Monoclonal antibodies are probably the most versatile and widespread class of reagents used in biology research, diagnostics, and therapy. Although IgG antibodies recovered from hybridoma cells still dominate the field, alternative formats know a constantly increasing success and recombinant Fab, single-chain (scFv) and single-domain (VHH) antibodies have demonstrated their reliability in basic research as well as in clinical applications [1],[2]. In the case of VHHs, their reduced mass is a decisive advantage to obtain faster clearance for in vivo imaging, a better penetration in solid tumors and even the permeation across the blood brain barrier [3]-[5]. The in vitro selection allows for the isolation of binders for toxic or scarcely antigenic targets as well as for epitopes correlated to specific functions. Whole cells have been successfully used for panning antibody fragments that recognize membrane proteins in their native membrane environment and for identifying new biomarkers [6]. Basic molecular biology techniques allows for VHH fusion to tags and larger carriers to obtain application-optimized reagents [7]. Single-domains can be easily reconstituted into the Camelidae IgG-like format by fusion to a Fc domain and Fc moieties with different characteristics can be selected to tune ADCC and CDC effects in different organisms.

In contrast to conventional antibodies, recombinant antibodies are routinely expressed also in prokaryotic systems. Bacteria can be used to display on their surface antibodies of different format for diagnostic applications [8],[9] and to obtain elevated VHH productions in both the periplasm and the cytoplasm [10]-[13]. In contrast, the yields of reconstituted IgG-like molecules and fusions with some valuable tags remain low [14],[15] due to either structural complexity or different redox requirements of the two partner polypeptides. Recently, it has been demonstrated that the cytoplasmic co-expression of disulfide-bond dependent proteins together with sulfhydryl oxidase and a disulfide bond isomerase increased significantly the production of the target proteins [16]. Such approach proved being effective to improve the cytoplasmic accumulation of full-length VHH-SNAP tag fusions [17]. Nevertheless, no proof of antibody functionality was shown in this preliminary communication. Now we demonstrate that several VHH-based constructs as complex as the IgG-like reconstituted VHH-Fc antibodies can be produced in bacterial cytoplasm at elevated yields and preserve their complete functionality. This opportunity represents a time and cost effective alternative to the conventional expression of IgG antibodies from hybridoma cells. Furthermore, it allows the production of fusion molecules such as the VHH-SNAP or VHH-GFP constructs that are difficult to obtain in oxidizing environments.

The in vitro recovery of recombinant antibodies represents an effective and rapid alternative for isolating binders against any antigen class. Furthermore, the access to the antibody sequence simplifies molecular engineering and opens the possibility to fuse suitable tags to the antibodies to develop them in reagents optimally suited to different applications. Such fusion constructs are often easy to produce in bacteria and we designed a collection of vectors for the parallel expression of application-friendly VHHs production in both bacterial periplasm and cytoplasm (Additional file 1: Figure S1). In a preliminary expression test we noticed that tags could significantly modify the antibody stability and yields. SNAP/CLIP and GFP were poorly folded and were prone to aggregate when expressed in the periplasm whereas the presence of trans disulfide bonds in structurally complex proteins such as alkaline phosphatase, peroxidases, and enzymatic toxins seemed to be compatible only with periplasmic expression. We designed a decision chart with the aim of optimizing the cytoplasmic expression of fusion antibodies that either failed to be expressed in the periplasm or, such as the Fc-fusions, accumulated in low amounts (Additional file 2: Figure S2). Bacterial mutants in which the cytoplasmic reducing metabolism is impaired have been sometimes successfully used to express disulfide-dependent proteins but the results are contradictory specially when molecules with multiple disulfide bonds must be produced [11],[18]. Therefore, we evaluated the promising alternative of inducing the formation of disulfide bonds in the cytoplasm of wild type bacteria in which a recombinant eukaryotic sulfhydryl oxidase is accumulated [16]. The expression of the oxidizing enzyme is triggered by arabinose addition to the culture medium and can be induced independently and before the expression of the antibodies by means of IPTG addition.

