According to pleiotropy, one gene contributes to multiple unrelated characteristics. For example, the gene coding for seed coat colour is not only responsible for seed coat colour, but it also contributes to flower and axil pigmentation as well. There are many examples of pleiotropic genes in humans, as well. Marfan syndrome is a disorder which shows pleiotropy. One gene is responsible for a constellation of symptoms, including thinness, joint hypermobility, limb elongation, lens dislocation, and increased susceptibility to heart disease.

Moreover, phenylketonuria PKU is one of the most widely cited examples of pleiotropy in humans. A defect in the gene coding for the enzyme phenylalanine hydroxylase results in the multiple phenotypes associated with PKU, including mental retardation, eczema, and pigment defects.

Epistasis occurs when the expression of a gene is controlled by the expression of another gene. Pleiotropy, on the other hand, occurs when a single gene controls many phenotypic traits. So, this is the key difference between epistasis and pleiotropy. According to epistasis, one gene can influence another gene for its expression. According to pleiotropy, some genes affect more than one trait.

Moreover, another difference between epistasis and pleiotropy is that gene interactions take place in epistasis while genes do not interact in pleiotropy. Epistasis is the phenomenon in which a gene at one particular locus modifies the phenotypic expression of a gene at another locus. Pleiotropy is the phenomenon in which a single gene controls or influences multiple phenotypic traits.

In epistasis, two or more genes affect one trait while in pleiotropy, one gene affects two or more trait. Thus, this is the key difference between epistasis and pleiotropy. Nevertheless, this example highlights the idea that antagonistic pleiotropy can be a trade-off between beneficial and detrimental effects. As touched upon earlier in this article, there are many examples of pleiotropic genes in humans, some of which are associated with disease.

For instance, Marfan syndrome is a disorder in humans in which one gene is responsible for a constellation of symptoms, including thinness, joint hypermobility, limb elongation, lens dislocation, and increased susceptibility to heart disease. Similarly, mutations in the gene that codes for transcription factor TBX5 cause the cardiac and limb defects of Holt-Oram syndrome , while mutation of the gene that codes for DNA damage repair protein NBS1 leads to microcephaly, immunodeficiency, and cancer predisposition in Nijmegen breakage syndrome.

One of the most widely cited examples of pleiotropy in humans is phenylketonuria PKU. This disorder is caused by a deficiency of the enzyme phenylalanine hydroxylase, which is necessary to convert the essential amino acid phenylalanine to tyrosine.

A defect in the single gene that codes for this enzyme therefore results in the multiple phenotypes associated with PKU, including mental retardation, eczema, and pigment defects that make affected individuals lighter skinned Paul, The phenotypic effects that single genes may impose in multiple systems often give us insight into the biological function of specific genes.

Pleiotropic genes can also provide us valuable information regarding the evolution of different genes and gene families, as genes are "co-opted" for new purposes beyond what is believed to be their original function Hodgkin, Quite simply, pleiotropy reflects the fact that most proteins have multiple roles in distinct cell types; thus, any genetic change that alters gene expression or function can potentially have wide-ranging effects in a variety of tissues.

Chang, B. Retinal degeneration mutants in the mouse. Vision Research 42 , — Fairbanks, D. Mendelian controversies: A botanical and historical review. American Journal of Botany 88 , — An analysis of the pleiotropic effects of a new lethal mutation in the rat Mus norvegicus. Hartl, D. Genetics: Analysis of Genes and Genomes , 6th ed. Boston, Jones and Bartlett, Hodgkin, J. Seven types of pleiotropy. International Journal of Developmental Biology 42 , — Landauer, W.

Weight and size of organs in frizzle fowl. Storrs Agricultural Experiment Station Bulletin Paul, D. A double-edged sword. Nature , doi Rodier, F. Two faces of p53 : Aging and tumor supression. Nucleic Acids Research 35 , — doi A delayed wave of death from reproduction in Drosophila. Science , — doi Sunquist, F.

Malaysian mystery leopards. National Wildlife 45 Accessed September 19, Williams, G. Pleiotropy, natural selection, and the evolution of senescence. Evolution 11 , — Chromosome Theory and the Castle and Morgan Debate. Discovery and Types of Genetic Linkage. Genetics and Statistical Analysis. Thomas Hunt Morgan and Sex Linkage. Developing the Chromosome Theory. Genetic Recombination. Gregor Mendel and the Principles of Inheritance. Mitosis, Meiosis, and Inheritance. Multifactorial Inheritance and Genetic Disease.

Non-nuclear Genes and Their Inheritance. Polygenic Inheritance and Gene Mapping. Sex Chromosomes and Sex Determination. Sex Determination in Honeybees. Consequently, the covariation of mutational epistatic effects cannot explain the modular covariation found among genotypes in the MEEP or SEQ G -matrices.

