One of the most common reasons that heterologous proteins fail to express in E. coli is the presence of “rare” codons in the target mRNA. Many proteins, especially human proteins, have mRNA sequences that include codons that are infrequently used in E. coli (rare codons) (Sharp and Li, 1987). This includes codons for arginine (AGA, AGG), isoleucine (AUA), leucine (CUA), and proline (CCC). Target genes that contain significant numbers of individual rare codons, or smaller numbers of tandem rare codons, are more likely to experience translational stalling in E. coli , and thus often either completely fail to express, express at very low levels, or are expressed as truncated proteins (Kane, 1995, Cruz‐Vera et al., 2004). Moreover, when they do express, these rare codon–rich genes can also be incorrectly translated, as high level misincorporation of lysine for arginine at AGA codons has been observed for protein targets expressed in E. coli (Calderone et al., 1996).

Fortunately, the codon bias of E. coli is straightforward to overcome. Multiple Web sites are now available that quantify the number and the location of rare codons in a gene (e.g., the rare codon calculator, RaCC; http://nihserver.mbi.ucla.edu/RACC/). These programs also often highlight the number of consecutive rare codons. If the target protein contains a significant number of rare codons, especially tandem rare codons, two approaches can be taken. In the first, changes are made to the gene. Specifically, the gene is “codon optimized,” i.e., the rare codons are replaced with those that are common to the host. This can be achieved using site‐directed mutagenesis. However, it is often faster and cheaper to simply have the codon‐optimized gene synthesized (this service is offered by multiple companies). Gene synthesis has the added benefit that most gene optimization algorithms optimize not only rare codons, but also mRNA secondary structure, which has also been shown to affect translation efficiency (Hatfield and Roth, 2007). In the second approach, changes are made to the expression host. Namely, genes encoding the rare tRNAs are co‐expressed with the wild‐type (non‐optimized) target gene (Wakagi et al., 1998). E. coli strains are now available that contain plasmids that encode rare tRNAs [i.e., BL21 (DE3)‐RIL/RP/RILP cells from Stratagene or Rosetta cells from Novagen; Schenk et al., 1995; Tegel et al., 2009].

Both approaches effectively overcome codon bias. For example, codon optimization of the human β‐defensin 2 (hBD2) gene led to a nine‐fold enhancement in the expression level (Peng et al., 2004) while co‐expression of inorganic phosphatase from Sulfolobus sp. with tRNA^Arg (AGA codon) more than doubled its already high expression level (Wakagi et al., 1998). Critically, recent large‐scale, comparative studies have shown that, for most targets, either approach is equally effective (Burgess‐Brown et al., 2008). Thus, many research groups, including those of structural genomics consortia, protein production facilities, and the authors (see the standard protocol in Fig. 2), use codon‐supplemented cells for all initial expression trials.

The evolution of eukaryotes has been characterized by a significant increase in the size and complexity of proteins; e.g., the average protein length in E. coli is 317 residues while in humans it is 510 residues (Netzer and Hartl, 1997, Sakharkar et al., 2006). This increase in size is due to an increase in the number of complex, multidomain proteins, in which individual domains have distinct and independent functions. Comparative studies have shown that the probability of soluble expression in E. coli decreases with increasing molecular weight, especially for proteins >60 kD (Canaves et al., 2004, Goh et al., 2004, Gråslund et al., 2008a). For example, a large‐scale study examining the protein properties of 95 recombinantly expressed mammalian proteins found that smaller proteins (average molecular weight of 22.8 kD) were often expressed solubly alone, while larger proteins (average molecular weight of 40.4 kD) were only solubly expressed when fused to solubility‐enhancing tags (Dyson et al., 2004). In a separate study, small proteins (100 amino acids or less) were expressed solubly in E. coli at levels suitable for purification 47% of the time, while large proteins (600 to 800 amino acids) were expressed solubly only 33% of the time (Gråslund et al., 2008a). Thus, when using E. coli as an expression host, it is typically advantageous to express individual protein domains, as opposed to the full‐length protein, whenever possible.

The starting and ending residues of the target domain can also greatly affect expression yield and solubility. For example, Klock et al. (2008) showed that deletion of just four residues at either the N‐ or C‐terminus can convert a solubly expressing protein into one that expresses insolubly. In a separate study, Gråslund et al. (2008b) generated 10 constructs of a single target domain of interest: full‐length and 9 deletion constructs that differed in length from one another by 2 to 10 residues at either the N‐ or C‐terminus. Thus, all available functional and structural data should be used to determine optimal boundaries for a protein domain construct. For a protein of unknown domain structure, threading the target protein sequence onto a homologous protein structure (i.e., SWISS‐MODEL; Arnold et al., 2006) or using structure‐based/fold recognition sequence alignment programs (i.e., FFAS; Jaroszewski et al., 2005) can aid in determining the optimal domain boundaries. When a homologous protein structure is not available, the prediction of secondary structural elements (i.e., PSIPRED; Jones, 1999) should be utilized. The disruption of predicted secondary structural elements must be avoided. Many of the bioinformatics tools needed to carry out these types of analyses are freely available at http://www.expasy.org. In addition, a program has been developed that integrates these bioinformatics tools to aid in protein boundary identification (the SGC Domain Boundary Analyzer; described in more detail in Gråslund et al., 2008b). Finally, it is typically advantageous to subclone four or more constructs with different N‐ and C‐terminal boundaries and test, in parallel, which construct results in the highest level of soluble expression (Brown et al., 2008).

