Limits...
Transcriptomic Heterogeneity in Cancer as a Consequence of Dysregulation of the Gene-Gene Interaction Network.

van Wieringen WN, van der Vaart AW - Bull. Math. Biol. (2015)

Bottom Line: Dysregulation of the regulatory network results in less control of transcript levels in the cell.Hence, dysregulation is reflected in the heterogeneity of the transcriptome: the more dysregulated the pathway, the more the transcriptomic heterogeneity.These mechanisms are statistically motivated, explored in silico, and their plausibility to occur in vivo illustrated by means of oncogenomics data of breast cancer studies.

View Article: PubMed Central - PubMed

Affiliation: Department of Epidemiology and Biostatistics, VU University Medical Center, P. O. Box 7057, 1007 MB, Amsterdam, The Netherlands. w.vanwieringen@vumc.nl.

ABSTRACT
Many pathways are dysregulated in cancer. Dysregulation of the regulatory network results in less control of transcript levels in the cell. Hence, dysregulation is reflected in the heterogeneity of the transcriptome: the more dysregulated the pathway, the more the transcriptomic heterogeneity. We identify four scenarios for a transcriptomic heterogeneity increase (i.e., pathway dysregulation) in cancer: (1) activation of a molecular switch, (2) a structural change in a regulator, (3) a temporal change in a regulator, and (4) weakening of gene-gene interactions. These mechanisms are statistically motivated, explored in silico, and their plausibility to occur in vivo illustrated by means of oncogenomics data of breast cancer studies.

No MeSH data available.


Related in: MedlinePlus

Number of edges present versus penalty parameter, for ER (solid lines) and ER+ (dashed lines) groups for three breast cancer studies (distinguished by color). Left panel Notch pathway; right panel TGF pathway (Color figure online)
© Copyright Policy - OpenAccess
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4644214&req=5

Fig8: Number of edges present versus penalty parameter, for ER (solid lines) and ER+ (dashed lines) groups for three breast cancer studies (distinguished by color). Left panel Notch pathway; right panel TGF pathway (Color figure online)

Mentions: It is left to assert whether gene–gene interactions may indeed weaken or vanish in cancer. Hereto we revisit the Notch and TGF pathway data of the previous sections. In Sect. 2, these data revealed a higher transcriptomic entropy in the ER group than in the ER+ group. Proposition 3 suggests that the entropy difference may be due to the weakening of gene–gene interactions. To investigate this, we compare between the two estrogen groups their number of conditional dependencies among the genes comprising the pathway. This is done in both the “transcriptome only” (as introduced in Sect. 2) and the integrative oncogenomics data (as introduced in Sect. 3). In the former setting, a standard Gaussian graphical model (as implied by the multivariate normal) describing the relations between the genes is fitted. From the thus fitted model, the nonzero partial correlations (reflecting the relations between the genes) are determined. For the integrative oncogenomics studies, comprising both DNA copy number and gene expression data, Model (3) relating the two molecular levels is fitted. The nonzero elements of the estimate of the matrix with gene-to-gene effects are then studied. For each pathway data set of the previous sections, we subsample repetitively (500 times) an equal number of samples from each estrogen group. This number of samples is set at 90 % of the sample size of the group with the smallest prevalence in the data set. For the “transcriptome only” data of Sect. 2, the number of edges (number of nonzero partial correlations) in each estrogen group is determined for a given penalty parameter using the method of Peng et al. (2009). Similarly, for the integrative oncogenomics studies, comprising both genomic and transcriptomic data, the number of edges (nonzero elements of ) in both groups is determined for a given penalty parameter using the method of Van Wieringen and Van de Wiel (2014) which fits a sparse version of Model (3). The number of edges found is averaged over the 500 subsamples. The above (for both the “transcriptome only” and the integrative oncogenomics data) is repeated for a grid of . In both cases, the penalty parameter grid is chosen such that the resulting number of edges (i.e., the number of nonzero partial correlations, or the number of nonzero off-diagonal elements in ) is between 1 and 10 % of the total number of possible edges. This range intends to capture only sparse networks, which are believed to be representative of realistic gene–gene interaction networks. The averaged number of selected edges is plotted against the penalty parameter in Fig. 8 for the integrative oncogenomics studies, whereas the plots for the “transcriptome only” data sets are in SM K. The latter suggests that there is no weakening of the gene–gene interaction pattern from one estrogen group to the other in either pathway. However, when taking into account DNA copy number aberrations, it becomes apparent that in both pathways, the number of selected edges in the ER samples is lower (over the selected range of ) than in the ER+ samples (confer the right panel of Fig. 8). This suggests a weaker gene–gene interaction pattern in the ER group, which may in turn contribute to the higher transcriptomic entropy/heterogeneity. Hence, it suggests that delineated mechanisms of transcriptomic entropy increase need not always act alone. For instance, in the illustration above only after correction for the effect of genomic abberations did the mechanism of weakened gene–gene interactions become apparent. Two (or more) transcriptomic entropy increasing mechanisms may thus be active simultaneously.


