Combining Mutational Signatures Clonal Fitness and Drug Affinity to Define Drug-Specific Resistance Mutations in Cancer

摘要

class="head no_bottom_margin" id="sec1title">IntroductionAlthough targeted cancer therapies, for example, against kinases, hormone receptors, or hormone-synthesizing enzymes, have shown clinical success, many patients develop resistance to treatment and subsequently relapse. Second- and third-generation drugs are being developed to target these resistant mutants; however, there is a significant time span between the detection of clinically validated resistance mutations and the availability of suitably targeted treatment options. Early identification of drug-specific mutations is therefore critical and the aim of this study.Several mechanisms underlying resistance to targeted drugs have been described (), including mutations directly affecting the drug target. Such mutations may, for example, increase affinity for the endogenous co-factor ATP, thereby decreasing the relative affinity of an ATP-competitive drug. Mutations within the binding site may also alter drug-protein interactions and directly interfere with drug binding ().Several computational studies have investigated the impact of protein mutations on drug efficacy (, , , ), mainly in an antiviral or antibacterial context (, , ). However, rarely do these studies prospectively identify drug resistance mutations (, , ). In predicting mutations that render Staphylococcus aureus resistant to an antifolate antibiotic, evaluated the likely effect of possible mutations on both binding of the inhibitor and on binding of the endogenous ligand — an important aspect since any mutation that significantly abrogates the native activity of the wild-type (WT) protein is unlikely to survive selective evolutionary pressure (, , href="#bib58" rid="bib58" class=" bibr popnode">Pandurangan et al., 2017). However, Reeve et al. do not consider the likelihood of whether each mutation can be formed in bacteria.In cancer, the mutation landscape of a tumor can be characterized by the mutational signatures operating in a particular cancer type (href="#bib2" rid="bib2" class=" bibr popnode">Alexandrov et al., 2013). These signatures describe the probability of a specific base exchange within a defined trinucleotide context. Some of these signatures have been associated with known mutagenic processes, such as UV irradiation or aging, while the mechanism of others still remains elusive (href="#bib2" rid="bib2" class=" bibr popnode">Alexandrov et al., 2013). These mutagenic processes can generate a single clone harboring the disease-causing “driver mutation,” which ultimately leads to the development of cancer (href="#bib31" rid="bib31" class=" bibr popnode">Greaves and Maley, 2012). In addition, non-transforming somatic mutations, so-called passenger mutations, are randomly created. While not oncogenic per se, passenger mutations can provide the substrate for an evolutionary advantage throughout cancer progression, for example, under the selective pressure of a targeted molecular therapy, leading to drug resistance. Known drug resistance mutations have not only been detected in treatment-naive patients (href="#bib36" rid="bib36" class=" bibr popnode">Inukai et al., 2006, href="#bib66" rid="bib66" class=" bibr popnode">Roche-Lestienne et al., 2002), but also in healthy individuals (href="#bib32" rid="bib32" class=" bibr popnode">Gurden et al., 2015). This suggests that small pools of viable treatment-resistant clones can pre-exist in patients and that drug treatment puts a selection pressure on a heterogeneous cancer cell population that selects for resistant sub-clones.Each drug interacts with its biological target in a unique way, and each protein target mutation will differentially affect diverse classes of drugs. As a consequence, each compound can be expected to exhibit a unique resistance mutation profile. Three factors contribute to the probability and functional impact of a residue change: (1) the probability that the protein mutation can be generated from a DNA mutational signature (signature-driven probability), (2) whether the mutation maintains protein function and clones harboring the mutation are still viable (fitness), and (3) whether the mutation confers lower drug affinity with respect to the endogenous ligand for the target protein (affinity). href="#bib50" rid="bib50" class=" bibr popnode">Martínez-Jiménez et al. (2017) recently reported a workflow classifying potential drug resistance mutations based on Random Forest models and mutation signatures. However, the effect of mutations on the fitness of the clone was not taken into account. In addition, only single-point mutations (SPMs) were considered, despite the notable detection of double-point mutations (DPMs) in cancer patients (href="#mmc1" rid="mmc1" class=" supplementary-material">Table S1).We report an in silico cascade that sequentially evaluates the probability of generating any mutant within 5 Å of a bound ligand, the clonal fitness of each mutation, and the effect of each mutation on drug affinity in order to systematically and objectively prioritize mutations that are highly likely to arise under drug treatment. Importantly, our workflow classifies the impact of a mutation on drug affinity relative to endogenous ligand and does not rely upon accurate calculation of binding free energies; it also ranks mutations according to their likelihood of being generated in particular cancer type. The workflow (href="/pmc/articles/PMC6242700/figure/fig1/" target="figure" class="fig-table-link figpopup" rid-figpopup="fig1" rid-ob="ob-fig1" co-legend-rid="lgnd_fig1">Figure 1 and described in detail below) is validated on a comprehensive benchmark dataset that describes the effect of nearly all possible extracellular signal-regulated kinase 2 (ERK2) missense mutations on sensitivity to the ERK2 inhibitor SCH772984 (href="#bib11" rid="bib11" class=" bibr popnode">Brenan et al., 2016). In addition, we apply the workflow to four well-established cancer targets and evaluate at least two US Food and Drug Administration-approved drugs (small molecules or biologics) per target.href="/pmc/articles/PMC6242700/figure/fig1/" target="figure" rid-figpopup="fig1" rid-ob="ob-fig1">class="inline_block ts_canvas" href="/core/lw/2.0/html/tileshop_pmc/tileshop_pmc_inline.html?title=Click%20on%20image%20to%20zoom&p=PMC3&id=6242700_gr1.jpg" target="tileshopwindow">target="object" href="/pmc/articles/PMC6242700/figure/fig1/?report=objectonly">Open in a separate windowclass="figpopup" href="/pmc/articles/PMC6242700/figure/fig1/" target="figure" rid-figpopup="fig1" rid-ob="ob-fig1">Figure 1WorkflowPotential mutations are evaluated based on their predicted effect on the affinity of both the drug and endogenous ligand (orange), the fitness of the resultant clone (blue), and the requirement for triple-point mutations to generate a mutant (lime green). Resistance hotspots are identified within the remaining set of resistant mutants; these resistance hotspots are protein residues where multiple amino acid changes are predicted to lead to resistance and which therefore have a high likelihood of functional relevance. Resistant mutations at these hotspots are prioritized based on the probability that they will be generated according to the known DNA mutational signatures operating in a particular cancer type.