VIRAL EVOLUTION

Destabilizing mutations encodenongenetic variation that drivesevolutionary innovationKatherine L. Petrie,1,2* Nathan D. Palmer,1 Daniel T. Johnson,1 Sarah J. Medina,1

Stephanie J. Yan,1 Victor Li,1 Alita R. Burmeister,3 Justin R. Meyer1*

Evolutionary innovations are often achieved by repurposing existing genes to perform newfunctions; however, the mechanisms enabling the transition from old to new remaincontroversial.We identified mutations in bacteriophage l’s host-recognition gene J that conferenhanced adsorption to l’s native receptor, LamB, and the ability to access a new receptor,OmpF.The mutations destabilize l particles and cause conformational bistability of J, whichyields progeny of multiple phenotypic forms, each proficient at different receptors.This workprovides an example of how nongenetic protein variation can catalyze an evolutionaryinnovation.We propose that cases where a single genotype can manifest as multiplephenotypesmay bemore common than previously expected and offer a general mechanism forevolutionary innovation.

The tree of life is punctuated with numerousbiological innovations, but explaining theorigin of novelty has been problematic.Natural selection, the predominant theoryfor adaptation, is unable to explain howmu-

tations encode new function (1, 2). This dilemma isepitomized by the progress made in understand-ing the evolution of themodel virus bacteriophagel. Under typical laboratory conditions, l evolvesto exploit a new receptor (OmpF) (3). This inno-

vation is achieved through four mutations inits host-recognition gene, J (3). The mutationsevolve by natural selection because they acceler-ate adsorption to the original receptor (LamB)(4) and later are coopted for OmpF function (3).Remaining to be explained is how J mutationsimprove LamB adsorption and open access toOmpF. Here we report studies of the molecularmechanisms of Jmutations and offer an answerto the question of how novelty arises.

Protein functional evolution is typically thoughtto occur through gene duplication and divergence(5). This model cannot explain l’s evolution be-cause the J gene did not duplicate (3). Instead, levolved a promiscuous J protein capable of usingLamB and OmpF receptors, which then divergedinto two receptor specialists (6). The leading ex-planation for the evolution of promiscuity isthroughprotein-destabilizingmutations that causestructural disorder, allowing proteins to occupymultiple conformations (7–10). Once a promis-cuous protein is evolved, additional evolutionaryrefinement of specific function leads to special-ization and restabilization (8, 9, 11) (Fig. 1A).We tested this hypothesis by measuring the

thermostability of five l genotypes that onlyvaried in the number of Jmutations. Each strainwas genetically engineered with mutations thathave been observed to evolve on the path toOmpF+

(3). Thermostability was assessed by measuringthe temperature dependence of phage survivalafter 1 hour of exposure to multiple temperatures,ranging from l’s natural 37°C to a destabilizing55°C. The number of mutations and the degreeof sensitivity were correlated [Spearman’s r =0.90, P = 0.0167; Fig. 1B and fig. S1 (12)]. Onecaveat is that nonsynonymous mutations tend

RESEARCH

Petrie et al., Science 359, 1542–1545 (2018) 30 March 2018 1 of 4

1Division of Biological Sciences, University of California (UC)San Diego, La Jolla, CA 92093, USA. 2Earth-Life ScienceInstitute (ELSI), Tokyo Institute of Technology, Tokyo, Japan.3Department of Ecology and Evolution, Yale University, NewHaven, CT 06520, USA.*Corresponding author. Email: kpetrie@ucsd.edu (K.L.P.);jrmeyer@ucsd.edu (J.R.M.)

Fig. 1. Evolutionary innovation through functionalbistability. (A) Schematic overview of l evolution.Mutations to improve ancestral receptor bindingreduce stability until a critical mutation produces amultifunctional genotype. Further evolution canrestabilize the protein. (B) Temperature sensitivityof genotypes with indicated numbers of mutationsrelative to the ancestor. Some have more thanfour mutations because l often evolves additionalmutations before achieving OmpF use. The y axisshows the fraction of infectious particles remainingafter incubation. Curves are logistic models fit todata from multiple trials; fig. S1 shows all data (12).OmpF+ phages are shown in teal, and LamB-reliantgenotypes are shown in black. (C) Temperaturesensitivity of the ancestor, an OmpF+ generalist, andthe generalist’s specialist descendants. Bars indicatethe data range; fig. S2 shows all data.

