NaLi-H1: A universal synthetic library of humanized nanobodies providing highly functional antibodies and intrabodies

1Antibodies expended as the biochemical tools of choice to label antigens in cells or tissues. Over the past 20 years, recombinant methods have been developed to quickly select and improve monoclonal antibodies from highly diverse libraries. Recombinant antibodies can be selected from immune or naïve libraries. Immune libraries provide in general high affinity binders but, depending on the antigen, diversity is sometimes limited. Because natural antibody selection requires animal immunization, very conserved or toxic antigens should be avoided and, in general, only limited control of the immune response is possible. On the contrary, non-immune (naïve) libraries provide a higher diversity of binders even for antigens highly conserved in mammals, but high specificity and affinity can be reached only when selecting from very large functional libraries. Immune and naïve libraries are based on the manipulation of antibody fragment that retain binding capacity and specificity of the entire immunoglobulin G (IgG).

2The smallest IgG portion capable of binding with high specificity an antigen is the Fv fragment consisting of the variable heavy (VH) and the variable light (VL) domains. In the case of single domain antibodies (sdAb), a VH or a VL alone, is able to bind its target antigen. Each variable domain contains four conserved framework regions (FRW or framework) and three regions called CDR (Complementarity Determining Regions) corresponding to hypervariable sequences which determine the specificity for the antigen. VH and VL can be fused together using a synthetic linker and produced as a single protein in the form of a single chain Fv (scFv). Easier to manipulate, they can be produced in several bacteria or eukaryote cell types, fused to various tags or functional domains. Interestingly, antibodies called HCAb in Camelidae or IgNAR in sharks have an antigen recognition part composed of only a VH domain. Camelid natural single domain VH, referred to as VHH or nanobodies, can be expressed as recombinant fragments and represent attractive alternatives over classical antibody fragments like scFvs because they are easy to manipulate and they are not limited by potential misfolding of the two domains. It is noteworthy that VHH FRWs show a high sequence and structural homology with human VH domains of family III and VHH have comparable immunogenicity as human VH. Thus, they further constitute very interesting agents for therapeutic applications, some of them are currently in phase II and Phase III clinical trials (Ablynx Nanobodies; http://clinicaltrials.gov/ct2/results?term=ablynx).

3Recombinant antibody fragments allowed not only to accelerate the identification of unique binders, but also the development of a novel type of tool: in this case, the antibodies are directly expressed in living cells as intracellular antibodies (intrabodies), to trace or perturb endogenous target at the protein level. Some scFv or sdAb have indeed been directly expressed in eukaryotic cell as intrabodies to target with high specificity intracellular antigens. Several intrabodies have been used as fluorescent protein fusion to highlight endogenous antigen in cells in a spatio-temporal manner. Intrabodies with intrinsic blocking activity have been reported, and several other approaches have been developed to allow a larger fraction of intrabodies to be used as inhibitory factors: forced co-localization, suicide through proteasome targeting, rerouting or sequestration to cell compartment, degradation. Depending on the target, such inhibitors may have potential in human therapy. Production of functional intrabodies depends on the stability of the antibody fragments in the reducing environment of the cytosol that does not allow disulfide bond formation between conserved cysteine. In this context, many advantages of the nanobody scaffold have been reported and, in particular, higher solubility, improved stability in a reducing environment, as well as higher expression yield and thermostability. For all these reasons, the nanobody scaffold represents an attractive option to generate functional intrabodies.

4Thus, we decided to create a non-immune recombinant antibody library of high diversity, based on a nanobody scaffold that would enable efficient in vitro antibody selection against virtually any antigen. Such a library should provide antibodies usable in conventional immune assays and be enriched in antibodies active in the intracellular environment. First, using a fusion assay in E. coli, a family of highly functional VHH scaffolds was isolated, optimized for intracellular expression and high stability. One particularly stable VHH scaffold consensus sequence was chosen from these selected antibodies. Additional changes were then introduced to reduce the distance between the Camelidae and human VH3 sequences. We confirmed by CDR grafting that this humanized synthetic scaffold (hs2dAb) was robust and functional. Statistics of amino-acid diversity in the CDRs were computed and these information were used to construct a high diversity phage display library of 3.109 independent hs2dAb. The library was then screened against diverse targets of various structures and origin. Highly specific antibodies were selected against EGFP, mCherry, β-tubulin, β-actin, heterochromatin protein HP1α, GTP-bound RHO, p53 and HER2. Affinity measurement indicated that affinities in the nM range can be obtained using this library. As expected from our design, we further showed that hs2dAb are frequently usable as fluorescent intrabodies to track antigens in cells. We also showed that they can be functionalized to induce antigen knockdown. This thus represents the first report of a large and diverse synthetic single domain antibody library enabling fully in vitro selection of highly functional antibodies and intrabodies.

