ARTICLE
Received 3 Jun 2016
| Accepted 30 Aug 2016 | Published 17 Oct 2016
In vivo continuous evolution of genes
and pathways in yeast
Nathan Crook
1,
*, Joseph Abatemarco
1,
*, Jie Sun
1,
*, James M. Wagner
1
, Alexander Schmitz
1
& Hal S. Alper
1,2
Directed evolution remains a powerful, highly generalizable approach for improving the
performance of biological systems. However, implementations in eukaryotes rely either on
in vitro diversity generation or limited mutational capacities. Here we synthetically optimize
the retrotransposon Ty1 to enable in vivo generation of mutant libraries up to 1.6 10
7
l
1
per
round, which is the highest of any in vivo mutational generation approach in yeast.
We demonstrate this approach by using in vivo-generated libraries to evolve single enzymes,
global transcriptional regulators and multi-gene pathways. When coupled to growth
selection, this approach enables in vivo continuous evolution (ICE) of genes and pathways.
Through a head-to-head comparison, we find that ICE libraries yield higher-performing
variants faster than error-prone PCR-derived libraries. Finally, we demonstrate transferability
of ICE to divergent yeasts, including Kluyveromyces lactis and alternative S. cerevisiae strains.
Collectively, this work establishes a generic platform for rapid eukaryotic-directed evolution
across an array of target cargo.
DOI: 10.1038/ncomms13051
OPEN
1
Department of Chemical Engineering, The University of Texas at Austin, 200 East Dean Keeton Street, Stop C0400, Austin, Texas 78712, USA.
2
Institute for
Cellular and Molecular Biology, The University of Texas at Austin, 2500 Speedway Avenue, Austin, Texas 78712, USA. * These authors contributed equally to
this work. Correspondence and requests for materials should be addressed to H.S.A. (email: halper@che.utexas.edu).
NATURE COMMUNICATIONS | 7:13051 | DOI: 10.1038/ncomms13051 | www.nature.com/naturecommunications 1
D
irected evolution
1,2
serves as a critical bridge between sub-
optimal and optimal biological components, even in light
of rational design approaches
3–5
. This approach has
generated solutions to engineering problems
6,7
, established novel
functions
8,9
and provided insights into evolution
10,11
. As opposed
to adaptive evolution, directed evolution aims to identify
beneficial mutations within a gene or pathway of interest.
Unfortunately, traditional in vitro mutagenesis is encumbered
by long and costly design–build–test cycles, restrictive
requirements for hands-on manipulation of nucleic acids and
intrinsic limitations of host transformation efficiency. These
limitations become especially poignant when attempting to
optimize larger genetic systems (for example, entire pathways
including regulatory DNA), especially in more industrially and
medically relevant eukaryotic systems. Indeed, the throughput of
novel microfluidics-based screening technologies currently
outpaces throughput for generation of genetic diversity in these
systems
12
. Next-generation evolution techniques aim to accelerate
the discovery of improved variants through continuous rounds of
mutagenesis/selection on specific DNA cargo with reduced costs
using in vivo diversity generation. It has been demonstrated that
mutational throughput can be increased in Escherichia coli in an
in vivo continuous process using phage, enabling the rapid
evolution of parts
13–16
. However, this approach is best suited for
phenotypes linkable to phage growth (for example, DNA-binding
proteins) and cannot be applied to eukaryotes. Genome-editing
technologies (such as MAGE
17
and CRISPR-Cas9 (refs 18,19))
have enabled discovery of sequence–function relationships across
a wide range of species but remain, at least at present, a method
for introducing finite, defined mutations across several base pairs
(ideal when important structural features are known) and are not
well-suited for kilobase-scale-directed evolution applications
(which are necessary when structural features are largely
unknown). In yeast, recent proof-of-concept demonstrations of
continuous evolution (1) have suffered from low mutagenic rates
and the necessity for mutant expression from weak promoters
20
,
or (2) require in vitro library generation
21
. Thus, none of these
new methods are well suited for the deep mesoscale optimization
(that is, generation of all single-nucleotide substitutions to
multi-kilobase pathways and gene networks) necessary for
evolution of complex multi-part systems or continuous
evolution in eukaryotes.