Hydrophobic residues, low complexity regions. In addition to molecular weight, other biophysical properties of the protein, such as hydrophobicity and sequence complexity, can influence expression yields. In the study by Dyson et al. (2004), 95 mammalian proteins were fused to a variety of N‐ and C‐terminal expression and purification tags in order to elucidate the properties of the proteins and fusion tags that facilitate soluble expression. They determined that contiguous hydrophobic residues (AILFWV) and low complexity regions (LCRs) negatively correlate with soluble expression. LCRs are regions of biased sequence composition, such as homopolymeric runs, short‐period repeats, and overrepresentations of one or a few residues that typically adopt disordered coil conformations (DePristo et al., 2006). This has also been reported by other groups (Canaves et al., 2004; Gråslund et al., 2008a). Thus, it is common to design protein expression constructs to avoid hydrophobic residues and low complexity segments in the extreme N‐ and C‐termini (Peti and Page, 2007). However, LCRs do not always inhibit soluble expression. Many intrinsically unstructured proteins (IUPs; proteins that do not adopt a single folded conformation yet are still biologically active) contain LCRs, yet have been robustly and solubly expressed in E. coli (Dyson and Wright, 2005; Dancheck et al., 2008). Because LCRs may play an active role in mediating protein function or protein‐protein interactions (Karlin et al., 2002), their inclusion in an expression protein construct must be determined on a protein‐by‐protein basis.

Disulfide bonds. The presence of disulfide bonds in a protein also negatively correlates with soluble expression in E. coli . The reducing environment of the bacterial cytoplasm makes the efficient production of disulfide‐containing proteins, such as growth factors and antibody FAB fragments, challenging (Stewart et al., 1998). Thus, the expression of disulfide‐bond containing proteins in E. coli commonly results in the production of insoluble protein (due to misfolding) sequestered into inclusion bodies (Veldkamp et al., 2007; Chen and Leong, 2009); for a review of refolding strategies, see Tsumoto et al., (2003). When refolding conditions cannot be successfully identified, the target protein must be produced solubly in vivo. The three most common strategies to express disulfide‐containing proteins, all of which are discussed later in this review, are to try the following: (1) target the expressed protein to the E. coli periplasm, which is highly oxidative (Leichert and Jakob, 2004), (2) fuse the protein to thioredoxin (Lefebvre et al., 2009a), and/or (3) express the protein in bacterial strains containing thioredoxin reductase and glutathione reductase mutants (Xu et al., 2008).

Transmembrane segments. Finally, the solubility of the recombinantly expressed protein will typically be compromised if the construct includes transmembrane‐spanning regions. Thus, the soluble expression of transmembrane‐containing proteins, especially integral membrane proteins, in E. coli is exceptionally challenging, requiring specialized materials and strategies (Mohanty and Wiener, 2004; Gordon et al., 2008; Dvir and Choe, 2009).

The origin of replication of a vector is the site where replication is initiated. It also determines copy number of the vector in the host. The copy number for common E. coli expression plasmids ranges from low (2 to 20) to high (20 to 40). Typically, high‐copy‐number plasmids are desired for protein expression in E. coli , as they result in the maximum protein yield for a given culture volume (Jing et al., 1993; Huang et al., 1994). The origin of replication is also important to consider when carrying out protein co‐expression experiments in which two different plasmids, each of which contains a different protein/biomolecule, are simultaneously transformed into the same expression cell (Johnston and Marmorstein, 2003). For these experiments, the origins of the two plasmids should be different to allow the cell to support both expression vectors.

Promoters are another element of the vector that can have a profound effect on the strength and duration of transcription and, in turn, protein yield. Synthesis of mRNA is initiated when RNA polymerase binds to a specific DNA sequence, the promoter, adjacent to the target gene. This sequence contains the transcription start site, as well as two hexanucleotide sequences approximately 10 and 35 bases upstream of the initiation site that direct the binding of essential elements of the polymerase machinery (Rosenberg and Court, 1979; Hawley and McClure, 1983; Harley and Reynolds, 1987). An effective promoter for expressing heterologous proteins in E. coli has three characteristic features. First, it is strong, resulting in robust expression of the target gene (typically 10% to 50% of the total cellular protein). Second, it exhibits low basal transcriptional activity to prevent unwanted transcription prior to induction. Third, induction is simple and cost‐effective.

When selecting a promoter system, the nature of the protein target and its desired downstream use must be considered. If the protein target is a toxic protein (like a ribonuclease), one should consider using promoter systems that have extremely low basal expression, such as the araBAD promoter (Lee et al., 1987). Alternatively, for maximal protein yields, a strong promoter should be selected, such as T7 or tac. Finally, for aggregation‐prone proteins, a cold‐shock promoter, in which expression is carried out at low temperatures, may be tested. Multiple promoters have been developed for expression in E. coli and are summarized in Table 2. The four most widely used promoters are the T7 RNA promoter, the araBAD promoter, hybrid promoters, and the cspA promoter.

T7 promoter (T7 RNA polymerase system). The T7 RNA polymerase system is the most commonly used promoter system in E. coli . Gene expression is driven by the T7 RNA polymerase (from the T7 bacteriophage), which transcribes DNA five times faster than the bacterial RNA polymerase (Studier et al., 1990). Because E. coli lacks this enzyme, the polymerase must be delivered to the cell, via an inducible plasmid or, more often, by using an E. coli strain that contains a chromosomal copy of the T7 polymerase gene (Studier and Moffatt, 1986b). In the absence of an inducer, the polymerase, which itself is under the control of the lacUV5 promoter, is not produced, and correspondingly, the gene of interest is not transcribed (Studier et al., 1990). Upon addition of the nonhydrolyzable lactose analog, IPTG, the T7 RNA polymerase is transcribed and synthesized. The polymerase then initiates transcription of the target gene by binding to a T7 polymerase–specific promoter.