5We reasoned that the usual lower quality of these non-immune libraries may come from (1) an antibody scaffold that may not allow robust folding and presentation of CDR region, (2) a lack of control of diversity in the CDR regions and (3) a frequent occurrence of incorrect clones due to the presence of unexpected mutation or empty clones. We designed a pipeline for the development of functional synthetic libraries that aims at overcoming these limitations (Figure 1A). As a first step to construct a single domain antibody library enriched in highly stable and functional antibody fragments, we screened for a robust sdAb scaffold (Figure 1—figure supplement 1). Previously, we selected from immune or naïve llama VHH libraries several hundreds of clones. From this population, we identified a set of robust scaffolds using an assay that discriminates highly stable clones from clones prone to aggregation, or unfolding, in the bacterial cytoplasm. This assay is based on the fusion of HA-tagged chloramphenicol acetyl transferase (CAT) to the carboxy-terminus of VHH sequences (Figure 1—figure supplement 1A). In these conditions, only bacteria expressing a functional VHH fusion in the reducing cytosol (non aggregating, non degraded) can grow in the presence of high antibiotic concentration, thus filtrating a sub-library of potential intrabodies. Expression yield in E.coli and apparent solubility as EGFP fusion in the mammalian cell cytoplasm were further assessed to select a set of robust antibody scaffolds. Strikingly, the consensus scaffold was matching the sequence of the most robust VHH framework, represented by a single domain antibody D10 (hereafter named sdAbD10). When compared to previously reported thermostable nanobodies or intrabodies obtained from immune libraries, sdAbD10 was found to provide higher antibiotic resistance in the chloramphenicol filter assay (Figure 1—figure supplement 1A). Its expression yield in E.coli periplasm was in the higher range of soluble llama VHH fragments, allowing efficient and quantitative purification (Figure 1—figure supplement 1B). Purified sdAbD10 showed excellent solubility, stability after treatment at 70°C and we did not observe aggregation when expressed as an intrabody fused to the EGFP in mammalian cells (Figure 1—figure supplement 1C).

6To test whether the sdAbD10 scaffold composed of the FRW 1 to 4 was robust independently on the CDR sequences, we grafted the CDR loops of the lam1 VHH directed against laminB into the framework of the sdAbD10 (sdAbD10-anti lamin). In parallel, partial humanization of the scaffold was also tested to figure out whether it affected dramatically its intrinsic properties. Seven residues of the synthetic scaffold sdAbD10 were altered towards the most represented in human VH3 while five other divergent residues were kept unchanged. The four llama VHH-specific amino acids hallmarks in the framework-2 region (positions 42, 49, 50, and 52), which are essential to increase intrinsic solubility properties, as well as the Glutamine in position 103, were maintained. We named the resulting hybrid single domain VH hs2dAb (humanized synthetic single domain antibody) (Figure 1B). After grafting the CDR of the anti-lamin into these two scaffolds, we analyzed their display on the M13 surface (Figure 1—figure supplement 2A) and their production by E. coli and by CHO cells (Figure 1—figure supplement 2B). Both scaffold showed similarly good efficacy. Importantly, both the sdAbD10 and hs2dAb scaffolds enabled functional display of the grafted CDRs and robust detection of endogenous lamin was observed by immunofluorescence staining of Hela cells (Figure 1—figure supplement 2C). Last, Rothbauer et al. showed that the original anti-lamin VHH recognized its target antigen upon intracellular expression. Similarly, we observed that both the hs2dAb and the sdAbD10 scaffolds allowed the efficient intrabody use of the synthetic anti-lamin antibodies (Figure 1—figure supplement 2D). This further indicated that the synthetic scaffolds were robust, non-aggregating and resisted to the reducing conditions found in the cytosol, while allowing the proper display of CDR loops. As partial humanization did not affect the properties of the sdAbD10 scaffold, we chose the hs2dAb scaffold as a unique framework to construct a diverse library of synthetic nanobodies endowed with the characteristic stability of these single domain antibodies while displaying an amino acid sequence closer to human VH3.

7A synthetic diversity was introduced in the three CDRs by rationally controlling each position of the CDR1 and CDR2 using a set of amino acids that recapitulates partially natural diversity (see the Appendix for more details) while reducing the presence of the most hydrophobic residues in order to avoid the aggregation propensity (see Material and methods). A constant length of 7 amino acids was selected for CDR1 and CDR2. A large spectrum of Camelidae VHH CDR3 length is naturally observed and this loop is known to contribute strongly to antigen binding selectivity. Thus, we chose to use four different lengths of CDR3 to cover this spectrum (9, 12, 15 or and 18 amino acids) and introduce random amino acid (except cysteine) at each position.