We thus speculate that it is the combination of main and epistatic effects within the observed sequences that contributes to the higher modularity seen in the MEEP and SEQ data. That combinatorial aspect of the data is given by the graph structure of the GP map and thus depends on the distribution of pleiotropy of the main and epistatic effects among the traits, which we evaluate next.

To measure the clustering of main and epistatic effects within the GP map, we use the parcellation statistic M P of Mezey et al. M P measures the lack of pleiotropic effects between nonoverlapping sets of traits modules. Among all possible clustering of the 15 drug traits, there will be some that maximize the parcellation of the ME data compared with randomized sets.

S7 , Supplementary Material online. Together, this shows that EP form well supported modules that match those found in the viral sequence fitness data, whereas ME cluster among a smaller partitioning of two modules, confirming that pleiotropy is less modular for main than for epistatic effects.

Fitness trade-offs among drug classes are not clearly apparent from the analysis of correlational patterns: the pairwise genetic correlations between drug environments are positive and often large.

Trade-offs might, however, exist and explain part of the variation in genetic correlations. Indeed, the proportion of antagonistic EP between two drug environments i. Main effects show lower proportions of antagonistic effects in drug pairs, with 1. The difference remains when accounting for the net effects of the mutations, calculated by summing main and epistatic effects of each interaction 1.

Interestingly, the higher proportion of antagonism between drug classes is largely caused by the NNRTI mutations with the other drug classes 8. By introducing variation in the sign of pleiotropy in addition to variation in its extent, epistasis causes lower correlations among drug environments because of larger trade-offs in net mutational fitness effects.

Is the correlation of fitness values within drug classes phenotypic integration indicative of the degree of pleiotropy of the mutations genetic integration? We looked at two aspects of mutation pleiotropy: 1 the proportion of significant mutations that are fully pleiotropic in each drug class i. Therefore, more phenotypically correlated traits are here also more genetically integrated traits, with regards to the among-trait density of pleiotropy of their underlying mutations.

Together with results from the previous section, phenotypic integration is supported by more pleiotropic links and less antagonistic mutational effects within than between drug classes. Until now we have kept all epistatic interactions within and between reverse transcriptase and protease. To our knowledge, it has not been investigated whether mutations within reverse transcriptase may affect the inhibitory effects of drugs targeting protease, and vice versa.

Hinkley et al. Our analysis of the genetic co-variation of fitness of HIV-1B among ARV drugs confirms that there exists high potential for the evolution of cross-resistance.

Evaluation of the response of HIV to selection imposed by multiple ARV treatments can be conducted using quantitative genetics theory based on the genetic variance—covariance matrix the G -matrix of fitness in multiple environments Falconer ; Lande ; Via and Lande As first noted by Falconer , considering the genetic correlation between trait values in two different environments is not different from the genetic correlation of two different traits in the same environment.

The differential environmental effects of the single and double mutations are thus rightly interpreted as pleiotropic effects structuring the GP map of fitness in multiple environments.

This is even more true for within drug class cross-resistance, especially within NNRTIs and PIs, which are more correlated and can be seen as near-identical traits.

From the same token, drug combinations with the lowest potential for cross-resistance evolution are combinations with lowest pairwise or three-way correlations, when mimicking combination therapies. Of course, medical recommendations take more criteria in consideration than the potential for cross-resistance evolution from in vitro data. For instance, each drug has its own pharmacokinetics and side effects, and may elicit specific combinations of resistance mutations, especially among NRTIs Ali et al.

This, in part, justifies the fact that we kept all drug environments separate, although NNRTIs and PIs could be considered as single trait-environments based on their high within module average correlations, which would lower the average pleiotropic degree. We have measured pleiotropy by performing an environment-wise exclusion of nonsignificant mutational effects under the rational that nearly neutral mutations of small effects can be discarded because not significantly contributing to fitness variation in a given environment.

This approach is expected to provide an acceptable proxy of the pleiotropic degree of a mutation, or a gene, when detection thresholds are not too low Wagner and Zhang The advantage of this measure is to correspond to the definition of pleiotropy used in theoretical studies as being the number of discrete phenotypic traits that are affected by a mutation Chevin et al.

Its evolutionary properties are thus well understood. It may, nevertheless, lead to a downward-biased estimate of pleiotropy resulting in the characteristic L-shaped PD distribution found previously because of low detectability of mutational effects of otherwise fully pleiotropic mutations Hill and Zhang Although our PD distributions may also be biased towards lower values, they drastically differ from previous distributions by their high frequency of fully pleiotropic mutations fig.