40 45 50 55

0.0

0.2

0.4

0.6

0.8

1.0

temperature (°C)

frac

tion

rem

aini

ng

0

3

4

5

7

ancestor

multifunctional intermediate(generalist)

specialist binds LamB binds OmpF

stab

ility

sequence space

mor

e st

able

le

ss s

tabl

e

temperature (°C)

frac

tion

rem

aini

ng

0.00001

0.0001

0.001

0.01

0.1

1

10

40 45 50 55

ancestor

OmpF+ generalist

OmpF specialists

LamB specialists

on October 31, 2020

http://science.sciencem

ag.org/D

ownloaded from

to destabilize proteins (11), so this relationshipis also expected under a null model of proteinmutagenesis.We testedwhether additional Jmutations that

cause specialization restabilize the protein. Wemeasured the thermostability of the ancestral land an OmpF+ generalist variant, as well as 10 ofthe generalist’s descendants that evolved to spe-cialize on either LamB or OmpF (table S1) (6).The null expectation is that all specialists shouldlose stability because each evolved additional Jmutations; however, none did. In fact, five out offive LamB specialists and three out of five OmpFspecialists regained ancestral stability (fig. S2).Onaverage, the specialistswere significantlymorestable than the OmpF+ generalist [Fisher’s meth-od, Χ2(4) = 22.6, P = 0.0001] and indistinguish-able from the ancestor [Fisher’s method, Χ2(4) =5.56, P = 0.234] [Fig. 1C; comparisons made at51.4°C (12)].An additional investigation of l particle decay

yielded a surprising discovery. We studied thetemporal dynamics of the most and least stablestrains observed (ancestor and seven mutations;Fig. 1B) decaying at 37°C for 18 days (Fig. 2A).The seven-mutation OmpF+ generalist phage de-cayed at a faster rate, but, unexpectedly, bothgenotypes decayed at decreased rates over time.If phage stocks contained just a single phagephenotype, then all particles should share a com-mon half-life. Instead, a model of fast-decayingand slow-decaying subpopulations fits the databetter than a model of single exponential decay[fig. S3 and table S3 (12)]. Thus, despite origi-nating from an isogenic stock, l particles ex-hibit phenotypic heterogeneity: Some are stableand some are unstable. In line with this obser-vation was the previous discovery that geneti-cally identical l particles can have different ratesof LamB adsorption as well (13).Destabilized promiscuous intermediates may

facilitate innovation by permitting proteins toflexibly accommodate different ligands (11, 14),or they may produce polypeptide chains that aresensitive to contingent folding processes thatresult in multiple isoforms with different capaci-ties (9, 15). Sometimes, the energy barrier between

these isoforms may be high enough to preventtheir refolding, making the protein appear con-formationally bistable (9, 16). The repeated ob-servations of nongenetic phenotypic heterogeneitysuggest that this mechanism may be acting dur-ing l evolution.To test whether the heterogeneity in decay was

related to the mutations spurring the innovation,we reexamined decay dynamics (12). First, westudied whether the accelerated OmpF+ decayis due to the emergence of a new, fast-decayingphenotypic class. We found that the decay rateof this strain was significantly faster during thefirst day of incubation than the second (t test, t8 =–14.2, P < 0.000001); there was no significantchange in the ancestral l decay rate over 2 days(t8 = –1.35, P = 0.214) (Fig. 2B). This pattern isconsistent with a new fast-decaying subgroupthat is mostly lost within the first day. Next, wepredicted that the newly evolvedunstable subgroupwas composed of OmpF+ particles. We repeatedthe decay experiment with theOmpF+ strain, butwe quantified the number of OmpF-binding andLamB-binding particles by plating subsampleson hosts that express only OmpF (lamB–) or onlyLamB (ompF–) (12). The number of particles thatproduced infections on OmpF lawns declinedsignificantly faster than the number on LamBlawns (t test, t3 = –7.31, P = 0.00528; Fig. 2C),indicating that among genetically identical l par-ticles, we observed unstable OmpF+ and stableLamB-reliant particles. We confirmed that the in-creased frequency of LamB-reliant particles wasdue to nongenetic heterogeneity and not geneticchanges by regrowing LamB-reliant phages andshowing that nearly all cultures regained functiononOmpF [fig. S4 (12)].We also confirmed that theinstability was not due to proteinase sensitivity[fig. S5 (12)].Further confirmation that only a fraction of