8Synthetic DNA was produced by the trinucleotide DNA assembly and amplification was carried out starting from 2.1011 different molecules, using only a few cycles of PCR (PCR linearity validated by Q-PCR) to prevent drift during amplification. The synthetic library was inserted into a modified pHEN2 phagemid vector containing a triple myc-tag and suicide gene (ccdB) that allowed positive selection of insert-bearing clones. Massive electroporation was carried out using E. coli TG1 cells and 430 large agar dishes (140 mm) were used to ensure proper plating of the library. About 3.109 individual recombinant hs2dAb clones were obtained. We named this library NaLi-H1 (for Nanobody Library-Humanized 1). We first evaluated the quality of the NaLi-H1 library by sequencing 315 random clones. Only 13 sequences were found to be incorrect (bearing an in-frame stop codon, missing one base, missing a large region [the CDR1 or CDR1-FWR1- CDR2], or being empty). Thus, a total of about 4% of potential default was observed, which is rather low and only marginal in comparison to the 3.109 clones obtained. The diversity was then evaluated by sequencing 5.6 105 inserts using ion Torrent chips (Life Technologies). This confirmed the quality of the library and showed that the four CDR3 lengths (9, 12, 15 or and 18 amino acids) were present in similar proportions. The diversity and statistical distribution of amino acids in the CDRs were found to be as expected (Figure 1—figure supplement 3). To estimate the overall folding of the hs2dAb present in the library, we picked randomly 24 clones and tested their solubility in bacteria medium after secretion. Medium were centrifuged at 250 000 g and the supernatant and pellet were analyzed. All tested clones showed essentially complete solubility, at least as good as a natural Lama antibody (GFP4). Even after warming at 90°C for 10 min, the hs2dAb showed good solubility (over 70%) (Figure 1—figure supplement 4).

9The NaLi-H1 library was screened against a set of various antigens. Several standard phage display methods were used (see Materials and methods for details): antigen adsorption on immunotube, native antigen captured on beads, direct selection at the cell surface. All conditions allowed the efficient recovery of diverse and functional antibodies.

10As a first screen to evaluate the quality of the library, we chose to select specific binders for the EGFP and mCherry fluorescent proteins. NaLi-H1 phages were panned against biotinylated EGFP or mCherry and 3 rounds of selection were carried out. Eighty clones were analyzed for each screening campaign. From the panning against EGFP, 37 non redundant nanobodies were shown to detect EGFP by phage ELISA. These antibodies were then used for immunofluorescence and 10 of them were found to efficiently stain EGFP in fixed HeLa cells (Figure 2A). Similarly, selection against mCherry led to 6 positive binders (Figure 2B). As shown in Figure 2A and B, no staining was obtained in untransfected cells.

11In the next screen, two highly and constitutively expressed components of cell cytoskeleton, tubulin and β-actin, were targeted. Antibodies against tubulin were selected in native conditions using commercial biotinylated tubulin (Cytoskeleton). After two rounds of selection, 3 out of 40 clones analyzed were shown to detect endogenous tubulin by immunofluorescence (Figure 2C). As expected, staining was lost in cells treated with the microtubule destabilizing drug nocodazole, with only a few stable microtubules being labelled in these conditions (Figure 2—figure supplement 1A). This antibody was recently used to stain microtubules by super-resolution imaging. Among phage display methods, selection on antigens directly adsorbed on the surface of immunotube is often used as a cheap and straightforward method, despite the low capacity and the strong denaturation imposed by non-specific adsorption. A screen against coated β-actin led to the identification of 16 unique binders positive in phage ELISA. Seven of these antibodies decorated endogenous actin stress fibers by immunofluorescence in MRC5 cells (Figure 2D) as well as in other cell lines. Treatment of cells with cytochalasin D disorganizes actin fibers. Accordingly, staining with the hs2dAb was strongly altered (Figure 2—figure supplement 1B). Three of the hs2dAb detected a single band at the molecular weight of β-actin in western blot from HeLa cell extract while one of the anti-tubulin detected a band at the correct molecular weight of tubulin (Figure 2—figure supplement 1C).

12Actin and tubulin are strongly expressed cellular proteins. Another screen was thus performed to select binders directed against proteins expressed at a lower level. A first screen was carried out against the tumor suppressor p53 protein. The 83 first amino acids of the NP_000537.3 isoform were produced in bacteria fused to a SNAP tag, biotinylated in vitro and used as a target in the phage display selection. Among 12 clones positive in phage ELISA, 6 selected hs2dAb were shown to label endogenous p53 in immunofluorescence on A431 cells (Figure 2E). The specificity of the staining was confirmed using RPE-1 cells (Figure 2—figure supplement 2). A low nuclear staining was observed in normal conditions, with some variability between cells. As expected, the intensity was enhanced upon UV-induced DNA damage. Such an increase was not observed in a RPE-1 cell line stably expressing an shRNA against p53. A second screen was carried out against the heterochromatin protein HP1α. HP1α was produced in bacteria fused to an avitag to obtain a biotinylated recombinant protein. Biot-HP1α was then immobilized on streptavidin beads and used as a target for 3 rounds of selection. 5 individual hs2dAb were directly identified by immunofluorescence staining of HeLa cells. Figure 2F shows that selected antibodies were efficiently staining endogenous HP1 in the nucleus. Overexpression of the different HP1 variant followed by western blot and immunofluorescence analysis actually suggested that HP1β/γ were also detected by the antibody (Figure 2—figure supplement 3). Together these results showed that the NaLi-H1 synthetic library can be screened in various conditions against very different purified targets while leading to the rapid identification of diverse and specific binders that can be used in classical antibody-based staining methods.

