Co-occurrence analysis of insertional mutagenesis data reveals cooperating oncogenes
1Information and Communication Theory Group, Faculty of EEMCS, Delft University of Technology, Delft, The Netherlands, 2Division of Molecular Genetics and 3Division of Molecular Biology, The Netherlands Cancer Institute, Amsterdam, The Netherlands
*To whom correspondence should be addressed.
 |
ABSTRACT |
|---|
 TOP ABSTRACT 1 INTRODUCTION2 METHODS3 RESULTS4 CONCLUSIONS AND DISCUSSIONACKNOWLEDGEMENTSREFERENCES |
|---|
Motivation: Cancers are caused by an accumulation of multiple independent mutations that collectively deregulate cellular pathways, e.g. such as those regulating cell division and cell-death. The publicly available Retroviral Tagged Cancer Gene Database (RTCGD) contains the data of many insertional mutagenesis screens, in which the virally induced mutations result in tumor formation in mice. The insertion loci therefore indicate the location of putative cancer genes. Additionally, the presence of multiple independent insertions within one tumor hints towards a cooperation between the insertionally mutated genes. In this study we focus on the detection of statistically significant co-mutations.
Results: We propose a two-dimensional Gaussian Kernel Convolution method (2DGKC), a computational technique that identifies the cooperating mutations in insertional mutagenesis data. We define the Common Co-occurrence of Insertions (CCI), signifying the co-mutations that are statistically significant across all different screens in the RTCGD. Significance estimates are made on multiple scales, and the results visualized in a scale space, thereby providing valuable extra information on the putative cooperation.
The multidimensional analysis of the insertion data results in the discovery of 86 statistically significant co-mutations, indicating the presence of cooperating oncogenes that play a role in tumor development. Since oncogenes may cooperate with several members of a parallel pathway, we combined the co-occurrence data with gene family information to find significant cooperations between oncogenes and families of genes. We show, for instance, the interchangeable cooperation of Myc insertions with insertions in the Pim family.
Availability: A list of the resulting CCIs is available at: http://ict.ewi.tudelft.nl/~jeroen/CCI/CCI_list.txt
Contact: m.j.t.reinders{at}tudelft.nl
|
1 INTRODUCTION |
|---|
|
TOPABSTRACT1 INTRODUCTION2 METHODS3 RESULTS4 CONCLUSIONS AND DISCUSSIONACKNOWLEDGEMENTSREFERENCES |
|---|
Cancers arise when the regulatory pathways that govern healthy cell proliferation (cell division) are disrupted. Moreover, one of the hallmarks of cancer is that multiple oncogenic events, disrupting multiple pathways, are required before the state of uncontrolled proliferation is reached (Hanahan and Weinberg, 2000). For instance, (mutational) activation of the Myc protooncogene together with the loss of the p53 tumor-suppressor gene in mice, is a commonly observed co-occurrence of mutations that can cause cancer. In this respect, these two genes can be considered to ‘cooperate’ in the development of the tumor.
In retroviral insertional mutagenesis experiments, genes involved in the development of cancer are identified by determining the loci of viral insertions from tumors induced by retroviruses in cancer-predisposed mice (reviewed in Mikkers and Berns, 2003; Uren et al., 2005). In van Lohuizen et al. (1991), for example, the cancer-predisposition is acquired by inserting an EµMyc transgene in the mouse DNA. After infecting a host cell, the retrovirus inserts its own DNA into the host cell's genome, mutating the host cell's DNA in the process. The mutation may cause alteration in expression of genes in the vicinity of the insertion or, when inserted within a gene, alteration of the gene product. When the affected gene is a cancer gene, activation of a proto-oncogene or inactivation of a tumor-suppressor gene can, in cooperation with the cancer predisposition, cause uncontrolled proliferation of cells. Eventually this may give rise to tumors. Throughout this text these cancer-causing insertions are referred to as oncogenic insertions.