particles can exploit OmpF was gained throughan OmpF+ pull-down experiment in which lsthat bind OmpF were selectively removed (12). Ifall l particles are equivalent, then this assaywould not affect the frequency of remainingparticles that exploit LamB or OmpF. However,if there is heterogeneity, then we would expect

a decline in the ratio of OmpF+ to LamB+ par-ticles. OmpF pull-downs preferentially depletedls that could form plaques on the OmpF lawns(t test, t2 = –5.15,P = 0.0357; Fig. 3A). As a control,we confirmed that this depletion by binding wasgreater than the decay observed when the incu-bation was done with no complementary recep-tors (t test, t3 = –4.14, P = 0.0255; Fig. 3B).Taken together, our results indicate that J

mutations that move l toward innovation de-stabilize the phage and allow it to access mul-tiple phenotypes. We reason that this observedheterogeneity originates from the stochastic fold-ing of individual J proteins, because Jmutationsalone are sufficient to alter the heterogeneity.However, because we studied the whole phage,we are unable to confirm whether the heteroge-neity is entirely explained by J protein variationor whether it stems from an interaction betweenJ and the rest of the particle.Next, we studied how natural selection pro-

motes the evolution of nongenetic phenotypicheterogeneity. We hypothesized that the OmpF+

subpopulation is a by-product of selection forimproved LamB adsorption and a result of pro-tein misfolding. To test this, we compared theadsorption rates of an OmpF+ genotype and itsOmpF–progenitor,which is onemutation removedfromOmpF function (3). As expected, even the lastmutation required for the innovation improvedbinding to the original receptor (t test, t4 = 11.7, P =0.0003; Fig. 3C). Next, we predicted that the LamB-reliant fraction of OmpF+ l particles should have afaster adsorption rate on LamB than themisfoldedOmpF+ particles. We measured the adsorption toLamB of the entire population versus that of thelamB-reliant subpopulation by removing OmpF+

particles with the OmpF pull-down technique. Thepull-down technique enriched the unbound frac-tion for particles that were 13.4% faster atadsorbing to LamB (t test, t8 = 2.77, P = 0.0244;Fig. 3D), indicating that there is a cost forproducing the subpopulation of OmpF+ particles.This demonstrates that a single gene can evolveto improve two functions simultaneously byexpressing multiple forms, even if the functionsare antagonistic.

Petrie et al., Science 359, 1542–1545 (2018) 30 March 2018 2 of 4

pfu/

mL

time (days)time (days)

pfu/

mL

0.0001

0.001

0.01

0.1

1

0 6 12 18

frac

tion

rem

aini

ng

time (days)

5

50

500

5000

50000

500000

0 1 2500

5000

50000

500000

5000000

50000000

0 1 2

LamB-only (ompF−) host

OmpF-only (lamB−) host

p = 0.00528(on rates)

Fig. 2. Decay rate heterogeneity. Decay of ancestor (black) or OmpF+

generalist (teal) phage over (A) 18 days or (B) 2 days. Four replicates (pale)and the average (bold) are shown in (A). Individual replicates are shown in (B).

(C) Number of phage particles able to form plaques on LamB-only (solid)or OmpF-only (dashed) bacterial lawns after 2 days. Individual replicates areshown. pfu, plaque forming units.

