PMC:4307189 / 49964-55814 JSONTXT

{"project":"2_test","denotations":[{"id":"25599599-23542173-14868175","span":{"begin":1653,"end":1654},"obj":"23542173"},{"id":"25599599-16452501-14868176","span":{"begin":2243,"end":2245},"obj":"16452501"},{"id":"25599599-16452501-14868177","span":{"begin":2461,"end":2463},"obj":"16452501"},{"id":"25599599-23940583-14868178","span":{"begin":4914,"end":4916},"obj":"23940583"},{"id":"25599599-19754359-14868179","span":{"begin":5205,"end":5207},"obj":"19754359"}],"text":"Discussion\nIn this study, we developed a computational framework to systematically predict signaling cross-talks between EGFR/ErbB and other signaling pathways that contribute to lapatinib (an EGFR and ErbB2/HER2 inhibitor) resistance. We hypothesized that gene-pairs (cross-talks) that can potentially cause drug-resistance have a high probability of occurring in the resistant condition(s) but a low probability in parental conditions. We employed a fully Bayesian statistical model: a special class of Exponential Random Graph Model known as the p1-model, to infer the posterior probabilities of such gene-pairs from corresponding networks inferred using gene expression values [17] of resistant and parental conditions. In selecting gene-pairs as putative cross-talks, threshold values for two parameters: odds and posterior probabilities of edges in resistant networks were empirically selected. However, more robust procedures for the selection of these two parameters can be made in future studies. All other parameters in the p1-model discussed above were estimated using Gibbs sampling (see Materials and method).\nOur results primarily focus on compensatory signaling pathways i.e. Notch signaling, Wnt signaling, GPCR signaling, and IR/IGF1R signaling, which cross-talk with EGFR/ErbB signaling to reduce the inhibiting effect of lapatinib. We present additional literature evidence that the identified cross-talks of the above compensatory signaling pathways with EGFR/ErbB signaling may contribute to drug-resistance by maintaining key cell survival and/or proliferation signals in common down-stream pathways, including PI3K/Akt signaling [1].\nKomurov et al. [17] hypothesized that cross-talks between EGFR/ErbB signaling and metabolic pathways contribute to resistance to lapatinib. More specifically, they identified that glucose deprivation reduces the inhibiting effects of lapatinib by up-regulating constituent genes and thus providing an EGFR/ErbB2-independent mechanism of glucose uptake and cell survival [17]. Here, by using the same gene expression datasets, we found MDM2:STK11 cross-talk between EGFR/ErbB and IGF1R signaling, where STK11 (also known as LKB1) phosphorylates and activates AMPK in absence of glucose [67]. Again, in the integrated signaling circuitry of pathways: p53-IGF-1-AKT-TSC2-mTOR, a positive feedback loop (p53-PTEN AKT-MDM2-p53) is formed which enhances p53-mediated apoptosis and senses nutrient deprivation [67]. Thus our results complement the findings of Komurov et al. by finding signaling cross-talks between EGFR/ErbB and IGF1R pathways.\nIn Netwalker analysis of our primary dataset (SKBR3 cell-line, GSE38376), we compared the expression changes of all the samples in parental conditions (basal, 0.1 μM and 1.0 μM) with those of all the samples in resistant conditions (basal, 0.1 μM and 1.0 μM). However, we conducted another experiment on both of our primary (SKBR3 cell-line, GSE38376) and secondary datasets (BT474 cell-line, GSE16179) in which we first identified genes dysregulated in treatment vs basal conditions in parental samples and then checked if those genes were reversely changed in treatment conditions in resistant samples. To that end, for each sample, first we calculated the fold-change(s) of parental treatment condition(s) compared to parental basal condition, and then we calculated the fold-changes of resistant basal and resistant treatment conditions compared to parental basal condition (Additional file 1: Figure S2A and S3A). Then, we chose only those genes for which, in any of the 3 samples, expressions were dysregulated (up-/down-regulated) in (all the) parental treatment condition(s) (log2 of fold-changes were positive/negative), and for that particular sample, expressions were reversely changed (the fold-change sign was opposite to that of parental condition) in all the resistant treatment conditions (Additional file 1: Figure S2B and S3B). This may be a strong indicator of sensitivity to an inhibitor in parental conditions and the development of acquired resistance. Next, we compared these selected genes to cross-talks found in results from GSE38379 (104, 188 and 299 EGFR/ErbB cross-talks from Reactome, KEGG and WikiPathway, respectively) and GSE16179 (83, 133 and 277 EGFR/ErbB cross-talks from Reactome, KEGG and WikiPathway, respectively). Although we didn’t find any such cross-talks overlapping with the results from the primary dataset (GSE38379), we found 401 from our secondary dataset (GSE16179) (Additional file 15: Table S14).\nCurrently, our network modeling only considers undirected edges among genes. In future we would like to generalise the approach to identify directed and indirect interactions among genes. In network modeling, a combination of both direct and indirect relationships among gene-pairs was found to provide better insights into biological systems in our previous studies [68]. The rationale for combining these two types of gene-gene relationships in signaling networks is that EGFR/ErbB and IGF1R can both cross-talk (EGFR/IGF1R heterodimerization) directly at the receptor level, and indirectly mediated by GPCR signaling, as reported by Van der Veeken et al. [62]. Other high-throughput datasets such as miRNA expression data, copy number aberration data, and methylation data could also be incorporated into our framework to obtain a better understanding of gene dependencies. Note that our methodology exploits a fully data-driven approach for finding putative drug-resistant cross-talks, without incorporating other prior information regarding gene-gene relationships, such as Protein-Protein Interactions. Hence, although our data-driven approach may inherently yield some false-positive predictions, it may also provide the possibilities of finding novel cross-talks contributing to drug- resistance."}