13One of the main advantages of full in vitro immunization using display technologies is the control of antigen conformation and concentration. It allows to drive selection towards the desired outcome. For example, selection schemes can be devised to improve the recovery of high affinity binders endowed with low off-rate kinetics, to target specific epitopes, or to identify conformation sensitive-binders. Recombinant antibody fragment library screening have, for example, provided several binders targeting selectively the active conformation of GTP binding proteins. We hypothesized that the NaLi-H1 synthetic library had enough diversity and functionality to enable the identification of selective conformational binders. A subtractive panning was performed to select conformation-specific antibodies directed against small GTPases from the RHO subfamily. Small GTPases are molecular switch that cycle between an inactive and an active state when bound to GDP or GTP nucleotides, respectively. Mutant of small GTPases can be designed that adopt stably an active or inactive conformation. A constitutively active (CA) mutant RHOA L63 was expressed in HEK293 as a bait then freshly pulled down for panning to preserve its native conformation. To enrich in phage specific for GTP-bound RHOA, a depletion step was introduced from the second round of panning using GDP-bound RHO proteins. After four rounds of selection, clones were analyzed using phage ELISA against either wild type RHOA loaded with GTPγS (a non-hydrolysable analogue of GTP) or GDP-loaded RHOA. Forty clones presenting a differential ELISA signal in favor of the GTP loaded RHOA were sequenced. One antibody, represented by clone H12, represented more than 50% of the population. We analyzed H12 binding specificity by ELISA on several purified RHO proteins expressed as GST fusion in E.coli. H12 recognized the constitutively active mutant RHOA L63 (RHOA-CA) which is bound to GTP due to impaired hydrolysis activity. A similar signal was obtained with wild type RHOA loaded with the non-hydrolysable GTP analogue GTPγS. In contrast, no binding was observed to the dominant negative RHOA N19 mutant RHOA-DN nor to GDP-loaded wild type RHOA (Figure 3A). The capacity of H12 to specifically immunoprecipitate GTP-loaded RHOA from mammalian cell extracts was then evaluated in comparison to the standard method to assay RHO activity. This pull down method is based on the RHO binding domain of Rhotekin fused to GST (GST-RBD) which is known to bind to the active conformation of RHO GTPase. The hs2dAb H12 bearing a carboxy-terminal CBD (Chitin-Binding Domain) was expressed in E. coli and immobilized on chitin beads. These beads were then incubated with HeLa cell extracts pre-treated with either GTPγS or GDP to load small GTPases with the respective nucleotide. The H12 hs2dAb was found to be highly selective of RHO loaded with GTPγS as it was unable to precipitate RHO from GDP loaded extract (Figure 3B). A similarly strong conformation-specificity was found when using H12 for immunofluorescence staining (Figure 3C). HeLa cells expressing the GFP-RHOA constitutively active mutant carrying the mutation Q63L (GFP-RHOA CA) or the dominant negative mutated T19N (GFP-RHOA DN), were fixed and stained with H12 hs2dAb. Expression of the dominant negative mutant GFP-RHOA DN, did not lead to an increased signal over the background of un-transfected cells. In contrast, a staining with H12 was correlated with the level of GFP-RHOA CA mutant expression. Note that the signal does not fully overlap GFP fluorescence and appeared stronger at the cell border and in large zone where cell shape is strongly retracted by large bundled actin stress fibers induced by a sustained activation of the RHOA/ROCK pathway. This CA mutant, like active mutant of many small GTPases related to RAS still need to be activated by guanine nucleotide exchange factors to be loaded with GTP and display the active conformation. Thus, we believe that the H12 staining revealed the active form of this mutant in cells (Figure 3C). All together, these results demonstrated that the H12 hs2dAb is selective for RHO GTPases in their active conformation, highlighting the performance and diversity of the NaLi-H1 library.

14The use of antibodies to target cells involved in pathologies like cancers or viral infection is one of the most promising therapeutic approach. Antibodies also represent a unique tool to identify novel targets at the cell surface. Using synthetic libraries like the NaLi-H1 library to carry out direct selection against the cell surface of tumor cells would strongly accelerate the identification of specific antigens while allowing extended control over the conditions of selection. A subtractive selection scheme was set up to identify antibodies selectively detecting the surface of breast tumor cells: phages displaying hs2dAbs were first depleted against a reference cell line before being selected against the target one (Figure 3D). As a target cell line, we used the SKBR3 line, which is known to overexpress the HER2 cell surface protein, while the MCF7 cell line, negative for HER2, was used to pre-adsorb the library. After the third round of bio-panning, 88 clones were analyzed by FACS and 58 were found to be positive when tested on SKBR3 cells and negative on MCF7 cells. Sequencing the 58 positive clones revealed that 15 independent binders had been selected. Although the subtractive selection was not performed on identical cell lines expressing or not HER2, we observed strikingly that 12 clones out of 15 recognized HER2 exoplasmic domain by ELISA using HER2-Fc as a target antigen (Figure 3E), suggesting that it behaves as a dominant differential epitope. As shown in Figure 4, these antibodies efficiently detected HER2 at the cell surface by immunofluorescence (Figure 3F) or by FACS (Figure 3G). These experiments demonstrated that the NaLi-H1 library will represent a unique tool to discover, in a rapid and cost effective manner, specific antibodies detecting antigens present at the surface of pathological cells. These antibodies may then be used to identify the corresponding target.

