Biomathematics / Computational Biology Colloquium

First of Two COB Talks for 11/14: Can Differential Geometry of Fitness Landscapes Solve the Genotype-Phenotype Problem? Hints from Bacteria, Cancer and Autism

Speaker: Olivier Lichtarge MD, PhD, Cullen Chair Professor of Molecular and Human Genetics, Baylor College of Medicine, Houston, TX

Location: Warren Weaver Hall 1314

Date: Tuesday, November 14, 2017, 11:15 a.m.


The genotype-phenotype relationship governs life in the short term and evolution over the long term, but its complexity has frustrated proper description by elementary equations. By defining differentiation and integration in mutational landscapes, the talk will present an evolutionary calculus to solve genotype-phenotype equations, predict the evolutionary action of mutations on fitness, and elucidate which genes drive a trait, or a disease, in a population. Across 20 cancer cohorts, integrating the evolutionary action of somatic missense mutations from 5,996 patients identified 460 genes with abnormal mutational displacements in the fitness landscape, and likely drivers of tumor. Many of these genes escape detection by current techniques, but most have literature, genetic or functional support, including several dozen with oncogenic integral profiles. Likewise, in autism, integrating de novo missense variants across 2,384 probands implicated 398 genes from 23 pathways, including axonogenesis, synaptic transmission, and neurodevelopmental pathways. The predicted fitness impact of de novo and inherited missense variants in these candidate genes correlated with patient IQ and identified phenotypic differences even for exclusively novel candidates. To test whether these ideas can unravel genotype-phenotype across evolutionary clades, other data show that integrating missense variants in ciprofloxacin pressured E. coli recovers precisely the genes known to confer resistance to that antibiotic. The evolutionary root of these results lies in the selection forces that constrain accepted mutations. These mutations, which are thought to be nearly neutral, can now be shown to have an exponential bias towards least evolutionary action. As a result, a normal population on average follows geodesics in the fitness landscape, but a population selected to carry a specific trait or with a given disease, steps off the geodesic.  Beyond biological insights in cancer, autism, or antibiotic resistance, this work shows that the explicit use of evolution to measure fitness gradients, pathogenic missteps and selective forces in mutational landscapes may elucidate which genes support a trait, phenotype, or disease common to a population. Thus, basic concepts from calculus, a technique to solve formulaically otherwise intractable problems, may support a new analytic paradigm in biology to solve the genotype-phenotype problem, as we sought to show.