The tumor tissue contains many copies of the cell bearing the oncogenic insertions, but only a few copies of cells carrying non-oncogenic (random, background) insertions. Consequently, cloning the flanking sequences of the inserted virus to determine the insertion loci, will result in a data set of insertion loci (the oncogenic insertions) that are indicative for the presence of nearby cancer genes contaminated with noise (the non-oncogenic insertions). This is schematically depicted in Figures 1A and B. The challenge is to find the regions in the genome that carry insertions in multiple independent tumors significantly more frequently than expected by chance. Such a region is called a Common Integration Site (CIS), and its location is highly correlated with the location of genes involved in tumor development. An important factor to consider is that viral insertions can disrupt gene functioning from various distances around or within the gene. It is therefore essential that significance estimates are made for a range of different CIS widths in order not to miss interesting loci. The discovery of CISs in insertion data will be referred to as a 1D analysis, for which recently a kernel convolution method has been developed (de Ridder et al., 2006).
|
Instead of revealing cooperation of insertionally targeted genes with the cancer-predisposition, this study focuses on revealing the cooperation between virally targeted genes (Nakamura et al., 1996; Kim et al., 2003). Ideally, for this purpose the insertions co-occurring in tumors from mice of a uniform genotype should be examined, but a data set that is large enough to acquire statistically significant results is currently absent. Therefore we focus on the co-mutations that are common across a number of different insertional mutagenesis screens from publicly available data. The genes that are targeted by the commonly co-occurring insertions in these tumors are likely to cooperate in the tumor development.
To find the cooperation between virally targeted genes, we propose to analyze the insertion data in the two dimensional co-occurrence space. We define an Insertion Co-occurrence (IC) as a unique combination of insertions within one tumor, and the Common Co-occurrence of Insertions (CCI) as observing the combination of two insertions significantly more frequently than expected by chance across multiple tumors (schematically depicted in Figure 1D). When compared to a 1D analysis, performing a 2D analysis on the insertion data will result in the discovery of new loci that play a role in tumorigenesis. This can be seen by considering a region that is not hit frequently enough to be labeled a CIS in the 1D analysis, but may still be called significant in the 2D analysis, because it co-occurs frequently enough with another inserted region. To ensure all different configurations of insertions around or within genes are taken into account, we evaluate the significance of the CCIs at various scales. Visualizing the CCIs at multiple widths will contribute essential additional information about how insertions disrupt the functioning of their target genes.
Another hallmark of tumorigenesis is the existence of many parallel pathways (Hanahan and Weinberg, 2000), and consequently, the many possibilities of reaching the state of uncontrolled proliferation. This is exemplified by a study using Pim1 deficient and Pim2 deficient mice. Pim1 is frequently hit in screens of EµMyc transgenic mice. When Pim1 is knocked out, Pim2 is frequently hit (van der Lugt et al., 1995), and when Pim1 and Pim2 are knocked out, Pim3 is hit (Mikkers et al., 2002), suggesting all three Pim genes promote tumors in cooperation with Myc. As a consequence, co-occurring mutations in the RTCGD may not occur frequently enough to be statistically significant, simply because there exist too many parallel possibilities for the cell to become malignant. In this study, we investigate this phenomenon by including gene family information, and assess whether there exists cooperation between genes and a certain gene family.
The data in the RTCGD are publicly available, and the screens in the database have been individually studied and published before. It is therefore likely that the most prominent CCIs will point to cooperations between genes that have been discovered before. However, since we are the first to analyze the combined set of screens in the RTCGD for the presence of statistically significant cooperations between virally targeted genes in a systematic fashion, we do expect to discover new interactions. As we expect a subset of our CCIs to be published, we can partially validate our method by showing that the pairs of genes predicted to cooperate by our method will co-occur in literature abstracts significantly more frequently than expected by chance.