RESEARCH | REPORTon O

ctober 31, 2020

http://science.sciencemag.org/

Dow

nloaded from

Nongenetic phenotypic heterogeneity may ariseby chance, or it could evolve by natural selection ifa distribution of phenotypes confers higher fitnessthan a single type. We tested this by mathemat-ical evaluations of the fitness of generalist ls thatonly vary in the width of their adsorption ratedistributions. We assumed that the l particles’adsorption rates to LamB and OmpF are nor-mally distributed but constrained by a trade-offcurve (6) and adsorption rates ≥0.We found thatsmall variances resulted in substantial loss ofmean population fitness; however, at large var-iances, this cost was recouped, and fitness ex-ceeded that at zero variance [Fig. 4 (12)].We show that the OmpF innovation evolved

by violating the “one sequence, one structure”dogma (17) with a DNA sequence that encodes

multiple protein forms. This allowed the virus tosimultaneously optimize binding to its originalreceptor and experiment with new functions.Nongenetic heterogeneity in gene expression lev-els is predicted to facilitate evolutionary novelty(18); here, we extend this hypothesis to the ex-pressed molecules themselves and demonstratethe role of nongenetic heterogeneity in action.Ourmathematical model establishes circumstanceswhere nongenetic heterogeneity is adaptive, andthus this mechanism is more likely than ifheterogeneity evolved by chance alone.

REFERENCES AND NOTES

1. S. Gavrilets, in Evolution: The Extended Synthesis, M. Pigliucci,G. B. Müller, Eds. (MIT Press, 2010), pp. 45–79.

2. H. DeVries, Species and Varieties, Their Origin by Mutation(Open Court Publishing Company, 1904).

3. J. R. Meyer et al., Science 335, 428–432 (2012).4. A. R. Burmeister, R. E. Lenski, J. R. Meyer, Proc. Biol. Sci. R.

283, 20161528 (2016).5. S. Ohno, Evolution by Gene Duplication (Springer, 1970).6. J. R. Meyer et al., Science 354, 1301–1304 (2016).7. N. Tokuriki, D. S. Tawfik, Science 324, 203–207 (2009).8. M. Münz, J. Hein, P. C. Biggin, PLOS Comput. Biol. 8, e1002749

(2012).9. T. Sikosek, H. S. Chan, J. R. Soc. Interface 11, 20140419 (2014).10. E. Dellus-Gur et al., J. Mol. Biol. 427, 2396–2409 (2015).11. N. Tokuriki, D. S. Tawfik, Curr. Opin. Struct. Biol. 19, 596–604

(2009).12. Supplementary materials.13. R. Gallet, T. Lenormand, I.-N. Wang, Evolution 66, 3485–3494 (2012).14. T. Zou, V. A. Risso, J. A. Gavira, J. M. Sanchez-Ruiz,

S. B. Ozkan, Mol. Biol. Evol. 32, 132–143 (2015).15. L. C. James, D. S. Tawfik, Trends Biochem. Sci. 28, 361–368

(2003).16. J. F. Sinclair, M. M. Ziegler, T. O. Baldwin, Nat. Struct. Biol. 1,

320–326 (1994).

Petrie et al., Science 359, 1542–1545 (2018) 30 March 2018 3 of 4

time incubated (minutes)

2x10-09

3x10-09

4x10-09

5x10-09

adso

rptio

n ra

te to

Lam

B(m

l/h)

OmpF−

genotype(all particles)

OmpF+

genotype(all particles)

OmpF+

genotype(all particles)

OmpF+

genotype(after removal

of particles that bind OmpF)

no-cellcontrol

p = 0.000301p = 0.0244

p = 0.0357

p = 0.333

LamB-only lawn OmpF-only lawn

50

500

5000

50000

500000

5000000

0 90 0 90

incubated with

OmpF-only cells

pfu/

ml

Fig. 3. Isolation and properties of l subpopulations. A phage pull-downassay shows the number of phage particles capable of infecting LamB-only(solid) or OmpF-only (dashed) hosts before and after incubation (A) in thepresenceofOmpF-onlycells or (B) alone (in the absenceof cells). (C) Adsorption

rate to LamB-only cells before or after OmpF innovation; genotypes differ byone mutation. (D) LamB adsorption rate of the entire OmpF+ culturecompared with that of the LamB-faithful subpopulation. P values are frompaired t tests [(A) and (B)] or standard t tests [(C) and (D)]. h, hours.

adsorption rate variance x 10-18 ml/h

rela

tive

mea

n fit

ness

prob

abili

ty d

ensi

ty

adsorption rate to OmpF (pl/h)adsorption rate to LamB (pl/h)

pareto front

1.0

0.5

0.0

0.5

0 10 20 30

Fig. 4. The fitness of ensembles of phenotypes. (A) Populationfitness as a function of phenotypic variance. (B) Example of bivariatenormal distribution of phenotypes constrained by a convex Paretofront (19). The variance of the colored points in (A) is the parameter

used to calculate corresponding colored outlines in (B). The outlinesrepresent the largest variance at which 50% of the population iscontained. Error bars in (A) represent numerical error in estimatingthe integrals.