Transcriptomic Heterogeneity in Cancer as a Consequence of Dysregulation of the Gene-Gene Interaction Network.

van Wieringen WN, van der Vaart AW - Bull. Math. Biol. (2015)

Number of edges present versus penalty parameter, for ER (solid lines) and ER+ (dashed lines) groups for three breast cancer studies (distinguished by color). Left panel Notch pathway; right panel TGF pathway (Color figure online)
© Copyright Policy - OpenAccess
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4644214&req=5

Fig8: Number of edges present versus penalty parameter, for ER (solid lines) and ER+ (dashed lines) groups for three breast cancer studies (distinguished by color). Left panel Notch pathway; right panel TGF pathway (Color figure online)
Mentions: It is left to assert whether gene–gene interactions may indeed weaken or vanish in cancer. Hereto we revisit the Notch and TGF pathway data of the previous sections. In Sect. 2, these data revealed a higher transcriptomic entropy in the ER group than in the ER+ group. Proposition 3 suggests that the entropy difference may be due to the weakening of gene–gene interactions. To investigate this, we compare between the two estrogen groups their number of conditional dependencies among the genes comprising the pathway. This is done in both the “transcriptome only” (as introduced in Sect. 2) and the integrative oncogenomics data (as introduced in Sect. 3). In the former setting, a standard Gaussian graphical model (as implied by the multivariate normal) describing the relations between the genes is fitted. From the thus fitted model, the nonzero partial correlations (reflecting the relations between the genes) are determined. For the integrative oncogenomics studies, comprising both DNA copy number and gene expression data, Model (3) relating the two molecular levels is fitted. The nonzero elements of the estimate of the matrix with gene-to-gene effects are then studied. For each pathway data set of the previous sections, we subsample repetitively (500 times) an equal number of samples from each estrogen group. This number of samples is set at 90 % of the sample size of the group with the smallest prevalence in the data set. For the “transcriptome only” data of Sect. 2, the number of edges (number of nonzero partial correlations) in each estrogen group is determined for a given penalty parameter using the method of Peng et al. (2009). Similarly, for the integrative oncogenomics studies, comprising both genomic and transcriptomic data, the number of edges (nonzero elements of ) in both groups is determined for a given penalty parameter using the method of Van Wieringen and Van de Wiel (2014) which fits a sparse version of Model (3). The number of edges found is averaged over the 500 subsamples. The above (for both the “transcriptome only” and the integrative oncogenomics data) is repeated for a grid of . In both cases, the penalty parameter grid is chosen such that the resulting number of edges (i.e., the number of nonzero partial correlations, or the number of nonzero off-diagonal elements in ) is between 1 and 10 % of the total number of possible edges. This range intends to capture only sparse networks, which are believed to be representative of realistic gene–gene interaction networks. The averaged number of selected edges is plotted against the penalty parameter in Fig. 8 for the integrative oncogenomics studies, whereas the plots for the “transcriptome only” data sets are in SM K. The latter suggests that there is no weakening of the gene–gene interaction pattern from one estrogen group to the other in either pathway. However, when taking into account DNA copy number aberrations, it becomes apparent that in both pathways, the number of selected edges in the ER samples is lower (over the selected range of ) than in the ER+ samples (confer the right panel of Fig. 8). This suggests a weaker gene–gene interaction pattern in the ER group, which may in turn contribute to the higher transcriptomic entropy/heterogeneity. Hence, it suggests that delineated mechanisms of transcriptomic entropy increase need not always act alone. For instance, in the illustration above only after correction for the effect of genomic abberations did the mechanism of weakened gene–gene interactions become apparent. Two (or more) transcriptomic entropy increasing mechanisms may thus be active simultaneously.

Bottom Line: Dysregulation of the regulatory network results in less control of transcript levels in the cell.Hence, dysregulation is reflected in the heterogeneity of the transcriptome: the more dysregulated the pathway, the more the transcriptomic heterogeneity.These mechanisms are statistically motivated, explored in silico, and their plausibility to occur in vivo illustrated by means of oncogenomics data of breast cancer studies.

View Article: PubMed Central - PubMed

Affiliation: Department of Epidemiology and Biostatistics, VU University Medical Center, P. O. Box 7057, 1007 MB, Amsterdam, The Netherlands. w.vanwieringen@vumc.nl.

ABSTRACT
Many pathways are dysregulated in cancer. Dysregulation of the regulatory network results in less control of transcript levels in the cell. Hence, dysregulation is reflected in the heterogeneity of the transcriptome: the more dysregulated the pathway, the more the transcriptomic heterogeneity. We identify four scenarios for a transcriptomic heterogeneity increase (i.e., pathway dysregulation) in cancer: (1) activation of a molecular switch, (2) a structural change in a regulator, (3) a temporal change in a regulator, and (4) weakening of gene-gene interactions. These mechanisms are statistically motivated, explored in silico, and their plausibility to occur in vivo illustrated by means of oncogenomics data of breast cancer studies.

No MeSH data available.


Related in: MedlinePlus