Patent Application
20120202207
This invention is related to the area of cancer diagnostics, prognostics, drug screening, and therapeutics. In particular, it relates to brain tumors in general, and glioblastoma multiforme, in particular.
Gliomas, the most common type of primary brain tumors, are classified as Grade I to Grade IV using histopathological and clinical criteria established by the World Health Organization (WHO)1. This group of tumors includes a number of specific histologies, the most common of which are astrocytomas, oligodendrogliomas, and ependymomas. Grade I gliomas, often considered to be benign lesions, are generally curable with complete surgical resection and rarely, if ever, evolve into higher-grade lesions2. However, tumors of Grades II and III are malignant tumors that grow invasively, progress to higher-grade lesions, and carry a correspondingly poor prognosis. Grade IV tumors (glioblastoma multiforme, GBM) are the most invasive form and have a dismal prognosis3,4. Using histopathologic criteria, it is impossible to distinguish a secondary GBM, defined as one which occurs in a patient previously diagnosed with a lower grade glioma, from a primary GBM which has no known antecedent tumor5,6.
A number of genes are known to be genetically altered in gliomas, including TP53, PTEN, CDKN2A, and EGFR7-12. These alterations tend to occur in a defined order in the progression to high grade tumors. TP53 mutation appears to be a relatively early event during astrocytoma development, while loss or mutation of PTEN and amplification of EGFR are characteristic of higher-grade tumors6,13,14. In oligodendrogliomas, allelic losses of 1p and 19q occur in many Grade II tumors while losses of 9p21 are largely confined to Grade III tumors15.
There is a continuing need in the art to identify the causes, identifiers, and remedies for glioblastomas and other brain tumors.
According to one aspect of the invention a method is provided of characterizing a glioblastoma multiforme (GBM) tumor in a human subject. A GBM tumor is analyzed to identify the presence or absence of a somatic mutation at codon 132 in isocitrate dehydrogenase 1 (IDH1) or at codon 172 in isocitrate dehydrogenase 2 (IDH2) in a GBM tumor of a human subject.
Also provided as another aspect of the invention is an isolated antibody which specifically binds R132H IDH1, or R132C IDH1, or R132S IDH1, or R132L IDH1, or R132G IDH1, but not R132 IDH1; or R172M IDH2, R172G IDH2, or R172K IDH2, but not R172; i.e., mutant forms of IDH1 or IDH2 which are found in GBM. Also provided is an isolated antibody which specifically binds R132 IDH1 or R172 IDH2, i.e., wild-type active sites of IDH1 or IDH2.
Another aspect of the invention is a method of immunizing a mammal. An IDH1 mutant polypolypeptide comprising at least 8 contiguous amino acid residues of a human IDH1 protein or an IDH2 mutant polypolypeptide comprising at least 8 contiguous amino acid residues of a human IDH2 protein found in a human tumor is administered to a mammal. The at least 8 contiguous amino acid residues comprise residue 132 or IDH1 or residue 172 of IDH2. Residue 132 or residue 172 is not arginine. Antibodies and/or T cells which are immunoreactive with epitopes found on the IDH1 or IDH2 mutant polypeptide but not found on normal IDH1 or IDH2 are produced.
Also provided as another aspect of the invention is an IDH1 or IDH2 mutant polypeptide comprising at least 8 but less than 200 contiguous amino acid residues of a human IDH1 or IDH2 protein found in a human tumor. The at least 8 contiguous amino acid residues comprise residue 132 of IDH1 or residue 172 of IDH2. Residues 132 or 172 are not R.
An additional aspect of the invention is an isolated polynucleotide comprising at least 18 but less than 600 contiguous nucleotide residues of a coding sequence of a human IDH1 or human IDH2 protein found in a human tumor. The at least 18 contiguous amino acid residues comprise nucleotides 394 and/or 395 of IDH1 or nucleotide 515 or IDH2. Nucleotides 394 and/or 395 of IDH1 are not C and/or G, respectively. Residue 515 of IDH2 is not G.
Another aspect of the invention is a method of immunizing a mammal. An IDH1 polypeptide comprising at least 8 contiguous amino acid residues of a human IDH 1 protein or an IDH2 polypeptide comprising at least 8 contiguous amino acid residues of a human IDH2 protein is administered to a mammal. The at least 8 contiguous amino acid residues comprise residue 132 of IDH1 or residue 172 of IDH2. Residue 132 or residue 172 is arginine. Antibodies and/or T cells which are immunoreactive with epitopes found on the IDH1 or IDH2 polypeptide are produced.
Also provided as another aspect of the invention is an IDH1 or IDH2 polypeptide comprising at least 8 but less than 200 contiguous amino acid residues of a human IDH1 or IDH2 protein. The at least 8 contiguous amino acid residues comprise residue 132 of IDH1 or residue 172 of IDH2. Residues 132 or 172 are R.
Still another aspect of the invention is a method of detecting or diagnosing glioblastoma multiforme (GBM) or minimal residual disease of GBM or molecular relapse of GBM in a human. A somatic mutation in a gene or its encoded mRNA or protein is determined in a test sample relative to a normal sample of the human. The gene is selected from the group consisting of those listed in
Yet another aspect of the invention is a method of characterizing a glioblastoma multiforme in a human. A CAN-gene mutational signature for a glioblastoma multiforme is determined by determining in a test sample relative to a normal sample of the human, a somatic mutation in at least one gene or its encoded cDNA or protein. The gene is selected from the group consisting of those listed in
Another method provided by the invention is for characterizing a glioblastoma multiforme tumor in a human. A mutated pathway selected from the group consisting of TP53, RB1, and PI3K/PTEN is identified in a glioblastoma multiforme tumor by determining at least one somatic mutation in a test sample relative to a normal sample of the human. The at least one somatic mutation is in one or more genes selected from the group consisting of TP53, MDM2, MDM4, RB1, CDK4, CDKN2A, PTEN, PIK3CA, PIK3R1, and IRS1. The glioblastoma multiforme is assigned to a first group of glioblastoma multiforme tumors that have a mutation in one of said pathways. The first group is heterogeneous with respect to the genes in the pathway that have a somatic mutation and homogeneous with respect to the pathway that has a somatic mutation.
Also provided is a method to detect or diagnose glioblastoma multiforme, or minimal residual disease of GBM or molecular relapse of GBM in a human. Expression is determined in a clinical sample of one or more genes listed in
Another aspect of the invention is a method to monitor glioblastoma multiforme burden. Expression in a clinical sample is determined of one or more genes listed in
Yet another aspect of the invention is a method to monitor glioblastoma multiforme burden. A somatic mutation is determined in a clinical sample of one or more genes listed in
Still another aspect of the invention relates to a method to detect or diagnose gliobastoma multiforme. Expression in a clinical sample of one or more genes listed in
A further aspect of the invention is a method to monitor gliobastoma multiforme burden. Expression in a clinical sample of one or more genes listed in
These and other embodiments which will be apparent to those of skill in the art upon reading the specification provide the art with new tools for analyzing, detecting, stratifying and treating GBM.
A sequence listing is part of this application.
In a genome-wide analysis of GBMs, we identified somatic mutations of codon 132 of the isocitrate dehydrogenase 1 gene (IDH1) in ˜12% of GBMs analyzed16. These mutations were found at higher frequency in secondary GBMs (5 of 6 patients evaluated). One interpretation of these data is that IDH 1 mutations occur in a subset of lower-grade gliomas, driving them to progress to GBMs. To evaluate this possibility, we have analyzed a large number of gliomas of various types. Remarkably, we found IDH1 mutations in the majority of early malignant gliomas. Furthermore, many of the gliomas without IDH1 mutations had analogous mutations in the closely related IDH2 gene. These results suggest that IDH mutations play an early and essential role in malignant glioma development.
Somatic mutations are mutations which occur in a particular clone of somatic cells during the lifetime of the individual organism. The mutation is thus not inherited or passed on. The mutation will appear as a difference relative to other cells, tissues, organs. When testing for a somatic mutation in a brain tissue suspected of being cancerous, a comparison can be made to normal brain tissue that appears to be non-neoplastic, or to a non-brain sample, such as blood cells, or to a sample from an unaffected individual.
