Comparison of two codon optimization strategies to enhance recombinant protein production in Escherichia coli.

Abstract

Variations in codon usage between species are one of the major causes affecting recombinant protein expression levels, with a significant impact on the economy of industrial enzyme production processes. The use of codon-optimized genes may overcome this problem. However, designing a gene for optimal expression requires choosing from a vast number of possible DNA sequences and different codon optimization methods have been used in the past decade. Here, a comparative study of the two most common methods is presented using calf prochymosin as a model.
Seven sequences encoding calf prochymosin have been designed, two using the "one amino acid-one codon" method and five using a "codon randomization" strategy. When expressed in Escherichia coli, the variants optimized by the codon randomization approach produced significantly more proteins than the native sequence including one gene that produced an increase of 70% in the amount of prochymosin accumulated. On the other hand, no significant improvement in protein expression was observed for the variants designed with the one amino acid-one codon method. The use of codon-optimized sequences did not affect the quality of the recovered inclusion bodies.
The results obtained in this study indicate that the codon randomization method is a superior strategy for codon optimization. A significant improvement in protein expression was obtained for the largely established process of chymosin production, showing the power of this strategy to reduce production costs of industrial enzymes in microbial hosts.

Full text

Page 1

RESEARCHOpen Access
Comparison of two codon optimization strategies
to enhance recombinant protein production in
Escherichia coli
Hugo G Menzella
Abstract
Background: Variations in codon usage between species are one of the major causes affecting recombinant
protein expression levels, with a significant impact on the economy of industrial enzyme production processes. The
use of codon-optimized genes may overcome this problem. However, designing a gene for optimal expression
requires choosing from a vast number of possible DNA sequences and different codon optimization methods have
been used in the past decade. Here, a comparative study of the two most common methods is presented using
calf prochymosin as a model.
Results: Seven sequences encoding calf prochymosin have been designed, two using the “one amino acid-one
codon” method and five using a “codon randomization” strategy. When expressed in Escherichia coli, the variants
optimized by the codon randomization approach produced significantly more proteins than the native sequence
including one gene that produced an increase of 70% in the amount of prochymosin accumulated. On the other
hand, no significant improvement in protein expression was observed for the variants designed with the one
amino acid-one codon method. The use of codon-optimized sequences did not affect the quality of the recovered
inclusion bodies.
Conclusions: The results obtained in this study indicate that the codon randomization method is a superior
strategy for codon optimization. A significant improvement in protein expression was obtained for the largely
established process of chymosin production, showing the power of this strategy to reduce production costs of
industrial enzymes in microbial hosts.
Background
Industrial enzymes, included those used in food indus-
try, are now traded as commodity products and there is
a continuing need to reduce manufacturing costs in
order to remain competitive in the global markets.
Escherichia coli is a preferred host for the production of
recombinant proteins because it combines fast growth
rate, inexpensive fermentation media and well under-
stood genetics; and the cost of production in this micro-
organism depends in large part upon the protein
expression levels [1,2].
Species-specific variations in codon usage are often
cited as one of the major causes impacting protein
expression levels [3,4]. The presence of rare codons,
which are correlated with low levels of their cognate
tRNAs species in the cell, can reduce the translation
rate and induce translation errors with a significant
impact on the economy of the production process [5,6].
In the past decade, a high number of genes have been
re-designed to increase their expression level [1-3,7-10].
However, designing a gene for optimal expression
requires choosing from a large space containing a vast
number of possible DNA sequences. Typically, two stra-
tegies have been used for codon optimization. The first
one, known as “one amino acid-one codon”, assigns the
most abundant codon of the host or a set of selected
genes to all instances of a given amino acid in the target
sequence [4,8,11-15]. The second one, designed here
“codon randomization”, uses translation tables, based on
the frequency distribution of the codons in an entire
Correspondence: hmenzella@fbioyf.unr.edu.ar
Genetic Engineering & Fermentation Technology. CONICET. Facultad de
Ciencias Bioquimicas y Farmacéuticas. Universidad Nacional de Rosario.
Suipacha 531 Rosario 2000. Republica Argentina
Menzella Microbial Cell Factories 2011, 10:15
http://www.microbialcellfactories.com/content/10/1/15
© 2011 Menzella; licensee BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons
Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in
any medium, provided the original work is properly cited.