|
2 METHODS |
|---|
|
TOPABSTRACT1 INTRODUCTION2 METHODS3 RESULTS4 CONCLUSIONS AND DISCUSSIONACKNOWLEDGEMENTSREFERENCES |
|---|
2.1 The data
Over the last few years an extensive amount of insertional mutagenesis data has been published (see e.g. Hansen et al., 2000; Hwang et al., 2002; Johansson et al., 2004; Joosten et al., 2002; Li et al., 1999; Lund et al., 2002; Mikkers et al., 2002; Suzuki et al., 2002). These data have been compiled in the Retroviral Tagged Cancer Gene Database (RTCGD) (Akagi et al., 2004) (URL: http://RTCGD.ncifcrf.gov, accessed January 4, 2007). Currently, the RTCGD contains 5473 retroviral insertions distributed over 1361 tumors. There are 1031 tumors that contain more than one insertion. The vast majority of the insertions have been acquired in twenty different screens, that used various experimental setups. Therefore, the number of insertions that are found in a tumor varies significantly per screen. Additionally, the mouse models used varied among screens. In this study we analyze the combined data from all the screens in the RTCGD, irrespective of the genetic background or cancer predisposition of the mice used in the screens. Also, we assume that background insertions are distributed uniformly across the genome, and all insertions are independent of each other.
2.2 Insertion Co-occurrence
To exploit the information contained in the joint occurrence of insertions within one tumor, we map the data to the co-occurrence space. In this space a point indicates the location of an IC, that is, two insertions co-occurring in one tumor. Finding the regions in the co-occurrence space that contain ICs more frequently than expected by chance will point to the genes in the genome that cooperate in the development of the tumor.
We propose to apply a 2D Gaussian Kernel Convolution (2DGKC) to determine the statistical significance of the regions with multiple ICs. The 2DGKC, which is very similar to Parzen density estimation, results in a smooth estimate for the number of ICs, 
, at a position 
in the co-occurrence space:
|
| (1) |
denotes the selection of the i-th element from the vector between brackets. By using the product of two univariate kernel functions local independence is assumed, but by summing multiple kernel functions complex correlation structures can still be discovered. In this study a Gaussian kernel function is used, given by: 
, where h is the kernel width. Note that the kernel function used in our study is not normalized, as is done in traditional density estimates (Parzen, 1962). As a result, the modified density estimate can be interpreted as a continuous estimation of the number of co-occurrences at a given position. The local maxima in 
(the peaks) will now indicate the location of putative CCIs. Since we are only interested in the local maxima, we reduce the number of evaluations of Equation (1) (required to find the maxima), by applying a standard non-linear optimization algorithm (fminunc, MATLAB Optimization toolbox) started from every IC in the data.
2.3 Significance estimates
Significance of the putative CCIs is evaluated by testing against the following null-hypothesis:
|
|
is the observed height of the peak at position g. The null-hypothesis is rejected if the observed height of the peak significantly exceeds the mean height of the peaks under the null-hypothesis.
The null-distribution is acquired by a permutation approach, schematically depicted in Figure 2. The kernel convolution is applied to the ICs that result from a random permutation of the insertions (Fig. 2A and B). This results in random peaks in the co-occurrence space. This is repeated K times, to obtain a set of random realizations (Fig. 2C). From this set, the height of all the peaks is collected, and the null-distribution is computed (Fig. 2D). Using the null-distribution we can convert the 
-level to a threshold for the real data. This threshold can now be applied to the smoothed estimate of the number of ICs, that was obtained by applying the 2DGKC to the real co-occurrence data (Fig. 2E). We correct for multiple testing using the Bonferroni multiple testing correction, by dividing the -level by the number of tests. Since we only evaluate the height of the peaks, we take the number of tests to be equal to the number of peaks in the co-occurrence density.
|
2.4 Scale space
The kernel width h can be considered as a scale parameter, thereby providing an excellent way of controlling at which scale the significance of the ICs are evaluated. By increasing h, the kernel functions cover a larger region, and, since potentially more kernel functions will contribute to the smoothed estimate of the number of ICs, this results in higher peaks in this estimate. This mechanism will ensure that the CCIs for which the ICs are confined to one or more very specific regions (narrow CCIs), will only become significant for small values of h (small scales), and conversely, the broad CCIs will only be present at larger scales. This motivates the definition of a cross scale CCI (csCCI), defined as the detection of a CCI at one or more scales.