The common amino acid at codon 132 of IDH1 and codon 172 of IDH2 in healthy tissues is arginine (R). Mutant codons have been found with substitutions of histidine (H), serine (S), and cysteine (C), leucine (L), and glycine (G) of IDH1 codon 132 and of methionine (M), lysine (K), and glycine (G) of codon 172 of IDH2. The mutations at codon 132 and codon 172 can be detected using any means known in the art, including at the DNA, mRNA, or protein levels. Antibodies which specifically bind to the arginine-132 form of the enzyme, the histidine-132 form of the enzyme, the serine-132 form of the enzyme, leucine-132 form of the enzyme, glycine-132 form of the enzyme, or the cysteine-132 form of the enzyme can be used in assays for mutation detection. Likewise antibodies which specifically bind to the arginine-172, methionine-172, lysine-172, or glycine-172 forms of IDH2 can be used in assays for mutation detection. Similarly, probes which contain codons for these amino acid residues in the context of the coding sequence of IDH1 or IDH2 can be used for detecting the gene or mRNA of the different forms. Primers which contain all or part of these codons can also be used for allele-specific amplification or extension. Primers hybridizing to regions surrounding these codons can be used to amplify the codons, followed by subsequent analysis of the amplified region containing codon 132 of IDH1 or codon 172 of IDH2.
Interestingly, the codon 132 mutations of IDH1 and codon 172 mutations of IDH2 have been found to be strongly associated with secondary GBM and with a favorable prognosis. Drugs can be tested against groups of glioblastoma patients that are stratified with regard to the 132nd amino acid residue of IDH1 and/or the 172nd amino acid residue of IDH2. The groups may comprise wild-type (arginine) and variants (combined) or variants (each separately). Drug sensitivity can be determined for each group to identify drugs which will or will not be efficacious relative to a particular mutation or wild-type (arginine). Both sensitivity and resistance information are useful to guide treatment decisions.
Once a codon 132 or 172 mutation is identified in a tumor, inhibitors of IDH 1 or IDH2 may be used therapeutically. Such inhibitors may be specific for a mutation in the tumor or may simply be an inhibitor of IDH1 or IDH2. Small molecule inhibitors as well as antibodies and antibody-derivatives can be used. Such antibodies include monoclonal and polyclonal antibodies, ScFv antibodies, and other constructs which comprise one or more antibody Fv moieties. Antibodies can be humanized, human, or chimeric, for example. Antibodies may be armed or unarmed. Armed antibodies may be conjugated to toxins or radioactive moieties, for example. Unarmed antibodies may function to bind to tumor cells and participate in host immunological processes, such as antibody-dependent cell-medicated cytotoxicity. Antibodies may preferentially bind to mutant versus wild-type IDH1 or IDH2, specifically bind to mutant versus wild-type IDH1 or IDH2, or bind equally to both mutant and wild-type IDH1 or IDH2. Preferably the antibodies will bind to an epitope in the active site which may include codon 132 or codon 172. Epitopes may be continuous or discontinuous along the primary sequence of the protein. Inhibitors may include alpha-methyl isocitrate, aluminum ions, or oxalomalate. Other inhibitors may be used and optionally identified using enzyme assays known in the art, including spectrophotometric assays (Kornberg, A., 1955) and bioluminescent assays (Raunio, R. et al., 1985). Inhibitors may be alternatively identified by binding tests, for example by in vitro or in vivo binding assays. Peptides and proteins which bind to IDH1 or IDH2 may also be used as inhibitors.
Inhibitory RNA molecules may be used to inhibit expression. These may be, for example, siRNA, microRNA or antisense oligonucleotides or constructs. These can be used to inhibit the expression of IDH1 or IDH2 as appropriate in a human.
Potential therapeutic efficacy can be tested for an antibody, polynucleotide, protein, small molecule, or antibody by contacting with cells, tissues, whole animals, or proteins. Indications of efficacy include modulation of enzyme activity, inhibition of cancer cell growth, prolongation of life expectancy, inhibition of cancer cell proliferation, stimulation of cancer cell apoptosis, and inhibition or retardation of tumor growth. Any assays known in the art can be used, without limitation. Combinations of candidates and combinations of candidates with known agents can be assessed as well. Known agents may include, for example, chemotherapeutic anti-cancer agents, biological anti-cancer agents, such as antibodies and hormones, radiation.
In order to raise or increase an immune response to a glioblastoma in a person or mammal with a tumor, in a person with a likelihood of developing a tumor, or in an apparently healthy individual, a polypeptide can be administered to the person or mammal. The polypeptide will typically comprise at least 6, at least 8, at least 10, at least 12, or at least 14 contiguous amino acid residues of human IDH1 protein including residue 132 or IDH2 including residue 172. Typically but not always, the polypeptide will contain a residue other than arginine at residue 132 of IDH1 or residue 172 or IDH2. In the situation where the person or mammal already has a tumor, the amino acid at residue 132 can be matched to the residue in the tumor. The polypeptide may comprise the whole of IDH1, but can comprise less than 200, less than 150, less than 100, less than 50, less than 30 amino acid residues. Although applicants do not wish to be bound by any mechanism of action, the polypeptide immunization may act though an antibody and/or T cell response. Polypeptides can be administered with immune adjuvants or conjugated to moieties which stimulate an immune response. These are well known in the art, and can be used as appropriate.
Antibodies which specifically bind to an epitope on IDH1 or IDH2 do so with a higher avidity or a higher association rate than they bind to other proteins. Preferably the higher avidity or rate of association is at least about 2-fold, 5-fold, 7-fold, or 10-fold relative to other proteins that do not contain the epitope.
An isolated polynucleotide can be used to encode and deliver the polypeptide for immunization. The polynucleotide can be used to manufacture the polypeptide in a host cell in culture, or may be used in a gene therapy context to raise an immune response in vivo upon expression in the vaccine recipient. Polynucleotides can also be used as primers or probes, which may or may not be labeled with a detectable label. Primers can be used for primer extension, for example, using a primer that is complementary to nucleotides adjacent to but not including either nt 394 or nt 395 of IDH1 or nucleotide 515 of IDH2. Products can be detected and distinguished using labeled nucleotides as reagents. Different labels may be used on different nucleotides so that the identity of the analyte can be readily determined. Typically the polynucleotide for use as a primer or probe will comprise at least 10, at least 12, at least 14, at least 16, at least 18, at least 20 contiguous nucleotides of IDH1 or IDH2 coding sequence. Typically the polynucleotide will comprise less than 600, less than 500, less than 400, less than 300, less than 200, less than 100 nucleotides of IDH1 or IDH2 coding sequence.
Our data identified IDH1 as a major target of genetic alteration in patients with GBM. All mutations in this gene resulted in amino acid substitutions at position 132, an evolutionarily conserved residue located within the isocitrate binding site (42). In addition, the only previously-reported mutation of IDH1 was another missense mutation affecting this same residue in a colorectal cancer patient (10). The functional effect of these IDH1 mutations is unclear. The recurrent nature of the mutations is reminiscent of activating alterations in other oncogenes such as BRAF, KRAS, and PIK3CA. The prediction that this mutation would be activating is strengthened by the lack of observed inactivating changes (i.e. frameshift or stop mutations, splice site alterations), the lack of alterations in other key residues of the active site, and by the fact that all mutations observed to date were heterozygous (without any evidence of loss of the second allele through LOH). Interestingly, enzymatic studies have shown that substitution of arginine at residue 132 with glutamate results in a catalytically inactive enzyme suggesting that this residue plays a critical role in IDH1 activity (46). However, the nature of the substitutions observed in GBMs is qualitatively different, with arginine changed to histidine or serine. Histidine forms hydrogen bonding interactions, with carboxylate as part of the catalytic activity of many enzymes (47), and could serve an analogous function to the known interaction of Arg132 and the α-carboxylate of isocitrate. It is conceivable that R132H alterations may lead to higher overall catalytic activity. Increased activity of IDH1 would be expected to result in higher levels of NADPH, providing additional cellular defenses against reactive oxygen species, preventing apoptosis and increasing cellular survival and tumor growth. Further biochemical and molecular analyses will be needed to determine the effect of alterations of IDH1 on enzymatic activity and cellular phenotypes.