Once induced, most of the cellular machinery is devoted to the production of the recombinant protein, comprising up to 50% of the total cellular protein (Studier and Moffatt, 1986b; Studier et al., 1990). However, such robust transcription can have undesirable effects. First, even minimal basal production of T7 RNA polymerase results in “leaky” expression (expression prior to induction) of the target protein (Moffatt and Studier, 1987). This can be detrimental if the protein is toxic to the host, resulting in cell death or growth arrest. To minimize leaky expression, several host strains [i.e., BL21(DE3)pLysS, Rosetta(DE3)pLysS, Rosetta‐gami(DE3)pLysS] have been developed that contain a plasmid encoding T7 lysozyme, an enzyme that binds and inhibits T7 polymerase (Moffatt and Studier, 1987). The concentrations of T7 lysozyme produced in these strains are sufficient to inhibit basal transcription of target genes. However, it continues to inhibit T7 RNA polymerase after expression has been induced, and thus, it is typical for the level of recombinant protein expression to be significantly reduced in pLysS versus non‐pLysS strains. For nontoxic proteins, this results in lower yields (Studier, 1991). However, for toxic proteins, this often results in higher yields because the cell growth is not arrested prior to induction by the premature expression of the toxic protein (Yeo et al., 2009). In addition, because T7 is such a strong promoter, some translated proteins aggregate and form inclusion bodies because they fail to fold before encountering another unfolded protein. In these cases, expression parameters can be changed (see section IV) to maximize the yield of soluble, folded protein. Alternatively, a weaker promoter can be used.

araBAD promoter. The arabinose promoter system (araBAD promoter) is a strong, titratable promoter which, unlike the T7 promoter, has almost no basal transcriptional activity (Lee et al., 1987). Thus, it is advantageous for the expression of highly toxic proteins. The induction agent for this promoter is l‐arabinose (Lee et al., 1987). In the absence of l‐arabinose, transcription is exceptionally low and, if needed, can be even further suppressed by the addition of glucose (Miyada et al., 1984). As reported by Guzman et al., protein expression levels increase linearly with increasing concentrations of l‐arabinose over two logarithms (Guzman et al., 1995). This allows the expression level to be titrated over a wide range of inducer concentrations, which can be important when trying to either maximize expression yields (higher l‐arabinose concentrations) or to increase the yield of soluble protein (lower l‐arabinose concentrations). It should be noted that although this system can efficiently repress gene expression, the repression level is not always zero and the efficiency of repression is gene dependent (Guzman et al., 1995). Finally, studies that have directly compared protein yields from the araBAD and the T7 promoters have found that T7 promoters generally result in higher expression yields (Goulding and Perry, 2003). For example, in our hands, we have seen protein expression yields increase by 2‐ to as much as 10‐fold by switching from the araBAD to the T7 promoter system.

Hybrid promoters: trc and tac promoters. The trc and tac promoters are hybrids of naturally occurring promoters, consisting of the −35 region of the trp promoter and the −10 region of the lacUV5 promoter (Amann et al., 1983; Deboer et al., 1983). The only difference between these two systems is the spacing between the −35 and −10 consensus sequences, with 16 bp and 17 bp separations in the tac and trc promoters, respectively (Brosius et al., 1985). Expression is induced with IPTG (Brosius et al., 1985). Trc and tac are both considered to be strong promoters, with the trc promoter ∼90% as active as the tac promoter, and both can result in the accumulation of up to 15% to 30% of the total cellular protein (Brosius et al., 1985). Because these promoters are leaky, they can be problematic when expressing proteins that are toxic to the cell (Brosius et al., 1985, Otto et al., 1995).

cspA promoter. Following a downshift in growth temperature from 37°C to 10°C, the bacterial cellular machinery is largely devoted to the production of 13 “cold‐shock” proteins (Jones et al., 1987). After 90 min, the major cold‐shock protein, CspA, accounts for 13% of total cellular protein (Goldstein et al., 1990). Accordingly, the cspA promoter has been exploited to direct the expression of recombinant proteins at low temperatures. For this system, induction is achieved by simply changing the growth temperature of the bacterial culture from 37°C to between 10°C and 25°C (Vasina and Baneyx, 1996); no chemical inductant is required. Moreover, because expression is carried out at low temperatures, this system has the added benefit that it promotes the soluble expression of aggregation‐prone proteins (Vasina and Baneyx, 1996, 1997). One drawback is that the cold shock promoter is not completely repressed at higher temperatures, which can result in basal expression of the target protein (Qing et al., 2004). Finally, expression levels using the cspA promoter are typically lower than that seen with the T7 promoter.

Fusion tags are proteins or peptides that are genetically fused to the target protein. They are useful because they can improve protein expression, promote folding, increase protein solubility, and facilitate downstream processes such as purification and detection. However, the “perfect” tag, i.e., one that can perform all of these tasks for every protein, still does not exist. Thus, it is often necessary to test multiple fusion tags to determine which tag results in the highest yields of soluble protein (Peti and Page, 2007; Brown et al., 2008) and to also use a combination of tags in order to facilitate both expression and purification (Nilsson et al., 1996; Pryor and Leiting, 1997; Routzahn and Waugh, 2002). For a comprehensive review of fusion tags, see Terpe (2003). Many of the most commonly used fusion tags, their biophysical characteristics, and their uses in expression and/or purification are listed in Table 3. The placement of the tag, either N‐terminal or C‐terminal, is also important as it can have a profound effect on soluble protein expression levels. Additionally, the presence of a fusion tag may interfere with the biological activity of the recombinantly expressed protein, and thus, in these cases, it may be important to enzymatically remove the tag after the fusion protein has been purified.

Common tags: hexahistidine (his₆). The his₆ tag does not enhance soluble expression, but it does facilitate purification and because of its widespread use, it is described here. His₆‐tagged fusion proteins are purified using immobilized metal affinity chromatography (IMAC; Porath et al., 1975). In IMAC, metal ions, typically nickel or cobalt, are immobilized to resin via a metal chelator, such as nitrilotriacetic acid, with only three or four of the six metal coordination sites occupied. The unprotonated histidines of the his₆‐tagged fusion protein coordinate the metal, allowing the expressed protein to be readily purified from E. coli lysate. The bound proteins are then eluted using either imidazole or by lowering the pH. Because this tag is small (<1 kD), it is frequently used in conjunction with solubility‐enhancing fusion tags to provide a bi‐functional tag that facilitates both solubility and purification. Increasing the length of the tag to eight or even ten histidine residues typically increases the purity of the purified fusion protein (Tanaka et al., 1999; Mobley et al., 2007).