Visualizing these phenomena will aid the biologist in determining the targeted genes. For this purpose we construct three-dimensional scale space diagrams (see e.g. Figs 5 and 6). In these diagrams the contour, defined by the intersection of the threshold with the smoothed estimate of the number of ICs (Fig. 2E), is plotted in the (
)-plane, as a function of the scale parameter (z-axis). The scale parameter is chosen to cover a range of biologically relevant scales (
). Since for every scale the - computationally intensive - permutation procedure has to be performed, the threshold value is computed only for eight log-uniformly spaced scales. For the 100 intermediate scales, that are used to build the scale space diagrams, the necessary threshold values are computed using a piecewise linear interpolation of the threshold values that were computed using the actual permutation procedure.
2.5 
2-ranking
In addition to ranking the csCCIs on their average peak height across the scales, it is also interesting to rank the csCCIs according to a one-tailed 
-test, which corrects for the frequency with which the individual co-occurring loci are hit. Using the P-value from the 
-test, it is possible to filter the csCCIs at a user-defined -level, which is an often employed pruning technique in the context of association rule mining (Liu et al., 2001). Note that, by filtering the results, statistically significant interactions (based on peak height) are lost, and should therefore only be employed in case too many interactions were discovered.
Per CCI and per scale a P-value is computed for the 
-test performed on the following table:
In this table, 
denotes an area in the co-occurrence space: 
, that is, an area of width 2h around 
, the g1 position of the CCI under investigation, and the height spanning the complete g2 axis. 
is defined in an analogous fashion. Now, 
can be defined as the number of ICs in the intersection of the areas 
and 
. Likewise, 
, 
and N are defined as the total number of ICs in the areas 
, 
and the complete co-occurrence space, respectively. The csCCIs can now be ranked according to their average P-value across the scales in which the CCI was found to be significant.
2.6 Family mapping
The presence of parallel pathways may prevent co-occurring insertions from reaching the significance threshold. A clear example is the previously mentioned cooperation of the Myc proto-oncogene and the Pim1 and Pim2 proto-oncogenes. Since more than one possibility exists to cooperate with Myc, the spatial correlation in the g2 direction of the ICs in the Myc locus will be diminished, that is, the ICs will be divided into two separate clusters: one near the Pim1/Myc locus on Chromosome 17/Chromosome 15 and one near the Pim2/Myc locus on Chromosome X/Chromosome 15. This results in lower peaks at these positions, and, because the data is far from saturated, possibly even causes one or both of these peaks to fail the significance test.
This problem is circumvented by increasing spatial correlation of the regions surrounding the genes that can substitute for each other. There is, however, no data source available that contains information on functional substitution. For this reason, we revert to Ensembl gene family information, which is based on sequence similarity (Hubbard et al., 2005), and is an indirect indication that the genes in such a family can act as functional substitutes. To increase the level of confidence that genes from one family can indeed substitute for each other, only families with up to ten family members are considered. The spatial correlation is increased by mapping the regions surrounding genes within the same family on top of each other, by aligning them with respect to a common reference (schematically depicted in Fig. 3). In this alignment the transcriptional direction of the genes is taken into account. The common reference, referred to as the pivot, is chosen to be the 5' end of the genes. A major advantage is that ICs that were previously separated now may be close enough to reach the significance threshold. Before the mapping is performed, a few conditions need to be satisfied: (1) ICs from the same tumor are not mapped, since common cooperations can only be called significant when encountered in more than one tumor. (2) Genes within one family that are close together are excluded, since the ICs in their neighborhood will already be spatially correlated. (3) ICs with a distance to the pivot exceeding five times the scale parameter are not mapped. These ICs will not contribute to the peak height, but may introduce false positives.
|
), the g1 dimension remains the same.After the family mapping is performed, the 2DGKC method is applied to the ICs in the family mapped space. A Family Mapped CCI (FM-CCI) is defined as a peak that exceeds the significance threshold. The FM-CCIs indicate the cooperation of a region in the g1 direction with one or more members of a certain gene family in the g2 direction. Note that the mapping and 2DGKC is applied per family.
By mapping the regions around the genes from a family onto each other, the peak height that is expected by chance will increase. As a consequence, the null-distribution, against which the resulting peaks are compared, should incorporate this effect. This is achieved by including the family mapping before the permutation procedure depicted in Figure 2. The number of regions that are mapped onto each other changes as a function of the family size, and therefore a null-distribution is computed per family size. The multiple testing correction factor is equal to the total number of peaks evaluated in the family mapped space, which is approximately equal to the one used in the detection of CCIs.