Regardless of the specific molecular consequences of IDH1 and IDH2 alterations, it is clear that detection of mutations in IDH1 and IDH2 will be clinically useful. Although significant effort has focused on the identification of characteristic genetic lesions in primary and secondary GBMs, the altered genes identified to date are far from perfect for this purpose. For example, in comparing primary versus secondary GBMs, TP53 is mutated in ˜30% vs. 65%, respectively, EFGR amplification is present in ˜35% vs. 5-10%, and PTEN mutation is present in ˜25% vs. ˜5% (5). Our study revealed IDH1 mutation to be a novel and significantly more specific marker for secondary GBM, with 5 of the 6 (83%) secondary GBM samples analyzed having a mutation in this gene, while only 7 of 99 (7%) primary GBM patients had such alterations (P<0.001, binomial test). The sole secondary GBM patient sample that did not have an IDH1 mutation was both genetically and clinically unusual, harboring mutations of PTEN but not TP53, and occurring in an older patient (age 56 years) with a prior diagnosis of ganglioglioma (which is rarely known to undergo malignant transformation) (48). It is possible that this patient had two distinct CNS tumors which were completely unrelated, and that the GBM in this case was actually a primary tumor.
One intriguing hypothesis is that IDH1 alterations identify a biologically-specific subgroup of GBM patients, including both patients who would be classified as having secondary GBMs as well as a subpopulation of primary GBM patients with a similar tumor biology and more protracted clinical course (Table 4). Interestingly, patients with IDH1 mutations had a very high frequency of TP53 mutation and a very low frequency of mutations in other commonly-altered GBM genes. For example, such patients had TP53 mutation without any detected mutation of EGFR, PTEN, RB1, or NF1 in 83% of cases (10 of 12 patients); in contrast, only 12% of patients with wildtype IDH1 (11 of 93) had the same mutation pattern (
Mutations that have been found in GBM tumors are shown in
CAN-gene signatures can be determined in order to characterize a GBM. A signature is a set of one or more somatic mutations in a CAN gene. The CAN genes for GBM are listed in
Analysis of the mutated genes in the analyzed GBM tumors has revealed interesting involvement of pathways. Certain pathways frequently carry mutations in GBMs. A single gene mutation appears to exclude the presence of a mutation in another gene in that pathway in a particular tumor. Frequently mutated pathways in GBMs are the TP53, RB1, PI3K/PTEN pathways. Pathways can be defined using any of the standard reference databases, such as MetaCore Gene Ontology (GO) database, MetaCore canonical gene pathway maps (MA) database, MetaCore GeneGo (GG) database, Panther, TRMP, KEGG, and SPAD databases. Groups can be formed based on the presence or absence of a mutation in a certain pathway. Such groups will be heterogeneous with respect to mutated gene but homogeneous with respect to mutated pathway. As with CAN gene signatures, these groups can be used to characterize a GBM. Once a mutation in a pathway has been determined, a GBM can be assigned to a group of GBMs sharing the mutated pathway. The group can be used to assign a prognosis, to assign to a clinical trial group, to assign to a treatment regimen, and/or to assign for further characterization and studies. In a clinical trial group, drugs can be assessed for the ability to differentially affect GBMs with and without the mutated pathway. Once a differential effect is determined, the pathway can be used to assign patients to drug regimens, or to avoid unnecessarily treating patients in whom the drug will not have a beneficial effect. The drug in a clinical trial can be one which is previously known for another purpose, previously known for treating GBM, or previously unknown as a therapeutic. Among the genes in the pathways which may be found mutant are: TP53, MDM2, MDM4, RB1, CDK4, CDKN2A, PTEN, PIK3CA, PIK3RI, and IRS 1. This list is not necessarily exhaustive.
Expression levels can be determined and overexpression may be indicative of a new GBM tumor, molecular relapse, or minimal residual disease of GBM. Highly increased expression found in GBM tumors are shown in
Tumor burden can be monitored using the mutations listed in
Analysis of the GBMs revealed certain genes which are homozygously deleted. These are listed in
The data resulting from this integrated analysis of mutations and copy number alterations have provided a novel view of the genetic landscape of glioblastomas. The combination of different types of genetic data, including point mutations, amplifications, and deletions allows for identification of individual CAN-genes as well as groups of genes that may be preferentially affected in complex cellular pathways and processes in GBMs. Identification of virtually all genes previously shown to be affected in GBMs by mutation, amplification, or deletion validates the comprehensive genomic approach we have employed.
It should be noted, however, that our approach, like all genome-wide studies, has limitations. First we did not assess chromosomal translocations, which is one type of genetic alteration that could play an important role in tumorigenesis. However, observations of recurrent chromosomal translocations have only rarely been reported in cyotogenetic studies of GBM. We also did not assess epigenetic alterations, though our large scale expression studies should have identified any genes that were differentially expressed through this mechanism (
Despite these limitations, our studies provide a number of important genetic and clinical insights into GBMs. The first of these is that the pathways known to be altered in GBMs affect a larger fraction of gene members and patients than previously anticipated. A majority of the tumors analyzed had alterations in members of each of the TP53, RB1, and P13K pathways. The fact that all but one of the cancers with mutations in members of a pathway did not have alterations in other members of the same pathway is significant and suggests that such alterations are functionally equivalent in tumorigenesis. These observations also point to distinct opportunities for potential therapeutic intervention in these pathways in GBMs. The second observation is that a variety of new genes and pathways not previously implicated in GBMs were identified. Among the new pathways detected, a number of these appear to be involved in brain specific ion transport and signaling processes and represent interesting and potentially useful aspects of GBM biology.
These data immediately raise questions with important implications for the treatment and counseling of patients with GBMs as well as those with lower-grade gliomas. For example, are mutations in IDH also present in a subset of patients diagnosed with lower-grade gliomas (WHO grades I-III)? If IDH1 mutations are indeed found to be a relatively early genetic event in glioma progression, are these patients at increased risk of progression to GBM? Given the significant clinical difficulty of deciding which low grade glioma patients will receive adjuvant radiation therapy or chemotherapy (and how aggressive treatment should be), the knowledge that a patient is at increased risk for malignant progression would significantly alter the risk-benefit analysis of such treatment decisions. For pediatric patients, in whom radiation therapy can have particularly devastating effects on neurocognitive development and function, these decisions are particularly difficult and any additional risk-classification would be especially useful. IDH mutations may also provide one biological explanation for the occasional long-term GBM survivor, and could help to identify patients that would receive particular benefit from specific currently-available therapies. The utility of IDH as a clinical marker is likely to be enhanced by the fact that only a single codon of the gene needs to be examined to determine mutation status. Finally, it is conceivable that new treatments may be designed to take advantage of these IDH alterations, either as monotherapy or in combination with other agents. Along these lines, inhibition of mitochondrial IDH2 has recently been shown to result in increased sensitivity of tumor cells to a variety of chemotherapeutic agents (49). In summary, this finding of IDH mutations in a subset of GBM patients and in at least one other cancer type opens a new avenue of research that could illuminate a previously unappreciated aspect of human tumorigenesis.
The above disclosure generally describes the present invention. All references disclosed herein are expressly incorporated by reference. A more complete understanding can be obtained by reference to the following specific examples which are provided herein for purposes of illustration only, and are not intended to limit the scope of the invention.
DNA was extracted from primary tumor and xenograft samples and patient-matched normal blood lymphocytes obtained from the Tissue Bank at the Preston Robert Tisch Brain Tumor Center at Duke and collaborating centers, as previously described17. All brain tumors analyzed were subjected to consensus review by two neuropathologists. The panel of brain tumors consisted of 21 pilocytic astrocytomas and 2 subependymal giant cell gliomas (WHO Grade I); 31 diffuse astrocytomas, 51 oligodendrogliomas, three oligoastrocytomas, 30 ependymomas, and seven pleomorphic xanthoastrocytomas (WHO Grade II); 43 anaplastic astrocytomas, 36 anaplastic oligodendrogliomas, and seven anaplastic oligoastrocytomas (WHO Grade III); 178 GBMs and 55 medulloblastomas (WHO Grade IV). The GBM samples included 165 primary and 13 secondary cases. Fifteen of the GBMs were from patients<20 years old). Secondary GBMs were defined as those that were resected>1 year after a prior diagnosis of a lower grade glioma (WHO Grades I-III). Sixty-six of the 178 GBMs, but none of the lower grade tumors, had been analyzed in our prior genome-wide mutation analysis of GBMs16. In addition to the brain tumors, 494 non-CNS cancers were examined: 35 lung cancers, 57 gastric cancers, 27 ovarian cancers, 96 breast cancers, 114 colorectal cancers, 95 pancreatic cancers, seven prostate cancers, 4 chronic myelogenous leukemias, 7 chronic lymphocytic leukemias, 7 acute lymphoblastic leukemias, and 45 acute myelogenous leukemias. All samples were obtained in accordance with the Health Insurance Portability and Accountability Act. Acquisition of tissue specimens was approved by the Duke University Health System Institutional Review Board and the corresponding IRBs at collaborating institutions.