To fill this gap, we establish a scalable, in vivo mutagenesis
system in yeast by engineering its native retroelement Ty1
(Fig. 1). By tuning the expression of key regulators of Ty1
transposition, we increase library size achievable using this system
and confirm its ability to impart a useful error rate to an encoded
cargo gene. Next, we apply this system to the directed evolution of
a variety of synthetic parts, including single enzymes, regulatory
factors and multi-enzyme pathways, realizing substantial and
significant improvements to performance in each case. We
further demonstrate that ICE enables the recovery of superior
mutants more quickly than error-prone PCR. Finally, we show
that ICE enables in vivo mutant generation across divergent
strains of yeast, indicating its applicability towards a wide range
of eukaryotic systems.
Results
Implementation of ICE. To establish this method, we turned to
the native yeast long terminal repeat (LTR) retrotransposon Ty1.
The replication cycle of Ty1 proceeds via an RNA intermediate that
is converted into complementary DNA through an encoded reverse
transcriptase
22
. Previous studies have demonstrated the potential
for heterologous gene expression from Ty1 when inserted between
Ty1RT and the 3
0
-LTR
23,24
. Thus, we reasoned that the
error-pronenatureofTy1replication
25,26
coupled with the
capacity for continuous retrotransposon cycling could enable a
unique mechanism for in vivo-directed mutagenesis of synthetic
DNA (denoted here as ‘cargo’) in eukaryotes a manner that is
scalablewithcellcount(Fig.1a).Insuchascheme,wedefineone
cycle of in vivo mutagenesis as the per-cell process of Ty1-cargo
transcription, reverse transcription and re-integration to a stable
genetic context (Fig. 1b). As yeast cell densities can routinely exceed
10
10
l
1
(and even 10
12
l
1
in controlled fermentations), library
size can easily exceed that of current in vitro techniques even with
low mutation or transposition rates. A complete round of in vivo
continuous evolution (ICE), in analogy to traditional directed
evolution, is achieved at the culture level by allowing multiple
cycles to occur through simple cell outgrowth, screening the
resulting in vivo library and isolating the best varian t.
As such, we hypothesized that this approach would enable high-
throughput, hands-off, scalable mutagenesis of desired parts and
pathways (Fig. 1c). For some applications, rounds may occur
continuously and growth-associated phenotypes can be selected in
tandem with mutagenesis, thus enabling ICE. In other applications,
independent rounds may be desirable to segregate dominant
mutations from background genetic drift. We demonstrate both
modes of operation in this work.
To implement and optimize this approach, we adapted a
previously described galactose-inducible Ty1 retrotransposon to
include a prototrophic marker containing an intron (URA3I)
between Ty1RT and the 3
0
-LTR in the reverse orientation relative
to Ty1 transcription
24,27
(Fig. 2a). This system enables rigorous
characterization and optimization of retroelement performance,
as transcription, splicing, reverse transcription and re-integration
are all necessary to confer uracil prototrophy. As our first
implementation and proof-of-concept, Saccharomyces cerevisiae
BY4741 containing this synthetic, inducible Ty1 on a plasmid
was exposed to galactose at low cell density to induce
retrotransposition and plated on selective media to measure
transposition rate via gained uracil prototrophy after 3 days of
growth (see Methods). This phenotype was seen with a frequency
of 6.1 10
4
per cell (Fig. 2b) and not observed when this
strain was grown in glucose (which represses pGAL1). After
demonstrating basal functionality of the plasmid-based synthetic
retroelement through induction at low cell density, we wished to
develop strategies for increasing transposition rate of Ty1. To this
end, we investigated cargo expression level, gene knockouts,
cell density, induction temperature and initiator methionine
transfer RNA expression level as potential drivers of increased
transposition rate. Taken together, this series of iterative
design cycles (Fig. 1e) increased the transposition rate
(and thus potential library size per round) by over 50-fold to
3.7 10
2
per cell in simple shake flasks (Fig. 2b).
Tuning cargo expression increases transposition rate. We first
investigated the effect of cargo transcription rate on Ty1 trans-
position. Although strong promoters (such as pTDH3) are
desirable for cargo overexpression, their high transcription rate
may interfere with that of pGAL1, thus lowering transposition
rate and library size. Out of three yeast promoters (pCYC1, pTEF1
and pTDH3, representing low, medium and high transcriptional
output, respectively
28
; Supplementary Fig. 1a), we observed the
highest transposition rate when pTEF1 drove expression of URA3
(Figs 2b and 3a). We used this promoter in future benchmarking
experiments.