By panning a pre-immune library, we isolated several VHHs specific for the ectodomain of the human HER2 receptor (data not shown) and expressed three of them (A10, C8, G3) using both the original periplasmic phagmid vector (pHEN2) which allows for the fusion to a His-myc tag and a cytoplasmic vector (pSNAP) for obtaining a His-SNAP fusion. The constructs were purified by IMAC and successively by gel filtration for separating aggregates and polymerization products (Additional file 3: Figure S3a). The yields of the SNAP-fusions were significantly higher than those of the myc-fusions (Table 1) and the degradation products/aggregates represented a negligible percentage of the affinity-purified antibodies (Figure 1 and Additional file 3: Figure S3a). We then compared the performance of the constructs expressed in the periplasm (His-myc, per) and cytoplasm (His-SNAP, cyt) at the FACS. In a first experiment, the detection was obtained in both cases using an anti-His labeled secondary antibody (indirect labeling). The A10-SNAP construct discriminated the HER2-positive SKBR3 and MCF10A negative cell populations even better than the A10-myc (Figure 1). However, the advantage of the SNAP tag is that it can be covalently labeled before immune detection (FACS or IF) enabling the direct detection of the antigen. The construct A10-SNAP coupled to the chromophore SNAP-Surface 549 successfully identified its antigen with no signal loss in FACS (Figure 1). The A10 clone fused to SNAP showed specificity in cell ELISA and in immunofluorescence when HER2-positive SKBR3 and HER2-negative MCF10A cells were compared (Figure 1). A10 myc and SNAP constructs showed comparable SPR-measured affinities for the purified ectodomain of HER2, a further clear indication that the cytoplasmic expression did not decrease the functionality of the recombinant antibody fused to the SNAP tag (Figure 2a). This first set of results demonstrated that the cytoplasmic expression of VHH-SNAP fusions in the presence of sulfhydryl oxidase enabled to produce functional antibodies the binding capacity of which was at least as good as that of the same VHHs expressed in the periplasm, but with significantly higher yields. Similarly, we successfully expressed, labeled, and validated by FACS a scFv-SNAP construct (anti-claudin-3, Additional file 4: Figure S4a), the A10 VHH fused to the tags CLIP and HALO (Additional file 4: Figure S4b), and produced a VHH-GFP fusion protein that provided a clear signal shift in flow cytometry without the necessity of using primary/secondary commercial antibodies (Figure 2b). Wild type GFP cannot be exported to the periplasm by the SEC pathway in an active form [19] and, although some fluorescent protein variants are able to accumulate in this oxidizing cell compartment necessary for the formation of the antibody disulfide bonds, the choice remains limited to few molecules [20],[21]. Here we demonstrate the feasibility of an approach to produce in the bacterial cytoplasm fluorescent immune reagents that should be not restricted to few specific sequences.

	A10 mg/L	C8 mg/L	G3 mg/L
Myc per IMAC	1-5	1-5	<1
Myc per SEC	1-2	<1	<0.5
SNAP cyt IMAC	15-20	20-25	10-15
SNAP cyt SEC	15-20	20-25	10-15
SORT-Tag cyt IMAC	6-8	4-7	4-7
SORT-Tag cyt SEC	4-6	3-5	3-5
SORT-Tag per IMAC	<0.1	<0.1	<0.1
Fc cyt IMAC	25-30	10-15	15-20
Fc cyt SEC	25-30	10-15	10-15

The recombinant antibodies were cloned in vectors that promote either periplasmic (per) or cytoplasmic (cyt) accumulation and folding, expressed in bacteria, and finally purified by metal affinity chromatography (IMAC) followed by gel filtration (Size Exclusion Chomatography - SEC). The yields are expressed in mg/L of culture and indicate the range obtained in at least three independent purification procedures.