Common tags: Thioredoxin (Trx). Trx is a small protein that catalyzes dithiol‐disulfide exchange reactions, an activity that is important for diverse cellular processes. When over‐expressed in E. coli , it can also accumulate to up to 40% of the total cellular protein, yet still remain soluble (LaVallie et al., 1993, Lauber et al., 2001). Thus, trx facilitates both expression and solubility. Its utility for enhancing soluble expression was recently demonstrated in two studies. In the first, 30 full‐length mammalian proteins were expressed with a series of fusion tags in order to identify those tags that promote soluble expression for the most diverse set of targets. It was found that two tags, Trx and maltose‐binding protein (MBP, discussed below), consistently promoted soluble expression when fused to the N‐terminus of the target protein (Dyson et al., 2004). In a second, similar study, in which the expression of 27 eukaryotic proteins of suitable size for NMR studies were examined, Trx was ranked highest for its ability to promote soluble protein expression (Hammarstrom et al., 2002).

Trx is also the fusion tag of choice if expressing a protein that contains disulfide bonds. To overcome the highly reductive environment of the bacterial cytosol, new bacterial strains, which contain mutations in two proteins that play key roles in maintaining the reducing environment of the cytosol, thioredoxin reductase (trxB) and glutathione reductase (gor), have been developed (Origami and Rosetta‐gami cells; discussed further in section IV). Because the formation of disulfide bonds appears to be dependent on the presence of thioredoxin in trxB mutant strains (Stewart et al., 1998), increased yields of soluble, active protein may be obtained by fusing the target protein to Trx and expressing the Trx‐fusion protein in trxB/gor mutant strains.

Common tags: Maltose‐binding protein (MBP). MBP is a soluble periplasmic protein that binds maltose and delivers it to the MalFGK2 transporter (Fox et al., 2001). As a fusion tag, MBP facilitates expression, solubility, and purification. It can also be used to target proteins to either the cytosol or periplasm (fusion proteins targeted to the periplasm include the endogenous malE signal sequence in the MBP gene), although it is most frequently used for cytosolic expression. Like Trx, MBP constantly ranks as one of the most effective fusion tags for promoting protein solubility (Dyson et al., 2004; Niiranen et al., 2007). In addition, MBP has been shown to passively promote the folding of its fused partner (Kapust and Waugh, 1999; Nallamsetty and Waugh, 2006, 2007). The mechanism by which MBP promotes folding is not well understood, but it is has been suggested that MBP possesses chaperone‐like qualities and may function by either interacting directly with its fusion protein, stabilizing folding intermediates, or inhibiting protein aggregation (Kapust and Waugh, 1999; Fox et al., 2001).

MBP is most commonly used as an N‐terminal tag. However, unlike Trx, which is only effective as an N‐terminal fusion tag, MBP has been shown to effectively enhance protein solubility as both an N‐ and a C‐terminal fusion tag (Dyson et al., 2004). Thus, MBP can be used effectively in both positions. Unlike Trx, MBP also functions as an affinity tag as it binds tightly to sugars, such as amylose or dextrin, which, when coupled to agarose or sepharose beads, can be used to purify MBP fusion proteins from the E. coli lysate (Diguan et al., 1988).

While the effective increase in solubility and affinity properties are highly favorable, MBP also has less desirable traits, which is why it is not the “perfect” tag. The protein is very large (42 kD) and its presence can interfere with biological activity of the recombinant protein if not removed. Moreover, MBP enhances solubility so strongly that it can even solubilize unfolded/misfolded proteins. In these cases, the target proteins often precipitate after enzymatic cleavage of MBP, resulting in wasted time and resources (Lee et al., 1993; Saavedraalanis et al., 1994; Kishore et al., 1998). In order to verify that MBP facilitates the production of folded, active protein, and does not simply solubilize misfolded proteins, it is advantageous to co‐express the MBP fusion protein with its site‐specific protease. In this system, the MBP fusion protein is proteolytically cleaved in vivo by the co‐expressed protease (Nallamsetty and Waugh, 2006). Proteins/protein constructs for which folding and solubility is enhanced by MBP will remain soluble following cleavage while those that do not will precipitate. Small‐scale expression tests can be used to rapidly identify which constructs benefit from fusion to MBP. Because the MBP tag is proteolytically removed in vivo, it is advantageous to add a purification tag, typically a his‐tag, after the protease site (i.e., MBP–TEV site–his₆tag–protein target).

Common tags: Glutathione‐S‐transferase (GST). GST, a protein that catalyzes the nucleophilic attack of glutathione on electrophilic substrates in order to reduce their reactivity with other biomolecules (Armstrong, 1997), is another commonly used fusion tag. GST facilitates both expression and purification, but is not generally considered a solubility‐enhancing tag as many studies have shown that it is a poor solubility enhancer (Esposito and Chatterjee, 2006; Brown et al., 2008). While GST (∼28 kDa) is smaller than MBP, it dimerizes. Thus, if the target protein also forms oligomers, the use of GST can lead to aggregation and precipitation of the expressed fusion protein, even though the target protein is folded and active (Kaplan et al., 1997). In spite of these disadvantages, GST is still widely used because it binds tightly and specifically to glutathione agarose, allowing the target protein to be purified in a single step (Smith and Johnson, 1988).

Recently developed tags: NusA. Recently, novel solubility tags have been developed with the common aim to enhance the solubility of a diverse set of targets with minimal drawbacks (Chong et al., 1998; Zhang et al., 2004; Butt et al., 2005; Ohana et al., 2009). N‐utilization substance A (NusA), a transcription termination/antitermination factor in E. coli , is one of these new tags. Its utility as a solubility‐enhancing fusion protein was identified through a systematic study of more than 4000 E. coli proteins. In this study, NusA was predicted to have a high solubility probability when expressed in E. coli (>90%), and subsequent experiments demonstrated that it was more effective at promoting solubility than both GrpE and baterioferritin (BFR; Davis et al., 1999). Since then, multiple studies have shown that NusA effectively enhances protein solubility for a diverse set of protein targets (De Marco et al., 2004; Douette et al., 2005; Niiranen et al., 2007); in fact, in some cases it has been shown to be even more effective than MBP (Kohl et al., 2008). However, NusA is not an affinity tag. Thus, it is typically coupled to the his₆‐tag to facilitate purification (de Marco, 2006). Finally, it is also large (55 kDa), which can sometimes make proteolytic removal of the NusA fusion protein difficult.