RESEARCH | REPORTon O

ctober 31, 2020

http://science.sciencemag.org/

Dow

nloaded from

17. S. C. Goodchild, P. M. G. Curmi, L. J. Brown, Biophys. Rev. 3,143–153 (2011).

18. A. Eldar, M. B. Elowitz, Nature 467, 167–173 (2010).19. O. Shoval et al., Science 336, 1157–1160 (2012).

ACKNOWLEDGMENTS

We thank A. Gupta and A. Tripathi for help in the laboratory.Funding: K.L.P. is supported by the ELSI Origins Network, fundedby the John Templeton Foundation. A.R.B. was supported by the

U.S. NSF (award DGE-1424871). The ideas expressed herein arenot necessarily those of the funders. Author contributions:K.L.P. and J.R.M. conceived the project, N.D.P. developedmathematical models, all authors ran experiments, and J.R.M.and K.L.P. wrote the manuscript. Competing interests: Theauthors have none. Data and materials availability: Data aredeposited at https://doi.org/10.5061/dryad.fj852. Biologicalmaterial is available from J.R.M. under a material transferagreement with UC San Diego (http://blink.ucsd.edu/research/conducting-research/mta/index.html).

SUPPLEMENTARY MATERIALS

www.sciencemag.org/content/359/6383/1542/suppl/DC1Materials and MethodsSupplementary TextFigs. S1 to S6Tables S1 to S6References (20–37)

12 October 2017; accepted 8 February 201810.1126/science.aar1954

Petrie et al., Science 359, 1542–1545 (2018) 30 March 2018 4 of 4

RESEARCH | REPORTon O

ctober 31, 2020

http://science.sciencemag.org/

Dow

nloaded from

Destabilizing mutations encode nongenetic variation that drives evolutionary innovation

and Justin R. MeyerKatherine L. Petrie, Nathan D. Palmer, Daniel T. Johnson, Sarah J. Medina, Stephanie J. Yan, Victor Li, Alita R. Burmeister

DOI: 10.1126/science.aar1954 (6383), 1542-1545.359Science 

, this issue p. 1542Sciencenongenetic heterogeneity to connect phenotypic heterogeneity, evolvability, and protein stability.can take on two shapes, this genotype can have two phenotypes. Natural selection may thus be able to act on this

's ability to exploit an additional host receptor for infection. Because the proteinλtheir genetic sequences, contributed to show how the evolution of variable folding conformations of isogenic proteins, which do not differ inet al.manner. Petrie

is a virus that infects bacteria by exploiting various membrane proteins in a well-characterizedλBacteriophage Nongenetic variation drives viral evolution

ARTICLE TOOLS http://science.sciencemag.org/content/359/6383/1542

MATERIALSSUPPLEMENTARY http://science.sciencemag.org/content/suppl/2018/03/28/359.6383.1542.DC1

REFERENCES

http://science.sciencemag.org/content/359/6383/1542#BIBLThis article cites 25 articles, 8 of which you can access for free

PERMISSIONS http://www.sciencemag.org/help/reprints-and-permissions

Terms of ServiceUse of this article is subject to the

is a registered trademark of AAAS.ScienceScience, 1200 New York Avenue NW, Washington, DC 20005. The title (print ISSN 0036-8075; online ISSN 1095-9203) is published by the American Association for the Advancement ofScience

Science. No claim to original U.S. Government WorksCopyright © 2018 The Authors, some rights reserved; exclusive licensee American Association for the Advancement of

on October 31, 2020

http://science.sciencem

ag.org/D

ownloaded from