Predicting evolution might sound like an impossible task. The immense complexity of biological systems and their interactions with the environment has meant that many biologists have abandoned the idea as a lost cause. But despite this, evolution often repeats itself. This repeatability offers hope for being able to spot in advance how evolution will happen. To make general predictions, it is necessary to understand the mechanisms underlying evolutionary pathways, and studying microbes in the laboratory allows for real-time experiments in evolution.

One of the best studied microbes for experimental evolution is Pseudomonas fluorescens, which repeatedly evolves flattened wrinkled colonies instead of round smooth ones when there is limited oxygen. The underlying molecular pathways that lead to this change have been studied in detail.

Lind et al. developed mathematical models to predict how often the three most common pathways would be used and which genes were most likely to be mutated. After controlling for the effects of natural selection and refining the models to take into account mutation hotspots, Lind et al. were able to accurately predict the genes that would be targeted by mutations.

The findings suggest that biologists need not lose hope when it comes to the goal of predicting evolution. A deep understanding of the molecular mechanisms of evolutionary changes are essential to predicting the mutations that lead to adaptive change. The results are an important first step towards forecasting organisms’ responses to changing conditions in the future. In the short term, this is important for medical issues, including antibiotic resistance, cancer and immune receptors. In the long term, predicting the course of evolution could be essential for survival of life on the planet.

Adaptation requires the realization of beneficial mutations. As self-evident as this may be, predicting the occurrence of beneficial mutations and their trajectories to improved fitness is fraught with challenges (Lässig et al., 2017). Nonetheless progress has been made for phenotypically diverse asexual populations subject to strong selection. Effective approaches have drawn upon densely sampled sequence data and equilibrium models of molecular evolution to predict amino acid preferences at specific loci (Luksza and Lässig, 2014). Predictive strategies have also been developed based on selection inferred from the shape of coalescent trees (Neher et al., 2014). In both instances the models are coarse-grained and sidestep specific molecular and mutational details.

There is reason to by-pass molecular details: mutation, being a stochastic process, means that for the most part details are likely to be idiosyncratic and unpredictable. But an increasing number of investigations give reason to think otherwise – that adaptive molecular evolution might follow rules (Yampolsky and Stoltzfus, 2001; Pigliucci, 2010; Stern, 2013; Laland et al., 2015). This is particularly apparent from studies of parallel molecular evolution (Colosimo et al., 2005; Woods et al., 2006; Ostrowski et al., 2008; Flowers et al., 2009; Meyer et al., 2012; Tenaillon et al., 2012; Zhen et al., 2012; Herron and Doebeli, 2013; Galen et al., 2015; Bailey et al., 2017; Kram et al., 2017; Stoltzfus and McCandlish, 2017), and particularly from studies that show parallel evolution to be attributable – at least in part – to factors other than selection (McDonald et al., 2009; Lind et al., 2015; Bailey et al., 2017).

A standard starting position for predicting adaptive evolution recognises the importance of population genetic parameters including mutation rate, generation time, population size, selection and more recently information on the distribution of beneficial fitness effects. An often used approach appeals to ‘origin-fixation’ models that estimate the probability that selection will realise particular mutational trajectories based on the assumption that the rate of evolution is a function of the rate at which mutations originate, multiplied by their fixation probability (for review, see (McCandlish and Stoltzfus, 2014), for application, see for example (Weinreich et al., 2006)). Leaving aside a number of simplifying and restrictive assumptions, population genetic approaches such as those encompassed by origin-fixation models require pre-existing knowledge of (or assumptions concerning) mutation rate to a particular phenotype and selection coefficients (obtained in the absence of frequency dependent effects) and thus have predictive power only in specific situations and with the benefit of hindsight (McCandlish and Stoltzfus, 2014).

Looking to the future, there is hope that one day it might be possible to predict the course of evolutionary change in response to specific environmental challenges in much the same way as gene function, regulation and interactions can be predicted today based on knowledge of nucleotide sequence data. A central issue is to define the necessary information. Mechanistic understanding of the connection between genotype and phenotype combined with knowledge of the genomic bases of mutational bias offer opportunities for progress.

Mutations arise randomly with respect to utility, but genetic architecture can influence the translation of mutation into phenotypic variation: the likelihood that a given mutation generates phenotypic effects depends on the genotype-to-phenotype map (Alberch, 1991; Gompel and Prud'homme, 2009; Stern and Orgogozo, 2009; Rainey et al., 2017). Thus, the functions of gene products and their regulatory interactions provide information on likely mutational targets underpinning particular phenotypes. For example, consider a hypothetical structural determinant subject to both positive and negative regulation and whose over-expression generates a given adaptive phenotype. Assuming a uniform distribution of mutational events, mutations in the negative regulator (and not the positive activator) will be the primary cause of the adaptive phenotype. This follows from the fact that loss-of-function mutations are more common than gain-of-function mutations. Indeed, an emerging rule indicates that phenotypes determined by genetic pathways that are themselves subject to negative regulation are most likely to arise by loss-of-function mutations in negative regulatory components (McDonald et al., 2009; Tenaillon et al., 2012; Lind et al., 2015; Fraebel et al., 2017).

Mutation is not equally likely at each nucleotide of a given genome (Lind and Andersson, 2008; Lynch, 2010; Seier et al., 2011; Foster et al., 2015; Reijns et al., 2015; Sankar et al., 2016; Stoltzfus and McCandlish, 2017). Numerous instances of mutational bias have been reported. Prime examples are simple sequence repeats such as homopolymeric nucleotide tracts or di-, tri- and tetrameric repeats that mutate at high frequency via slipped strand mispairing (Levinson and Gutman, 1987). These readily identifiable sequences define contingency loci in obligate human pathogens and commensals (Moxon et al., 1994) and are widespread in eukaryotic genomes (Tautz and Renz, 1984). The behaviour of contingency loci can be further modulated by defects in components of methyl-directed mismatch repair systems (Richardson and Stojiljkovic, 2001; Martin et al., 2004; Hammerschmidt et al., 2014; Heilbron et al., 2014).

Certain palindromic structures also lead to mutational bias (Viswanathan et al., 2000; Lovett, 2004) and promote amplification events that can increase target size for additional mutations (Roth et al., 1996; Kugelberg et al., 2010; Reams and Roth, 2015) Transition-transversion bias (Stoltzfus and McCandlish, 2017) and elevated mutation rates at CpG sites (Galen et al., 2015) can further skew the distributions of mutational effects. Additional bias arises from the chromosomal neighbourhood of genes under selection (Steinrueck and Guet, 2017), the location of genes with regard to interactions with DNA replication/transcription machineries (Sankar et al., 2016), and environmental factors that affect not only mutation rate but also the spectra of mutational events (Krašovec et al., 2017; Maharjan and Ferenci, 2017; Shewaramani et al., 2017).

Beyond the genotype-to-phenotype map and mutational biases, predicting adaptive evolution requires ability to know a priori the fitness effects of specific mutations. At the present time there is much theoretical and empirical interest in the distribution of fitness effects (DFE) (Eyre-Walker and Keightley, 2007) — and particularly the DFE of beneficial mutations (Orr, 2005) — because of implications for predicting the rate of adaption and likelihood of parallel evolution (de Visser and Krug, 2014), but knowledge of the shape of the distribution is insufficient to connect specific mutations to their specific effects, or to their likelihood of occurrence. Such connections require a means of knowing the connection between mutations and their environment-specific fitness effects. This is a tall order. A starting point is to understand the relationship between all possible mutational routes to a particular phenotype and the set that are realised by selection.

Here we take a bacterial system in which the genetic pathways underpinning evolution of the adaptive ‘wrinkly spreader’ (WS) type are known and use this to explore the current limits on evolutionary forecasting. Pseudomonas fluorescens SBW25 growing in static broth microcosms rapidly depletes available oxygen establishing selective conditions that favour mutants able to colonise the air-liquid interface. The most successful mutant-class encompasses the WS types (Ferguson et al., 2013; Lind et al., 2017b). These types arise from mutational activation of diguanylate cyclases (DGCs) that cause over-production of the second messenger c-di-GMP (Goymer et al., 2006; McDonald et al., 2009), over-production of an acetylated cellulose polymer (Spiers et al., 2002; Spiers et al., 2003) and ultimately formation of a self-supporting microbial mat (Figure 1A).

Figure 1

Download asset Open asset

McDonald et al. (McDonald et al., 2009) showed that each time the tape of WS evolution is re-run mutations generating the adaptive type arise in one of three DGC-encoding pathways (Wsp, Aws, or Mws) (Figure 1A). Subsequent work revealed that when these three pathways are eliminated from the ancestral type evolution proceeds along multiple new pathways (Lind et al., 2015) resulting in WS types whose fitnesses are indistinguishable from those arising via mutations in Wsp, Aws, or Mws. Parallel evolution of WS involving preferential usage of the Wsp, Aws and Mws pathways is therefore not explained by selection: repeated use of Wsp, Aws and Mws stems from the fact that these pathways are subject to negative regulation and thus, relative to pathways subject to positive regulation, or requiring promoter-activating mutations, gene fusion events, or other rare mutations, present a large mutational target (Lind et al., 2015).

Given repeatability of WS evolution, knowledge of the Wsp/Aws/Mws pathways, plus genetic tools for mechanistic investigation — including capacity to obtain WS mutants in the absence of selection — the WS system offers a rare opportunity to explore the extent to which knowledge of the genotype-to-phenotype map can lead mechanistic models for evolutionary forecasting. Our findings show that short-term mechanistic-level predictions of mutational pathways are possible, but also draw attention to challenges that stem from current inability to a priori predict locus-specific mutational biases and environment-specific fitness effects.

The issue of evolutionary predictability and the relative importance of stochastic events compared to deterministic processes have a long history in evolutionary biology (Darwin, 1872; Simpson, 1949; Jacob, 1977; Gould, 1989; Conway Morris, 2003; Orgogozo, 2015). Recent interest has been sparked by an increasing number of observations that both phenotypic and molecular evolution, under certain circumstances, can be remarkably repeatable (Colosimo et al., 2005; Shindo et al., 2005; Jost et al., 2008; Barrick et al., 2009; Lee and Marx, 2012; Meyer et al., 2012; Zhen et al., 2012; Herron and Doebeli, 2013), but whether these cases are representative for evolutionary processes in general remains to be determined. A related question, with greater potential for practical applications, is whether it is possible to forecast short-term evolutionary events and if so, then the challenge is to stipulate the data necessary to make successful predictions.