Exon 4 of the IDH1 gene was PCR-amplified and sequenced in the matched tumor and normal DNAs for each patient as previously described16. In selected patients without an R132 IDH1 mutation (those with Grade II or III lesions or secondary GBM), the remaining seven exons of IDH1 and all 11 exons of IDH2 were sequenced and analyzed for mutations. All coding exons of TP53 and PTEN were also sequenced in the panel of oligodendrogliomas, anaplastic oligodendrogliomas, anaplastic astrocytomas, and GBMs. EGFR amplification and CDKN2A/CDKN2B deletion were analyzed by quantitative real-time PCR in the same tumors18. Oligodendroglioma and anaplastic oligodendroglioma samples were evaluated for loss of heterozygosity (LOH) at 1p and 19q as previously described 15,19.
Clinical information included date of birth, date the study sample was obtained, date of pathologic diagnosis, date and pathology of any preceding diagnosis of a lower grade glioma, administration of radiation therapy and/or chemotherapy prior to the date that the study sample was obtained, date of last patient contact, and patient status at last contact Clinical information for survival analysis was available for all 482 primary brain tumor patients. Kaplan-Meier survival curves were plotted and the survival distributions were compared by the Mantel Cox log-rank test and the Wilcoxon test. Overall survival was calculated by using date of GBM diagnosis and date of death or last patient contact. The correlations between the occurrence of IDH1/IDH2 mutations and other genetic alterations were examined using Fisher's exact test.
The top CAN-gene list included a number of individual genes which had not previously been linked to GBMs. The most frequently mutated of these genes, IDH1, encodes isocitrate dehydrogenase 1, which catalyzes the oxidative carboxylation of isocitrate to α-ketoglutarate, resulting in the production of NADPH. Five isocitrate dehydrogenase genes are encoded in the human genome, with the products of three (IDH3 alpha, IDH3 beta, IDH3 gamma) forming a heterotetramer (α2βγ in the mitochondria and utilizing NAD(+) as an electron acceptor to catalyze the rate-limiting step of the tricarboxylic acid cycle. The fourth isocitrate dehydrogenase (IDH2) is also localized to the mitochondria, but like IDH1, uses NADP(+) as an electron acceptor. The IDH1 product, unlike the rest of the IDH proteins, is contained within the cytoplasm and peroxisomes (41). The protein forms an asymmetric homodimer (42), and is thought to function to regenerate NADPH and α-ketoglutarate for intraperoxisomal and cytoplasmic biosynthetic processes. The production of cytoplasmic NADPH by IDH1 appears to play a significant role in cellular control of oxidative damage (43) (44). None of the other IDH genes, other genes involved in the tricarboxylic acid cycle, or other peroxisomal proteins were found to be genetically altered in our analysis.
IDH1 was found to be somatically mutated in five GBM tumors in the Discovery Screen. Surprisingly, all five had the same heterozygous point mutation, a change of a guanine to an adenine at position 395 of the IDH1 transcript (G395A), leading to a replacement of an arginine with a histidine at amino acid residue 132 of the protein (R132H). In our prior study of colorectal cancers, this same codon had been found to be mutated in a single case through alteration of the adjacent nucleotide, resulting in a R132C amino acid change (10). Five additional GBMs evaluated in our Prevalence Screen were found to have heterozygous R132H mutations, and an additional two tumors had a third distinct mutation affecting the same amino acid residue, R132S (
Several important observations were made about IDH1 mutations and their potential clinical significance. First, mutations in IDH1 preferentially occurred in younger GBM patients, with a mean age of 33 years for IDH1-mutated patients, as opposed to 53 years for patients with wildtype IDH1 (P<0.001, t-test, Table 4. In patients under 35 years of age, nearly 50% (9 of 19) had mutations in IDH1. Second, mutations in IDH1 were found in nearly all of the patients with secondary GBMs (mutations in 5 of 6 secondary GBM patients, as compared to 7 of 99 patients with primary GBMs, P<0.001, binomial test), including all five secondary GBM patients under 35 years of age. Third, patients with IDH1 mutations had a significantly improved prognosis, with a median overall survival of 3.8 years as compared to 1.1 years for patients with wildtype IDH1 (P<0.001, log-rank test). Although younger age and mutated TP53 are known to be positive prognostic factors for GBM patients, this association between IDH1 mutation and improved survival was noted even in patients<45 years old (
Tumor DNA was obtained from GBM xenografts and primary tumors, with matched normal DNA for each case obtained from peripheral blood samples, as previously described (1). All samples were given the histologic diagnosis of glioblastoma multiforme (GBM; World Health Organization Grade IV), except for two Discovery Screen samples who were recorded as “high grade glioma, not otherwise specified”. Samples were classified as recurrent for patients in whom a GBM had been diagnosed at least 3 months prior to the surgery when the study GBM sample was obtained. There were 3 recurrent GBMs in the Discovery Screen, and 15 in the Prevalence Screen. Samples were classified as secondary for patients in whom a lower grade glioma (WHO grade I-III) had been histologically confirmed at least 1 year prior to the surgery when the study GBM sample was obtained. One Discovery Screen sample and 5 Prevalence Screen samples were classified as secondary.
Pertinent clinical information, including date of birth, date study GBM sample obtained, date of original GBM diagnosis (if different than the date that the GBM sample was obtained, as in the case of recurrent GBMs), date and pathology of preceding diagnosis of lower grade glioma (in cases of secondary GBMs), the administration of radiation therapy and/or chemotherapy prior to the date that the GBM sample was obtained, date of last patient contact, and patient status at last contact. All samples were obtained in accordance with the Health Insurance Portability and Accountability Act (HIPAA). All samples were obtained in accordance with the Health Insurance Portability and Accountability Act (HIPAA). As previously described, tumor-normal pair matching was confirmed by typing nine STR loci using the PowerPlex 2.1 System (Promega, Madison, Wis.) and sample identities checked throughout the Discovery and Prevalence screens by sequencing exon 3 of the HLA-A gene. PCR and sequencing was carried out as described in (1).
Paired normal and malignant tissue from 105 GBM patients were used for genetic analysis. Complete clinical information (i.e. all pertinent clinical information such as date of initial GBM diagnosis, date of death or last contact) was available for 91 of the 105 patients. Of these 91 patients, five (all IDH1-wildtype) died within the first month after surgery and were excluded from analysis (Br308T, Br246T, Br23X, Br301 T, Br139X), as was a single patient (Br119X) with a presumed surgical cure (also IDH1-wildtype) who was alive at last contact ˜10 years after diagnosis. Kaplan Meier survival curves were compared using the Mantel Cox log-rank test. Hazard ratios were computed using the Mantel-Haenszel method. The following definitions were used in the GBM patient grouping and survival analysis computations: 1) Patient age refered to the age at which the patient GBM sample was obtained. 2) Recurrent GBM designates a GBM which was resected >3 months after a prior diagnosis of GBM. 3) Secondary GBM designates a GBM which was resected>1 year after a prior diagnosis of a lower grade glioma (WHO I-III) Overall survival was calculated using date of GBM diagnosis and date of death or last patient contact. All confidence intervals were calculated at the 95% level.