Rrm3 deletion increases transposition rate. Ty1 replication is
known to be highly regulated by various host factors
29
.
To evaluate a coupling between genotype and function,
ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms13051
2 NATURE COMMUNICATIONS | 7:13051 | DOI: 10.1038/ncomms13051 | www.nature.com/naturecommunications
we performed an extensive literature search and used the yeast
haploid knockout collection to identify knockout phenotypes,
which enabled increased rates of Ty1 replication
29–31
. Of the
various genotypes tested (Fig. 3b), deletion of rrm3 most
significantly increased transposition rates in S. cerevisiae
BY4741 (Fig. 2b). Rrm3p plays a role in DNA repair, which
may influence retrotransposition, as it is dependent on
homologous recombination
29
. Several other combinations of
targets were evaluated, but these combinations did not exceed the
transposition rate beyond that of Drrm3 alone (Fig. 3b).
Therefore, S. cerevisiae BY4741 Drrm3 was used for all
subsequent experiments.
Genomic integration of Ty1 decreases transposition. Subse-
quently, we elected to move the retroelement into the genome, to
gain a more accurate picture of retroelement behaviour in its final
context. We used BY4741 Drrm3 with a genomically integrated,
pTEF1-containing retroelement at high optical density (OD) and
determined that the transposition rate was significantly inhibited
compared with the plasmid-based retroelement (Fig. 2b).
Transposition at high cell density increases library size. In all
initial experiments, Ty1 transposition was induced when cells
were at a low OD and continued as cells divided. However,
this growth can significantly reduce effective library sizes,
as mutations that occur early during growth can dominate the
resulting culture during outgrowth
32
. We aimed to increase
library sizes by inducing transposition for the same length of time
(3 days), but at a much higher initial cell density (OD
600
¼ 1).
In this condition, additional cell growth would have a greatly
reduced effect on library size. This condition significantly
increased the retrotransposition rate (Fig. 2b) and all
subsequent inductions were carried out at high cell density.
Reducing induction temperature increases transposition.
We next made use of the known temperature sensitivity of Ty1
(ref. 33) by inducing transposition at a lower temperature (22 °C).
This modification greatly improved transposition rate (Fig. 2b).
Interestingly, it also increased basal activation of our inducible
Ty1 retroelement in the absence of a cis-encoded reverse
transcriptase (Supplementary Fig. 1b), which could be due to the
Synthetic retroelement
a
b
d
ef
c
Intron
Genome
Host generality
Re-integration
cDNA
Reverse transcription
/ mutation
mRNA
Induction Selection or screen
CARGO
Transcription factor
Catalytic gene
Xylose pathway
(any expression cassette)
Transcription
S. cerevisiae
S. cerevisiae
BY4741
S. cerevisiae
K. lactis
CEN.PK2
BY4741
ICE optimization
1) Genome modification
2) High cell-density induction
a) Growing culture:
b) Non-growing culture:
3) Optimized induction temperature
4) tRNA
iMet
overexpression
Δrrm3
CARGO
GAL1p
Gag-Pol
ICE round
pTEF1
SPT15
URA3
XylA
XKS1
pTEF1
pTDH3 *
P2A
5′-LTR
3′-LTR
3
2
1
ICE cycle: 1 to 3
Figure 1 | Rationale and schematic for ICE in yeast. In the operational scheme of ICE: (a) genetic cargo of interest is cloned into the genome of an
inducible Ty1 retrotransposon; (b) on induction of retroelement transcription, the encoded reverse transcriptase is expressed, converts the Ty1 genome
(including the cargo) into cDNA in an error-prone manner and then this cassette is re-integrated into a stable genomic locus. This process is defined as one
cycle; (c) the procedure of inducing mutagenesis to a bulk culture and selecting for improved variants is analogously defined as one round. In this work,
we (d) apply this approach to several divergent strains and species of yeast, (e) iteratively improve the efficiency of Ty1 retrotransposition through deletion
of rrm3, reducing temperature, increasing cell density and increasing expression of limiting cellular components, and (f) apply this improved system to the
evolution of transcriptional activators, single enzymes and multi-enzyme pathways.