Our uniquely detailed knowledge of the WS experimental evolution system has provided a rare opportunity to determine the contributions of mutational bias and genetic architecture to the generation of new adaptive phenotypes, and consequently explore the limits of evolutionary forecasting. A thorough understanding of the function of the molecular species and their interactions allowed development of null models (see especially Model IV) that capture essential features of the genotype-to-phenotype map sufficient to predict the relative likelihood that evolution will follow each of the three principle pathways, along with specific mutational (genetic) targets. Even the most sophisticated of these models failed to forecast outcomes that matched the experimental data shown in Figure 2, however, the reason became apparent upon characterisation of the set of WS mutants obtained without selection: the presence of mutational hotspots. Armed with knowledge of sources of mutational bias it was a simple matter to refine Model IV leading to predictions matching those observed by experiment.

Problematic at the current time is inability to a priori detect all causes of mutational bias, however, it is likely that this will improve as understanding of the biochemical causes of bias improves and algorithms trained to recognise and detect nucleotide patterns indicative of bias are implemented. One specific deletion (ΔY77-Q87) in awsX was found to account for nearly half (20/41) of the mutations in the Aws pathway. Thus, despite the existence of hundreds of possible mutations leading to WS (this work and (McDonald et al., 2009; McDonald et al., 2011; Lind et al., 2015)) one single mutation accounts for more than one quarter of all WS mutations. While the six base pair direct repeat flanking the deletion provides a convincing explanation for its increased rate, it is not clear why this deletion would be ten times more common than the ΔP34-A46 deletion in the same gene that is flanked by ten base pair repeats and contains five base pairs identical to those from the ΔY77-Q87 deletion (Figure 6—source data 1). There are also instances where single base pair substitutions are overrepresented: the AwsR T27P mutation is found in nine cases, while eight other single pair substitutions in Aws were found only once. Consider further the fact that wspE (a gene of ~2.3 kb), where changes to only four specific amino acids repeatedly cause WS, and wspF (a gene of ~1 kb) where any mutation that disrupts function results in WS (Figure 6A) contribute equally to the rates of at which new WS types arise.

It is evident from these findings and from related studies (Pollock and Larkin, 2004) that there is need for detailed experimental measurement of local mutation rates in specific systems. Such investigations stand to contribute to understanding of the causes of mutational bias and the extent to which biases might be conserved among related or even unrelated organisms. If local nucleotide sequence is the major determinant, an estimate of mutation rate will apply strictly to very closely related species, but if the dynamics of molecular processes, such as transcription and replication (Sankar et al., 2016), are major influences then estimates might be applicable to a wider range of species.

Forecasting mutational routes to new phenotypes is one component of a comprehensive forecasting strategy. The second requirement is ability to a priori predict fitness effects. Solving this problem is the Holy Grail for predicting evolution, but at the current time this is not possible. That it matters is made clear by the observation that a subset of all possible WS mutants was found after experimental evolution (Figure 8) due to inability of some to successfully compete with high fitness WS types (Figure 9).

Direct measurement of the fitness effects of large numbers of mutations is difficult, time-consuming, typically only possible for microbial species and against the spirit of identifying a priori predictors of fitness effects. Future success likely rests on ability to infer fitness from parameters such as estimated effects of mutations on thermodynamic stability (Capriotti et al., 2005; Dean and Thornton, 2007; Rodrigues et al., 2016), molecular networks, evolutionary conservation of amino acid residues (Glaser et al., 2003; Ng and Henikoff, 2003; Landau et al., 2005; Kumar et al., 2009; Celniker et al., 2013; Choi and Chan, 2015) or machine-learning methods combining several types of information (Bromberg and Rost, 2007; Li et al., 2009; Capriotti et al., 2013; Yates et al., 2014; Hecht et al., 2015). Possibilities likely also exist to extrapolate findings from a small number of mutations that are either directly constructed and assayed in the laboratory or through fitness estimates of polymorphism data from natural populations. Recent work on the prediction of the fitness effects of random mutations in several genes suggests that in many cases large effect mutations can be predicted using methods based on evolutionary conservation (Lind et al., 2017a).

On a less ambitious scale it may be possible to take advantage of the fact that the distribution of fitness effects associated with mutations in single genes is often bimodal – a consequence of many mutations causing complete loss-of-function rather than intermediate deleterious; effects (Sanjuan, 2010; Hietpas et al., 2011; Jacquier et al., 2013; Firnberg et al., 2014; Lind et al., 2017a; Sarkisyan et al., 2016; Lundin et al., 2017). As such it may be sufficient to know gene function plus consequences arising from loss of function mutations and thus treat mutations within single genes as having equivalent fitness (Sanjuan, 2010; Jacquier et al., 2013; Sarkisyan et al., 2016; Lind et al., 2017a). Interestingly WS mutations occurring in the same gene typically show similar fitness effects (Figure 9). While still requiring experimental data, this simplification may fuel exploration of the relationship between large numbers of mutations and their fitness effects and thus understanding of the extent to which fitness effects are transferable between strains with different genetic backgrounds or closely related species (Ungerer et al., 2003; Pearson et al., 2012; Wang et al., 2014).

Taken at face value, our findings give reason to question the value of aspiring to forecast evolution from first principles based on mechanistic understanding. But we argue against such pessimism and point firstly to the value that stems from a clear understanding of current limitations on forecasting: defining what is known, what is not known, and what needs to be known. That it is possible to take knowledge of the genotype-to-phenotype map and forecast with good accuracy targets and rates is an important advance. There is good reason to suspect that the principles outlined here and previously (Lind et al., 2015) are transferable to other systems and even generalizable (Lind, 2018).

Evolutionary forecasting is in its infancy. In the short term it is likely to be most successful for biological systems where there are experimental data on a large number of independent evolutionary events, such as is the case for influenza, HIV and cancer (Kouyos et al., 2012; Fraser et al., 2014; Lawrence et al., 2014; Luksza and Lässig, 2014; Neher et al., 2014; Eirew et al., 2015). Evolution might appear idiosyncratic indicating that every specific system requires detailed investigation, but our hope is that deeper knowledge of the genotype-to-phenotype map, distribution of fitness effects and mutational biases will allow short term forecasts to be produced using modelling without the need for large-scale experimental studies. A major boost to further refinement of evolutionary forecasting is likely to come from combining coarse (top down) and fine-grained (bottom-up) approaches. Our demonstration that simple null models of functional networks can produce quantitative predictions is a step forward allowing predictions to be directly tested in other experimental systems (Lind, 2018).

All data generated or analysed during this study are included in the manuscript and supporting files. Source data files have been provided for Figures 6 and 9.

1. 1. Alberch P

   (1991) From genes to phenotype: dynamical systems and evolvability

   Genetica 84:5–11.

   https://doi.org/10.1007/BF00123979

   • PubMed
   • Google Scholar
2. 1. Bailey SF
   2. Blanquart F
   3. Bataillon T
   4. Kassen R

   (2017) What drives parallel evolution?: How population size and mutational variation contribute to repeated evolution

   BioEssays : News and Reviews in Molecular, Cellular and Developmental Biology 39:1–9.

   https://doi.org/10.1002/bies.201600176

   • PubMed
   • Google Scholar
3. 1. Bantinaki E
   2. Kassen R
   3. Knight CG
   4. Robinson Z
   5. Spiers AJ
   6. Rainey PB

   (2007) Adaptive divergence in experimental populations of Pseudomonas fluorescens. III. Mutational origins of wrinkly spreader diversity

   Genetics 176:441–453.

   https://doi.org/10.1534/genetics.106.069906

   • PubMed
   • Google Scholar
4. 1. Barrick JE
   2. Yu DS
   3. Yoon SH
   4. Jeong H
   5. Oh TK
   6. Schneider D
   7. Lenski RE
   8. Kim JF

   (2009) Genome evolution and adaptation in a long-term experiment with Escherichia coli

   Nature 461:1243–1247.

   https://doi.org/10.1038/nature08480

   • PubMed
   • Google Scholar
5. 1. Bromberg Y
   2. Rost B

   (2007) SNAP: predict effect of non-synonymous polymorphisms on function

   Nucleic Acids Research 35:3823–3835.

   https://doi.org/10.1093/nar/gkm238

   • PubMed
   • Google Scholar
6. 1. Capriotti E
   2. Fariselli P
   3. Casadio R

   (2005) I-Mutant2.0: predicting stability changes upon mutation from the protein sequence or structure

   Nucleic Acids Research 33:W306–W310.

   https://doi.org/10.1093/nar/gki375

   • PubMed
   • Google Scholar
7. 1. Capriotti E
   2. Calabrese R
   3. Fariselli P
   4. Martelli PL
   5. Altman RB
   6. Casadio R

   (2013) WS-SNPs&GO: a web server for predicting the deleterious effect of human protein variants using functional annotation

   BMC Genomics 14 Suppl 3:S6.

   https://doi.org/10.1186/1471-2164-14-S3-S6

   • PubMed
   • Google Scholar
8. 1. Celniker G
   2. Nimrod G
   3. Ashkenazy H
   4. Glaser F
   5. Martz E
   6. Mayrose I
   7. Pupko T
   8. Ben-Tal N

   (2013) ConSurf: Using Evolutionary Data to Raise Testable Hypotheses about Protein Function

   Israel Journal of Chemistry 53:199–206.

   https://doi.org/10.1002/ijch.201200096

   • Google Scholar
9. 1. Choi Y
   2. Chan AP

   (2015) PROVEAN web server: a tool to predict the functional effect of amino acid substitutions and indels

   Bioinformatics 31:2745–2747.

   https://doi.org/10.1093/bioinformatics/btv195

   • PubMed
   • Google Scholar
10. 1. Colosimo PF
    2. Hosemann KE
    3. Balabhadra S
    4. Villarreal G
    5. Dickson M
    6. Grimwood J
    7. Schmutz J
    8. Myers RM
    9. Schluter D
    10. Kingsley DM

    (2005) Widespread parallel evolution in sticklebacks by repeated fixation of Ectodysplasin alleles

    Science 307:1928–1933.

    https://doi.org/10.1126/science.1107239

    • PubMed
    • Google Scholar
11. Book

    1. Conway Morris S

    (2003)

    Life's Solution Inevitable Humans in a Lonely Universe

    New York: Cambridge University Press.

    • Google Scholar
12. Book

    1. Darwin C

    (1872)

    The Origin of Species by Means of Natural Selection, Or, the Preservation of Favoured Races in the Struggle for Life

    London: John Murray.