In these first examples of cytoplasmic antibody expression, no direct comparison between the same construct expressed either as cytoplasmic or secreted protein was possible because we did not manage to recover soluble VHH-GFP and VHH-SNAP fusion proteins from the bacterial periplasm. Therefore, we used another tag that could be expressed in both bacterial cell compartments to obtain comparable samples. The SORT-tag is a short peptide (LPTEG) that allows for the sortase-dependent covalent conjugation to N-terminal available glycine residue(s) [22] and has been used for obtaining fusions between several classes of molecules [23],[24]. Surprisingly, the presence of this short tag instead of the conventional myc strongly decreased the solubility of all three tested anti-HER2 antibodies when they were expressed in the periplasm (Table 1). In contrast, the cytoplasmic yields were elevated and, in terms of molarity, comparable with those obtained using SNAP as a fusion tag (Table 1). The SORT-tagged antibodies produced in the periplasm and in the cytoplasm labeled at similar intensity HER2 pos and HER2 neg cells in FACS experiments, with a reproducible stronger signal for the cytoplasmic constructs (Figure 2c). Unexpectedly, the cytoplasmic A10-SORT construct showed higher SPR-measured affinity for its antigen than the control A10-myc construct recovered from the periplasm (Figure 2a). The difference was totally due to the improved association rate. This result confirmed the binding functionality of the construct expressed in the cytoplasm and suggests that despite having similar overall mass, the relative length and/or folding of the different tags can influence the accessibility and actual binding of the two constructs to the antigen epitope, as already noticed by other authors [25],[26]. We cannot rule out that the small observed differences in terms of antibody binding capacity were due to minimal stability variation of the different constructs. If this is the case, cytoplasmic expression resulted more convenient than the periplasmic production of the same construct because it enabled both higher yields and slightly but repeatedly measured "improved" functionality.

Recombinant antibodies with tags such as SNAP/CLIP/HALO/GFP and SORT are suitable for several attractive biological applications, but still represent a niche in comparison to IgG antibodies. In some cases, introducing new protocols based on non-conventional antibody formats for performing immunological analyses might be demanding for the lab management. Therefore, it can be more practical to reconstitute antibody fragments into IgG-like molecules that can be used and detected as conventional monoclonal antibodies. Consequently, we expressed in the cytoplasm the same anti-HER2 set of antibodies used in the previous experiments but as fusions with a rabbit Fc domain. This tag offers some specific advantages such as the possibility to affinity purify the molecules using Protein A/G, to increase the avidity for the antigen by inducing dimerization, and to perform multiple parallel staining by exploiting Fc domains issued from different organisms [27].

All the constructs were well expressed (Table 1) and highly pure already after IMAC step (Additional file 3: Figure S3b). Size exclusion chromatography (SEC) confirmed that the antibodies were monodispersed (Additional file 3: Figure S3b). The functionality of the bivalent Fc-VHHs was confirmed by cell ELISA and FACS. In this last case, the reconstituted IgG-like molecules were at least as effective as the therapeutic antibody trastuzumab (Figure 3). Although VHHs usually bind to conformational epitopes that are lost after antigen denaturation, C8 recognized its antigen by western blot in a total cell lysate and immunoprecipitated the full-length as well as many of the shedding products [28] of the HER2 ectodomain (Figure 3). These results indicate that the epitope recognized by the binder is probably very basal and that C8 could be useful for evaluating the metal protease-dependent HER2 isoform distribution in different cell populations. The IF staining pattern on HER2 positive and negative cells was identical for the monoclonal trastuzumab, the cytoplasmic produced IgG-like A10-Fc construct, and the monovalent periplasmic produced A10-myc construct (Figure 4).

The measurement of the apparent affinity (avidity) of bivalent constructs confirmed that the tag characteristics can influence the antigen/antibody binding, as previously observed for the monovalent constructs (Additional file 5: Table S1). The avidity effect enabled the construct A10-Fc to bind its antigen one log stronger (Figure 5) than the monovalent constructs (Figure 2c). At the same time, there is a substantial difference between the avidities of the Fc and alkaline phosphatase (AP) fusions (Figure 5). Although expressed in the periplasm, the dimer A10-AP bound HER2 with a K _D in the lower nanomolar range, a log worse than the A10-Fc construct produced in the cytoplasm (Figure 5). This result confirms that functional binding capacity is not a simple matter of active domain number [29].