Page 2

genome or a subset of highly expressed genes, to attach
weights to each codon. In this case, codons are assigned
randomly with a probability given by the weights
[2,7,8,16,17].
Many transgenic proteins expressed in E. coli are
recovered as insoluble aggregates in the form of inclu-
sion bodies. The formation of these aggregates seems to
be independent of the type of protein, and this draw-
back has been proven difficult to overcome [18,19].
Nevertheless, the fact that inclusion bodies are easy to
isolate and mainly composed by the over-expressed pro-
tein facilitates product recovery at industrial scale
[20,21]. Thus, the production of recombinant proteins
as inclusion bodies represent a cost-effective alternative
for enzyme manufacturing, provided that efficient large
scale refolding methods are available. This is the case
for calf prochymosin, the precursor of chymosin, which
is widely used in cheese making and mostly obtained
from recombinant microorganisms [18,22].
It has been recently shown that proteins contained in
inclusion bodies possess a degree of secondary structure
and exhibit biological activity in many cases [23-28].
Moreover, the quality of the inclusion bodies, deter-
mined by the degree of folding of the aggregated pro-
teins, depends on different factors; many of them
related to the translation rate of the corresponding
mRNA [23,24]. In addition, it has been recently demon-
strated that synonymous codon replacement is not
always silent [29]. Codon optimization affects translation
rate which, in turn, may alter protein structure and
function and thus the efficiency of refolding of proteins
recovered from inclusion bodies [29-31]. Therefore,
codon-optimized genes may lead to the formation of
inclusion bodies from which the recovered proteins
could be difficult to refold.
In this study, calf prochymosin was used as a model to
perform a quantitative and qualitative evaluation of the
inclusion bodies obtained from the expression of a set
of synthetic genes encoding this protein. Seven genes
were designed and synthesized using two different
codon optimization strategies, and the amount of
recombinant protein and the refolding yield of the inclu-
sion bodies obtained with each gene were compared.
Results
Gene design, synthesis and expression vector
construction
Analysis of the native calf prochymosin gene revealed
that almost 59% of codons are not the preferred for E.
coli (Table 1). For example, seven out of the eight
codons encoding for arginine found in the native
sequence are represented with a frequency below 6% in
the E. coli genome. Six of these codons are AGA or
AGG, which have been shown to cause low levels of
expression and mistranslational errors [1,32]. Thus, it
was surmised that codon optimization of the calf pro-
chymosin gene might result in an increase in protein
expression.
Seven variants of the calf prochymosin gene were
designed using two different codon optimization strategies.
In the first approach, only one codon was assigned for
each amino acid to create two sequences named V0 and
V1. For the V0 gene, the preferred codon found in the
entire genome of E. coli W3110 was assigned to each
amino acid. For the design of the V1 gene, a similar strat-
egy was used; but in this case the favorite codon found in
a set of highly expressed genes was employed to encode
for each amino acid.
The second strategy used for codon optimization con-
sisted on randomly assigning a triplet for each amino
acid using a preference table http://www.kazusa.or.jp/
codon/cgi-bin/showcodon.cgi?species=316407, with a
probability based on the weight of each codon within
the set encoding a given amino acid. Using this algo-
rithm, five sequences were independently designed using
the GeMS software package [16], and named V2-V6.
The codon distribution for sequences V0, V1 and V2 is
shown in Table 1 and the sequences of the seven genes
and their codon usage is shown in Additional file 1.
The codon-optimized synthetic genes were created by
using single strand 5´phosphorylated complementary
primers. In all the cases 27 primers with a length ran-
ging from 38 to 42 bases were used to create the leading
strand and 27 primers with a length between from 38 to
43 bases were used to create the lagging strand. For all
the genes the designed single stranded oligonucleotides
overlapped each other by a minimum of 18 bases to
ensure annealing. Two TAA stop codons in tandem
were added at the end of each coding region followed
by an EcoRI site. Additionally, an NdeI site overlapping
the initial ATG was included in all the genes to mobilize
the ORFs. Finally, all the synthetic genes were inserted
into the expression vector pBru [17], where gene expres-
sion is driven by the PBADpromoter, inducible by the
addition of L-arabinose.
Protein expression and yield
Recombinant plasmids were transformed into the E. coli
W3110 strain for expression tests. Cell cultures were
induced by the addition of L-arabinose when OD600
reached 0.5 and cells were harvested after 5 h. As pre-
viously reported, analysis of soluble and insoluble fractions
of cell lysates by SDS-PAGE showed that all detectable
prochymosin was located in the insoluble fraction in the
form of inclusion bodies [18,22]. The amount of prochy-
mosin produced by the expression of each gene variant
was quantified by densitometry of the stained gels and is
shown in Figure 1 and Table 2. The variants V2-V6
Menzella Microbial Cell Factories 2011, 10:15
http://www.microbialcellfactories.com/content/10/1/15
Page 2 of 8

