Open Collections

Using high-dimensional yeast-cell phenotyping to screen for polymorphisms influencing trait variability

Banff International Research Station for Mathematical Innovation and Discovery

Identifying which genetic variants contribute to complex phenotypes and disease is a major goal of modern biology. Screens for such variants are no longer limited to single traits. Instead, an increasing number of studies survey hundreds to thousands of phenotypes at once. Often these phenotypes are not independent, yet very few high-dimensional screens have addressed this issue; most report results assuming independence among phenotypes. We encounter this problem while performing high-dimensional phenotyping to search for polymorphisms that influence trait variability, which are not often studied despite their demonstrable fitness effects. Increased trait variability can be beneficial, for example, microorganisms with increased growth heterogeneity stand a better chance to survive antibiotics. On the other hand, phenotypic variability can be highly undesirable and even buffered during development. Although countless screens have identified polymorphisms that alter the mean values of traits, only a few studies have identified polymorphisms that alter trait variances. Because the prevalence of polymorphisms that influence variability is unknown, we utilize high-dimensional phenotyping to assay many traits for genetic effects on variability. We measure 200 morphological parameters in 374 segregating yeast strains from a cross between morphologically divergent parents. Using strict statistical cutoffs, we identity 20 genetic regions that contribute to the average morphology of each strain, but we detect none that independently influence morphological variability of each strain. The apparent rarity of polymorphisms influencing variability may be less striking given the 200 phenotypes we measure are partially redundant; for example, cell size correlates strongly with cell width. Correlation between phenotypes poses other problems for our study as well. For example, redundant phenotypes introduce bias when we ask what fraction of overall morphological diversity is heritable. An exploration of various techniques to reduce redundancies (e.g. principal component analysis, partitioning around medoids, GFLasso) presents several solutions, but none are yet ideal for our purposes. Methods that simultaneously model means and variances of multiple, potentially correlated traits are needed to understand how genetic variation shapes the phenotypes we observe in nature.

Mathematics; Biology and other natural sciences; Systems theory; control; Mathematical biology

video/mp4

Author affiliation: New York University

2014-08-07

Attribution-NonCommercial-NoDerivs 2.5 Canada

http://hdl.handle.net/2429/49616

Unreviewed

http://creativecommons.org/licenses/by-nc-nd/2.5/ca/