That shape is not predicted as an outcome of detection bias, even when genetic correlations among traits are high Hill and Zhang Furthermore, the shape of the PD distributions is robust to our significance filtering because not affected when using only the most or the least frequent mutations. Yet, PD ME of the most frequent mutations i. S1 , Supplementary Material online. Similar PD distributions are found for less frequent mutations and for interactions supplementary figs.

S2 and S3 , Supplementary Material online. In sum, given that nonpleiotropic mutations have lower absolute effect sizes than mutations with large PDs, we have effectively eliminated mutations that would, on an average, less affect fitness variation in the presence of selection than more pleiotropic mutations supplementary fig. S4 , Supplementary Material online. Altogether, this gives us confidence that our PD estimates are robust and biologically meaningful.

Studying mutations within two essential genes of a virus allowed us to uncover a direct link between mutational variation in pleiotropy and the pattern of covariation among the phenotypes. This relationship is not expected to be as direct in higher organisms where the ontogeny of phenotypes involves multiple loci and intermingled developmental processes, which may rewrite the character relationships in slightly different ways at each developmental stage.

The best known example is the mouse cranial and mandibular morphological traits, where pleiotropic effects of mapped QTL show significant modularity relative to basal developmental units Mezey et al.

In a simpler organism like HIV, the influence of the genes is directly integrated into phenotypic variation avoiding being drowned out by a succession of developmental processes. Nevertheless, a similar direct relationship between GP map and phenotypic structures may be expected in higher organisms for more basal phenotypes. An example would be molecular phenotypes, such as gene expression traits where co-variation in expression may be a direct function of the activity of pleiotropic gene regulatory elements.

Recent transcriptomics studies support that idea and have uncovered heritable variation of gene expression in a few organisms [e. The extent of epistatic pleiotropy in other organisms is little known beside a few QTL studies in mice Wolf et al. In contrast, we show a large effect of epistatic pleiotropy on trait covariation.

Moreover, we found that although a large proportion of single mutations do not significantly affect fitness in any environments, they will gain significant effects in interaction and change the pleiotropic effects of their interacting partners.

Such mutations can be viewed as modifiers of pleiotropy with potentially large effects on trait covariances Pavlicev and Wagner Similar effects have been found by mapping loci that affect the relationship between traits but remain hidden to the analysis of single effects.

Such relationship QTL have been shown to affect the trait covariation of mice morphological traits Cheverud et al. Interestingly, although our study drastically differs in scope and scale, we show that the same phenomena take place in organisms as different as a virus and a mouse. Our study contributes to better understand the relationship between M - and G -matrices. Little is known about their correspondence in living species, and direct estimates of M -matrices are scarce but see Camara and Pigliucci ; Estes et al.

The correlation of the allelic effects of pleiotropic genes encapsulated in the M -matrix has important evolutionary consequences because, if nonrandom, it creates stable patterns of genetic and phenotypic correlations among phenotypic characters Jones et al.

The correlational pattern of phenotypes then depends on how much the structure of the M -matrix translates into heritable genetic covariation of the traits, encapsulated in the G -matrix. We show that the correlational structure within the M -matrices of main and epistatic effects are poor predictors of the covariation of fitness among drug environments. Therefore, patterns of trait integration here depend on the distribution of the pleiotropic effects among the traits within the GP map.

This indicates that the precise structure of the GP map cannot be ignored as a cause of trait covariation or genetic constraints, and, hence, as a means of their evolution by modification of gene pleiotropy Guillaume and Otto ; Pavlicev and Wagner By studying phenotypic variation and its underlying mutational variation, we were able to link phenotypic with pleiotropic modularity. The main effects also play an important role by quantifying the effects that are shared among environments.

They, however, lack modularity in the correlations and in the patterning of their pleiotropic effects within the GP map. Epistasis modulates these effects and although it increases pleiotropy, it does so in a modular fashion, and provides more drug-specific and class-specific profiles to viral fitness.

Our findings have strong implications for our understanding of the evolutionary potential of drug resistance against multiple drugs in HIV. More generally, they suggest that epistasis may play a fundamental role in shaping the evolvability of living organisms by directly affecting the structure of the GP map and the strength of genetic correlations of phenotypic traits. Our analyses are based on a sample of 70, patient-derived HIV-1 sequences assayed for fitness in 16 different environments: one drug-free and 15 in presence of one antiretroviral drug Petropoulos et al.

The replicative capacity fitness of each sample was measured by inserting the full sequence of the protease gene 99 amino acids and most of the reverse transcriptase gene amino acids 1— of the patient-derived virion into the backbone of an HIV-derived test vector used for routine drug resistance testing.