Page 3

optimized by the codon randomization approach pro-
duced significantly more proteins than the native
sequence. The best result was obtained for the V2
sequence with an increase of 70% in the amount of pro-
chymosin accumulated. On the other hand, no significant
improvement in protein expression was observed for the
V0 and V1 variants. In both cases the amount of prochy-
mosin produced was similar to that obtained upon the
expression of the wild type gene. Inclusion bodies were
washed and isolated and the yield of recombinant protein
was also measured by the Lowry method [33]. Consistent
with the results obtained from gel scanning quantification,
the yields for V2 and the wild type sequence were 490 mg/
L and 282 mg/L respectively. As reported by other
authors, no correlation was found between the codon bias,
measured using the codon adaptation index [34], and the
Table 1 Codon distribution of V0, V1, V2 and wild type (WT) sequences
AACodonfa
WT V0 V1V2AACodonfa
WTV0 V1 V2
AlaGCG
GCC
GCA
GCU
0.36
0.27
0.21
0.16
1
15
0
1
17
0
0
0
17
0
0
0
6
5
4
2
Leu CUG
UUA
UUG
CUC
CUU
CUA
0.50
0.13
0.13
0.10
0.10
0.04
23
0
0
5
1
0
29
0
0
0
0
0
29
0
0
0
0
0
16
0
5
4
4
0 ArgCGC
CGU
CGG
CGA
AGA
AGG
0.40
0.38
0.10
0.06
0.04
0.02
1
0
0
1
1
5
8
0
0
0
0
0
0
8
0
0
0
0
4
3
1
0
0
0
Lys AAA
AAG
0.76
0.24
6
9
15
0
15
0
9
6
MetATG1.009999
AsnAAC
AAU
0.55
0.45
11
4
15
0
15
0
8
7
PheUUU
UUC
0.57
0.43
619
0
011
81319
AspGAU
GAC
0.63
0.37
3
20
23
0
014
9
ProCCG
CCA
CCU
CCC
0.53
0.19
0.16
0.12
3
1
1
16
0
0
0
16
0
0
0
9
2
3
2
23
CysUGC
UGU
0.56
0.44
3
3
6
0
6
0
3
3
11
SerAGC
UCG
AGU
UCC
UCU
UCA
0.28
0.15
0.15
0.15
0.15
0.12
13
4
4
9
4
1
35
0
0
0
0
0
0
0
0
0
9
5
5
5
6
5
GlnCAG
CAA
0.65
0.35
24
1
25
0
25
0
17
8
GluGAA
GAG
0.69
0.31
2
12
14
0
14
0
10
4
35
0
GlyGGC
GGU
GGG
GGA
0.41
0.34
0.15
0.11
15
2
13
1
31
0
0
0
011
10
5
5
Thr ACC
ACG
ACU
ACA
0.44
0.27
0.16
0.13
13
2
4
5
24
0
0
0
24
0
0
0
10
6
4
4
31
0
0
HisCAU
CAC
0.57
0.43
4
2
6
0
6
0
3
3
TrpUGG1.004444
TyrUAU
UAC
0.57
0.43
522
0
013
9IleAUU
AUC
AUA
0.51
0.42
0.07
2
19
1
22
0
0
013
8
1
1722
22
0ValGUG
GUU
GUC
GUA
0.37
0.26
0.22
0.15
14
3
7
2
26
0
0
0
0 10
7
4
5
26
0
0
aRelative frequency of each codon in E. coli W3110.
Sequences V0 and V1 were designed using the one amino acid- one codon method; and V2 was designed with the codon randomization algorithm.
Menzella Microbial Cell Factories 2011, 10:15
http://www.microbialcellfactories.com/content/10/1/15
Page 3 of 8