15Various antibody fragments have long been proposed to represent powerful tools when expressed in cells as intrabodies. Although several studies indeed report efficient use of intrabodies (reviewed in), this is limited to antibody scaffolds that resist to the reducing environment of the cytosol. We evaluated the use of the hs2dAb scaffold to develop intrabodies. Randomly chosen hs2dAbs were fused to a fluorescent protein and observed in living cells. In comparison to our previous experience with scFv or nanobody libraries in which a majority of tested antibodies showed aggregation when expressed in the cytosol as an EGFP-fusion, most of the hs2dAbs tested here gave a monodispersed fluorescence, and very few were showing aggregates. Monodispersed fluorescence of the EGFP has been proposed to be a convenient indicator of stability and potential use as intrabody. Several hs2dAb directed against actin, tubulin, EGFP or p53 were tested for their ability to trace intracellular antigens in living cells. None of the anti-tubulin or anti-actin antibodies tested were found localized on microtubule or microfilament, respectively, in living cells. We reasoned that this poor efficiency may be linked to the antigen denaturating conditions used during the selection of antibodies directed against actin. The existence of a large pool of unpolymerized actin and tubulin may also prevent efficient recruitment on the polymers. In contrast, several anti-GFP and anti-mCherry hs2dAbs were found to efficiently label their targets, like for example GFP-Rab6 or Myr-Palm-mCherry fusion proteins, in living cells (Figure 4A). Similarly, the anti-p53 hs2dAb fused to mCherry were clearly accumulated in the nucleus of SKBR3 cells where endogenous p53 is also localized. This signal was enhanced in cells overexpressing p53-EGFP (Figure 4B). This further confirmed the binding specificity of the hs2dAb anti-p53 while expressed in the reducing cytosol. Effective intrabodies were also obtained against HP1. Figure 4C (lower panel) shows that anti-HP1 expressed as an EGFP tagged protein in HeLa cells localized in the nucleus where it labels condensed structures similar to HP1 usual staining. In contrast, a diffuse cytoplasmic and nuclear staining was obtained using non relevant hs2dAbs fused to EGFP (Figure 4C, upper panel).

16These results indicate that functional intrabodies can be obtained at high frequency using the NaLi-H1 library. Intrabodies can be used, upon fusion with a fluorescent protein, to track the dynamics of their target in living cells (see for example ). Such an application is illustrated in the Video 1 where an hs2dAb directed against mCherry is used to track mCherry-fused Rab6 in living cells. Intrabodies may allow not only to track the dynamics of their cellular target, but also to perturb, or block, their activity. Our results indeed indicate that the H12 antibody was able to perturb endogenous RHO activity when expressed in the cytosol. We could not directly image enrichment of the H12 antibodies in cellular sub-domains in living cells but we observed that H12 may behave as an efficient intrabody carrying out co-immunoprecipitation experiments. H12 carrying a carboxy-terminal myc tag was expressed in HeLa cells together with either the CBD-fused RHOA DN or with the CBD-RHOA CA mutants. H12 was pulled-down by RHOA CA but not by inactive RHOA DN (Figure 5—figure supplement 1). H12 thus worked as an intrabody and kept its conformation sensitivity in the cytosol. Because RHO GTPases are involved in signaling pathways that promotes the actin cytoskeleton polymerization, we looked at functional effects induced by H12 overexpression. In contrast to un-transfected cells or cells transfected with various non-relevant EGFP fused hs2dAb (Figure 5A, upper panel), we observed that cells expressing H12-EGFP were totally devoid of actin stress fibers (Figure 5A, lower panel). This alteration in actin filament organization was associated with marked changed in cell shape characteristic of loss of intracellular mechanical forces and tension. As RHOA plays a major role in activating myosin II and actin cytoskeleton reorganization, our results suggested that H12 efficiently perturbed RHO-dependent signaling, mimicking the phenotype induced by the C3 exoenzyme RHO inhibitor.