Recently developed tags: SET. It was also recently shown that small highly acidic peptide tags, which are based on the C‐terminal portion of the T7 phage gene 10B, promote the soluble expression of folded, active protein for highly aggregation prone proteins. These tags are known as solubility enhancing tags (SET tags). Namely, Zhang and colleagues showed that a series of T7 phage gene 10B‐based peptide tags, with net negative charges of greater or equal than −6, fully solubilized the highly aggregation‐prone Ig variable‐type domain, CAR D1 (Zhang et al., 2004). Moreover, it was folded. Significantly, in the same study, a peptide tag with a net charge of −12 partially solubilized the myelin P‐zero protein, P₀ex, a protein that is highly recalcitrant to solubilization (Zhang, Y.B. et al., 2004). An advantage of the SET tags is that they are small, typically <40 amino acids. They are also highly acidic, with net charges between −6 and −18. Although the mechanism by which SET tags improve solubility is still under investigation, it has been suggested that the negatively charged fusion peptide increases electrostatic repulsion between folding‐intermediates (Zhang, Y.B. et al., 2004). This effectively delays polypeptide aggregation, increasing the time available for the target protein to properly fold.

Recently developed tags: SUMO. Another novel and increasingly routinely used tag is the small ubiquitin‐like modifier (SUMO) tag. Its development as a prokaryotic protein expression system was based on the observation that the addition of ubiquitin to recombinant proteins increases their solubility (Butt et al., 1989; Wittliff et al., 1990). The utility of SUMO as both an expression and solubility‐enhancing tag has been demonstrated in multiple studies. One of the first showed that fusion of SUMO to the N‐terminus of GFP and MMP13, a protein highly recalcitrant to soluble expression, dramatically enhanced their expression and solubility (Malakhov et al., 2004). Subsequent studies have shown SUMO also effectively promoted the soluble expression of SARS coronavirus proteins (Zuo et al., 2005b) and membrane proteins (Zuo et al., 2005a). One of the advantages of SUMO is that it is small (11.2 kD). It also has its own, highly specific protease (Ulp), which functions by recognizing the tertiary structure of the SUMO protein, rather than a short, specific amino acid sequence. It cleaves the peptide chain immediately after the C‐terminal residue of the SUMO protein. This is an advantage because, unlike when using other proteases, no non‐native residues are left on the target protein following cleavage (Butt et al., 2005). The only restriction of the SUMO protease is that the N‐terminal amino acid of the target protein cannot be a proline, as this residue restricts access to the SUMO protease active site.

Facilitating the formation of disulfide bonds: export signal sequences. An additional method to produce disulfide bond–containing proteins in E. coli is to export them to the periplasm, the space between the inner cytoplasmic membrane and the external outer membrane of Gram‐negative bacteria like E. coli . The bacterial periplasm is highly oxidative and thus promotes the formation of disulfide bonds (for a review of methods used to produce disulfide containing proteins in E. coli see de Marco, 2009). The E. coli maltose‐binding protein is targeted to the periplasm in vivo via an N‐terminal export signal sequence. Thus, when a construct of MBP that includes the export sequence is used as a fusion tag, the fusion protein is exported to the periplasm and, in turn, the oxidative environment promotes disulfide bond formation (Riggs, 2000). Because MBP is used to target proteins both to the cytoplasm and to the periplasm, it is important to verify that the expression vector used localizes the target protein to the desired cellular compartment. In addition to MBP, DsbA, a Trx homolog, has also been used as a fusion partner to target and promote the proper folding of proteins in the periplasm (Collinsracie et al., 1995; Couprie et al., 2000; Winter et al., 2000). Finally, it is important to point out that proteins targeted to the periplasm have specialized protocols for purification, which differ from those used for proteins targeted to the bacterial cytosol (Malik et al., 2007).

Position of fusion tag. All described solubility and affinity tags can be fused either to the N‐ or C‐terminus of the target protein. However, N‐terminal fusions are the most common and have the added benefit that they often enhance soluble protein expression more successfully than C‐terminal fusions (Sati et al., 2002; Dyson et al., 2004; Busso et al., 2005). One of the reasons N‐ versus C‐terminally fused tags effectively improve protein expression and solubility is because many of the fusion tags are native E. coli proteins. Thus, they provide an efficient 5′ sequence for the initiation of transcription (Esposito and Chatterjee, 2006). Correspondingly, the N‐terminal mRNA sequence and structure is efficiently recognized by the ribosome and compatible with robust translation (Cebe and Geiser, 2006). In contrast, the transcription and translation initiation sites of target genes and proteins are variable and can impede these processes when tags are fused to the target protein C‐terminus. Notably, there are exceptions, as MBP and SET tags have been shown to be equally effective in enhancing solubility regardless of which terminus they are fused to (Dyson et al., 2004; Zhang, Y.B. et al., 2004), although this is still protein dependent. One tag that is often fused to the protein C‐terminus is the his₆‐tag. Typically, it is equally effective at either location, but has the advantage that when fused to the C‐terminus, only fully translated proteins are purified as incompletely translated target proteins do not include the purification tag.

Removal of the fusion tag. All fusion tags have the potential to influence the behavior of the expressed protein, and thus it is typically desirable to remove the tag following purification. This is achieved by engineering a protease‐specific cleavage site (a sequence of ∼7 amino acids that is specifically recognized by the protease) between the tag and target protein. Following expression and purification, the corresponding protease is used to cleave the tag from the protein in vitro.