NATURE COMMUNICATIONS | DOI: 10.1038/ncomms13051 ARTICLE
NATURE COMMUNICATIONS | 7:13051 | DOI: 10.1038/ncomms13051 | www.nature.com/naturecommunications 3
activation of endogenous Ty1 elements that are natively repressed
at 30 °C.
Increasing tRNA
iMet
expression increases transposition. Based
on real-time PCR experiments (Fig. 3c,d), we noted that induc-
tion of Ty1 RNA levels by pGAL1 greatly exceeded resulting
cDNA levels produced by Ty1RT. As Ty1 replication is primed by
the yeast initiator methionine tRNA (tRNA
iMet
), we hypothesized
that the concentration of this tRNA may be limiting transposition
rates. By overexpressing tRNA
iMet
from several promoters, we
observed greatly improved transposition rates (Figs 2b and 3e). In
particular, by overexpressing tRNA
iMet
using its native promoter
and terminator on a high-copy plasmid, transposition rate could
be improved by B3.5-fold and this increase was accompanied by
a corresponding increase in cDNA levels (Fig. 3f). All subsequent
experiments used this overexpression strategy.
Characterizing the effect of cargo length on transposition.
As it is highly desirable to include long sequences consisting of
multi-gene pathways in the inducible retroelement, the effect
of transcript length on transposition rate was characterized.
Specifically, we inserted gene fragments as additional cargo
between the URA3 reporter gene and the reverse transcriptase
gene, and then measured the resulting transposition rates. In
addition, we measured transposition rates after 3, 5 and 7 days of
high cell-density induction in S. cerevisiae BY4741 Drrm3. These
experiments clearly revealed a negative correlation between
cargo length and retrotransposition rate. However, lengthening
the induction time from 3 to 7 days increased the number of
retrotransposition events, especially for constructs containing the
longest sequences (Fig. 3g). Importantly, relatively high trans-
position rates were maintained within approximately an order of
magnitude as cargo size increased to roughly 5 kb, indicating that
Ty1 is capable of generating diversity to a multi-gene pathway.
It should be noted that this experiment combined each pathway
element on the same mutagenesis cassette. However, it is also
possible to distribute several multi-gene mutagenesis cassettes
across the genome to enable simultaneous evolution on multiple
segments of longer cargo.
Characterizing the effect of terminators on transposition.
When expressing multi-gene pathways in yeast, it is common to
include a promoter before each gene and a terminator afterward.
When inserting a multi-gene pathway into the Ty1 mutagenesis
cassette, however, a terminator with bidirectional activity can
significantly affect transposition, as the entire retroelement must
be transcribed before reverse transcription. To characterize this
effect, several native and synthetic terminators were inserted
after the URA3 reporter gene in the synthetic retroelement
34
.
These experiments showed that including terminators inside
the retroelement can lower the rate of transposition, with several
terminators eliminating activity altogether (Supplementary Fig. 1c).
Although we did identify several terminators that reduced
transposition rate to within one order of magnitude, we instead
opted to use ribosome-cleavable 2A sites, which allow a single
promoter to drive expression of a fusion peptide that then self-
cleaves during translation
35
. This strategy allowed the evolution
of multi-gene pathways, such as the xylose pathway evolved here,
without including any terminators between genes. In addition, it
allows multi-gene pathways to be expressed from a single
promoter, reducing the length of DNA needed in the cargo and
thus increasing the rate of transposition (Fig. 3g). Including 2A
sequences as opposed to terminators thus allows our approach to
attain a significantly higher library size for multi-gene pathways
through two mechanisms: it avoids terminators and it reduces