2.7 Validation from literature
In order to validate the most prominent csCCIs that resulted from our analysis, we evaluated how often the two genes, close to a csCCI, co-occurred in the same MEDLINE abstract according to the online database PubGene (http://www.pubgene.org) (Jenssen et al., 2001). This required a non-trivial mapping of the csCCI to their target genes. Although it has been shown that viral insertions most frequently target their closest neighboring gene (Erkeland et al., 2006), it is likely that this simple heuristic will introduce some false negatives, thereby diluting the number of discovered co-occurring gene pairs in the PubGene database. To overcome this problem we evaluate all nine combinations of the three nearest genes surrounding the region marked by a csCCI in the g1 direction against their three counterparts in the g2 direction, and use only the combination that resulted in the maximum number hits in PubGene. We compare the results obtained by this procedure against the result obtained by repeating the same procedure with 2500 random combinations with the genes in our list.
|
3 RESULTS |
|---|
|
TOPABSTRACT1 INTRODUCTION2 METHODS3 RESULTS4 CONCLUSIONS AND DISCUSSIONACKNOWLEDGEMENTSREFERENCES |
|---|
3.1 Common co-occurrence of insertions
We have applied the proposed 2DGKC method to the combined data from the screens in the RTCGD. We evaluated the data at the following eight log-uniformly spaced scales: [10000, 17487, 30579, 53472, 93506, 163512, 285930, 500000] at a significance level of
= 0.05. This resulted in the discovery of 86 csCCIs, that is, we find 86 pairs of loci that cooperate with each other in the development of the tumor. An overview of the results are given in Figure 4 and the top ten csCCIs are listed in Table 1 (a complete list is available online).
|
|
-test, described in the Methods section. The radius of the csCCI marker is proportional to the score obtained by normalizing the peak heights of the csCCIs per scale, and averaging this normalized peak height across the scales at which the csCCI was found to be significant. The arrows indicate the gene pairs discussed in the Results section.A number of interactions identified in retroviral mutagenesis screens have previously been characterized. Myc collaborates with Pim1 (Verbeek et al., 1991), Myb (Davies et al., 1999), Gfi1 (Schmidt et al., 1998), and Cyclin D1 (Lovec et al., 1994) and Hoxa9/Hoxa7 collaborate with Meis1 (Kroon et al., 1998). The majority of co-occurences however, have not been studied in mouse models of lymphoma, but in some cases the literature provides supporting evidence for their cooperation. For instance, the csCCI near Rasgrp1/Cebpb ranked 43rd in the list. Rasgrp1 is a guanine nucleotide exchange factor that activates Ras signalling. Cebpb (CCAAT/enhancer-binding protein beta) is a transcription factor that mediates interleukin-6 (IL-6) signalling. Cebpb is also an important mediator of Ras induced oncogenesis (Zhu et al., 2002).
Interestingly, when ranking the csCCIs according to the 
-test, a rather different top 10 is found (Table 2). These interactions are of special interest, since the individual loci are inserted in relatively few tumors, which makes it more likely that the combination of the two mutations is causal for development of the tumor. Figure 2 shows the result after applying an additional 0.05 threshold to the P-value resulting from the 
-test. Indeed, it can be seen that 12 csCCIs (colored blue in Fig. 4) do not reach this additional threshold, and may therefore be of less interest. Notably, they mainly represent interactions with either Sox4 or Gfi1, which, by themselves, are both frequently targeted in insertional mutagenesis screens.
|
3.2 Validation from literature
Table 1 lists the candidate target gene pairs, as indicated by the top ten of the 86 csCCIs. By searching the PubGene database we found six of these ten gene pairs to co-occur in the literature abstracts. This is statistically significant (
), when compared to the 322 hits that resulted from querying 2500 random, and therefore mostly unrelated, combinations in our set. Also when considering the complete list of 86 gene pairs indicated by the csCCIs, we find a statistically significant overrepresentation in the literature abstracts (
), since 23 of these co-occurred in the PubGene database. For the ten gene pairs listed in Table 2, no significant overrepresentation in literature abstracts was established. This is not surprising, since these genes are hit relatively infrequently, and are therefore less likely to be well-characterized in literature.