17Identification of blocking antibodies is a challenging task and not all functional intrabodies are inhibitory. However, it is possible to functionalize non-blocking intrabodies to inhibit their target function. One approach relies on the ubiquitinylation and degradation of the recognized target as described by. This approach is based on the fusion of intrabodies to an F-box domain which allows interaction with Skip1, a member of the SCF complex, an E3 ubiquitin ligase of the complex E1/E2/E3 ubiquitinylation machinery, that targets proteins to proteasome-dependent cellular degradation (Figure 5—figure supplement 2). This approach was efficiently developed to target several EGFP fusion proteins in Drosophila using a single anti EGFP intrabody, named GFP4, which is a robust and high affinity EGFP llama intrabody originally isolated from an immune library. To get insight into the relative functionality of hs2dAb for such a protein interference approach, several of the anti EGFP hs2dAb selected from the NaLi-H1 library were fused at their amino terminus to the Fbox domain and their efficacy was compared to the efficacy of the Fbox-GFP4 nanobody. To detect cells expressing Fbox-intrabody fusion proteins (F-Ib), we constructed a bicistronic vector driving the co-expression of F-Ib together with a mitochondria-targeted mCherry (Mito-mCherry) (Figure 5B). We expressed the F-Ib antibodies in a HeLa clone stably expressing EGFP fused to histone H2B and looked for EGFP-H2B depletion. As expected, F-GFP4, also known as degradFP, induced a strong reduction of H2B-EGFP expression as analyzed by western blot (Figure 5—figure supplement 2). Accordingly, a strong reduction in nuclear fluorescence intensity was observed in cells expressing F-GFP4 (see mito-mCherry positive cells, Figure 5B; Figure 5—figure supplement 2). No effect was observed when expressing either GFP4 alone or a GFP4 fused to a truncated, nonfunctional, Fbox domain (Figure 5—figure supplement 2). When anti-EGFP clones selected from the NaLi-H1 library were tested, we observed that some of the hs2dAb that were found to active as fluorescent intrabodies failed to degrade H2B-EGFP when expressed as F-Ib. This highlights the fact that not all intrabodies can efficiently be functionalized with the F-box and that in vivo binding to a target is not the only parameter to consider. However, several hs2dAb anti-EGFP induced a complete disappearance of nuclear H2B-EGFP signal when expressed as F-Ib (F-R3TF3, Figure 5B) while no reduction was observed when using an anti-EGFP that cannot be used as an intrabody (F-R3TG4, Figure 5B). FACS analysis showed a decreased of fluorescence intensity by as much as 70% (Figure 5C). As expected, this effect was reversed in the presence of a proteasome inhibitor (Figure 5D). Altogether, these experiments show that the hs2dAb scaffold enables the frequent selection of antibodies that can be expressed in the mammalian cell cytoplasm to be used as functional fluorescent or inhibitory intrabodies.

18Here, we report the construction of the first large fully synthetic single domain antibody library based on a humanized scaffold derived from llama VHH. A set of robust nanobody scaffolds was first identified using a positive expression screening in E. coli cytosol. One very robust scaffold (sdAbD10) was identified and was used as a base. After introduction of several modifications that aimed at humanizing its primary sequence, we designed the hs2dAb scaffold which is as stable as sdAbD10 while being closer to human VH3. Our data indicate that the hs2dAb displays partial resistance and/or refolding after treatment for 10 min at 90°C. Using CDR grafting experiments we confirmed the efficacy and the stability of the synthetic scaffold to display CDR regions. Based on our prior experience on phage display libraries, immune or naïve llama VHH libraries or from scFv libraries we then rationally designed CDR diversity with fixed CDR1 and CDR2 size and four CDR3 sizes (9, 12, 15 or and 18 amino acids). The power of modern gene synthesis approach permits to reach very high genetic diversity while controlling codon bias and cloning features. Fully random codon combination using NNN or NNK trinucleotide cannot prevent stop codon, undesired cysteine or hydrophobic residues to be incorporated, and it does not lead to the controlled probability of amino acid occurrence at a given position. Therefore a more rational design was implemented with defined set of codons for each CDR amino acid position so that it does not mimic natural diversity, in contrast to recently developed Fab synthetic libraries, but is rather optimized for intrinsic hydrophilicity or solubility. After a large-scale cloning of synthetic fragments, 3 billion independent clones were transformed in the bacteria. Library quality was confirmed by Sanger and next generation sequencing.

19The library was validated by screening against various targets and in each case specific and highly functional antibodies were obtained (Table 1). Various selection schemes yielded a large diversity of high affinity and high selectivity binders. Selections were carried out using purified antigen coated on polystyrene, on magnetic beads or directly on the cell surface. In many cases, two rounds of selection were sufficient to obtain selective binders. We usually analyzed only 80 randomly picked clones because the diversity of specific binders was systemically high. Only a few selections led to antibodies usable in western blotting (anti-actin, ant-tubulin) probably because most screenings were done using natively folded targets. Accordingly, selected hs2dAb performed very well in other conventional immunoassays like ELISA, FACS, immunoprecipitation or immunofluorescence. Affinity measurements done by surface plasmon resonance revealed KD values in the order of 10 nanomolar and up to 50 picomolar. Such high affinities are rather good and usually rarely observed for monovalent binders obtained without in vivo immunization or in vitro affinity maturation steps (Figure 6; Table 2).

20The NaLi-H1 library thus enables the rapid selection of diverse and highly functional binders. Because it is a fully synthetic, non immune, library, it does not depend on animal experimentation, it is not limited by natural immunogenicity or toxicity of antigens and allows to develop and adjust the selection without ethic consideration. In addition, because all steps are carried out in vitro, conditions can be tightly controlled. This allowed to develop powerful differential selection and to identify conformation-specific antibodies. This also allowed to directly screen for antibodies directed against antigens specifically present at the surface of a particular cell type. Such a differential selection will be a powerful approach to identify novel antigen at the surface of tumor or infected cells. Such antibodies may also represent powerful tools for diagnostic and therapeutic applications to target cells in human pathologies. For example, after dimerization using Fc domains, hs2dAb antibodies may be used to target tumor cells and benefit from antibody-dependent cell-mediated cytotoxicity (ADCC) for example. They may also be used directly as the smallest antibody-derived domain naked as an agonist or antagonist or armed for enhanced toxicity. Similarly, it may be labeled using radioactive compounds (e.g. 99mTc, 111In, 64Cu) and used to image tumors in patient using positron emission tomography. Altogether, he NaLi-H1 library may accelerate the identification of novel potent tools to be used in human clinical applications.