Several proteases are widely used for fusion tag removal (see Table 4). However, many enzymes, such as thrombin, factor Xa, and enterokinase, can promiscuously cleave the target protein at nonspecific sites (Chang, 1985; Choi et al., 2001). Today, one of the most widely used proteases is tobacco etch virus (TEV) protease. It has a number of advantages. First, it is more stringent and thus efficiently cleaves fusion tags without nonspecific secondary cleavage (Phan et al., 2002). Second, it can accommodate a number of different residues on the C‐terminal side of the sessile bond, eliminating nonnative residues on the N‐terminus of the target protein (Kapust et al., 2002). TEV protease is also active over a diverse set of experimental conditions, such as pH, buffers, and temperatures (Kapust et al., 2001, 2002). A disadvantage is that TEV protease is inhibited by a large number of detergents, making the removal of fusion tags from integral membrane proteins, which require protein‐detergent complexes for solubilization, challenging (Mohanty et al., 2003). If detergents are an essential component of the protein buffer, thrombin protease, which retains its activity in a wide range of detergents, can be used to cleave the fusion tag.

Sometimes a protease will fail to cleave the fusion protein. Most often, this is due to steric hindrance in which the protease site is not accessible to the enzyme (Kapust and Waugh, 2000; Lee et al., 2008). Including a short linker of glycine, asparagine, or alanine residues between the protease site and the fusion protein often alleviates this problem (Esposito and Chatterjee, 2006). Inefficient cleavage may also be overcome by altering the cleavage conditions, such as increasing protease concentrations, extending the cleavage time, and altering temperature, among other parameters.

E. coli BL21 and its derivatives are most frequently used for routine protein expression. These strains are deficient of ompT and lon proteases. OmpT is a bacterial endoprotease that readily cleaves T7 RNA polymerase (Grodberg and Dunn, 1988), while the Lon (La) protease, encoded by the lon gene, is an ATP‐dependent enzyme that rapidly degrades misfolded and recombinant proteins (Phillips et al., 1984). Correspondingly, deletion of lon and ompT is correlated with increased expression and stability of recombinant proteins (Gottesman, 1990). Since, the T7 RNA polymerase system is the most widely used promoter system for protein expression, most BL21 strains, designated (DE3) strains, contain a chromosomal copy of the T7 RNA polymerase gene, allowing for simple and efficient expression of genes from T7‐based expression vectors.

BL21Star (DE3) (Invitrogen), a derivative of the BL21 (DE3) strain, contains an additional mutation in the rne gene (Lopez et al., 1999). This gene encodes RNase E, an enzyme that functions as an essential part of the “degradosome” to actively degrade mRNA within the cell (Grunberg‐Manago, 1999; Lopez et al., 1999; Carpousis, 2007). Consequently, the use of this strain results in an increase in mRNA stability, and, in turn, protein expression. The strain developers have routinely noted a 2‐ to 10‐fold increase in heterologous protein expression when compared to non‐RNase E defective BL21 strains (Invitrogen). However, the basal expression level of the target gene is also increased when using these strains, and therefore, a variant that exhibits repressed basal expression, BL21Star (DE3)pLysS (see description of pLysS strains below), should be used when expressing toxic proteins.

As described in section II, differences in codon frequency between the target gene and the expression host can lead to translational stalling, premature translation termination, and amino acid mis‐incorporation (Kane, 1995). Instead of generating a codon‐optimized gene, this disparity may be overcome by supplying the rare tRNAs during expression (Burgess‐Brown et al., 2008). Numerous bacterial strains (BL21‐RP, BL21‐RIL, BL21‐RPIL, Rosetta, and Rosetta‐gami) are available that contain plasmids that encode rare tRNAs to promote the efficient expression of genes that contain high frequencies of rare codons. The tRNAs that are supplemented in each strain differ, and therefore, the appropriate host strain should be determined on a protein‐to‐protein basis. Additionally, the Rosetta‐gami cell line, which contains trxB/gor mutants, has the added benefit of facilitating cytoplasmic disulfide bond formation (see next section) (Milisavljevic et al., 2009).

As discussed in section III, some host strains (BL21 trxB, Origami, Rosetta‐gami) have mutations in thioredoxin reductase (trxB) and/or glutathione reductase (gor) genes, two proteins that maintain the reducing environment of the bacterial cytosol (Prinz et al., 1997). Consequently, these strains aid the formation of cytosolic disulfide bonds, greatly enhancing the solubility of folded, active disulfide bond–containing proteins produced in the cytosol. Moreover, these strains can also be more effective than exporting the target protein to the periplasm. For example, in one study, a trxB/gor mutant strain (FA113; a suppressor strain that overcomes the normally slow growth of trxB/gor mutant strains in the absence of reductant) increased the active yield of multiple proteins [ E. coli alkaline phosphatase, a truncated form of the human tissue plasminogen activator (vtPA) and full‐length human tPA, which contained two, nine, and seventeen disulfide bonds, respectively] when compared to yields obtained by secretion of the proteins to the periplasm (Bessette et al., 1999). Because inter‐protein disulfide bonds can also form, leading to high‐molecular‐weight oligomers, these strains should only be used to express proteins that require disulfide bonds for proper folding.

The BL21‐AI host strain, in which the T7 RNA polymerase gene is under the control of the araBAD promoter, was developed to allow the expression of toxic genes from any T7‐based expression vector. As described earlier, the low basal activity of the araBAD promoter system is optimal for the expression of proteins that are toxic to the host. In BL21‐AI cells, basal expression of the T7 RNA polymerase, and the subsequent target gene, is highly repressed in the absence of arabinose and the presence of glucose. This cell line has been used for the robust expression of several heterologous enzymes, with soluble yields exceeding 30 mg/liter (Chen et al., 2009; Yao et al., 2009).

As an alternative to the BL21‐AI strain, several BL21 derivatives, known as pLysS strains (also described in section III), express T7 phage lysozyme, an enzyme that effectively inhibits T7 RNA polymerase activity. Inhibition of T7 RNA polymerase, in turn, decreases basal expression of the target protein. Like BL21‐AI strains, pLysS strains have been used to successfully express toxic proteins under the control of T7 RNA polymerase (Jensen et al., 1999; Cinquin et al., 2001).