3.3 Scale space diagrams
The list in Table 2 contains some interesting putative cooperations between genes, but by plotting the csCCIs in the scale space, valuable extra information about the cooperation can be gained. From Figure 5 it is clear that, at the largest scales, insertions near Myb clearly co-occur with Gfi1 insertions. Gfi1 and Myb are transcription factors with roles in hematopoiesis (Mucenski et al., 1991: Zeng et al., 2004). At the smaller scales however, inserts surrounding Myb can be divided into two separate clusters, and independently associate with the Gfi1 locus. This suggests that inserts from both clusters are functionally equivalent, thereby strengthening the case for grouping them into a single CCI at larger scales, but possibly also indicates a different mechanism by which they disrupt functioning of Myb. This diagram can thus give valuable insight in the mechanisms that disrupts gene functioning. Other examples exist where csCCIs are only significant at a certain range of scales, for instance the previously mentioned csCCI near Rasgrp1 and Cebpb (Fig. 6). Clearly, when evaluating this csCCIs at a single scale or subset of scales, one runs the risk to miss this significant cooperation if the scale at which it is evaluated does not match the scale of the CCI.
|
|
3.4 Family mapping
Figure 7A shows the previously mentioned example of the possible substitution of insertions near Pim2 for Pim1 mutations. The figure exemplifies that, by performing the family mapping, indeed meaningful extra interactions are found. The IC near Pim2 and Myc would have gone undetected in the normal co-occurrence analysis, the family mapping proves capable of exploiting the additional information contained in this IC.
|
direction. (A) The interaction between Myc and the Pim family (ENSF00000001108: SERINE/THREONINE KINASE PIM) in the scale space. The red triangles mark ICs near Pim2, and yellow triangles mark ICs near Pim1. (B) The interaction between Sox4 and the Cyclin dependent kinases Family (ENSF00000000186: CELL DIVISION). The coloring of the ICs indicate near which seperate family member it occurred. Notably, seven of the nine genes in this family are hit.Similarly interesting is the discovered FM-CCI indicating cooperation between Sox4 and the Cyclin dependent kinases family. Seven from the nine genes in this family are hit in eight independent tumors. Figure 7B shows the scale space diagram for this interaction. Apparently, Sox4 insertions cooperate interchangeably with one of the members of the Cyclin dependent kinases family. Figure 8 shows how the ICs targeting the Sox4/Cyclin dependent kinases family are distributed over the tumors. Notably, none of the genes in the Cyclin dependent kinases family is hit frequently enough to reach significance on its own account (the two ICs near Sox4/Cdk6 are too far from each other to reach significance). It is only by applying the family mapping that cooperation between Sox4 and the Cyclin dependent kinases family can be discovered.
|
|
4 CONCLUSIONS AND DISCUSSION |
|---|
|
TOPABSTRACT1 INTRODUCTION2 METHODS3 RESULTS4 CONCLUSIONS AND DISCUSSIONACKNOWLEDGEMENTSREFERENCES |
|---|
Until now, the main focus of analysis on insertional mutagenesis data has been one-dimensional, that is, discovering regions in the genome that are causal for tumor development, the CISs. In this article we analyzed the data from publicly available retroviral insertional mutagenesis screens in the 2D co-occurrence space. By evaluating the significance of co-occurring insertions we found 86 statistically significant csCCIs, that indicate cooperation between insertionally targeted genes. By analyzing the data in a scale space we are able to detect csCCIs that are only significant at a limited subset of the scales, for instance the putative cooperation between Rasgrp1 and Cebpb. In addition, the scale space provides essential information about mechanisms that underlie the viral disruption of gene functioning. This was exemplified by the putative cooperation between Myb and Gfi1, where the scale space showed two sub-CCIs at low scales, indicating two confined regions of integration.
