Methods blog · Cancer Systems Biology & Drug Resistance

Signal rewiring

Signal rewiring and V-structures: a Bayesian view of how breast cancer cells bypass lapatinib

Differential posterior edge probabilities between resistant and parental networks expose dysregulated pathways, and a new motif, the V-structure, captures the switch from targeted to bypass signalling.

#signal rewiring#acquired resistance#p1-model#V-structures#EGFR-TKI#breast cancer

Small-molecule inhibitors such as lapatinib are effective in ErbB2-positive breast cancer, yet tumour cells eventually adapt. A growing body of evidence pointed to adaptive reprogramming of signalling circuitry, or signal rewiring, as a major cause. In this PLOS ONE paper we built a Bayesian framework to model that rewiring directly, and found a recurring network pattern that captures how cells switch to bypass routes.

Classical pathway analyses are node-centric: they ask whether the genes of a pathway are differentially expressed (SPIA, DAVID, GATHER). Signal rewiring is about edges: two genes may each change little while their dependency switches on or off. We therefore used the p₁-model (an exponential random graph model; see our 2015 cross-talk paper) to infer, for every gene pair, a posterior probability of interaction in the resistant network and in the parental network.

A pair is aberrant when those probabilities differ substantially:

\[\Delta_{ij} \;=\; \Pr(x_{ij}=1 \mid D_{\text{resistant}}) \;-\; \Pr(x_{ij}=1 \mid D_{\text{parental}}).\]

Pathways that accumulate many aberrant pairs are flagged as dysregulated, an edge-centric view that is complementary to node-centric tools.

Two cell lines, one story

Using matched expression profiles from two lapatinib-treated ErbB2-positive lines, SKBR3 and BT474, the method identified similar sets of dysregulated pathways: EGFR-related pathways and other receptor-related pathways, many previously reported as compensatory routes after EGFR inhibition. A manual literature survey provided strong support for their role in acquired resistance to EGFR tyrosine-kinase inhibitors.

The predictions were complementary to both node-centric methods (SPIA, DAVID, GATHER) and edge-centric methods (ESEA, PAGI), picking up literature-supported pathways that the others missed.

A new motif: V-structures of aberrant signalling

The key conceptual contribution is the V-structure. Consider three genes where a gene \(j\) is linked to \(i\) in one condition but to \(k\) in the other:

We observed that genes were dysregulated between resistant and sensitive conditions precisely when they took part in such a switch of dependencies from targeted to bypass signalling events. Literature evidence for several important V-structures suggested roles in breast cancer metastasis and in acquired resistance to EGFR-TKIs. In particular, mRNA changes of TGFBR2, LEF1 and TP53 were tied to dependency switches from targeted to bypass links.

Take-home message

Many signalling pathway structures are compromised in acquired resistance, and V-structures within and among those pathways offer a mechanistic handle on how bypass happens, not just which genes change. This work formed the core of my PhD thesis at Monash University, Computational modelling and characterisation of cell signalling cross-talks in acquired drug resistance (2017).

Read the papers

  1. Bayesian model of signal rewiring reveals mechanisms of gene dysregulation in acquired drug resistance in breast cancer

    AKM Azad, Alfons Lawen, Jonathan Macgregor Keith

    PLOS ONE 12 (3), e0173331 · 2017

    Q1First author OA 20
  2. Computational modelling and characterisation of cell signalling cross-talks in acquired drug resistance

    AKM Azad

    PhD thesis, Monash University · 2017

    ThesisSole author OA
    DOI Blog

BibTeX