Page 4

quantity of recombinant protein produced by the codon-
optimized genes [2,35].
Cell growth rates were determined for E. coli cells har-
boring each expression plasmid. No significant differ-
ence was observed among the growth curves of the
recombinant strains indicating that none of the genes
had a toxic effect. As expected, the growth rate of the
strain harboring the empty vector was higher after the
addition of the inducer (data not shown).
It has been reported that the 5’coding region is parti-
cularly important in modulating translation initiation
[2,35-37]. Predicted mRNA secondary structures did not
correlate strongly with expression when the sequences
were analyzed using the online RNA folding program
“mfold” http://mfold.rna.albany.edu/[38]. In order to
investigate whether the reason for the low expression of
the V0 variant was due to the presence of a 5’ local ele-
ment, the first 10 codons of this gene were replaced by
those of the V2 gene; the variant showing the greatest
level of prochymosin production. Expression of the chi-
meric gene V2/0 showed a modest increase in prochy-
mosin production (Table 2), indicating that downstream
elements account for the lower production of prochy-
mosin observed for the V0 variant.
In order to investigate whether a further increase in
protein expression level could be obtained by using a
???????????
???????
???????
????????????????????????????????????????????????????????????????????????????????????????????????????????????????
??
?
Figure 1 Expression analysis of the synthetic gene variants by SDS-PAGE. Lane 1, molecular weight marker. Lane 2, lysate of E. coli W3100
culture harboring the pWT expression vector for the expression of wild type calf prochymosin gene grown in the absence of L-arabinose; lanes
3-10, lysate of E. coli W3100 culture harboring the pV0, pV1, pV2, pV3, pV4, pV5 and pV6 expression vector for the expression of V0-V6 synthetic
versions of calf prochymosin grown in the with 2 g/l of L-arabinose. In all cases, cell cultures were brought to OD600= 3 and 20 μl were used
for the analysis.
Table 2 Amount of prochymosin produced by E coli W3110 cells expressing gene variants created using different
codon optimization methods
Sequence -
expression
vector
Codon optimization
method
CAIa
Prochymosin
produced
(mg/l)b
Prochymosin
produced
(relative to wild type)
Wild type - pWT
V0 - pV0
V1 - pV1
V2 - pV2
V3 - pV3
V4 - pV4
V5 - pV5
V6 - pV6
V2/0 - pV2/0
- 0.66
1
0.82
0.72
0.70
0.72
0.73
0.74
0.97
262 ±19
248 ± 17
246 ± 21
448 ± 31
366 ± 26
330 ± 16
311 ± 20
309 ± 29
296 ± 12
1
0.95
0.94
1.71
1.50
1.36
1.29
1.21
1.13
One amino acid- one codon
One amino acid- one codon
Codon randomization
Codon randomization
Codon randomization
Codon randomization
Codon randomization
Hybrid construct
aCodon adaptation index.
bDetermined by scanning and densitometry analysis of stained gels. Values shown represent mean and standard error of three independent experiments.
Cells were grown at 30°C in supplemented LB medium to OD600= 0.5 and induced by the addition of 2 g/l of L-arabinose for 5 h.
Menzella Microbial Cell Factories 2011, 10:15
http://www.microbialcellfactories.com/content/10/1/15
Page 4 of 8