Typically, protein expression is accomplished using host strains that have a combination of the elements described above. For example, Origami (DE3) pLysS cells have trxB/gor mutations, promoting the production of disulfide‐bonds, contain a chromosomal copy of T7 RNA polymerase for the efficient expression of genes under the control of this enzyme, and express pLysS to decrease basal expression and aid production of toxic target proteins. This strain was essential for the soluble production of BSPH1, a novel human sperm‐binding protein that contains four disulfide‐bonds and is completely insoluble when expressed in BL21 (DE3) cells (Lefebvre et al., 2009b). Therefore, all properties of the protein should be considered when choosing the optimal host strain to express the protein of interest.

Lowering the expression temperature routinely improves the solubility of recombinantly expressed proteins (Shirano and Shibata, 1990; Kataeva et al., 2005; Volonte et al., 2008; Piserchio et al., 2009). At lower temperatures, cell processes slow down, leading to reduced rates of transcription, translation, and cell division (Chou, 2007), while also leading to reduced protein aggregation (Vasina and Baneyx, 1997; Sahdev et al., 2008). Moreover, most proteases are less active at lower temperatures, and thus lowering the expression temperature also results in a reduction in the degradation of proteolytically sensitive proteins (Spiess et al., 1999; Hunke and Betton, 2003; Pinsach et al., 2008).

Because of the profound increase in the yield of soluble protein at low temperatures, it is strongly suggested to use a low induction temperature as the default (see typical protocol in Fig. 2). The bacterial culture should be cultivated at 37°C until mid‐to‐late log phase (Chou, 2007). The culture is then induced and the recombinant protein synthesized between 15°C to 25°C (Peti and Page, 2007; Gråslund et al., 2008a). Due to the reduced protein synthesis rate, longer induction times are necessary to obtain a sufficient protein yield (typical induction time: 4 hr at 37°C, 16 to 20 hr at 18°C).

In addition to lowering the growth temperature, a reduction in transcription rate can also be achieved by lowering the concentration of the induction agent. For example, the araBAD promoter system is titratable, and thus lower concentrations of the inducer, l‐arabinose, result in the production of less protein (Guzman et al., 1995). Additionally, decreasing the concentration of IPTG can also enhance the production of soluble protein. For example, the solubility of recombinant cyclomaltodextrinase (CDase) was shown to be highly sensitive to the concentration of the inducer. When the protein was induced using 0.05 mM IPTG, the protein was soluble and active; however, when the inducer concentration was doubled to 0.1 mM the expressed protein was insoluble and inactive (Turner et al., 2005). Thus, although the most common IPTG concentrations for protein induction range from 0.1 to 1.0 mM, a decrease to even lower levels can effect solubility. Finally, a derivative of the BL21 (DE3) expression strain, containing a lacY1 deletion mutation, has been developed that allows the T7 promoter system to be titratable over a range of IPTG concentrations (Turner et al., 2005). This cell line, called Tuner (DE3), permits accurate tuning of the induction level and can, in some cases, promote the production of soluble protein (Turner et al., 2005).

Although fermentation has its clear advantages, batch culture is the most common method to cultivate cells for recombinant protein expression. Since there is limited control over the growth parameters using this approach, all nutrients that are required for growth must be supplied from the beginning by inclusion in the growth medium. Luria broth (LB) is the standard for the expression of proteins. This broth is composed of bacto‐tryptone, which provides peptides, peptones and essential amino acids, yeast extract, which provides vitamins and trace elements, and sodium chloride, which provides sodium ions to maintain osmotic balance (Sahdev et al., 2008). Terrific broth (TB), the second most widely used expression medium, is formulated to increase protein solubility and yield (Sahdev et al., 2008). M9, or minimal medium, contains the minimum nutrients essential for bacterial species to grow. Minimal medium usually lacks amino acids, and thus is most widely used for selective labeling, i.e., isotope labeling for proteins studied using NMR spectroscopy (Reilly and Fairbrother, 1994; Pryor and Leiting, 1997; Chen et al., 2006).

Expressing proteins whose activities disrupt E. coli growth and/or signaling are often toxic to the cell. While host cells can be cultivated to a high density under repressive conditions, induction of the toxic protein may result in rapid growth arrest and/or cell death. For some proteins, this toxicity can be mitigated when the toxin is coexpressed with a second protein that either binds and inhibits the toxin or does not bind yet counteracts its activity. Both strategies allow the toxic protein to be successfully solubly expressed to high levels. Two families of proteins that are routinely associated with cell toxicity are endogenous E. coli toxins, such as RelE, which inhibits translation through mRNA cleavage (Pedersen et al., 2003), and protein tyrosine kinases (PTK), which can promiscuously phosphorylate nonnative E. coli targets (Wang, Y.H. et al., 2006). However, multiple groups have successfully expressed these proteins in E. coli by coexpressing them with an appropriate protein partner. Thus, if the target protein is toxic to the host cell, coexpression with a protein that antagonizes the toxic effect should be considered.

Toxin‐antitoxin pairs. Toxin:antitoxin pairs (Gerdes et al., 2005) are composed of an unstable antitoxin and a stable toxin. Under normal conditions, the toxin and antitoxin associate to form a tight, nontoxic complex. However, under conditions of stress, the antitoxins are degraded, leading to cell growth arrest due to the cellular effects of the toxin (inhibiting replication by blocking DNA gyrase or, more commonly, inhibiting translation via mRNA cleavage) (Gerdes et al., 2005). One such toxin:antitoxin system is mqsRA, which encodes the MqsR toxin, a ribonuclease, and its cognate MqsA antitoxin (Brown et al., 2009). We showed that overexpression of MqsR alone results in growth arrest, as translation is inhibited via mRNA cleavage by MqsR. However, coexpression of MqsR with MqsA, which binds and mitigates MqsR‐mediated toxicity, resulted in the robust expression of both proteins. In this case, MqsR and MqsA were induced simultaneously using IPTG from separate T7 polymerase‐based vectors (MqsR: pet28A, Kan resistance, pBR322 promoter; MqsA: pAC21a, Amp resistance, pACYC promoter) (Brown et al., 2009). There are also other examples in which other bacterial toxins have been solubly produced in E. coli by coexpression with their cognate antitoxins (Gerdes et al., 2005).