    • Google Scholar
13. 1. De N
    2. Pirruccello M
    3. Krasteva PV
    4. Bae N
    5. Raghavan RV
    6. Sondermann H

    (2008) Phosphorylation-independent regulation of the diguanylate cyclase WspR

    PLoS Biology 6:e67.

    https://doi.org/10.1371/journal.pbio.0060067

    • PubMed
    • Google Scholar
14. 1. De N
    2. Navarro MV
    3. Raghavan RV
    4. Sondermann H

    (2009) Determinants for the activation and autoinhibition of the diguanylate cyclase response regulator WspR

    Journal of Molecular Biology 393:619–633.

    https://doi.org/10.1016/j.jmb.2009.08.030

    • PubMed
    • Google Scholar
15. 1. de Visser JA
    2. Krug J

    (2014) Empirical fitness landscapes and the predictability of evolution

    Nature Reviews Genetics 15:480–490.

    https://doi.org/10.1038/nrg3744

    • PubMed
    • Google Scholar
16. 1. Dean AM
    2. Thornton JW

    (2007) Mechanistic approaches to the study of evolution: the functional synthesis

    Nature Reviews Genetics 8:675–688.

    https://doi.org/10.1038/nrg2160

    • PubMed
    • Google Scholar
17. 1. Dykhuizen DE

    (1990) Experimental studies of natural selection in bacteria

    Annual Review of Ecology and Systematics 21:373–398.

    https://doi.org/10.1146/annurev.es.21.110190.002105

    • Google Scholar
18. 1. Eirew P
    2. Steif A
    3. Khattra J
    4. Ha G
    5. Yap D
    6. Farahani H
    7. Gelmon K
    8. Chia S
    9. Mar C
    10. Wan A
    11. Laks E
    12. Biele J
    13. Shumansky K
    14. Rosner J
    15. McPherson A
    16. Nielsen C
    17. Roth AJ
    18. Lefebvre C
    19. Bashashati A
    20. de Souza C
    21. Siu C
    22. Aniba R
    23. Brimhall J
    24. Oloumi A
    25. Osako T
    26. Bruna A
    27. Sandoval JL
    28. Algara T
    29. Greenwood W
    30. Leung K
    31. Cheng H
    32. Xue H
    33. Wang Y
    34. Lin D
    35. Mungall AJ
    36. Moore R
    37. Zhao Y
    38. Lorette J
    39. Nguyen L
    40. Huntsman D
    41. Eaves CJ
    42. Hansen C
    43. Marra MA
    44. Caldas C
    45. Shah SP
    46. Aparicio S

    (2015) Dynamics of genomic clones in breast cancer patient xenografts at single-cell resolution

    Nature 518:422–426.

    https://doi.org/10.1038/nature13952

    • PubMed
    • Google Scholar
19. 1. Eyre-Walker A
    2. Keightley PD

    (2007) The distribution of fitness effects of new mutations

    Nature Reviews Genetics 8:610–618.

    https://doi.org/10.1038/nrg2146

    • PubMed
    • Google Scholar
20. 1. Ferguson GC
    2. Bertels F
    3. Rainey PB

    (2013) Adaptive divergence in experimental populations of Pseudomonas fluorescens. V. Insight into the niche specialist fuzzy spreader compels revision of the model Pseudomonas radiation

    Genetics 195:1319–1335.

    https://doi.org/10.1534/genetics.113.154948

    • PubMed
    • Google Scholar
21. 1. Firnberg E
    2. Labonte JW
    3. Gray JJ
    4. Ostermeier M

    (2014) A comprehensive, high-resolution map of a gene's fitness landscape

    Molecular Biology and Evolution 31:1581–1592.

    https://doi.org/10.1093/molbev/msu081

    • PubMed
    • Google Scholar
22. 1. Flowers JM
    2. Hanzawa Y
    3. Hall MC
    4. Moore RC
    5. Purugganan MD

    (2009) Population genomics of the Arabidopsis thaliana flowering time gene network

    Molecular Biology and Evolution 26:2475–2486.

    https://doi.org/10.1093/molbev/msp161

    • PubMed
    • Google Scholar
23. 1. Foster PL
    2. Lee H
    3. Popodi E
    4. Townes JP
    5. Tang H

    (2015) Determinants of spontaneous mutation in the bacterium Escherichia coli as revealed by whole-genome sequencing

    PNAS 112:E5990–E5999.

    https://doi.org/10.1073/pnas.1512136112

    • PubMed
    • Google Scholar
24. 1. Fraebel DT
    2. Mickalide H
    3. Schnitkey D
    4. Merritt J
    5. Kuhlman TE
    6. Kuehn S

    (2017) Environment determines evolutionary trajectory in a constrained phenotypic space

    eLife 6:e24669.

    https://doi.org/10.7554/eLife.24669

    • PubMed
    • Google Scholar
25. 1. Fraser C
    2. Lythgoe K
    3. Leventhal GE
    4. Shirreff G
    5. Hollingsworth TD
    6. Alizon S
    7. Bonhoeffer S

    (2014) Virulence and pathogenesis of HIV-1 infection: an evolutionary perspective

    Science 343:1243727.

    https://doi.org/10.1126/science.1243727

    • PubMed
    • Google Scholar
26. 1. Fukami T
    2. Beaumont HJ
    3. Zhang XX
    4. Rainey PB

    (2007) Immigration history controls diversification in experimental adaptive radiation

    Nature 446:436–439.

    https://doi.org/10.1038/nature05629

    • PubMed
    • Google Scholar
27. 1. Galen SC
    2. Natarajan C
    3. Moriyama H
    4. Weber RE
    5. Fago A
    6. Benham PM
    7. Chavez AN
    8. Cheviron ZA
    9. Storz JF
    10. Witt CC

    (2015) Contribution of a mutational hot spot to hemoglobin adaptation in high-altitude Andean house wrens

    PNAS 112:13958–13963.

    https://doi.org/10.1073/pnas.1507300112

    • PubMed
    • Google Scholar
28. 1. Glaser F
    2. Pupko T
    3. Paz I
    4. Bell RE
    5. Bechor-Shental D
    6. Martz E
    7. Ben-Tal N

    (2003) ConSurf: identification of functional regions in proteins by surface-mapping of phylogenetic information

    Bioinformatics 19:163–164.

    https://doi.org/10.1093/bioinformatics/19.1.163

    • PubMed
    • Google Scholar
29. 1. Gompel N
    2. Prud'homme B

    (2009) The causes of repeated genetic evolution

    Developmental Biology 332:36–47.

    https://doi.org/10.1016/j.ydbio.2009.04.040

    • PubMed
    • Google Scholar
30. Book

    1. Gould SJ

    (1989)

    Wonderful Life: The Burgess Shale and the Nature of History

    New York: W.W. Norton.

    • Google Scholar
31. 1. Goymer P
    2. Kahn SG
    3. Malone JG
    4. Gehrig SM
    5. Spiers AJ
    6. Rainey PB

    (2006) Adaptive divergence in experimental populations of Pseudomonas fluorescens. II. Role of the GGDEF regulator WspR in evolution and development of the wrinkly spreader phenotype

    Genetics 173:515–526.

    https://doi.org/10.1534/genetics.106.055863

    • PubMed
    • Google Scholar
32. 1. Griswold IJ
    2. Zhou H
    3. Matison M
    4. Swanson RV
    5. McIntosh LP
    6. Simon MI
    7. Dahlquist FW

    (2002) The solution structure and interactions of CheW from Thermotoga maritima

    Nature Structural Biology 9:121–125.

    https://doi.org/10.1038/nsb753

    • PubMed
    • Google Scholar
33. 1. Güvener ZT
    2. Harwood CS

    (2007) Subcellular location characteristics of the Pseudomonas aeruginosa GGDEF protein, WspR, indicate that it produces cyclic-di-GMP in response to growth on surfaces

    Molecular Microbiology 66:1459–1473.

    https://doi.org/10.1111/j.1365-2958.2007.06008.x

    • PubMed
    • Google Scholar
34. 1. Hall BM
    2. Ma CX
    3. Liang P
    4. Singh KK

    (2009) Fluctuation analysis CalculatOR: a web tool for the determination of mutation rate using Luria-Delbruck fluctuation analysis

    Bioinformatics 25:1564–1565.

    https://doi.org/10.1093/bioinformatics/btp253

    • PubMed
    • Google Scholar
35. 1. Hammerschmidt K
    2. Rose CJ
    3. Kerr B
    4. Rainey PB

    (2014) Life cycles, fitness decoupling and the evolution of multicellularity

    Nature 515:75–79.

    https://doi.org/10.1038/nature13884

    • PubMed
    • Google Scholar
36. 1. Hecht M
    2. Bromberg Y
    3. Rost B

    (2015) Better prediction of functional effects for sequence variants

    BMC Genomics 16 Suppl 8:S1.

    https://doi.org/10.1186/1471-2164-16-S8-S1

    • PubMed
    • Google Scholar
37. 1. Heilbron K
    2. Toll-Riera M
    3. Kojadinovic M
    4. MacLean RC

    (2014) Fitness is strongly influenced by rare mutations of large effect in a microbial mutation accumulation experiment

    Genetics 197:981–990.

    https://doi.org/10.1534/genetics.114.163147

    • PubMed
    • Google Scholar
38. 1. Herron MD
    2. Doebeli M

    (2013) Parallel evolutionary dynamics of adaptive diversification in Escherichia coli

    PLOS Biology 11:e1001490.

    https://doi.org/10.1371/journal.pbio.1001490

    • PubMed
    • Google Scholar
39. 1. Hietpas RT
    2. Jensen JD
    3. Bolon DN

    (2011) Experimental illumination of a fitness landscape

    PNAS 108:7896–7901.

    https://doi.org/10.1073/pnas.1016024108

    • PubMed
    • Google Scholar
40. 1. Huangyutitham V
    2. Güvener ZT
    3. Harwood CS

    (2013) Subcellular clustering of the phosphorylated WspR response regulator protein stimulates its diguanylate cyclase activity

    mBio 4:e00242.

    https://doi.org/10.1128/mBio.00242-13

    • PubMed
    • Google Scholar
41. 1. Jacob F

    (1977) Evolution and tinkering

    Science 196:1161–1166.

    https://doi.org/10.1126/science.860134

    • PubMed
    • Google Scholar
42. 1. Jacquier H
    2. Birgy A
    3. Le Nagard H
    4. Mechulam Y
    5. Schmitt E
    6. Glodt J
    7. Bercot B
    8. Petit E
    9. Poulain J
    10. Barnaud G
    11. Gros PA
    12. Tenaillon O

    (2013) Capturing the mutational landscape of the beta-lactamase TEM-1

    PNAS 110:13067–13072.

    https://doi.org/10.1073/pnas.1215206110

    • PubMed
    • Google Scholar
43. 1. Jost MC
    2. Hillis DM
    3. Lu Y
    4. Kyle JW
    5. Fozzard HA
    6. Zakon HH

    (2008) Toxin-resistant sodium channels: parallel adaptive evolution across a complete gene family

    Molecular Biology and Evolution 25:1016–1024.

    https://doi.org/10.1093/molbev/msn025

    • PubMed
    • Google Scholar
44. 1. Kelley LA
    2. Mezulis S
    3. Yates CM
    4. Wass MN
    5. Sternberg MJ

    (2015) The Phyre2 web portal for protein modeling, prediction and analysis

    Nature Protocols 10:845–858.

    https://doi.org/10.1038/nprot.2015.053

    • PubMed
    • Google Scholar
45. 1. King EO
    2. Ward MK
    3. Raney DE

    (1954)

    Two simple media for the demonstration of pyocyanin and fluorescin

    The Journal of Laboratory and Clinical Medicine 44:301–307.