To assess whether also known cooperation between genes are found, we showed that the set of candidate gene pairs, resulting from our study, is significantly overrepresented in the PubGene database, a literature network containing gene-to-gene co-citations. In addition to known cooperations, our study also revealed previously unknown putative cooperations, that are interesting targets for possible follow-up studies. We have presented two rankings of the resulting csCCIs, one based on average peak height and one based on the average P-value resulting from a 
-test. The latter ranking takes into account the possibility that a csCCI is caused by frequent insertion of one or both of the individual loci. We can conclude that, by analyzing the data in the co-occurrence space, and at multiple scales, we can find new statistically significant regions in the genome that play a role in tumor development.
To deal with the possibility that cells choose alternative pathways to become malignant, we have incorporated information about gene families in the analysis. By remapping the data according to putative substitutions derived from gene family membership, we were able to discover significant cooperations between genes and genes from a gene family. Examples of the known substitution of Pim2 insertions for insertions near Pim1 in tumors with virally activated Myc, as well as the putative cooperation between Sox4 and the Cyclin dependent kinases family were given. These examples show that much is to be gained by integrating insertional mutagenesis data with other data sources, such as gene family information, since the insertion data in itself is far from saturated.
The methods presented are especially beneficial for data from high throughput screens with many insertional mutations per tumor. Therefore, the methods may be applied to other types of genome wide mutagenesis data as well, for example data from transposon screens (Collier and Largaespada, 2005). As the amount of data increases, extensions to a multi-occurrence analysis become interesting. For the proposed 2DGKC method, these extensions are fairly straightforward.
|
ACKNOWLEDGEMENTS |
|---|
|
TOPABSTRACT1 INTRODUCTION2 METHODS3 RESULTS4 CONCLUSIONS AND DISCUSSIONACKNOWLEDGEMENTSREFERENCES |
|---|
This work was part of the BioRange programme of the Netherlands Bioinformatics Centre (NBIC), which is supported by a BSIK grant through the Netherlands Genomics Initiative (NGI).
Conflict of interest: none declared.
|
REFERENCES |
|---|
|
TOPABSTRACT1 INTRODUCTION2 METHODS3 RESULTS4 CONCLUSIONS AND DISCUSSIONACKNOWLEDGEMENTSREFERENCES |
|---|
Akagi K, et al. RTCGD: retroviral tagged cancer gene database. Nucleic Acids Res, ( (2004) ) 32, (Database issue): D523–D527.
Collier LS, Largaespada DA. Hopping around the tumor genome: transposons for cancer gene discovery. Cancer Res, ( (2005) ) 65, : 9607–9610.
Davies J, et al. Cooperation of myb and myc proteins in t cell lymphomagenesis. Oncogene, ( (1999) ) 18, : 3643–3647.[CrossRef][ISI][Medline].
de Ridder J, et al. Detecting statistically significant common insertion sites in retroviral insertional mutagenesis screens. PLoS Comput. Biol, ( (2006) ) 2, : e166.[CrossRef][Medline].
Erkeland SJ, et al. Significance of murine retroviral mutagenesis for identification of disease genes in human acute myeloid leukemia. Cancer Res, ( (2006) ) 66, : 622–626.
Hanahan D, Weinberg RA. The hallmarks of cancer. Cell, ( (2000) ) 100, : 57–70.[CrossRef][ISI][Medline].
Hansen GM, et al. Genetic profile of insertion mutations in mouse leukemias and lymphomas. Genome Res, ( (2000) ) 10, : 237–243.
Hubbard T, et al. Nucleic Acids Res, ( (2005) ) 33, : D447–D453.
Hwang HC, et al. Identification of oncogenes collaborating with p27Kip1 loss by insertional mutagenesis and high-throughput insertion site analysis. Proc. Natl Acad. Sci. USA, ( (2002) ) 99, : 11293–11298.
Jenssen TK, et al. A literature network of human genes for high-throughput analysis of gene expression. Nat. Genet, ( (2001) ) 28, : 21–28.[CrossRef][ISI][Medline].
Johansson FK, et al. Identification of candidate cancer-causing genes in mouse brain tumors by retroviral tagging. Proc. Natl Acad. Sci. USA, ( (2004) ) 101, : 11334–11337.