Sequence analysis of IDH1 in 976 tumor samples revealed a total of 167 somatic mutations at residue R132, including R132H (148 tumors), R132C (8 tumors), R132S (2 tumors), R132L (8 tumors) and R132G (1 tumor) (
If IDH1 were critical to the development or progression of oligodendrogliomas and astrocytomas, we reasoned that alterations in other genes with similar functions to IDH1 might be found in those in those tumors without IDH1 mutations. We therefore analyzed the IDH2 gene, which encodes the only human protein homologous to IDH1 that utilizes NADP+ as an electron acceptor. Sequence evaluation of all IDH2 exons in these samples, revealed eight somatic mutations, all at residue R172: R172M in three tumors, R172K in three tumors, and R172G in two tumors (
To further evaluate the timing of IDH alterations in glioma progression, we assessed IDH1 mutations in seven patients with progressive gliomas in which both low- and high-grade tumor samples were available. Sequence analysis identified IDH1 mutations in both the low and high-grade tumors in all seven cases (
We also examined the oligodendrogliomas, anaplastic oligodendrogliomas, anaplastic astrocytomas, and a subset of GBMs for mutations of TP53 and PTEN, amplification of EGFR, deletion of CDKN2A/CDKN2B, and LOH of 1p/19q (
Comparison of these alterations with those in IDH1 and IDH2 revealed several striking correlations. Nearly all of the anaplastic astrocytomas and GBMs with mutated IDH1/IDH2 also had mutation of TP53 (82%), but only 5% had any alteration of PTEN, EGFR, or CDKN2A/CDKN2B (
Patients with anaplastic astrocytomas and GBMs with IDH1 or IDH2 mutations were significantly younger than those with wild-type IDH 1 and IDH2 genes (median age of 34 years vs. 58 years, p<0.001, Student's t-test). Interestingly, despite the lower median age of patients with IDH1 or IDH2 mutations, no mutations were identified in GBM from patients who were less than 20 years old (0 of 18 patients,
Our prior observation of improved prognosis for GBM patients with mutated IDH1 16 was confirmed in this larger data set and extended to include patients with mutations in IDH2. Patients with IDH1 or IDH2 mutations had a median overall survival of 39 months, significantly longer than the 13.5 month survival in patients with wild-type IDH1 (
The disclosure of each reference cited is expressly incorporated herein. The references in the following list are cited in the text with superscript reference numerals.
The references in the following list are cited in the text with reference numerals in parentheses. The disclosure of each is expressly incorporated herein.
We extended our previously-developed sequencing strategy for identification of somatic mutations to include 23,219 transcripts from 20,583 genes. These included 2783 additional genes from the Ensembl databases that were not present in the CCDS or RefSeq databases analyzed in previous studies (10, 11). In addition, we redesigned PCR primers for regions of the genome that (i) were difficult to PCR amplify and had been sub-optimally analyzed in prior studies; or (ii) were found to share significant identity with other human or mouse sequences. The combination of these new, redesigned, and existing primers sequences resulted in a total of 208,311 primer pairs (table S1; available on-line at Science 26 Sep. 2008: Vol. 321. no. 5897, pp. 1807-1812) that were successfully used for sequence analysis of the coding exons of these genes.
Twenty-two GBM samples (
In the first stage of this analysis, called the Discovery Screen, the primer pairs were used to amplify and sequence 175,471 coding exons and adjacent intronic splice donor and acceptor sequences in the 22 GBM samples and in one matched normal sample. The data were assembled for each amplified region and evaluated using stringent quality criteria, resulting in successful amplification and sequencing of 95.0% of targeted amplicons and 93.0% of targeted bases in the 22 tumors. A total of 689 Mb of sequence data were generated through this approach. The amplicon traces were analyzed using automated approaches to identify changes in the tumor sequences that were not present in the reference sequences of each gene, then alterations present in the normal control sample and in single nucleotide polymorphism (SNP) databases were removed from further analyses. The remaining sequence traces of potential alterations were visually inspected to remove false-positive mutation calls generated through our automated software. All exons containing putative mutations were then re-amplified and sequenced in the affected tumor and matched normal DNA samples. This process allowed confirmation of the mutation in the tumor sample and determined whether the alteration was somatic (i.e. tumor-specific) or was present in the germline. All putative somatic mutations were examined computationally and experimentally to confirm that the alterations did not arise through the aberrant co-amplification of related gene sequences (12).
#Fraction of nucleotides having PHRED20 scores or better (see Supporting Online Materials for additional information).
We found that 2043 genes (10% of the 20,661 genes analyzed) contained at least one somatic mutation that would be expected to alter the protein sequence. The vast majority of these alterations were single-base substitutions (94%), while the others were small insertions, deletions, or duplications. The tumor sample Br27P obtained from the patient previously treated with radiation therapy and chemotherapy (including temozolomide), had 1332 total somatic mutations, 17-fold higher than any of the other 21 patients (
After removing Br27P from consideration, the remaining 993 mutations were observed to be distributed relatively evenly among the 21 remaining tumors (
Both the total number and frequency of sequence alterations in GBMs were substantially smaller than the number and frequency of such alterations observed in cancers of the colon or breast, and slightly less than in pancreas (10, 11, 17). The most likely explanation for this difference is the reduced number of cell generations in glial cells prior to the onset of neoplasia. It has been suggested that up to half of the somatic mutations observed in colorectal cancers occur in epithelial stem cells during the normal cell renewal processes (16). As normal glial stem cells turn over much less frequently than mammary or colon epithelial cells, they would be expected to contain many fewer mutations when the tumor-initiating mutation occurred (18).
We further evaluated a set of 20 mutated genes identified in the Discovery Screen in a second screen, called a Prevalence Screen, comprising an additional 83 GBMs with well-documented clinical histories (table S2, available on line at Science 26 Sep. 2008: Vol. 321. no. 5897, pp. 1807-1812). These genes were mutated in at least two tumors and had mutation frequencies>10 mutations per Mb of tumor DNA sequenced. Nonsilent somatic mutations were identified in 15 of these 20 genes in the additional tumor samples (
In addition to the frequency of mutations in a gene, the type of mutation can provide information useful for evaluating its potential role in disease (19). Nonsense mutations, out-of-frame insertions or deletions, and splice site changes generally lead to inactivation of the protein products. The likely effect of missense mutations can be assessed through evaluation of the mutated residue by evolutionary or structural means. To evaluate missense mutations, we developed a new algorithm that employs machine learning of 56 predictive features based on the physical-chemical properties of amino acids involved in the substation and their evolutionary conservation at equivalent positions of conserved proteins (12). Approximately 15% of the missense mutations identified in this study were predicted to have a statistically significant effect on protein function when assessed by this method (
The same tumors were then evaluated for copy number alterations through genomic hybridization of DNA samples to Illumina high density oligonucleotide arrays containing ˜1 million SNP loci probes (21). We have recently developed a sensitive and specific approach for the identification of focal amplifications resulting in 12 or more copies per nucleus (6-fold or greater amplification compared to the diploid genome) as well as deletions of both copies of a gene (homozygous deletions) using such arrays (22). Such focused alterations can be used to identify underlying candidate genes in these regions. It is impossible to reliably identify such candidate genes in regions with larger chromosomal aberrations, such as those involving gains or losses of entire chromosomal arms, which occur frequently in tumors and are of unknown significance.
We identified a total of 147 amplifications (
Mutations that arise during tumorigenesis may provide a selective advantage to the tumor cell (driver mutations) or have no net effect on tumor growth (passenger mutations). The mutational data obtained from sequencing and analysis of copy number alterations were integrated in order to identify GBM candidate cancer genes (CAN-genes) that would be most likely to be drivers and therefore worthy of further investigation. The bioinformatic approach employed to determine if a gene was likely to harbor driver mutations involved comparison of the number and type of mutations observed in each gene to the number that would be expected due to the passenger mutation rates. For sequence alterations, we calculated upper and lower bounds of passenger rates. The upper bound was conservatively calculated as the total number of observed alterations minus those mutations occurring in known cancer genes divided by the amount of tumor DNA sequenced, while the lower bound was determined on the basis of the observed silent mutations and estimates of expected NS:S ratios (12). For copy number changes, we made the very conservative assumption that all amplifications and deletions were passengers when determining the background rate. For analysis of each gene, all types of alterations (sequence changes, amplifications and homozygous deletions), were then combined to estimate the passenger probability for that gene (see (12) for a more detailed description of the statistical methods).
The top-ranked CAN-genes, together with their passenger probabilities, are listed in
{circumflex over ( )}Fraction of tumors with point mutations indicates the fraction of mutated GBMs out of the 105 samples in the Discovery and Prevalence Screens.
&Fraction of tumors with amplifications and deletions indicates the number of tumors with these types of alterations in the 22 Discovery Screen samples.