    • PubMed
    • Google Scholar
46. 1. Kouyos RD
    2. Leventhal GE
    3. Hinkley T
    4. Haddad M
    5. Whitcomb JM
    6. Petropoulos CJ
    7. Bonhoeffer S

    (2012) Exploring the complexity of the HIV-1 fitness landscape

    PLOS Genetics 8:e1002551.

    https://doi.org/10.1371/journal.pgen.1002551

    • PubMed
    • Google Scholar
47. 1. Kram KE
    2. Geiger C
    3. Ismail WM
    4. Lee H
    5. Tang H
    6. Foster PL
    7. Finkel SE

    (2017) Adaptation of Escherichia coli to Long-Term Serial Passage in Complex Medium: Evidence of Parallel Evolution

    mSystems 2:16.

    https://doi.org/10.1128/mSystems.00192-16

    • Google Scholar
48. 1. Krašovec R
    2. Richards H
    3. Gifford DR
    4. Hatcher C
    5. Faulkner KJ
    6. Belavkin RV
    7. Channon A
    8. Aston E
    9. McBain AJ
    10. Knight CG

    (2017) Spontaneous mutation rate is a plastic trait associated with population density across domains of life

    PLOS Biology 15:e2002731.

    https://doi.org/10.1371/journal.pbio.2002731

    • PubMed
    • Google Scholar
49. 1. Kugelberg E
    2. Kofoid E
    3. Andersson DI
    4. Lu Y
    5. Mellor J
    6. Roth FP
    7. Roth JR

    (2010) The tandem inversion duplication in Salmonella enterica: selection drives unstable precursors to final mutation types

    Genetics 185:65–80.

    https://doi.org/10.1534/genetics.110.114074

    • PubMed
    • Google Scholar
50. 1. Kumar P
    2. Henikoff S
    3. Ng PC

    (2009) Predicting the effects of coding non-synonymous variants on protein function using the SIFT algorithm

    Nature Protocols 4:1073–1081.

    https://doi.org/10.1038/nprot.2009.86

    • PubMed
    • Google Scholar
51. 1. Laland KN
    2. Uller T
    3. Feldman MW
    4. Sterelny K
    5. Müller GB
    6. Moczek A
    7. Jablonka E
    8. Odling-Smee J

    (2015) The extended evolutionary synthesis: its structure, assumptions and predictions

    Proceedings of the Royal Society B: Biological Sciences 282:20151019.

    https://doi.org/10.1098/rspb.2015.1019

    • Google Scholar
52. 1. Landau M
    2. Mayrose I
    3. Rosenberg Y
    4. Glaser F
    5. Martz E
    6. Pupko T
    7. Ben-Tal N

    (2005) ConSurf 2005: the projection of evolutionary conservation scores of residues on protein structures

    Nucleic Acids Research 33:W299–W302.

    https://doi.org/10.1093/nar/gki370

    • PubMed
    • Google Scholar
53. 1. Lässig M
    2. Mustonen V
    3. Walczak AM

    (2017) Predicting evolution

    Nature Ecology & Evolution 1:0077.

    https://doi.org/10.1038/s41559-017-0077

    • Google Scholar
54. 1. Lawrence MS
    2. Stojanov P
    3. Mermel CH
    4. Robinson JT
    5. Garraway LA
    6. Golub TR
    7. Meyerson M
    8. Gabriel SB
    9. Lander ES
    10. Getz G

    (2014) Discovery and saturation analysis of cancer genes across 21 tumour types

    Nature 505:495–501.

    https://doi.org/10.1038/nature12912

    • PubMed
    • Google Scholar
55. 1. Lee MC
    2. Marx CJ

    (2012) Repeated, selection-driven genome reduction of accessory genes in experimental populations

    PLOS Genetics 8:e1002651.

    https://doi.org/10.1371/journal.pgen.1002651

    • PubMed
    • Google Scholar
56. 1. Levinson G
    2. Gutman GA

    (1987) Slipped-strand mispairing: a major mechanism for DNA sequence evolution

    Molecular Biology and Evolution 4:203–221.

    https://doi.org/10.1093/oxfordjournals.molbev.a040442

    • PubMed
    • Google Scholar
57. 1. Li B
    2. Krishnan VG
    3. Mort ME
    4. Xin F
    5. Kamati KK
    6. Cooper DN
    7. Mooney SD
    8. Radivojac P

    (2009) Automated inference of molecular mechanisms of disease from amino acid substitutions

    Bioinformatics 25:2744–2750.

    https://doi.org/10.1093/bioinformatics/btp528

    • PubMed
    • Google Scholar
58. 1. Lind PA
    2. Andersson DI

    (2008) Whole-genome mutational biases in bacteria

    PNAS 105:17878–17883.

    https://doi.org/10.1073/pnas.0804445105

    • PubMed
    • Google Scholar
59. 1. Lind PA
    2. Farr AD
    3. Rainey PB

    (2015) Experimental evolution reveals hidden diversity in evolutionary pathways

    eLife 4:e07074.

    https://doi.org/10.7554/eLife.07074

    • Google Scholar
60. 1. Lind PA
    2. Arvidsson L
    3. Berg OG
    4. Andersson DI

    (2017a) Variation in Mutational Robustness between Different Proteins and the Predictability of Fitness Effects

    Molecular Biology and Evolution 34:408–418.

    https://doi.org/10.1093/molbev/msw239

    • PubMed
    • Google Scholar
61. 1. Lind PA
    2. Farr AD
    3. Rainey PB

    (2017b) Evolutionary convergence in experimental Pseudomonas populations

    The ISME Journal 11:589–600.

    https://doi.org/10.1038/ismej.2016.157

    • PubMed
    • Google Scholar
62. Preprint

    1. Lind PA

    (2018) Evolutionary forecasting of phenotypic and genetic outcomes of experimental evolution in Pseudomonas

    bioRxiv.

    https://doi.org/10.1101/342261

    • Google Scholar
63. 1. Lovett ST

    (2004) Encoded errors: mutations and rearrangements mediated by misalignment at repetitive DNA sequences

    Molecular Microbiology 52:1243–1253.

    https://doi.org/10.1111/j.1365-2958.2004.04076.x

    • PubMed
    • Google Scholar
64. 1. Luksza M
    2. Lässig M

    (2014) A predictive fitness model for influenza

    Nature 507:57–61.

    https://doi.org/10.1038/nature13087

    • PubMed
    • Google Scholar
65. 1. Lundin E
    2. Tang PC
    3. Guy L
    4. Näsvall J
    5. Andersson DI

    (2017) Experimental determination and prediction of the fitness effects of random point mutations in the biosynthetic enzyme HisA

    Molecular Biology and Evolution.

    https://doi.org/10.1093/molbev/msx325

    • PubMed
    • Google Scholar
66. 1. Luria SE
    2. Delbrück M

    (1943)

    Mutations of Bacteria from Virus Sensitivity to Virus Resistance

    Genetics 28:491–511.

    • PubMed
    • Google Scholar
67. 1. Lynch M

    (2010) Rate, molecular spectrum, and consequences of human mutation

    PNAS 107:961–968.

    https://doi.org/10.1073/pnas.0912629107

    • PubMed
    • Google Scholar
68. 1. Maharjan RP
    2. Ferenci T

    (2017) A shifting mutational landscape in 6 nutritional states: Stress-induced mutagenesis as a series of distinct stress input-mutation output relationships

    PLOS Biology 15:e2001477.

    https://doi.org/10.1371/journal.pbio.2001477

    • PubMed
    • Google Scholar
69. 1. Malone JG
    2. Jaeger T
    3. Spangler C
    4. Ritz D
    5. Spang A
    6. Arrieumerlou C
    7. Kaever V
    8. Landmann R
    9. Jenal U

    (2010) YfiBNR mediates cyclic di-GMP dependent small colony variant formation and persistence in Pseudomonas aeruginosa

    PLOS Pathogens 6:e1000804.

    https://doi.org/10.1371/journal.ppat.1000804

    • PubMed
    • Google Scholar
70. 1. Malone JG
    2. Jaeger T
    3. Manfredi P
    4. Dötsch A
    5. Blanka A
    6. Bos R
    7. Cornelis GR
    8. Häussler S
    9. Jenal U

    (2012) The YfiBNR signal transduction mechanism reveals novel targets for the evolution of persistent Pseudomonas aeruginosa in cystic fibrosis airways

    PLOS Pathogens 8:e1002760.

    https://doi.org/10.1371/journal.ppat.1002760

    • PubMed
    • Google Scholar
71. 1. Martin P
    2. Sun L
    3. Hood DW
    4. Moxon ER

    (2004) Involvement of genes of genome maintenance in the regulation of phase variation frequencies in Neisseria meningitidis

    Microbiology 150:3001–3012.

    https://doi.org/10.1099/mic.0.27182-0

    • PubMed
    • Google Scholar
72. 1. McCandlish DM
    2. Stoltzfus A

    (2014) Modeling evolution using the probability of fixation: history and implications

    The Quarterly Review of Biology 89:225–252.

    https://doi.org/10.1086/677571

    • PubMed
    • Google Scholar
73. 1. McDonald MJ
    2. Gehrig SM
    3. Meintjes PL
    4. Zhang XX
    5. Rainey PB

    (2009) Adaptive divergence in experimental populations of Pseudomonas fluorescens. IV. Genetic constraints guide evolutionary trajectories in a parallel adaptive radiation