21The synthetic scaffold we defined was based on the selection of a set of VHH able to fold properly in the bacteria cytosol. The goal was not only to define a robust scaffold that would be efficiently produced without aggregation but also to allow frequent selection of functional intrabodies. Intrabodies have been isolated from various antibody libraries as well as other protein scaffold like Darpin or FN3 which are devoid of cysteine. A peculiar feature of the NaLi-H1 library is that it is based on a humanized nanobody-like robust scaffold, stable in a reducing environment, while it still contains the two canonical cysteine residues. Stabilized nanobodies, human single domain scaffolds and libraries were reported before. However, to our knowledge, no synthetic library producing at high frequency functional intrabodies was developed before based on such stabilized scaffolds. Almost every hs2dAb antibodies we expressed in mammalian cytosol showed no sign of aggregation, which further supported the idea that the synthetic scaffold we designed is robust and highly resistant to reduction. Previous studies showed that functional intrabody identification often relied on additional steps of selection in a protein–protein interaction reporter system such as PCA or bacterial 2 hybrid, yeast IACT or F2H assays. Using the NaLi-H1 library, we observed that without using particular selection schemes, functional intrabodies were frequently obtained. Although we did not formally compare the NaLi-H1 library to previous llama naïve or semi synthetic libraries, the functionality of selected hs2dAb was compared to a sub-nanomolar affinity intrabody, the GFP4 nanobody, which has been extensively used. We observed by monitoring its signal-to-noise ratio and by using it in a protein knockdown assay that NaLi-H1 can provide highly functional hs2dAb which appeared as good as intrabodies from immune libraries.

22Intrabodies can be used in several applications like tracking of intracellular dynamics of endogenous proteins but the most appealing application is to use them for rapid protein inactivation in living cells. Intrabodies may be used to directly block their target proteins in cells. One of the conformation-sensitive antibody that was selected can be used to inhibit active RHO GTPase signaling in living cells and is as potent as the C3 exoenzyme toxin. Only few intrabodies have been described to be intrinsically inhibitors of protein activity, and our results suggest that the NaLi-H1 library may enable rapid selection of inhibitory antibodies. The next challenge will be to select conformational sensors specifically directed against a particular member of closely related RHO subtypes (RHOA/B/C) which share more than 90% similarity in primary sequence. But in any case, our results show that the NaLi-H1 library allows the selection of efficient, conformation-specific, inhibitory intrabodies.

23Another way intrabodies may be used to inactivate their targets in living cells is to fuse the intrabody to a dominant inhibitory domain. Following an idea pioneered by Affolter and colleagues, we showed here that intrabodies selected from the NaLi-H1 library can be fused to a proteasome-targeting domain to impose the specific degradation of their respective targets. This protein interference approach was validated using anti-EGFP hs2dAb and we believe that this approach will bring disruptive tools to generate rapid protein knockdown both in cell culture and in the animal.

24In summary, we have designed a novel nanobody scaffold endowed with improved stability and created a highly diverse library, the NaLi-H1 library, that was successfully screened to identify highly functional binders directed against very diverse targets. We believe that this library will allow the fast, and fully in vitro, identification of immunological tools usable both for fundamental and medical applications.

25Artificial gene synthesis (Mr Gene, GmbH, Germany) composed of a 6His-Tag and a triple c-myc Tag was inserted into the pHEN2 phagemid vector (Griffin 1. library) between NotI and BamHI sites. CcdB gene from pENTR4 vector (Invitrogen - ThermoFisher Scientific, France) was inserted into the pHEN2 vector between NcoI and NotI sites. This vector allows to express antibody fragments in fusion, upstream, with the pelB leader to drive secretion in the periplasm and downstream with the PIII protein of M13 phages. An amber stop codon is present between the antibody and the pIII. This stop codon is partially suppressed in SupE E. coli. For expression and purification of dimeric antibodies, hs2dAb were inserted in vectors derived from pFuse (Invivogen, France) as described in. For intrabody expression in mammalian cells, hs2dAb were digested by NcoI and NotI and ligated into the pIb-mEGFP, pEGFP or the pmCherry vectors (Clontech - Takara, USA). (See the Appendix for more details).

26Previously selected VHHs from naïve or immune libraries were subcloned into pAOCAT using the NcoI and NotI restriction sites. Chloramphenicol resistance assay was performed using BL21 (DE3) cells transformed with the pAOCAT-VHH fusion constructs. (See the Appendix for more details).

27Details about the construction of the library can be found in the Appendix. In short, a synthetic design was ordered based on a statistic analysis of the diversity found in natural VHH and aiming at reducing hydrophobicity at some position. The size of the CDR1 and CR2 was fixed at 7 amino acids while 4 sizes of CDR3 were chosen (9, 12, 15 and 18 amino acids). Large -scale PCR was then carried out ensuring that at least 1010 DNA molecules were used as a matrix. Fragments were then cut and inserted into the pHEN2-3myc plasmid. The ligated DNA material was used to transform electrocompetent E. coli TG1 cells (Lucigen Corp., Middleton, United States). Serial dilution was used to count the total number of bacteria transformed. A potential diversity of 3 × 109 was calculated. Transformed bacteria plated on 430 2xYT-ampicillin agar dishes (140 mm), grown overnight at 37°C, scrapped and stored in 30% of glycerol at −80°C.