Analysis of additional gene members within pathways affected by these genes identified alterations of critical genes in the TP53 pathway (TP53, MDM2, MDM4), the RB1 pathway (RB1, CDK4, CDKN2A), and the PI3K/PTEN pathway (PIK3CA, PIK3R1, PTEN, IRS1). These alterations resulted in aberrant pathways in a majority of tumors (64%, 68%, and 50%, respectively) and in all cases but one, mutations within each tumor affected only a single member of each pathway in a mutually exclusive manner (P<0.05) (Table 3). Systematic analyses of functional gene groups and pathways contained within the well-annotated MetaCore database (35) identified enrichment of mutated genes in additional members of the TP53 and PI3K/PTEN pathways as well as in a variety of other cellular processes, including those regulating cell adhesion as well as brain specific cellular pathways such those involving synaptic transmission, transmission of nerve impulses, and channels involved in transport of sodium, potassium and calcium ions (
#Fraction of affected tumors in 22 Discovery Screen samples
Gene expression patterns can inform the analysis of pathways because they can reflect epigenetic alterations not detectable by sequencing or copy number analyses. They can also point to downstream effects on gene expression resulting from the altered pathways described above. To analyze the transcriptome of GBMs, we performed SAGE (serial analysis of gene expression) (36) on all GBM samples used for mutation analysis for which RNA was available (total of 18 samples) as well as two independent normal brain RNA controls. When combined with massively parallel sequencing-by-synthesis methods (37-40), SAGE provides a highly quantitative and sensitive measure of gene expression.
The transcript analysis was first used to help identify target genes from the amplified and deleted regions that were identified in this study. Though some of these regions contained a known tumor suppressor gene or oncogene, many contained several genes that had not previously been implicated in cancer. In tables S5 and S6, a candidate target gene could be identified within several of these regions through the use of the mutational as well as transcriptional data.
Second, we attempted to identify genes that were differentially expressed in GBMs compared to normal brain. There was a high number (143) of genes that were expressed at an average 10-fold higher level in 18 GBMs analyzed (compared to normal brain samples). Among the 143 over-expressed genes, there were 16 that were secreted or expressed on the cell surface. Many of these were over expressed in the xenografts as well as in the primary brain tumors, suggesting new opportunities for diagnostic and therapeutic applications.
The top CAN-gene list (
was found to be somatically mutated in five GBM tumors in the Discovery Screen. Surprisingly, all five had the same heterozygous point mutation, a change of a guanine to an adenine at position 395 of the IDH1 transcript (G395A), leading to a replacement of an arginine with a histidine at amino acid residue 132 of the protein (R132H). In our prior study of colorectal cancers, this same codon had been found to be mutated in a single case through alteration of the adjacent nucleotide, resulting in a R132C amino acid change (10). Five additional GBMs evaluated in our Prevalence Screen were found to have heterozygous R132H mutations, and an additional two tumors had a third distinct mutation affecting the same amino acid residue, R132S (
#Recurrent GBM designates a GBM which was resected >3 months after a prior diagnosis of GBM.
{circumflex over ( )}Secondary GBM designates a GBM which was resected >1 year after a prior diagnosis of a lower grade glioma (WHO I-III).
&Overall survival was calculated using date of GBM diagnosis and date of death or last patient contact: patients Br10P and Br11P were alive at last contact. Median survival for IDH1 mutant patients and IDH1 wildtype patients was calculated using logrank test. Previous pathologic diagnoses in secondary GBM patients were oligodendroglioma (WHO grade II) in Br123X, low grade glioma (WHO grade I-II) in Br237T and Br211T, anaplastic astrocytoma (WHO grade III) in Br27P, and anaplastic oligodendroglioma (WHO grade III) in Br129X,
Several important observations were made about IDH1 mutations and their potential clinical significance. First, mutations in IDH1 preferentially occurred in younger GBM patients, with a mean age of 33 years for IDH1-mutated patients, as opposed to 53 years for patients with wildtype IDH1 (P<0.001, t-test, Table 4). In patients under 35 years of age, nearly 50% (9 of 19) had mutations in IDH 1. Second, mutations in IDH1 were found in nearly all of the patients with secondary GBMs (mutations in 5 of 6 secondary GBM patients, as compared to 7 of 99 patients with primary GBMs, P<0.001, binomial test), including all five secondary GBM patients under 35 years of age. Third, patients with IDH1 mutations had a significantly improved prognosis, with a median overall survival of 3.8 years as compared to 1.1 years for patients with wildtype IDH1 (P<0.001, log-rank test). Although younger age and mutated TP53 are known to be positive prognostic factors for GBM patients, this association between IDH1 mutation and improved survival was noted even in patients<45 years old (
The disclosure of each reference cited is expressly incorporated herein.
The protein coding exons from 23,781 transcripts representing 20,735 unique genes were targeted for sequencing. This set comprised 14,554 transcripts from the highly curated Consensus Coding Sequence (CCDS) database (http://www.ncbi.nlm.nih.gov/CCDS/), a further 6,019 transcripts from the Reference Sequence (RefSeq) database (http://www.ncbi.nlm.nih.gov/projects/RefSeq/) and an additional 3,208 transcripts with intact open reading frames from the Ensembl database (http://www.ensembl.org/). We excluded transcripts from genes that were located on the Y chromosome or were precisely duplicated within the genome. As detailed below, 23,219 transcripts representing 20,661 genes were successfully sequenced.
Consensus Coding Sequence (Release 1), RefSeq (release 16, March 2006) and Ensembl (release 31) gene coordinates and sequences were acquired from the UCSC Santa Cruz Genome Bioinformatics Site (http://genome.ucsc.edu). The positions listed in the Supplementary Tables correspond to UCSC Santa Cruz hg17, build 35.1. The single nucleotide polymorphisms used to filter-out known SNPs were those present in dbSNP (release 125) that had been validated by the HapMap project. BLAT and In Silico PCR (http://genome.ucsc.edu/cgi-bin/hgPcr) were used to perform homology searches in the human and mouse genomes.
Primer 3 software (http://frodo.wi.mit.edu/cgi-bin/primer3/primer3_www.cgi) was used to generate primers no closer than 50 by to the target boundaries, producing products of 300 to 600 by. Exons exceeding 350 by were divided into several overlapping amplicons. In silico PCR and BLAT were used to select primer pairs yielding a single PCR product from a unique genomic position. Primer pairs for duplicated regions giving multiple in silico PCR or BLAT hits were redesigned at positions that were maximally different between the target and duplicated sequences. A universal primer (M13F, 5′-GTAAAACGACGGCCAGT-3′; SEQ ID NO: 136) was added to the 5′ end of the primer with the smallest number of mono- or dinucleotide repeats between itself and the target region. The primer sequences used in this study are listed in table Si available on line at Science 26 Sep. 2008: Vol. 321. no. 5897, pp. 1807-1812.
Tumor DNA was obtained from GBM xenografts and primary tumors, with matched normal DNA for each case obtained from peripheral blood samples, as previously described (1). The Discovery Screen consisted of 22 tumor samples (15 xenografts and 7 primary tumors), with the Prevalence screen including another 83 samples (53 xenografts and 30 primary tumors). Additional clinical information regarding Discovery and Prevalence Screen samples is available in table S2, available on line at Science 26 Sep. 2008: Vol. 321. no. 5897, pp. 1807-1812. All samples were given the histologic diagnosis of glioblastoma multiforme (GBM; World Health Organization Grade IV), except for two Discovery Screen samples who were recorded as “high grade glioma, not otherwise specified”. Samples were classified as recurrent for patients in whom a GBM had been diagnosed at least 3 months prior to the surgery when the study GBM sample was obtained. There were 3 recurrent GBMs in the Discovery Screen, and 15 in the Prevalence Screen. Samples were classified as secondary for patients in whom a lower grade glioma (WHO grade I-III) had been histologically confirmed at least 1 year prior to the surgery when the study GBM sample was obtained. One Discovery Screen sample and 5 Prevalence Screen samples were classified as secondary.
Pertinent clinical information, including date of birth, date study GBM sample obtained, date of original GBM diagnosis (if different than the date that the GBM sample was obtained, as in the case of recurrent GBMs), date and pathology of preceding diagnosis of lower grade glioma (in cases of secondary GBMs), the administration of radiation therapy and/or chemotherapy prior to the date that the GBM sample was obtained, date of last patient contact, and patient status at last contact. All samples were obtained in accordance with the Health Insurance Portability and Accountability Act (HIPAA). All samples were obtained in accordance with the Health Insurance Portability and Accountability Act (HIPAA). As previously described, tumor-normal pair matching was confirmed by typing nine STR loci using the PowerPlex 2.1 System (Promega, Madison, Wis.) and sample identities checked throughout the Discovery and Prevalence screens by sequencing exon 3 of the HLA-A gene. PCR and sequencing was carried out as described in (1).