    Genetics 183:1041–1053.

    https://doi.org/10.1534/genetics.109.107110

    • PubMed
    • Google Scholar
74. 1. McDonald MJ
    2. Cooper TF
    3. Beaumont HJ
    4. Rainey PB

    (2011) The distribution of fitness effects of new beneficial mutations in Pseudomonas fluorescens

    Biology Letters 7:98–100.

    https://doi.org/10.1098/rsbl.2010.0547

    • PubMed
    • Google Scholar
75. 1. Meyer JR
    2. Dobias DT
    3. Weitz JS
    4. Barrick JE
    5. Quick RT
    6. Lenski RE

    (2012) Repeatability and contingency in the evolution of a key innovation in phage lambda

    Science 335:428–432.

    https://doi.org/10.1126/science.1214449

    • PubMed
    • Google Scholar
76. 1. Micali G
    2. Endres RG

    (2016) Bacterial chemotaxis: information processing, thermodynamics, and behavior

    Current Opinion in Microbiology 30:8–15.

    https://doi.org/10.1016/j.mib.2015.12.001

    • PubMed
    • Google Scholar
77. 1. Moxon ER
    2. Rainey PB
    3. Nowak MA
    4. Lenski RE

    (1994) Adaptive evolution of highly mutable loci in pathogenic bacteria

    Current Biology 4:24–33.

    https://doi.org/10.1016/S0960-9822(00)00005-1

    • PubMed
    • Google Scholar
78. 1. Neher RA
    2. Russell CA
    3. Shraiman BI

    (2014) Predicting evolution from the shape of genealogical trees

    eLife 3:e03568.

    https://doi.org/10.7554/eLife.03568

    • Google Scholar
79. 1. Ng PC
    2. Henikoff S

    (2003) SIFT: Predicting amino acid changes that affect protein function

    Nucleic Acids Research 31:3812–3814.

    https://doi.org/10.1093/nar/gkg509

    • PubMed
    • Google Scholar
80. 1. O'Connor JR
    2. Kuwada NJ
    3. Huangyutitham V
    4. Wiggins PA
    5. Harwood CS

    (2012) Surface sensing and lateral subcellular localization of WspA, the receptor in a chemosensory-like system leading to c-di-GMP production

    Molecular Microbiology 86:720–729.

    https://doi.org/10.1111/mmi.12013

    • PubMed
    • Google Scholar
81. 1. Orgogozo V

    (2015) Replaying the tape of life in the twenty-first century

    Interface Focus 5:20150057.

    https://doi.org/10.1098/rsfs.2015.0057

    • PubMed
    • Google Scholar
82. 1. Orr HA

    (2005) The genetic theory of adaptation: a brief history

    Nature Reviews Genetics 6:119–127.

    https://doi.org/10.1038/nrg1523

    • PubMed
    • Google Scholar
83. 1. Ostrowski EA
    2. Woods RJ
    3. Lenski RE

    (2008) The genetic basis of parallel and divergent phenotypic responses in evolving populations of Escherichia coli

    Proceedings of the Royal Society B: Biological Sciences 275:277–284.

    https://doi.org/10.1098/rspb.2007.1244

    • Google Scholar
84. 1. Parkinson JS

    (2010) Signaling mechanisms of HAMP domains in chemoreceptors and sensor kinases

    Annual Review of Microbiology 64:101–122.

    https://doi.org/10.1146/annurev.micro.112408.134215

    • PubMed
    • Google Scholar
85. 1. Pearson VM
    2. Miller CR
    3. Rokyta DR

    (2012) The consistency of beneficial fitness effects of mutations across diverse genetic backgrounds

    PLoS ONE 7:e43864.

    https://doi.org/10.1371/journal.pone.0043864

    • PubMed
    • Google Scholar
86. 1. Phippen CW
    2. Mikolajek H
    3. Schlaefli HG
    4. Keevil CW
    5. Webb JS
    6. Tews I

    (2014) Formation and dimerization of the phosphodiesterase active site of the Pseudomonas aeruginosa MorA, a bi-functional c-di-GMP regulator

    FEBS Letters 588:4631–4636.

    https://doi.org/10.1016/j.febslet.2014.11.002

    • PubMed
    • Google Scholar
87. 1. Pigliucci M

    (2010) Genotype-phenotype mapping and the end of the 'genes as blueprint' metaphor

    Philosophical Transactions of the Royal Society B: Biological Sciences 365:557–566.

    https://doi.org/10.1098/rstb.2009.0241

    • PubMed
    • Google Scholar
88. 1. Pollock DD
    2. Larkin JC

    (2004) Estimating the degree of saturation in mutant screens

    Genetics 168:489–502.

    https://doi.org/10.1534/genetics.103.024430

    • PubMed
    • Google Scholar
89. 1. Rainey PB
    2. Remigi P
    3. Farr AD
    4. Lind PA

    (2017) Darwin was right: where now for experimental evolution?

    Current Opinion in Genetics & Development 47:102–109.

    https://doi.org/10.1016/j.gde.2017.09.003

    • PubMed
    • Google Scholar
90. 1. Rainey PB
    2. Travisano M

    (1998) Adaptive radiation in a heterogeneous environment

    Nature 394:69–72.

    https://doi.org/10.1038/27900

    • PubMed
    • Google Scholar
91. 1. Reams AB
    2. Roth JR

    (2015) Mechanisms of gene duplication and amplification

    Cold Spring Harbor Perspectives in Biology 7:a016592.

    https://doi.org/10.1101/cshperspect.a016592

    • PubMed
    • Google Scholar
92. 1. Reijns MAM
    2. Kemp H
    3. Ding J
    4. de Procé SM
    5. Jackson AP
    6. Taylor MS

    (2015) Lagging-strand replication shapes the mutational landscape of the genome

    Nature 518:502–506.

    https://doi.org/10.1038/nature14183

    • PubMed
    • Google Scholar
93. 1. Rice MS
    2. Dahlquist FW

    (1991)

    Sites of deamidation and methylation in Tsr, a bacterial chemotaxis sensory transducer

    The Journal of Biological Chemistry 266:9746–9753.

    • PubMed
    • Google Scholar
94. 1. Richardson AR
    2. Stojiljkovic I

    (2001) Mismatch repair and the regulation of phase variation in Neisseria meningitidis

    Molecular Microbiology 40:645–655.

    https://doi.org/10.1046/j.1365-2958.2001.02408.x

    • PubMed
    • Google Scholar
95. 1. Rodrigues JV
    2. Bershtein S
    3. Li A
    4. Lozovsky ER
    5. Hartl DL
    6. Shakhnovich EI

    (2016) Biophysical principles predict fitness landscapes of drug resistance

    PNAS 113:E1470–E1478.

    https://doi.org/10.1073/pnas.1601441113

    • PubMed
    • Google Scholar
96. 1. Römling U
    2. Galperin MY
    3. Gomelsky M

    (2013) Cyclic di-GMP: the first 25 years of a universal bacterial second messenger

    Microbiology and Molecular Biology Reviews 77:1–52.

    https://doi.org/10.1128/MMBR.00043-12

    • PubMed
    • Google Scholar
97. 1. Rosche WA
    2. Foster PL

    (2000) Determining mutation rates in bacterial populations

    Methods 20:4–17.

    https://doi.org/10.1006/meth.1999.0901

    • PubMed
    • Google Scholar
98. Book

    1. Roth JR
    2. Benson N
    3. Galitsky T
    4. Haack K
    5. Lawrence JG
    6. Miesel L

    (1996)

    Escherichia Coli and Salmonella: Cellular and Molecular Biology

    Neidhardt F. C, editors. Washington: Cell Press.

    • Google Scholar
99. 1. Sanjuan R

    (2010) Mutational fitness effects in RNA and single-stranded DNA viruses: common patterns revealed by site-directed mutagenesis studies

    Philosophical Transactions of the Royal Society B: Biological Sciences 365:1975–1982.

    https://doi.org/10.1098/rstb.2010.0063

    • Google Scholar
100. 1. Sankar TS
     2. Wastuwidyaningtyas BD
     3. Dong Y
     4. Lewis SA
     5. Wang JD

     (2016) The nature of mutations induced by replication–transcription collisions

     Nature 535:178–181.

     https://doi.org/10.1038/nature18316

     • PubMed
     • Google Scholar
101. 1. Sarkisyan KS
     2. Bolotin DA
     3. Meer MV
     4. Usmanova DR
     5. Mishin AS
     6. Sharonov GV
     7. Ivankov DN
     8. Bozhanova NG
     9. Baranov MS
     10. Soylemez O

     (2016)

     Local fitness landscape of the green fluorescent protein

     Nature 533:397–401.

     • Google Scholar
102. 1. Seier T
     2. Padgett DR
     3. Zilberberg G
     4. Sutera VA
     5. Toha N
     6. Lovett ST

     (2011) Insights into mutagenesis using Escherichia coli chromosomal lacZ strains that enable detection of a wide spectrum of mutational events

     Genetics 188:247–262.

     https://doi.org/10.1534/genetics.111.127746

     • PubMed
     • Google Scholar
103. 1. Shewaramani S
     2. Finn TJ
     3. Leahy SC
     4. Kassen R
     5. Rainey PB
     6. Moon CD

     (2017) Anaerobically Grown Escherichia coli Has an Enhanced Mutation Rate and Distinct Mutational Spectra

     PLOS Genetics 13:e1006570.

     https://doi.org/10.1371/journal.pgen.1006570

     • PubMed
     • Google Scholar
104. 1. Shindo C
     2. Aranzana MJ
     3. Lister C
     4. Baxter C
     5. Nicholls C
     6. Nordborg M
     7. Dean C

     (2005) Role of FRIGIDA and FLOWERING LOCUS C in Determining Variation in Flowering Time of Arabidopsis

     Plant physiology 138:1163–1173.

     https://doi.org/10.1104/pp.105.061309

     • Google Scholar
105. 1. Silby MW
     2. Cerdeño-Tárraga AM
     3. Vernikos GS
     4. Giddens SR
     5. Jackson RW
     6. Preston GM
     7. Zhang X-X
     8. Moon CD
     9. Gehrig SM
     10. Godfrey SAC
     11. Knight CG
     12. Malone JG
     13. Robinson Z
     14. Spiers AJ
     15. Harris S
     16. Challis GL
     17. Yaxley AM
     18. Harris D
     19. Seeger K
     20. Murphy L
     21. Rutter S
     22. Squares R
     23. Quail MA
     24. Saunders E
     25. Mavromatis K
     26. Brettin TS
     27. Bentley SD
     28. Hothersall J
     29. Stephens E
     30. Thomas CM
     31. Parkhill J
     32. Levy SB
     33. Rainey PB
     34. Thomson NR

     (2009) Genomic and genetic analyses of diversity and plant interactions of Pseudomonas fluorescens

     Genome Biology 10:R51.

     https://doi.org/10.1186/gb-2009-10-5-r51

     • Google Scholar
106. Book

     1. Simpson GG

     (1949)

     The meaning of evolution: a study of the history of life and of its significance for man

     New Haven: Yale University Press.