More recently, these toxin‐antitoxin systems have been used as plasmid‐stabilization systems that can effectively increase recombinant protein production levels. While the antitoxin protein is constitutively expressed from a plasmid that encodes the protein of interest, the toxin gene is chromosomally incorporated and highly repressed in the presence of the antitoxin (StabyExpress, Delphi Genetics). If the plasmid is lost from the host cell, the highly labile antitoxin is readily degraded and the activity of the toxin is no longer inhibited, leading to cell growth arrest. This plasmid‐stabilization strategy, using the CcdA/CcdB toxin‐antitoxin pair, often increases the final recombinant protein production levels by three‐ to five‐fold (Stieber et al., 2008).

Kinase‐phosphatase partners. Several, but not all, tyrosine kinases have proven difficult to express in E. coli . The expression of an active kinase is often accompanied by low yield, extensive degradation, insolubility, and a heterogeneously autophosphorylated sample (Y.H. Wang et al., 2006). In contrast, catalytically inactive kinases can be expressed to high yields with little difficulty. Thus, the activity of the kinase is not well tolerated by the host cell. To counteract the negative effects of the kinase, multiple groups have coexpressed tyrosine kinases with an opposing tyrosine phosphatase. For example, Src kinase, a tyrosine kinase, either expresses insolubly or spontaneously mutates to generate an inactive kinase when expressed alone. However, when coexpressed with PTP1B, a tyrosine phosphatase, milligram amounts of soluble, active kinase were produced (Y.H. Wang et al., 2006). In this case, rather than using two different plasmids, an MBP‐PTP1B‐thrombin site‐Src fusion protein was expressed from a single plasmid. Following purification, the protein fusion tag (MBP‐PTP1B) was removed by thrombin cleavage (Y.H. Wang et al., 2006). This kinase‐phosphatase coexpression method has also proven effective for the soluble expression of Met kinase, which was solubly coexpressed with PTP1B in which both proteins were expressed using two compatible vectors (Seeliger et al., 2005), and Abl kinase, which was solubly coexpressed with YopH using a single bicistronic vector (W.R. Wang et al., 2006).

Small bacterial proteins typically fold rapidly, due to fast folding kinetics. However, larger bacterial and heterologous proteins fold more slowly, and thus require protein chaperones and folding catalysts to prevent aggregation and facilitate folding in E. coli (Baneyx and Mujacic, 2004). While molecular chaperones are ubiquitous in the cell, they are rapidly titrated during overexpression of recombinant proteins. Thus, supplementation of these folding‐partners using coexpression methods has proven to effectively aid solubilization of target proteins.

DnaK‐DnaJ‐GrpE and GroEL‐GroES. The DnaK‐DnaJ‐GrpE and GroEL‐GroES systems are the most extensively characterized molecular chaperones in E. coli . Both groups are folding chaperones that bind to solvent‐exposed hydrophobic domains, preventing nascent peptide aggregation, and through iterative rounds of ATP‐driven conformational changes, mediate the folding/unfolding of their substrates (for an extensive review of cytosolic molecular chaperones see Hartl and Hayer‐Hartl, 2002). While the soluble expression of certain proteins is improved by coexpression with DnaK‐DnaJ‐GrpE (Chen et al., 2003), other proteins result in higher yields of soluble protein when coexpressed with GroEL‐GroES (Ayling and Baneyx, 1996; Park et al., 2004; Sahu et al., 2009). Unfortunately, it is not possible to predict which, if any, molecular chaperone system will improve protein solubility. Due to the reduced metabolic burden, it is recommended to test protein expression with the two chaperone systems separately (for a thorough coexpression protocol see Baneyx and Palumbo, 2003).

In our laboratory, we have found that coexpression of protein phosphatase 1 (PP1α), a metal‐dependent serine/threonine phosphatase, with GroEL and GroES results in the maximum yield of soluble protein (Kelker et al., 2009). To ensure ample chaperones are available to aid protein folding at the start of recombinant protein production, GroEL and GroES expression, which is under the control of the araBAD promoter (induced with l‐arabinose), is induced at a culture optical density of λ₆₀₀ = 0.5, while PP1α, which is under the control of the T7 promoter (induced with IPTG), is induced later, at a culture optical density of λ₆₀₀ = 1.0. After 20 hr (10°C), the cultivated cells are pelleted and then resuspended in fresh medium that contains 200 µg/ml chloroamphenical to block all ribosome activity and the cultures are allowed to shake for an additional 2 hr. This last step allows for in vivo refolding and results in a substantial increase in the amount of folded, active PP1α (Kelker et al., 2009).

cpn10‐cpn60. Unfortunately, most molecular chaperone systems display reduced activity at reduced temperatures. For example, while GroEL/GroES is most efficient at 30°C, this system is only 30% active at 12°C. As discussed throughout the review, there are extensive advantages to expressing recombinant proteins at low temperatures. A second system is available that is maximally functional at low temperatures: the cold‐adapted chaperonins Cpn10 and Cpn60 from psychrophilic bacterium, Oleispira antarctica. These chaperonins, which have 74% and 54% amino acid identity with GroEL and GroES, respectively, are effective folding‐modulators at low temperatures (4°C to 12°C) (Amada et al., 1995). Using this system, a temperature‐sensitive esterase expressed at 10°C with Cpn10 and Cpn60 exhibited a 180‐fold increase in activity over expression at 37°C (Ferrer et al., 2004). A derivative of the BL21 host strain, ArcticExpress (Stratagene), that coexpresses these two chaperonins, has been developed and used to successfully express several proteins, including interleukin‐2 tyrosine kinase (Joseph and Andreotti, 2008).