Paired normal and malignant tissue from 105 GBM patients were used for genetic analysis. Complete clinical information (i.e. all pertinent clinical information such as date of initial GBM diagnosis, date of death or last contact) was available for 91 of the 105 patients. Of these 91 patients, five (all IDH1-wildtype) died within the first month after surgery and were excluded from analysis (Br308T, Br246T, Br23X, Br301T, Br139X), as was a single patient (Br119X) with a presumed surgical cure (also IDH 1-wildtype) who was alive at last contact ˜10 years after diagnosis. Kaplan Meier survival curves were compared using the Mantel Cox log-rank test. Hazard ratios were computed using the Mantel-Haenszel method. The following definitions were used in the GBM patient grouping and survival analysis computations: 1) Patient age refered to the age at which the patient GBM sample was obtained. 2) Recurrent GBM designates a GBM which was resected>3 months after a prior diagnosis of GBM. 3) Secondary GBM designates a GBM which was resected>1 year after a prior diagnosis of a lower grade glioma (WHO I-III) Overall survival was calculated using date of GBM diagnosis and date of death or last patient contact. All confidence intervals were calculated at the 95% level.
CCDS, RefSeq and Ensembl genes were amplified in 22 GBM samples and one control samples from normal tissues of one of the GBM patients. All coding sequences and the flanking 4 by were analyzed using Mutations Surveyor (Softgenetics, State College, Pa.) coupled to a relational database (Microsoft SQL Server). For an amplicon to be further analyzed, at least three quarters of the tumors were required to have 90% or more of bases in the region of interest with a Phred quality score of 20. In the amplicons that passed this quality control, mutations identical to those observed in the normal sample as well as known single nucleotide polymorphisms were removed. The sequencing chromatogram of each detected mutation was then visually inspected to remove false positive calls by the software. Every putative mutation was re-amplified and sequenced in tumor DNA to eliminate artifacts. DNA from normal tissues of the same patient in which the mutation was identified was amplified and sequenced to determine whether the mutations were somatic. When a mutation was found, BLAT was used to search the human and mouse genomes for related exons to ensure that putative mutations were the result of amplification of homologous sequences. When there was a similar sequence with 90% identity over 90% of the target region, additional steps were performed. Mutations potentially arising from human duplications were re-amplified using primers designed to distinguish between the two sequences. Mutations not observed using the new primer pair were excluded. The remainder were included as long as the mutant base was not present in the homologous sequence identified by BLAT. Mutations originally observed in mouse xenografts were re-amplified in DNA from primary tumors and included either if the mutation was present in the primary tumors or if the mutant was not identified in the homologous mouse sequence identified by BLAT.
We further evaluated a set of 20 mutated genes that had been identified in the Discover Screen in a second (Prevalence) screen, which included an additional 83 GBMs (tabl S2). The genes selected were mutated in at least two tumors and had mutatio frequencies>10 mutations per Mb of tumor DNA sequenced. The primers used (table S1, available on line at Science 26 Sep. 2008: Vol. 321. no. 5897, pp. 1807-1812) and methods of analysis and duration of potential mutations were the same as in the Discovery screen. All somatic mutations observed in the Prevalence screen are reported in
The Illumina Infinium II Whole Genome Genotyping Assay employing the BeadChip platform was used to analyze tumor samples at 1,072,820 (1M) SNP loci. All SNP positions were based on the hg18 (NCBI Build 36, March 2006) version of the human genome reference sequence. The genotyping assay begins with hybridization to a 50 nucleotide oligo, followed by a two-color fluorescent single base extension. Fluorescence intensity image files were processed using Illumina BeadStation software to provide normalized intensity values (R) for each SNP position. For each SNP, the normalized experimental intensity value (R) was compared to the intensity values for that SNP from a training set of normal samples and represented as a ratio (called the “Log R Ratio”) of log2(Rexperimental/Rtraining set).
The SNP array data were analyzed using modifications of a previously described method (2). Homozygous deletions (HDs) were defined as three or more consecutive SNPs with a Log R Ratio value of −2. The first and last SNPs of the HD region were considered to be the boundaries of the alteration for subsequent analyses. To eliminate chip artifacts and potential copy number polymorphisms, we removed all HDs that were included in copy number polymorphism databases. Adjacent homozygous deletions separated by three or fewer SNPs were considered to be part of the same deletion, as were HDs within 100,000 by of each other. To identify the target genes affected by HDs, we compared the location of coding exons in the RefSeq, CCDS and Ensembl databases with the genomic coordinates of the observed HDs. Any gene with a portion of its coding region contained within a homozygous deletion was considered to be affected by the deletion.
As outlined in (2), amplifications were defined by regions containing three SNPs with an average LogR ratio 0.9, with at least one SNP having a LogR ratio 1.4. As with HDs, we excluded all putative amplifications that had identical boundaries in multiple samples. As focal amplifications are more likely to be useful in identifying specific target genes, a second set of criteria were used to remove complex amplifications, large chromosomal regions or entire chromosomes that showed copy number gains. Amplifications>3 Mb in size and groups of nearby amplifications (within 1 Mb) that were also >3 Mb in size were considered complex. Amplifications or groups of amplifications that occurred at a frequency of 4 distinct amplifications in a 10 Mb region or 5 amplifications per chromosome were deemed to be complex. The amplifications remaining after these filtering steps were considered to be focal amplifications and were the only ones included in subsequent statistical analyses. To identify protein coding genes affected by amplifications, we compared the location of the start and stop positions of each gene within the RefSeq, CCDS and Ensmbl databases with the genomic coordinates of the observed amplifications. As amplifications containing only a fraction of a gene are less likely to have a functional consequence, we only considered genes whose entire coding regions were included in the observed amplifications.
From the synonymous mutations observed in the Discovery Screen, we estimated a lower bound of the passenger rate. The lower bound was defined as the product of the synonymous mutation rate and the NS:S ratio (1.02) observed in the HapMap database of human polymorphisms. The calculated rate of 0.38 mutations/Mb successfully sequenced is likely an underestimate because selection against nonsynonymous mutations may be more stringent in the germline than in somatic cells. An upper bound was calculated from the total observed number of non-synonymous mutations/Mb after excluding the most highly mutated genes known to be drivers from previous studies (TP53, PTEN, and RB1). The resultant passenger mutation rate of 1.02 non-synonymous mutations/Mb represents an over-estimate of the background rate as some of the mutations in genes other than TP53, PTEN, and RB1 were likely to be drivers. A ‘Mid” measure of 0.70 mutations/Mb was obtained from the average of the lower and upper bound rates. For comparisons of the number and type of somatic mutations identified in the Discovery and Prevalence Screens, two sample t-tests between percents were used.
Expression analysis
SAGE tags were generated using a Digital Gene Expression-Tag Profiling preparation kit (Illumina, San Diego, Calif.) as recommended by the manufacturer. In brief, RNA was purified using guianidine isothiocyanate and reverse transcription with oligo-dT magnetic beads was performed on ˜1 ug of total RNA from each sample. Second strand synthesis was accomplished through RNAse H nicking and DNA polymerase I extension. The double-stranded cDNA was digested with the restriction enonuclease Nla III and ligated to an adapter containing a Mme I restriction site. After Mme I digestion, a second adapter was ligated, and the adapter-ligated cDNA construct was enriched by 18 cycles of PCR and fragments of 85 by were purified from a polyacrylamide gel. The library size was estimated using real-time PCR and the tags sequenced on a Genome Analyzer System (Illumina, San Diego, Calif.).
The statistical analyses focused on quantifying the evidence that the mutations in a gene or a biologically defined set of genes reflect an underlying mutation rate that is higher than the passenger rate. In both cases, the analysis integrates data on point mutations with data on copy number alterations (CNA). The methodology for the analysis of point mutations is based on that described in (3) while the methodology for integration across point mutations and CNA's is based on (2). We provide a self-contained summary herein, as several modifications to the previously described methods were required.
The mutation profile of a gene refers to the number of each of the twenty-five context-specific types of mutations defined earlier (3). The evidence on mutation profiles is evaluated using an Empirical Bayes analysis (4) comparing the experimental results to a reference distribution representing a genome composed only of passenger genes. This is obtained by simulating mutations at the passenger rate in a way that precisely replicates the experimental plan. Specifically, we consider each gene in turn and simulate the number of mutations of each type from a binomial distribution with success probability equal to the context-specific passenger rate. The number of available nucleotides in each context is the number of successfully sequenced nucleotides for that particular context and gene in the samples studied. When considering nonsynonymous mutations other than indels, we focus on nucleotides at risk, as defined previously (3).