     • Google Scholar
107. 1. Spiers AJ
     2. Kahn SG
     3. Bohannon J
     4. Travisano M
     5. Rainey PB

     (2002)

     Adaptive divergence in experimental populations of Pseudomonas fluorescens. I. Genetic and phenotypic bases of wrinkly spreader fitness

     Genetics 161:33–46.

     • PubMed
     • Google Scholar
108. 1. Spiers AJ
     2. Bohannon J
     3. Gehrig SM
     4. Rainey PB

     (2003) Biofilm formation at the air-liquid interface by the Pseudomonas fluorescens SBW25 wrinkly spreader requires an acetylated form of cellulose

     Molecular Microbiology 50:15–27.

     https://doi.org/10.1046/j.1365-2958.2003.03670.x

     • PubMed
     • Google Scholar
109. 1. Steinrueck M
     2. Guet CC

     (2017) Complex chromosomal neighborhood effects determine the adaptive potential of a gene under selection

     eLife 6:e25100.

     https://doi.org/10.7554/eLife.25100

     • PubMed
     • Google Scholar
110. 1. Stern DL
     2. Orgogozo V

     (2009) Is genetic evolution predictable?

     Science 323:746–751.

     https://doi.org/10.1126/science.1158997

     • PubMed
     • Google Scholar
111. 1. Stern DL

     (2013) The genetic causes of convergent evolution

     Nature Reviews Genetics 14:751–764.

     https://doi.org/10.1038/nrg3483

     • PubMed
     • Google Scholar
112. 1. Stoltzfus A
     2. McCandlish DM

     (2017) Mutational Biases Influence Parallel Adaptation

     Molecular Biology and Evolution 34:2163–2172.

     https://doi.org/10.1093/molbev/msx180

     • PubMed
     • Google Scholar
113. 1. Tautz D
     2. Renz M

     (1984) Simple sequences are ubiquitous repetitive components of eukaryotic genomes

     Nucleic Acids Research 12:4127–4138.

     https://doi.org/10.1093/nar/12.10.4127

     • PubMed
     • Google Scholar
114. 1. Tenaillon O
     2. Rodríguez-Verdugo A
     3. Gaut RL
     4. McDonald P
     5. Bennett AF
     6. Long AD
     7. Gaut BS

     (2012) The molecular diversity of adaptive convergence

     Science 335:457–461.

     https://doi.org/10.1126/science.1212986

     • PubMed
     • Google Scholar
115. 1. Ungerer MC
     2. Linder CR
     3. Rieseberg LH

     (2003)

     Effects of genetic background on response to selection in experimental populations of Arabidopsis thaliana

     Genetics 163:277–286.

     • PubMed
     • Google Scholar
116. 1. Viswanathan M
     2. Lacirignola JJ
     3. Hurley RL
     4. Lovett ST

     (2000) A novel mutational hotspot in a natural quasipalindrome in Escherichia coli

     Journal of Molecular Biology 302:553–564.

     https://doi.org/10.1006/jmbi.2000.4088

     • PubMed
     • Google Scholar
117. 1. Wang AD
     2. Sharp NP
     3. Agrawal AF

     (2014) Sensitivity of the distribution of mutational fitness effects to environment, genetic background, and adaptedness: a case study with Drosophila

     Evolution 68:840–853.

     https://doi.org/10.1111/evo.12309

     • PubMed
     • Google Scholar
118. 1. Weinreich DM
     2. Delaney NF
     3. Depristo MA
     4. Hartl DL

     (2006) Darwinian evolution can follow only very few mutational paths to fitter proteins

     Science 312:111–114.

     https://doi.org/10.1126/science.1123539

     • PubMed
     • Google Scholar
119. 1. Woods R
     2. Schneider D
     3. Winkworth CL
     4. Riley MA
     5. Lenski RE

     (2006) Tests of parallel molecular evolution in a long-term experiment with Escherichia coli

     PNAS 103:9107–9112.

     https://doi.org/10.1073/pnas.0602917103

     • PubMed
     • Google Scholar
120. 1. Xu M
     2. Yang X
     3. Yang XA
     4. Zhou L
     5. Liu TZ
     6. Fan Z
     7. Jiang T

     (2016) Structural insights into the regulatory mechanism of the Pseudomonas aeruginosa YfiBNR system

     Protein & Cell 7:403–416.

     https://doi.org/10.1007/s13238-016-0264-7

     • PubMed
     • Google Scholar
121. 1. Yampolsky LY
     2. Stoltzfus A

     (2001) Bias in the introduction of variation as an orienting factor in evolution

     Evolution and Development 3:73–83.

     https://doi.org/10.1046/j.1525-142x.2001.003002073.x

     • PubMed
     • Google Scholar
122. 1. Yates CM
     2. Filippis I
     3. Kelley LA
     4. Sternberg MJ

     (2014) SuSPect: enhanced prediction of single amino acid variant (SAV) phenotype using network features

     Journal of Molecular Biology 426:2692–2701.

     https://doi.org/10.1016/j.jmb.2014.04.026

     • PubMed
     • Google Scholar
123. 1. Zhen Y
     2. Aardema ML
     3. Medina EM
     4. Schumer M
     5. Andolfatto P

     (2012) Parallel molecular evolution in an herbivore community

     Science 337:1634–1637.

     https://doi.org/10.1126/science.1226630

     • PubMed
     • Google Scholar

[Editors’ note: the authors were asked to provide a plan for revisions before the editors issued a final decision. What follows is the editors’ letter requesting such plan.]

Thank you for submitting your article "Disentangling the effects of genetic architecture, mutational bias and selection on evolutionary forecasting" for consideration by eLife. Your article has been reviewed by Patricia Wittkopp as the Senior Editor, a Reviewing Editor, and three reviewers. The following individuals involved in review of your submission have agreed to reveal their identity: James Bull (Reviewer #3).

Overall, we were impressed by the experimental data presented, but had concerns about the predictions and model testing. These concerns included both the clarity of presentation and the substance of the modeling. Different reviewers had different interpretations of some elements of the work because key elements (including Figure 3) were not clearly explained. The mutation model was one element identified as particularly difficult to understand. The lack of statistical tests comparing the predictions to data was also identified as a missing element. Especially in light of the title of the paper; making sure these predictions are made and tested appropriately, is key - we agree that this is where some of the most exciting aspects of the paper lie. Critical points raised during our discussion are provided below. Minimally, we request that you make it clear that you are evaluating two different prediction models and clarify them. Ideally, we'd also like you to make the evolution model explicit and quantitative as well as evaluate each model separately by computing numerical predictions and reporting statistics on the fit with results.

We recognize that the refinement of the modeling work we are requesting may be substantial. Given the list of essential revisions, the editors and reviewers invite you to respond within the next two weeks with an action plan and timetable for the completion of the additional work. We plan to share your responses with the reviewers and then issue a binding recommendation.

This paper reports important experimental work that could easily become a paradigm copied by others. We appreciated the study's span of known biochemical pathways for a phenotype, of documented mutations and mutation counts in those pathways leading to that phenotype, and the attempt to model/quantify the mutation rates in light of some genomic details. The topic is of sufficient interest to publish in eLife.

1) A major strength of the study was that there is an in-depth analysis of different mutational pathways to the WS phenotype. This is a data-intensive part of the paper with lots of detail about gene identity of mutations that were found in WS mutants. However, descriptions of gene functions would be better presented in a table with the in text discussion streamlined more.

2) A much better description of Figure 3 is needed. Not only are the colors not explained, but the meaning of the arrows isn't given either (are they quantitative, qualitative; how do they plug into the model?). I thus cannot easily understand how the figure connects to their models. We ultimately found this information in the supplementary materials, but the reader shouldn't need to look there to understand this key element of the paper. As one reviewer wrote: the manuscript describes Figure 3 as a diagram of a "pathway" or "molecular network" but it does not use ordinary conventions for biochemical or regulatory pathways. It is better described as a graph of mathematical dependencies between reaction rates (red) and gene-product concentrations (blue). Once you know this, you can read out some of the implied reactions like F + E* ==(r6)==> E + F*. In the supplement, see reaction 6, Figure 3—figure supplement 1, "Phosphorylated WspE can phosphorylate WspF". It would be more understandable to just list the reactions, I think.

3) Most importantly, we felt that the modeling elements were presented superficially and were not well-integrated. There are two different prediction models developed and evaluated in the paper: one of mutation rates and one of evolutionary tendencies. Regarding mutational predictability, we have (1) a model for predicting relative mutability for 3 pathways, (2) measured mutation rates for the 3 pathways, and (3) a revised model. But there are no statistics telling us how well the model fits and how much it was improved.

In theoretical population genetics there is a simple "origin-fixation" or SSWM framework for considering the relative chances of different discrete evolutionary changes. Whether or not one wishes to criticize this formalism for being too simplistic, it is the obvious framework for making a prediction when one has measured mutation rates and selection coefficients. It allows precise quantitative statements about the relative contributions of selection and mutational heterogeneity to parallel evolution, as per "What drives parallel evolution?" by Bailey et al., 2017, which uses the theory explained more thoroughly in "Parallel evolution: what does it (not) tell us and why is it (still) interesting?" by Lenormand et al., 2016. The analysis also needs statistical tests, e.g., tests of differences where the authors are making claims about differences.

Regarding the modeling framework, let me explain this concretely. The relevant facts on mutation rate, selection coefficient, and frequency evolved are (subsection “Obtaining an unbiased measure of pathway-specific mutation rates to WS”, Figure 6 and Figure 7):

Wsp: u = 3.7E-9; s ~ x; E = 15

Aws: u = 6.5E-9; s ~ x – 0.2; E = 7

Mws: u = 7.4E-10; s ~ x – 0.05; E = 3

Here x is the selection coefficient of the WspF control used in growth competitions, and I have estimated s by eye from Figure 6.