Page 5

stronger promoter, an additional experiment was carried
out. For this, the V2 sequence was cloned into the pET
24b expression vector, where the expression is driven by
the T7 promoter. The resulting plasmid was trans-
formed into the BL21(DE3) strain and three colonies
grown on LB medium. In all the cases a fall in the opti-
cal density with a concomitant increase in the viscosity
of the cultures was observed within 6 h after the inocu-
lation, indicating cell lysis. This observation suggests a
toxic effect of the V2 gene when controlled by the
strong T7 promoter.
In vitro refolding efficiency
Codon optimization affects translation rate which, in
turn, may alter protein structure and function. It has
been described that inclusion bodies formed in E. coli
under different expression conditions may differ in their
quality and, therefore, their ability to yield active pro-
teins [25,39,40]. To the best of my knowledge, the
impact of codon optimization on the quality of inclusion
bodies has not been previously studied. Thus, I decided
to investigate the ability of inclusion bodies obtained
from the expression of different gene variants to yield
functional prochymosin.
Inclusion bodies were prepared from cultures of E. coli
W3110 strain harboring expression plasmids for the
seven prochymosin gene variants described above. In all
cases, inclusion bodies were washed after cell lysis and
recovered as a white paste. The paste was dissolved in 8
M urea and the total protein concentration adjusted to
20 g/L. For all the preparations, PAGE analysis showed
that more than 90% of the protein contained in the urea
solution corresponds to calf prochymosin (data not
shown). The urea solution was rapidly diluted into
refolding buffer supplemented with 0.5 M arginine and
10 μM Cu++since these additives were previously
shown to increase the refolding efficiency of prochymo-
sin [18].
Figure 2 shows the refolding efficiency obtained for
the different inclusion body preparations. No significant
differences were found in the recovery of calf prochy-
mosin for inclusion bodies prepared using the different
gene variants. In the refolding method employed, air
oxidation was used to promote the formation of disul-
phide bonds and the activity recovered was similar to
that previously reported for the native sequence [18,41]
Discussion
Codon optimization of the calf prochymosin gene was
chosen due to the commercial value of improving its
expression and to study the impact of codon optimiza-
tion in an established production process. Even when
the wild type sequence has been reported to express
well in E. coli [18,22], the presence of some rare codons
led me to investigate codon optimization strategies in
order to increase the expression of this protein. In the
present study, seven gene variants were designed and
synthesized to evaluate the effect of the two most com-
mon gene design strategies on the production of calf
prochymosin in E. coli. The five sequences designed
using the codon randomization strategy yielded higher
protein quantities than those designed with the one
amino acid-one codon method. These results suggest
that the former method is a superior strategy for codon
?
??
??
??
??
??
???????????? ??
???????????????????
?????????????
Figure 2 Refolding efficiency of prochymosin prepared by expressing codon-optimized genes in E. coli W3110. Renaturation was carried
out by diluting the urea-solubilized inclusion bodies in renaturation buffer at a final protein concentration of 1 mg/ml and incubating the
mixture at 4°C for 12 h. Other experimental details are provided in the methods section. Values shown are means of three independent
determinations. The standard deviations were in all the cases less than 10% of the corresponding means.
Menzella Microbial Cell Factories 2011, 10:15
http://www.microbialcellfactories.com/content/10/1/15
Page 5 of 8