The test vector is based on the NL molecular HIV clone and has been modified such that it undergoes only one round of replication full details are in Petropoulos et al. The fitness of each sequence is the number of virus progeny produced after one replication cycle relative to that of the base test vector, and provides an estimate of viral fitness in each environment.

Amino acid sequences of the protease gene and the partial reverse transcriptase gene were obtained by population sequencing for all virus samples included in this analysis Petropoulos et al. Because of this population sequencing, sequences are defined in terms of probabilities of allele occurrences for each locus.

The effects of all single and double mutations present in the sequence dataset were then estimated using a generalized kernel ridge regression GKRR procedure Hinkley et al. In brief, the GKRR is a machine learning approach fitting a general linear model in which the number of parameters to estimate by far outnumbers observation points.

It avoids over-fitting of the model by using a training sub-dataset completely independent from the sub-dataset from which the parameters are estimated. The model fitted by Hinkley et al.

The s parameters may be different from 0 or 1 because of population sequencing of the HIV viral sequences and the possible ambiguities caused by site-specific amino-acid polymorphism. I , the intercept, represents the log fitness of the reference sequence of the test vector see details in Hinkley et al.

In total, 1, alleles are present at the amino-acid positions, with 1,, possible interactions. Among these, we excluded nonindependent interactions between nonpolymorphic sites and those interactions present in less than 10 sequences, resulting in a net total of , epistatic effects. Amino-acid variants with less than ten copies in the whole sequence dataset were also excluded in Hinkley et al. To provide an estimate of the accuracy of the GKRR method to detect small effects, we bootstrapped the MEEP model by resampling the fitness of each viral sequence from the observed values, with replacement.

Because the bootstrap confidence intervals CI are strongly correlated among drug environments, we generated bootstrap estimates in the EFV, 3TC, and LPV drug environments, hence bootstrapping one drug per drug class. The number of bootstraps is limited because of the extremely high computational demand of the procedure, and because little variation in confidence interval sizes is detected above bootstrap replicates.

The size of the CI of a mutation thus depends on the amount of information available in the dataset instead of its effect size. We assessed the significance of the observed distributions of pleiotropy and epistatic pleiotropy using different randomization tests.

First, we tested for the significance of the distributions of PD ME and PD EP by randomizing the significant single and double mutation effects among the 16 environments.

Each matrix was shuffled 10, times from which we calculated the mean PD distributions. Second, to test for the significance of the observed pattern of epistatic pleiotropy, we permuted the significant MEs and EPs within each environment to generate random expectations of repertoire composition while keeping the significance levels constant per environment for main and epistatic effects.

This randomization scheme allows us to test for the null hypothesis of universal pleiotropy under which the epistatic pleiotropic effects the private traits would be caused by the failure of detecting the corresponding main effects in the interacting mutations.

It thus tests whether the degree of nonadditive epistatic pleiotropy we observe is an artifact of the significance filtering we used. For this, we generated a null distribution of the average number of private traits per interaction by performing 10, permutations of the 1, main and , epistatic effects in each environment. A P value associated with our point estimate of privateness can be readily obtained by finding the empirical percentile from that bootstrapped distribution.

Finally, we also test whether the additivity of the interaction departs from its null expectation under the same assumption. GP maps with more integrated modules of traits that are genetically less related to other such modules will thus have higher autonomy.

The random skewers approach measures the average vector correlation average angle of the selection response vectors of two covariance matrices submitted to the same set of random selection vectors Cheverud We used 10, random selection vectors random skewers to provide an estimate of the similarity in the orientation of the covariance matrices in phenotype space.

The covariance matrices M and G -matrices were themselves generated by measuring all pairwise environment covariances of within environment genotypic values G -matrices or of all within environment significant mutational effects M -matrices. To compare the parcellation of two networks of different sizes, we defined the following standardized parcellation index:.

All tests and analyses were performed with R v3. PCA and hierarchical cluster analysis were performed with prcomp and pvclust functions in R, respectively. The quality of the article was greatly improved after full peer-review at Axios Review Vancouver, Canada and comments of two anonymous reviewers. National Center for Biotechnology Information , U. Journal List Mol Biol Evol v. Mol Biol Evol. Published online Sep Robert Polster , 1 Christos J. Christos J. Author information Copyright and License information Disclaimer.

For commercial re-use, please contact journals. This article has been cited by other articles in PMC. Abstract The genotype—phenotype GP map is a central concept in evolutionary biology as it describes the mapping of molecular genetic variation onto phenotypic trait variation.

Keywords: mutation, genetic integration, M-matrix, G-matrix. Introduction A central goal of evolutionary biology is to understand how genetic variation maps onto phenotypic variation, and ultimately fitness. Open in a separate window. Table 4 Abbreviations used in the Text. GP map genotype—phenotype map ARV antiretroviral a.