The authors talk about predictability, but they do not compute a prediction using a model, nor compute the fit between results and predictions. One way to do this is to invoke origin-fixation dynamics, yielding 3 relative predictions of u * Pr_fixation(s), which can be compared to the 3 E values using a chi-squared test (or, better, define expected Wsp and Aws as sums over the 5 and 3 relevant gene targets). Given all the numbers and Kimura's formula for Pr_fixation(s, N), this can be done in a spreadsheet in 15 minutes.

If the authors do not want to commit to the assumptions underlying origin-fixation dynamics, they can just do a multinomial regression on the assumption that E(p) = f(u_p, s_p) where p is the pathway.

In summary, we all agreed that under the origin-fixation framework, having estimates of mutation rates for different phenotypes is important and that these are not easy to obtain in most systems. We are excited about the opportunity to test how well a model based on known information can recapitulate observed mutation rates. It is interesting that it does a relatively poor job unless they incorporate information about mutational hotspots. We think that making this analysis more formal in addition to formally testing your evolutionary predictions using an origin-fixation framework (or similar approach) would both strengthen the paper and be in the spirit of what the title claims the paper does.

[Editors’ note: formal revisions were requested, following approval of the authors’ plan of action.]

Thank you for describing your proposed revisions. This summary has been reviewed and discussed by the reviewers and myself. We all agree that your proposed responses to everything except point 3 is satisfactory. For point 3, the reviewers were disappointed in the response. They all agreed that their recommendation to take advantage of your rich and powerful dataset to model the evolutionary process more formally had been dismissed too quickly.

After considering your response and their comments, it seems to me that you are thinking about "predictability" in the way that David Stern does (i.e., which genes/pathways are most likely to be changed to give rise to a particular phenotypic change given the molecular/developmental functions of all genes in the genome), and they are thinking about predictability in a more population/quantitative genetics sense (i.e., given that you know the relative mutation rates and fitness effects for different genes, you should be able to make predictions using theoretical models about how the trait should evolve, which you can then compare to the observed data). In the reviewers’ opinion, you are focused more on the effects of mutations as they arise and they want to see these combined with other evolutionary forces to make more robust evolutionary predictions.

I do not want to be so heavy handed in the review process as to try force you to make your paper something you didn't intend it to be. However, I also see the potential for modeling with your data that has the reviewers excited. It is rare to have the type of data that you have, and there are well-developed models that could make your data even more impactful.

Ultimately, I will leave it up to you to decide whether or not to go beyond the response you propose and fit a specific evolutionary model and compare it to your observed data with statistical tests. If you choose not to do this, however, it is essential that the title, abstract, and other parts of the paper are modified so that readers do not think this type of analysis is included within the paper.

[Editors’ note: what follows is the authors’ plan to address the revisions.]

Overall, we were impressed by the experimental data presented, but had concerns about the predictions and model testing. These concerns included both the clarity of presentation and the substance of the modeling. Different reviewers had different interpretations of some elements of the work because key elements (including Figure 3) were not clearly explained. The mutation model was one element identified as particularly difficult to understand. The lack of statistical tests comparing the predictions to data was also identified as a missing element. Especially in light of the title of the paper; making sure these predictions are made and tested appropriately, is key – we agree that this is where some of the most exciting aspects of the paper lie. Critical points raised during our discussion are provided below. Minimally, we request that you make it clear that you are evaluating two different prediction models and clarify them. Ideally, we'd also like you to make the evolution model explicit and quantitative as well as evaluate each model separately by computing numerical predictions and reporting statistics on the fit with results.

We recognize that the refinement of the modeling work we are requesting may be substantial. Given the list of essential revisions, the editors and reviewers invite you to respond within the next two weeks with an action plan and timetable for the completion of the additional work. We plan to share your responses with the reviewers and then issue a binding recommendation.

This paper reports important experimental work that could easily become a paradigm copied by others. We appreciated the study's span of known biochemical pathways for a phenotype, of documented mutations and mutation counts in those pathways leading to that phenotype, and the attempt to model/quantify the mutation rates in light of some genomic details. The topic is of sufficient interest to publish in eLife.

1) A major strength of the study was that there is an in-depth analysis of different mutational pathways to the WS phenotype. This is a data-intensive part of the paper with lots of detail about gene identity of mutations that were found in WS mutants. However, descriptions of gene functions would be better presented in a table with the in text discussion streamlined more.

We propose including a table similar to that pasted below into which we move information about the probable molecular effects and the effects of reaction rates on different classes of mutations. This will allow streamlining of the text and improved focus on the motivation for determining the molecular effects on mutations. This is important because it forms the link between targeted genes and reaction rates. It will also allow more attention to be given to the proposed effects of mutations on gene function and reaction rates and their agreement (or not) with the model. However, in removing this information to a table we are concerned that we may leave the reader interested in important detail frustrated. To accommodate this possibility we propose including a supplementary text that complements the table in which we will include relevant detail and where necessary, nuanced interpretations (removed from the current text). This is readily achievable.

2) A much better description of Figure 3 is needed. Not only are the colors not explained, but the meaning of the arrows isn't given either (are they quantitative, qualitative; how do they plug into the model?). I thus cannot easily understand how the figure connects to their models. We ultimately found this information in the supplementary materials, but the reader shouldn't need to look there to understand this key element of the paper. As one reviewer wrote: the manuscript describes Figure 3 as a diagram of a "pathway" or "molecular network" but it does not use ordinary conventions for biochemical or regulatory pathways. It is better described as a graph of mathematical dependencies between reaction rates (red) and gene-product concentrations (blue). Once you know this, you can read out some of the implied reactions like F + E* ==(r6)==> E + F*. In the supplement, see reaction 6, Figure 3—figure supplement 1, "Phosphorylated WspE can phosphorylate WspF". It would be more understandable to just list the reactions, I think.

We agree that a better description of Figure 3 is required. In response to your point 3 (below) we propose to include null models for each of the pathways. These will allow a better introduction / motivation and description of our mathematical model that will lead to a revision of Figure 3. We will take care to ensure that the caption serves as a complete description of the figure and its significance. This will require one month.

3) Most importantly, we felt that the modeling elements were presented superficially and were not well-integrated. There are two different prediction models developed and evaluated in the paper: one of mutation rates and one of evolutionary tendencies. Regarding mutational predictability, we have (1) a model for predicting relative mutability for 3 pathways, (2) measured mutation rates for the 3 pathways, and (3) a revised model. But there are no statistics telling us how well the model fits and how much it was improved.

In theoretical population genetics there is a simple "origin-fixation" or SSWM framework for considering the relative chances of different discrete evolutionary changes. Whether or not one wishes to criticize this formalism for being too simplistic, it is the obvious framework for making a prediction when one has measured mutation rates and selection coefficients. It allows precise quantitative statements about the relative contributions of selection and mutational heterogeneity to parallel evolution, as per "What drives parallel evolution?" by Bailey et al., 2017, which uses the theory explained more thoroughly in "Parallel evolution: what does it (not) tell us and why is it (still) interesting?" by Lenormand et al., 2016. The analysis also needs statistical tests, e.g., tests of differences where the authors are making claims about differences.

We have given this particular attention. It seems, especially from the suggestion that we borrow from origin-fixation models, there is some misunderstanding of what we have done: our motivation, aims and overall long-term aspiration. Clearly, we need to improve.

Ultimately – although perhaps not for some time – we think that it will be possible to take a given genome sequence and, armed with knowledge of the genotype-to-phenotype map underpinning a particular phenotype, combined with knowledge of localised mutational biases, predict the genetic bases of phenotypic evolution in the absence of additional data. Our work is a small step in this direction and perhaps its most important contribution is to outline a way forward and uniquely (because of our experimental system) assess its validity in terms of what is currently known, what is not known, and what needs (in the future) to be known.

Importantly, our approach aids understanding of why evolution follows a particular route. Indeed in our previous work (McDonald et al.,2009; Lind et al.,2015) we have shown the genotype-to-phenotype map to be a significant contributor to parallel evolution (it’s not all about selection and mutation rate). Our work thus contributes to development of a body of knowledge that deepens understanding of the mechanisms of evolution.

Our approach is readily extendable to other organisms that harbour Wsp, Aws and Mws pathways. Precisely this transferability is demonstrated in Lind 2018 (https://www.biorxiv.org/content/early/2018/06/08/342261). By extension to other systems, this stands to feedback to affect the value and predictive power of bottom-up approaches in the future. Further, our approach is amenable to the study of future evolutionary change, for example, that involving reverse evolution of the WS phenotype as evident in studies of lifecycle evolution (Hammerschmidt et al., 2014). In contrast, approaches involving measurement of mutation rate and selection coefficient will be — in absence of mechanistic understanding — specific to each particular instance of adaptive evolution.

We appreciate that origin-fixation models can be useful in predicting the rise and fixation of new mutants. As you and the referees point out such models require estimation of (1) the probability that a mutant will arise and (2) the selection coefficient of that mutant. These parameters, particularly the first, are difficult to predict a priori and usually require fitting to experimental data or an assumption about the probability distribution of mutations. The framework introduced in our paper is different, in that it attempts to estimate the probability that a particular mutant (Wsp, Aws, Mws) will arise in the absence of experimental data (with only knowledge of the genes in a pathway and how they interact to produce a particular phenotype). While we compare predictions arising from our models to experimental data, we do not use these data in our predictions.

Initially we make incorrect predictions based solely on knowledge of the genotype-to-phenotype map (although we are not out by much) but show that we are able to refine predictions through incorporation of information concerning mutational hotspots. The capacity to weigh the influences of gene interactions with gene mutation rates avoids lumping these factors into a single term as is the norm for origin-fixation models. In the revised version of this paper, we propose to spell out these differences more clearly.

In addition to the comments above we believe that using an origin-fixation framework is not appropriate here. Although the selection coefficients could of course be used in combination with mutation rates to predict the probability of fixation this is not what we are trying to do, moreover to do so would be to mislead the reader. There are three reasons. Firstly, the fitness of WS is frequency-dependent and thus a mutant’s selection coefficient is very different at the time it originates compared to when it is established at the air-liquid interface (Rainey and Travisano,1998; Lind et al.,2015). Secondly, the selection coefficient is also dependent on the frequency of other WS types and other phenotypes selected at the air-liquid interface (Lind et al., 2017, Ferguson et al.,2013). Thirdly, fitness in our study is measured against a wspF (WS) mutant and not the wild type ancestor, which is needed for use in origin-fixation models. In our experiment, it is not the case that fitness can be estimated by using s=x-0.2 for aws as suggested in the comments.