Finally we evaluated the capacity of the IgG-like reconstituted antibodies produced in the bacterial cytoplasm to target a HER2-positive tumor in vivo using a xenografted murine model. C8 was preferred because it was the only of the three characterized antibodies that worked in both WB and IP and, therefore, we expected that it could recognize its antigen in a wider range of structural conformations. Animals were treated with infra-red labeled anti-HER2 C8-Fc and trastuzumab (positive control), as well as with an anti-GFP antibody (negative control). Tumors were recovered after both 3 and 48 hours, sectioned along the median axis to have access to the inner section, and the emitted infrared fluorescence was recorded (Figure 6). The mass (80 kDa) of reconstituted IgG-like Abs impairs direct kidney filtration and the rapid clearence specific of monovalent VHHs (14 kDa) [5],[30]. Consequently, C8-Fc accumulated in the tumor tissues with the same kinetic of the positive control trastuzumab. Although fluorescence does not allow for accurate quantitative comparisons, the recombinant antibodies seem as effective as trastuzumab in concentrating inside the tumor. The tumors treated with the control anti-GFP recombinant antibody did not show any fluorescence signal at both 3 and 48 hours.

Single-domain antibodies can be produced at elevated yields and their structural stability allows for the accumulation of functional molecules even in the cytoplasm of wild-type bacteria [13]. Here we show that even formats as complex as Camelidae IgG-like structures can be produced rapidly and cost-effectively in bacteria. Functional IgG recombinant antibodies have been already successfully produced in bacterial periplasm in the past [14] but yields of correctly assembled full-length molecules rarely exceeded 1 mg/L when shake flasks were used. This condition limited antibody discovery programs that rely on the availability of large batches of antibodies. Recently, engineered systems based on mutagenized over-expressing strains and optimized plasmids succeeded in yielding up to 4 mg/L of IgGs [15] and similar yields were obtained expressing scFv-Fc fusions in Leishmania tarantolae[31]. The protocol described in this article shows that it is possible to increase the yields to tens of mg/L and, according to recently published data [13], there is still the possibility to enhance the yields per liter of culture by increasing the cell density using optimized media and fermentation conditions. However, the main achievement of our approach is the demonstration -obtained by means of several independent techniques [32]- that the functionality of the IgG-like constructs is totally preserved when they are purified after expression in the cytoplasm of bacteria co-expressing sulfhydryl oxidase and the bacterial protein disulfide isomerase DsbC. The combined oxidative and isomerase activities apparently prevent the accumulation of non-functional antibodies. When the comparison was possible, the constructs expressed in the cytoplasm showed both lower aggregation and degradation than those measured in the periplasm. This approach, initially proposed by Ruddock's group for producing disulfide-bond dependent secreted proteins [16],[33], perfectly fits to antibody fragments fused to tags with different folding requirements [17],[34]. The bottleneck represented by the saturation of the SEC-translocon [35] would explain the remarkable higher yields obtained when the antibodies were produced in the cytoplasm rather than in the periplasm. Lately we expressed tens of other VHH-Fc constructs using both rabbit and human Fc and clearly remarked yield difference among antibodies. This condition is probably due to the intrinsic stability of the different sequences but the cytoplasmic expression was constantly more productive than the periplasmic.

The present trend in both academic and industrial labs seems to be focused on using single domain antibodies from Camelidae and human origin as well as Fab fragments to identify new reagents and therapeutic molecules [36],[37]. We believe that our approach might represent a break-through technology for researchers who need to produce significant amounts of antibodies and antibody-based reagents for preliminary in vitro and in vivo characterization. We produced the described antibodies in wild type E. coli that cannot provide eukaryotic-like glycosylation but our results should encourage expressing IgG-like constructs in modified bacteria [38] with the aim of obtaining antibodies with native post-translational modifications.