Computational protocol: An ID2-dependent mechanism for VHL inactivation in cancer

Similar protocols

Protocol publication

[…] To infer if ID2 modulates the interactions between HIF2α and its transcriptional targets we used a modified version of MINDy algorithm, called CINDy. CINDy uses adaptive partitioning method to accurately estimate the full conditional mutual information between a transcription factor and a target gene given the expression or activity of a signaling protein. Briefly, for every pair of transcription factor and target gene of interest, it estimates the mutual information i.e. how much information can be inferred about the target gene when the expression of the transcription factor is known, conditioned on the expression/activity of the signaling protein. It estimates this conditional mutual information by estimating the multidimensional probability densities after partitioning the sample distribution using adaptive partitioning method. We applied CINDy algorithm on gene expression data for 548 samples obtained from The Cancer Genome Atlas (TCGA). Since the activity level and not the gene expression of ID2 is the determinant of its modulatory function i.e. the extent to which it modulates the transcriptional network of HIF2α, we used an algorithm called Virtual Inference of Protein-activity by Enriched Regulon analysis (VIPER) to infer the activity of ID2 protein from its gene expression profile. VIPER method allows the computational inference of protein activity, on an individual sample basis, from gene expression profile data. It uses the expression of genes that are most directly regulated by a given protein, such as the targets of a transcription factor (TF), as an accurate reporter of its activity. We defined the targets of ID2 by running ARACNe algorithm on 548 gene expression profiles and use the inferred 106 targets to determine its activity ().We applied CINDy on 277 targets of HIF2α represented in Ingenuity pathway analysis (IPA) and for which gene expression data was available (). Of these 277 targets, 77 are significantly modulated by ID2 activity (p-value ≤ 0.05). Among the set of target genes whose expression was significantly positively correlated (p-value ≤ 0.05) with the expression of HIF2α irrespective of the activity of ID2, i.e. correlation was significant for samples with both high and low activity of ID2, the average expression of target genes for a given expression of HIF2α was higher when the activity of ID2 was high. The same set of target gene were more correlated in high ID2 activity samples compared to any set of random genes of same size (), whereas they were not in ID2 low activity samples (). We selected 25% of all samples with the highest/lowest ID2 activity to calculate the correlation between HIF2α and its targets.To determine whether regulation of ID2 by hypoxia might impact the correlation between high ID2 activity and HIF2α shown in we compared the effects of ID2 activity versus ID2 expression for the transcriptional connection between HIF2α and its targets. We selected 25% of all patients (n=548) in TCGA with high ID2 activity and 25% of patients with low ID2 activity and tested the enrichment of significantly positively correlated targets of HIF2α in each of the groups. This resulted in significant enrichment (p-value < 0.001) in high ID2 activity but showed no significant enrichment (p-value = 0.093) in low ID2 activity samples. Moreover, the difference in the enrichment score (ΔES) in these two groups was statistically significant (p-value < 0.05). This significance is calculated by randomly selecting the same number of genes as the positively correlated targets of HIF2α, and calculating the ΔES for these randomly selected genes, giving ΔESrand. We repeated this step 1000 times to obtain 1000 ΔESrand that are used to build the null distribution (). We used the null distribution to estimate p-value calculated as (# ΔES > ΔESrand)/1000. Enrichment was observed only when ID2 activity was high but not when ID2 activity was low, thus suggesting that ID2 activity directionally impacts the regulation of targets of HIF2α by HIF2α. Consistently, the significant ΔES using ID2 activity suggests that ID2 activity is determinant of correlation between HIF2α and its targets.Conversely, when we performed similar analysis using ID2 expression instead of ID2 activity, we found significant enrichment of positively correlated targets of HIF2α both in samples with high expression (p-value=0.025) and low expression of ID2 (p-value=0.048). Given the significant enrichment in both groups, we did not observe any significant difference in the enrichment score in the two groups (p-value of ΔES = 0.338). Thus, while the determination of the ID2 activity and its effects upon the HIF2α-targets connection by VIPER and CINDy allowed us to determine the unidirectional positive link between high ID2 activity and HIF2α transcription, a similar analysis performed using ID2 expression contemplates the dual connection between ID2 and HIF2α. […]

Pipeline specifications

Software tools VIPER, ARACNE, IPA
Databases TCGA Data Portal
Application scRNA-seq analysis
Organisms Homo sapiens
Chemicals Oxygen