists the candidates identified by this criterion. Many gene pairs are frequently mutated but never cosequenced in any sample. For instance, in thyroid, BRAF is mutated in 35% of samples (1752 of 4936), and RET—a receptor tyrosine kinase (52)
—is mutated in 36% of samples (205 of 568). Yet BRAF and RET have not been cosequenced in any sample. Other examples include TP16-PTCH (a receptor in the hedgehog signaling pathway; ref. 53
) in skin, FLT3-ABL1 (a tyrosine kinase) in AML, PDGFRA-SMARCB1 (a chromatin structure regulator; ref. 54
) in gastrointestinal soft-tissue sarcomas, and many more.
|
An alternative criterion for selecting target genes for cosequencing derives from the assumption that mutually exclusive relations are transitive. If A and B are in the same pathway, and B and C are in the same pathway, then A and C should be in the same pathway. Following the pathway hypothesis, the mutually exclusive relations of A-B and B-C are passed to A-C as well. Therefore, we can validate the transitivity of mutual exclusions by selecting the triplets A-B-C where two pairwise patterns (say A-B and B-C) are both mutually exclusive and the third pair (A-C) is cosequenced in less than 10 samples.
The only triplet passing this criterion is CEBPA-FLT3-KIT in AML. Both CEBPA-FLT3 and FLT3-KIT pairs are mutually exclusive in AML (CEBPA-FLT3: LR 0.36, P<0.015; FLT3-KIT: LR 0, P<0.042). Yet CEBPA and KIT have not been cosequenced in any AML sample. We predict they are mutually exclusive. In fact, all three of the genes are in the Ras pathway.
|
CONCLUSIONS AND FUTURE DIRECTIONS |
|---|
1) The cell or tissue type that will develop into a tumor determines which oncogene and tumor-suppressor gene combinations of mutations will be selected in that tumor. There is a strong tissue specificity to the pattern of mutations selected in a tumor, and this has little to do with selective expression of those genes in that tissue type. This result suggests that signal transduction pathways often act in a tissue-specific fashion, playing different roles in different tissues.
2) Certain combinations of gene mutations repeatedly occur in tissue-specific cancers but are rarely found in 100% of those cancers, suggesting other pathways to produce the same cancer types. An exception to this are some of the leukemias and lymphomas, which may have up to 100% of the same mutational patterns and therefore appear to be more simple. When combinations of mutations in three genes do occur in a tumor, they most commonly reside in genes that are located in three separate signal transduction pathways involved in the cell cycle, stress responses, and cell growth and division.
3) Some combinations of gene mutations are rarely or never found in the same tumor, even though both mutations occur at a high frequency in that tumor type. These combinations are commonly in the same signal transduction pathway, such as APC and β-catenin mutations. These types of observations confirm our present knowledge of signal transduction pathways by fulfilling the expectations for mutually exclusive mutations.
4) Combinations of mutations in genes appear to occur together in one tissue type that do not occur together in another tissue type. For example, a common co-occurrence of either KRAS and β-catenin or APC appears in tumors from the large intestine, while tumors from the pancreas have a high occurrence of KRAS and β-catenin mutations (22/78 and 33/78, respectively), but these two mutations almost never occur in the same tumor (2/78 tumors). These data demonstrate that similar signal transduction pathways act differently in different cell or tissue types.
5) Clear examples can be found of either one gene (APC) or another gene (β-catenin) in the same pathway being mutated but at different frequencies in a tumor type. Whether this finding is do to the cross-sectional size of a gene, the nature of the mutation required to inactivate or activate the gene product, or some other variable is not clear.
6) The three RAS family genes (H, K, and N) can be mutated in pairs in some tumors and exhibit mutations that are mutually exclusive in other tumors. At times, mutating two or more of these three genes in the same tumor confers a selective phenotypic advantage on the tumor. In some tissue types these genes can partially substitute for each other, whereas in other cell types they cannot substitute.
7) Signal transduction and stress response pathways are often connected, yet the co-occurring mutations between these pathways still exist. PIK3CA is regulated by RAS genes, but the comutations of KRAS and PIK3CA are observed in the large intestine, and the comutations of PTEN and RAS genes are found in skin, ALL, and endometrium. PTEN is regulated by TP53 in some tissue types (41)
, but their comutations are found in six tissue types. This finding suggests that the connectivity alone does not suffice to determine the mutational patterns. Tissue-specific gene expressions and effects on other genes may also play roles in selection.
8) The patterns of mutations in cell lines when compared to tumors have some clear differences (increased p53 and ARF mutations, etc.), but many similarities between the cells growing in these very different environments remain.
9) It has been possible to examine the frequencies of different mutations in a tumor group and the preferred coexistence of two mutations in that group of tumors to discern an order of the mutations that were selected during the development of the tumor. The detected ordering of these random mutations does imply that this process indeed plays a role in tumor formation. Some types of mutations may be "gatekeepers," or the first mutation could raise the mutation rate for subsequent mutations. If this concept is correct and common, we will need to develop models to test these ideas.
The results of this analysis point out both the limitations and the advantages in carrying out whole-genome sequencing of cancer genomes vs. selected genome sequencing. Clearly we have learned some things from our progress to date. It will be interesting to see what the NCI pilot project adds to this information.
Received for publication March 3, 2008. Accepted for publication March 20, 2008.
|
REFERENCES |
|---|
This article has been cited by other articles:
|
|
 |
|
  G. Weissmann Monumental Revolutions: Scientific, Sanitary and 'Omic FASEB J, November 1, 2009; 23(11): 3639 - 3643. [Full Text] [PDF]
|
|
|
|
 |
|
 P. Pajer, V. Karafiat, V. Pecenka, D. Prukova, J. Dudlova, J. Plachy, P. Kasparova, and M. Dvorak Industasis, a Promotion of Tumor Formation by Nontumorigenic Stray Cells Cancer Res., June 1, 2009; 69(11): 4605 - 4612. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
TOP
ABSTRACT
INTRODUCTION