Using these simulated datasets, we evaluated the passenger probabilities for each of the genes that were analyzed in this study. These passenger probabilities represent statements about specific genes rather than about groups of genes. Each passenger probability is obtained via a logic related to that of likelihood ratios: the likelihood of observing a particular score in a gene if that gene is a passenger is compared to the likelihood of observing it in the real data. The gene-specific score used in our analysis is based on the Likelihood Ratio Test (LRT) for the null hypothesis that, for the gene under consideration, the mutation rate is the same as the passenger mutation rate. To obtain a score, we simply transform the LRT to s=log(LRT). Higher scores indicate evidence of mutation rates above the passenger rates. This general approach for evaluating passenger probabilities follows that described by Efron and Tibshirani (4). Specifically, for any given score s, F(s) represents the proportion of simulated genes with scores higher than s in the experimental data, F0 is the corresponding proportion in the simulated data, and p0 is the estimated overall proportion of passenger genes (discussed below). The variation across simulations is small but nonetheless we generated and collated 100 datasets to estimate FO. We then numerically estimated the density functions f and f0 corresponding to F and FO and calculated, for each score s, the ratio p0·f0(s)/f(s), also known as “local false discovery rate” (4). Density estimation was performed using the function “density” in the R statistical programming language with default settings. The passenger probability calculations depend on an estimate of p0, the proportion of true passengers. Our implementation seeks to give an upper bound to p0 and thus provide conservatively high estimates of the passenger probability. To this end we set p0=1. We also constrained the passenger probability to change monotonically with the score by starting with the lowest values and recursively setting values that decrease in the next value to their right. We similarly constrain passenger probabilities to change monotonically with the passenger rate.
An open source package for performing these calculations in the R statistical environment, named CancerMutationAnalysis, is available at http://astor.som.jhmi.edu/˜gp/software/CancerMutationAnalysis/cma.htm. A detailed mathematical account of our specific implementation is provided in (5) and general analytic issues are discussed in (6).
Statistical Analysis of CNA. For each of the genes involved in amplifications or deletions, we further quantified the strength of the evidence that they drive tumorigenesis through estimations of their passenger probabilities. In each case, we obtain the passenger probability as an a posteriori probability that integrates information from the somatic mutation analysis of (3) with the data presented in this article. The passenger probabilities derived from the point mutation analysis serve as a priori probabilities. These are available for three different scenarios of passenger mutation rates and results are presented separately for each in
This combination approach makes an approximating assumption of independence of amplifications and deletions. In reality, amplified genes cannot be deleted, so independence is technically violated. However, because of the relatively small number of amplification and deletion events, this assumption is tenable for the purposes of our analysis. Inspection of the likelihood, in a logarithmic scale, suggests that it is roughly linear in the overall number of events, supporting the validity of this approximation as a scoring system.
Four types of data were obtained from the MetaCore database (GeneGo, Inc., St. Joseph, Mich.): pathway maps, Gene Ontology (GO) processes, GeneGo process networks, and protein-protein interactions. The memberships of each of the 23,781 transcripts in these categories were retrieved from the databases using RefSeq identifiers. In GeneGo pathway maps, 22,622 relations were identified, involving 4,175 transcripts and 509 pathways. For Gene Ontology processes, a total of 66,397 pairwise relations were identified, involving 12,373 transcripts and 4,426 GO groups. For GeneGo process networks, a total of 23,356 pairwise relationships, involving 6,158 transcripts and 127 processes, were identified. The predicted protein products of each mutated gene were also evaluated with respect to their physical interactions with proteins encoded by other mutated genes as inferred from the MetaCore database.
For each of the gene sets considered, we quantified the strength of the evidence that they included a higher-than-average proportion of drivers of carcinogenesis after consideration of set size. For this purpose, we sorted the genes by a score based on the combined passenger probability described above (taking into account mutations, homozygous deletions, and amplifications). We compared the ranking of the genes contained in the set with the ranking of those outside, using the Wilcoxon test, as implemented by the Limma package in Bioconductor (8), then corrected for multiplicity by the q-value method with an alpha of 0.2 (9).
We have developed a novel bioinformatics software pipeline (depicted below) to compute: (1) a score for ranking somatic missense mutations by the likelihood that they are passengers (LSMUT). The scores are based on properties derived from protein sequences, amino acid residue changes and positions within the proteins; and (2) qualitative annotations of each mutation, based on protein structure homology models.
We tested several supervised machine learning algorithms to identify one that would reliably distinguish between presumably neutral polymorphisms and cancer-associated mutations. The best algorithm was a Random Forest (12), which we trained on 2,840 cancer-associated mutations and 19,503 polymorphisms from the SwissProt Variant Pages (13) using parallel Random Forest software (PARF) [http://www.irb.hrien/cir/projects/info/parf]. Cancer-associated mutations were identified by parsing for the keywords “cancer”, “carcinoma”, “sarcoma”, “blastoma”, “melanoma”, “lymphoma”, “adenoma” and “glioma”. For each mutation or polymorphism, we computed 58 numerical and categorical features (see table below). Two mutations present in the GBM tumor samples were found in the SwissProt Variant Pages and removed from the training data. Because the training set contained ˜7 times as many polymorphisms as cancer- associated mutations, we used class weights to upweight the minority class (cancer-associated mutation weight was 5.0 and polymorphism weight was 1.0). The mtry parameter was set to 8 and the forest size to 500 trees. Missing feature values were filled in using the Random Forest proximity-based imputation algorithm (12) with six iterations. Full parameter settings and all data used to build the Random Forest are available upon request.
We then applied the trained forest to 594 GBM missense mutations and to a control set of 142 randomly generated missense mutations in transcripts of 78 genes that were found to be non-mutated in 11 colorectal cancers (5). For each mutation, the 58 predictive features were computed as described above and the trained forest was used to compute a predictive score for ranking the mutations. Specifically, the scores used are the fraction of trees that voted in favor of the “Polymorphic” class for each mutation.
To test the hypothesis that the scores of missense mutations in top-ranked CAN-genes were distributed differently than random missense mutations, we applied a modified Kolmogorov-Smirnov (KS) test, in which ties are broken by adding a very small random number to each score. The scores of missense mutations in the top 13 CAN genes were found to be significantly different from the mutations in the control set (P<0.001).
We estimate that mutations with scores<0.7 (˜15% of the missense mutations) are unlikely to be passengers. The threshold is based on the putative similarity of passengers to the neutral polymorphisms in the SwissProt Variant set, of which only ˜2% have scores<0.7. Scores of SwissProt Variants were obtained by randomly partitioning them into two folds, training a Random Forest on each (as described above) and then scoring each fold with the Random Forest trained on the other one.
The protein translations of mRNA transcripts found to have somatic missense mutations were input into ModPipe 1.0/MODELLER 9.1 homology model building software (14, 15). For each mutation, we identified all models that included the mutated position. If more than one model was produced for a mutation, we selected the model having the highest sequence identity with its template structure. The resulting model was used to compute the solvent accessibility of the wild type residue at the mutated position, using DSSP software (16). Accessibility values were normalized by dividing by the maximum residue solvent accessibility for each side chain type in a Gly-X-Gly tri-peptide (17). Solvent accessibilities greater than 36% were considered to be “exposed”, those between 9% and 35% were considered “intermediate”, and those <9% were considered “buried”. DSSP was also used to compute the secondary structure of the mutated position. We used the LigBase (18) and PiBase (19) databases to identify mutated residue positions in the homology models that were close to ligands or domain interfaces in the equivalent positions of their template structures. Finally, for each mutation, we generated an image of the mutation mapped onto its homology model with UCSF Chimera (20). The images and associated information for each mutation are available at http://karchinlab.org/Mutants/CAN-genes/pancreatic/Pancreatic_cancer.html). Model coordinates are available on request.
Bioconductor V. Gentleman, S. Carey, R. Dudoit, W. H. Irizarry, Eds. (Springer, New York, 2005) pp. 397-420.
This application was made using funds from the United States government. Therefore the U.S. government retains certain rights in the invention under the terms of NIH grants CA 43460, CA 57345, CA 62924, R01CA118822, NS20023-21, R37CA11898-34, and CA 121113.
| Number | Date | Country | |
|---|---|---|---|
| 61093739 | Sep 2008 | US | |
| 61110397 | Oct 2008 | US | |
| 61162737 | Mar 2009 | US |
| Number | Date | Country | |
|---|---|---|---|
| Parent | 13060191 | Jun 2011 | US |
| Child | 13412696 | US |