cargo length by only requiring a single promoter. However,
Method for detecting transpositions
pGAL1 Gag-Pol
URA3
5′-LTR
3′-LTR
Intron
Genome
Genome
ab
BY4741 Δrrm3 Ty1RT: pTEF1-URA3l,
22 °C induction, high OD
tRNA
iMet
overexpression
BY4741 Δrrm3 Ty1RT: pTEF1-URA3l,
22 °C induction, high OD
BY4741 Δrrm3 Ty1RT:
pTEF1-URA3l, high OD
BY4741 Δrrm3 Ty1RT:
pTEF1-URA3l
BY4741 Δrrm3 Ty1RT:
pTEF1-URA3l
BY4741 Ty1RT:
pTEF1-URA3l
BY4741 Ty1RT:
pHIS3-URA3l
Ura3p
Ura3p
Ura3p
Ura3p
1.0E–06 1.0E–04
1.0E–02 1.0E+00
Plasmid version
Genomic version
Total number of distinct transposants per cell
mRNA
Cap
5′
AAAAA-3′
cDNA
Transcription/splicing
1
Reverse transcription
/ mutation
2
Re-integration
3
Figure 2 | Iterative improvement of synthetic Ty1 transposition rate and scheme for detection of retrotransposition. (a) URA3 is inserted into the
retroelement in the reverse orientation relative to transcription from the pGAL1 promoter. The presence of an intron in the same transcriptional direction as
pGAL1 prevents mRNA originating from the URA3 promoter from being correctly spliced and initiating Ura3p synthesis. On transcription from pGAL1,
the intron is spliced. This mRNA cannot give rise to Ura3p due to URA3 being present in the reverse orientation on this transcript. However, once mRNA is
converted into cDNA, a functional URA3 expression cassette is formed and integration of this cDNA into the genome ensures a heritable URA3
þ
phenotype. (b) Strain background, induction conditions and expression of critical Ty1 components were modified to improve transposition rates of the
synthetic retroelement. Error bars for the plasmid version represent 95% confidence intervals obtained via fluctuation analysis of biological triplicates and
error bars for the genomic version represent the s.d. of biological triplicates.
ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms13051
4 NATURE COMMUNICATIONS | 7:13051 | DOI: 10.1038/ncomms13051 | www.nature.com/naturecommunications
it should be noted that 2A sites are as of yet unoptimized for use
in yeast, such that cleavage efficiency may not be 100% in all
contexts and thus may pose an issue for pathways in which the
generation of fusion proteins would be undesirable. For these
cases, we recommend the integration of multiple distinct
mutagenesis cassettes into the same strain to enable the
simultaneous directed evolution of pathway components.
Measurement of mutation rate enabled by ICE. Next, we
undertook a mutation reversion experiment to investigate the
ab
cd
e
g
f
1.4E–03
1.2E–03
1.0E–03
8.0E–04
6.0E–04
4.0E–04
2.0E–04
0.0E–04
Wild type
apl2mre11
hir3cac3
hir3
mre11
hir3apl2
hir3
cac2
cac3
rrm3
ice2
ckb2
hir3
mrc1
cac2
mre11
apl2
Total number of distinct transposants
per cell
Total number of distinct transpositions
per cell
1.2E–03
1.0E–03
8.0E–04
6.0E–04
4.0E–04
2.0E–04
0.0E–04
Total number of distinct transposants
per cell
Number per cell
1.0E+00
1.0E–01
1.0E–02
1.0E–03
1.0E–04
0 2,000 4,000 6,000 8,000
Size of exogenous ‘cargo’ DNA insert (bp)
Low copy High copy
Low copy High copy
100
10
1
0
0.0E–02
1.0E–02
2.0E–02
3.0E–02
4.0E–02
5.0E–02
mRNA generation cDNA generation
Genotype
Transposition
repressed
Transposition
induced
Transposition
repressed
Transposition
induced
No tRNA overexpression
No tRNA Glc
No tRNA Gal
With tRNA Gal
With tRNA Glc
Distinct transposants (3 day induction)
Distinct transposants (5 day induction)
Distinct transposants (7 day induction)
Max distinct mutants
pSUP4-tRNA
iMet
-RPR1t
pSUP4-tRNA
iMet
-IMT4t
pIMT4-tRNA
iMet
-IMT4t
1,000
100
10
1
0.1
Ty1RT WT
No RT
Ty1RT ΔRRM3
Ty1RT Δrrm3
Ty1RT WT
No RT
Ty1RT ΔRRM3
Ty1RT Δrrm3
URA3 mRNA (arbitrary units)
3.0
2.5
2.0
1.5
1.0
0.5
0.0
URA3 DNA (arbitrary units)
Retroelement cDNA production
(arbitrary units)
BY4741
Ty1RT
BY4741
Ty1RT
pCYC1
BY4741
Ty1RT
pTEF1
BY4741
Ty1RT
pTDH3
NATURE COMMUNICATIONS | DOI: 10.1038/ncomms13051 ARTICLE
NATURE COMMUNICATIONS | 7:13051 | DOI: 10.1038/ncomms13051 | www.nature.com/naturecommunications 5