Joosten M, et al. Large-scale identification of novel potential disease loci in mouse leukemia applying an improved strategy for cloning common virus integration sites. Oncogene, ( (2002) ) 21, : 7247–7255.[CrossRef][ISI][Medline].
Kim R, et al. Genome-based identification of cancer genes by proviral tagging in mouse retrovirus-induced T-cell lymphomas. J Virol, ( (2003) ) 77, : 2056–2062.
Kroon E, et al. Hoxa9 transforms primary bone marrow cells through specific collaboration with meis1a but not pbx1b. EMBO J, ( (1998) ) 17, : 3714–3725.[CrossRef][ISI][Medline].
Li J, et al. Leukaemia disease genes: large-scale cloning and pathway predictions. Nat. Genet, ( (1999) ) 23, : 348–353.[CrossRef][ISI][Medline].
Liu B, et al. Identifying non-actionable association rules. In: KDD '01: Proceedings of the Seventh ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, ( (2001) ) New York, NY, USA: ACM Press. 329–334..
Lovec H, et al. Cyclin d1/bcl-1 cooperates with myc genes in the generation of b-cell lymphoma in transgenic mice. EMBO J, ( (1994) ) 13, : 3487–3495.[ISI][Medline].
Lund AH, et al. Genome-wide retroviral insertional tagging of genes involved in cancer in Cdkn2a-deficient mice. Nat. Genet, ( (2002) ) 32, : 160–165.[CrossRef][ISI][Medline].
Mikkers H, Berns A. Retroviral insertional mutagenesis: tagging cancer pathways. Adv. Cancer Res, ( (2003) ) 88, : 53–99.[CrossRef][ISI][Medline].
Mikkers H, et al. High-throughput retroviral tagging to identify components of specific signaling pathways in cancer. Nat. Genet, ( (2002) ) 32, : 153–159.[CrossRef][ISI][Medline].
Mucenski ML, et al. A functional c-myb gene is required for normal murine fetal hepatic hematopoiesis. Cell, ( (1991) ) 65, : 677–689.[CrossRef][ISI][Medline].
Nakamura T, et al. Cooperative activation of Hoxa and Pbx1-related genes in murine myeloid leukaemias. Nat. Genet, ( (1996) ) 12, : 149–153.[CrossRef][ISI][Medline].
Parzen E. On estimation of a probability density function and mode. The Ann. Math. Stat, ( (1962) ) 33, : 1065–1076..
Schmidt T, et al. Zinc finger protein gfi-1 has low oncogenic potential but cooperates strongly with pim and myc genes in t-cell lymphomagenesis. Oncogene, ( (1998) ) 17, : 2661–2667.[CrossRef][ISI][Medline].
Suzuki T, et al. New genes involved in cancer identified by retroviral tagging. Nat Genet, ( (2002) ) 32, : 166–174.[CrossRef][ISI][Medline].
Uren AG, et al. Retroviral insertional mutagenesis: past, present and future. Oncogene, ( (2005) ) 24, : 7656–7672.[CrossRef][ISI][Medline].
vander Lugt NM, et al. Proviral tagging in e mu-myc transgenic mice lacking the pim-1 proto-oncogene leads to compensatory activation of pim-2. EMBO J, ( (1995) ) 14, : 2536–2544.[ISI][Medline].
van Lohuizen M, et al. Identification of cooperating oncogenes in E mu-myc transgenic mice by provirus tagging. Cell, ( (1991) ) 65, : 737–752.[CrossRef][ISI][Medline].
Verbeek S, et al. Mice bearing the e mu-myc and e mu-pim-1 transgenes develop pre-b-cell leukemia prenatally. Mol. Cell. Biol, ( (1991) ) 11, : 1176–1179.
Zeng H, et al. Transcription factor gfi1 regulates self-renewal and engraftment of hematopoietic stem cells. EMBO J, ( (2004) ) 23, : 4116–4125.[CrossRef][ISI][Medline].
Zhu S, et al. Ccaat/enhancer binding protein-beta is a mediator of keratinocyte survival and skin tumorigenesis involving oncogenic ras signaling. Proc. Natl Acad. Sci. USA, ( (2002) ) 99, : 207–212.
CiteULike 
Connotea 
Del.icio.us What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||