The way forward: we propose as a vehicle for improvement, inclusion of null models for each of the pathways in which we firstly ignore genetic architecture. This means the first line of prediction – as to which pathway evolution will follow – will be based solely on target size. Preliminary analyses show that these null models are hopelessly inaccurate. Next, as we better develop the argument for the importance of genetic architecture, we propose a set of three null models that incorporate genetic architecture with the same number of components and interactions but with randomly assigned architectures. Comparison between these arbitrary models and those informed by current knowledge can then be made – and will be statistically appraised. Together these comparisons will demonstrate the value of incorporating valid architectural information. Finally, as we move to introduce known mutational bias into our models (based on our experimental data) we will compare these to null models where the hotspot is placed randomly in the pathway. Again, we will subject the comparisons to statistical testing. The proposed new analyses will lead to clarifications in the Introduction and Discussion concerning our objectives and the differences with current approaches and particularly the difference between our strategy and origin-fixation models.

This analysis will be completed within two months and will offer many points of comparison and demonstrate the value of the predictive power of our approach.

Additionally, we propose: (1) to perform a statistical analysis of differences between all mutation rates (Figure 2) and fitness (for example wspA and wspR vs. the control in Figure 6). The number of replicates will be added to all figures. (2) Provide a better motivation for our measurements of fitness: an explanation of why mutations in certain genes have not previously been found (wspA and wspE) and why Aws is is more rarely used than Wsp despite a higher mutation rate to WS. (3) Add information concerning controls for fitness assays in order to account for effects of the kanamycin marker and double deletions and clarify that the number of mutants per independent culture for the mutation rate assay is <0.5 so that differences in fitness between different WS mutants thus making the point that these have minimal influence on results. These modifications will be completed within a month.

Regarding the modeling framework, let me explain this concretely. The relevant facts on mutation rate, selection coefficient, and frequency evolved are (subsection “Obtaining an unbiased measure of pathway-specific mutation rates to WS”, Figure 6 and Figure 7):

Wsp: u = 3.7E-9; s ~ x; E = 15

Aws: u = 6.5E-9; s ~ x – 0.2; E = 7

Mws: u = 7.4E-10; s ~ x – 0.05; E = 3

Here x is the selection coefficient of the WspF control used in growth competitions, and I have estimated s by eye from Figure 6.

The authors talk about predictability, but they do not compute a prediction using a model, nor compute the fit between results and predictions. One way to do this is to invoke origin-fixation dynamics, yielding 3 relative predictions of u * Pr_fixation(s), which can be compared to the 3 E values using a chi-squared test (or, better, define expected Wsp and Aws as sums over the 5 and 3 relevant gene targets). Given all the numbers and Kimura's formula for Pr_fixation(s, N), this can be done in a spreadsheet in 15 minutes.

If the authors do not want to commit to the assumptions underlying origin-fixation dynamics, they can just do a multinomial regression on the assumption that E(p) = f(u_p, s_p) where p is the pathway.

In summary, we all agreed that under the origin-fixation framework, having estimates of mutation rates for different phenotypes is important and that these are not easy to obtain in most systems. We are excited about the opportunity to test how well a model based on known information can recapitulate observed mutation rates. It is interesting that it does a relatively poor job unless they incorporate information about mutational hotspots. We think that making this analysis more formal in addition to formally testing your evolutionary predictions using an origin-fixation framework (or similar approach) would both strengthen the paper and be in the spirit of what the title claims the paper does.

[Editors’ notes: the authors’ response after being formally invited to submit a revised submission follows.]

Thank you for describing your proposed revisions. This summary has been reviewed and discussed by the reviewers and myself. We all agree that your proposed responses to everything except point 3 is satisfactory. For point 3, the reviewers were disappointed in the response. They all agreed that their recommendation to take advantage of your rich and powerful dataset to model the evolutionary process more formally had been dismissed too quickly.

After considering your response and their comments, it seems to me that you are thinking about "predictability" in the way that David Stern does (i.e., which genes/pathways are most likely to be changed to give rise to a particular phenotypic change given the molecular/developmental functions of all genes in the genome), and they are thinking about predictability in a more population/quantitative genetics sense (i.e., given that you know the relative mutation rates and fitness effects for different genes, you should be able to make predictions using theoretical models about how the trait should evolve, which you can then compare to the observed data). In the reviewers’ opinion, you are focused more on the effects of mutations as they arise and they want to see these combined with other evolutionary forces to make more robust evolutionary predictions.

I do not want to be so heavy handed in the review process as to try force you to make your paper something you didn't intend it to be. However, I also see the potential for modeling with your data that has the reviewers excited. It is rare to have the type of data that you have, and there are well-developed models that could make your data even more impactful.

Ultimately, I will leave it up to you to decide whether or not to go beyond the response you propose and fit a specific evolutionary model and compare it to your observed data with statistical tests. If you choose not to do this, however, it is essential that the title, abstract, and other parts of the paper are modified so that readers do not think this type of analysis is included within the paper.

We have changed the title to “Predicting mutational routes to new adaptive phenotypes”. This we have done after realizing that our previous title gave the unintended message that we were combining mutation and selection into our model of evolutionary forecasting. This, as the referees have correctly understood, is not what we have done. What we have done (by virtue of being able to obtain a set of mutants with and without selection) is to show that there really is a need to link mutations to their fitness effects. We have made sure that this message is consistently made in the Abstract and elsewhere. In the revised Discussion section we emphasize that construction of models that link mutations to their fitness effects is the yet to be achieved Holy Grail of forecasting.

We have added new material to the Introduction to acknowledge origin-fixation models. We point out briefly their limitations, but importantly, we make clear that what we aspire to is quite different. Our hope is that at some future time it will be possible to read a genome sequence and be able to forecast future evolutionary change based on mechanistic understanding of rules underpinning expression of the phenotype from genotype, knowledge of mutational biases and understanding of the connection between mutations and their fitness effects, just as today we use the genome to interrogate various databases to infer much about gene function and regulation.

We have removed from the text the detailed descriptions of the likely effect of mutations on reaction rates of each pathway – leaving just a short summary – and placed this in Table 1.

We have substantially revised the modeling section as per our proposed plan. Specifically, in order to show the thinking behind our modeling approach we have included three new models, that begin with a simple (and intentionally naïve) expression of the relationship between genotype and the probability of generating WS phenotypes and conclude with a comprehensive model that expresses the relationship between genotype and WS phenotype based on interaction rates among pathway components (the model previously presented). We describe this as a graph of mathematical dependencies and provide full details of the models in terms of functional descriptions and system of differential equations in the child figures (Figure 4—figure supplement 1, Figure 4—figure supplement 2 and Figure 4—figure supplement 3). We have also clarified the use of multiple changes to reaction rates by single mutations and modified the model to disregard multiple changes that are likely to require several mutations. This has led to minor changes in the ratios of the different pathways and in also in the Figure 4 (now Figure 5), but they do not change our conclusions.

We have added a new Figure 3 showing the predictions of the different null models compared to the experimental data.

The complex graph depicting interactions among pathway components (previously Figure 3), is now Figure 4 and we have taken care to explain its significance fully both in the text and in the caption.

A new Figure 7 has been added that presents an analysis of the results of the different models the experimentally supported mutational hotspot and the predicted relationship between mutation and likelihood of generating WS via each of the three focal pathways. For the two comprehensive models we include the range of predicted states arising from extremes of all enabling and all disabling reactions allowing the reader to assess the full range of predictions and not just values from a single hotspot as previously. This was previously Figure 4D,E and part of the Figure 4—figure supplement 2B.

We have reorganized the end of the Results section to make it clear to the reader that we do not attempt to predict the outcome of experimental evolution using mutation rates and selection coefficients. Instead we now first present the differences in mutational spectrum with and without selection. We have provided a better motivation for the fitness assays, which is to provide an explanation for these differences. We show that there are significant differences in fitness between different mutants and that the direction of these fitness differences explains the underrepresentation of certain types of mutations in the experimental evolution data thus highlighting the importance of incorporating selection in forecasting evolution. We deal with the origin-fixation framework in the Introduction (see above) touching on various assumptions, in particular the need for interactions between mutants to show no frequency dependence. As our decades long experience with the WS system has shown that it is replete with frequency dependent interactions we make no further mention in the Results section. To attempt to place our data in context of origin-fixation models would constitute misuse of the fitness data.

We have included statistical tests for all claims of differences as well as number of replicates. A table of key resources has been added to the Results section. Information about controls for fitness assays have been added to the methods section and we have clarified that the number of mutants per independent culture is so low that differences in fitness between different WS mutants will have minimal influence on results.

Finally, we have substantially revised the Discussion section with much clearer focus on the issues surrounding mechanistic (bottom-up) evolutionary forecasting.

Author details

1. Peter A Lind

   1. New Zealand Institute for Advanced Study, Massey University at Albany, Auckland, New Zealand
   2. Department of Molecular Biology, Umeå University, Umeå, Sweden

   Contribution

   Conceptualization, Data curation, Investigation, Visualization, Methodology, Writing—original draft, Writing—review and editing

   For correspondence

   peter.lind@umu.se

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0003-1510-8324

2. Eric Libby

   1. New Zealand Institute for Advanced Study, Massey University at Albany, Auckland, New Zealand
   2. Santa Fe Institute, New Mexico, United States
   3. Department of Mathematics, Umeå University, Umeå, Sweden

   Contribution

   Conceptualization, Software, Formal analysis, Investigation, Visualization, Methodology, Writing—original draft, Writing—review and editing

   Competing interests

   No competing interests declared

3. Jenny Herzog

   New Zealand Institute for Advanced Study, Massey University at Albany, Auckland, New Zealand

   Contribution

   Data curation, Investigation, Methodology

   Competing interests

   No competing interests declared

4. Paul B Rainey

   1. New Zealand Institute for Advanced Study, Massey University at Albany, Auckland, New Zealand
   2. Department of Microbial Population Biology, Max Planck Institute for Evolutionary Biology, Plön, Germany
   3. Ecole Supérieure de Physique et de Chimie Industrielles de la Ville de Paris, ESPCI Paris-Tech, CNRS UMR 8231, PSL Research University, Paris, France

   Contribution

   Conceptualization, Supervision, Funding acquisition, Validation, Visualization, Methodology, Writing—original draft, Project administration, Writing—review and editing

   Competing interests

   Reviewing editor, eLife

   "This ORCID iD identifies the author of this article:" 0000-0003-0879-5795

• 3,451

  Page views

• 413

  Downloads

• 35

  Citations

Article citation count generated by polling the highest count across the following sources: Scopus, Crossref, PubMed Central.