28IonTorrent sequencing library was prepared with the Ion Plus Fragment Library kit for AB Library Builder System (Life Technologies - TermoFisher Scientific, France) following manufacturer's instructions and was controlled on the Agilent 2100 Bioanalyzer (Agilent Technologies, France) with the High Sensitivity DNA Kit (Agilent Technologies). The sequencing template was prepared by emulsion PCR with the Ion OneTouch 2 system and the Ion PGM Template OT2 400 Kit (Life Technologies). Sequencing was performed on a IonTorrent Personal Genome Machine using the Ion PGM Sequencing 400 Kit and a 314v2 Ion chip (Life Technologies).

29Human βActin was purchased from Sigma-Aldrich (France). RHOA GTPase fused to either an amino terminal Chitin Binding Domain or a streptactin binding peptide were produced in HEK293 cells. EGFP (as mCherry) in fusion with a streptavidine binding peptide (SBP) were produced through in vitro translation system (Roche Life Science, France) and used directly for screening without the need for purification.

30Biotinylated Tubulin was purchased from Cytoskeleton, Inc. (Denver, United States). For p53, the 83 first amino acids of the NP_000537.3 isoform were produced in bacteria with a SNAP and His Tag, purified using Talon resin (Takara - Clontech) and biotinylated in vitro. HP1α was produced in bacteria with an avitag and a His Tag, and purified using Talon resin.

31For HER2, the natural receptor was used as membrane protein target on SKBR3 cells.

32For more details see the Appendix for more details.

33Screening for ßactin was performed by panning in immunotubes as described.Screening for EGFP, Tubulin and p53 were performed in native condition as described. Screening for HER2 was performed on surface cells as described. Screening on RHO was performed in native condition. (See the Appendix for more details).

34Individual clones were screened by monoclonal phage ELISA as described. (See the Appendix for more details).

35After boiling in SDS-PAGE loading buffer, the samples were separated on a 12% SDS-PAGE and transferred to nitrocellulose membranes (Whatman GmbH, Germany). Membranes were blocked in 3% non-fat milk-PBS with 0.2% Tween 20 for 1 hr at room temperature or overnight at 4°C. unpurified hs2dAb were used at 1/100 from culture supernatant and added to the membranes with an anti-hisTag antibody at 1/3000 (Sigma-Aldrich) for 90 min. Blots were then washed and incubated 1 hr with secondary anti-Mouse HRP labeled antibodies (diluted at 1/10000 in PBS 0.1% Tween 20) (Jakson ImmunoResearch Laboratories). After 5 washes with PBS 0.1% Tween 20, secondary antibodies were then revealed using the SuperSignal chemoluminescent reagent (Pierce) and Hyperfilm ECL (GE HealthCare). For RHO-GTP pull down, the primary anti RHOA mAb was used (Cell Signaling Technology; 1/1000). For protein knockdown experiments, 500 000 of transfected cells (mCherry positive cells) were sorted with a MoFlo Astrios flow cytometer (Beckman Coulter). Cells were lysed with SDS-Tris lysis buffer (Tris pH7.4 10 mM, SDS 1% supplemented with phosphatase and protease inhibitors). 20 µg of cell extracts were separated on 12.5% SDS-PAGE and electro transferred onto PVDF membranes. Blots were probed with a rabbit polyclonal anti-EGFP full length (Santa Cruz, sc-8334, 1:500), a mouse monoclonal anti-α-tubulin (Sigma, T5168, 1:25000) and an anti-myc HRP antibody (Novus Biologicals, NB600-341, 1:40000). Detection was performed using peroxydase conjugated secondary antibodies and Pierce ECL Western Blotting Substrate (Thermo Scientific Pierce).

36Immunofluorescence screenings were performed on HeLa cells as described before. (See the Appendix for more details).

37Hela or HeLa S3 H2B-EGFP Cells cultured on coverslips were transfected according to the CaPO4 or jet prime procedure with 1 µg DNA per well (24 wells plate) or 10 µg DNA (10 cm2 diameter dish). Cells can be observed from 12 hr post-transfection on.

38For HER2 immunoassay, cell surface staining were performed in phosphate-buffered saline (PBS) supplemented with 1% SFV. 100 µL of supernatant (80 µL phages + 20 µL PBS/milk1%) were incubated on 1.105 cells for 1 hr on ice. Phage binding was detected by a 1:300 dilution of anti-M13 antibody (GE healthcare, France) for 1 hr on ice followed by a 1:1000 dilution of PE-conjugated anti-Mouse antibody (BD Bioscience, France) for 45 min. Samples were analyzed by flow cytometry on a FACSCalibur using CellQuest Pro software (BD Biosciences,France).

39In the protein knockdown experiments, 48 hr after transfection, at least 10000 HeLa S3 H2B-GFP cells were analyzed on a MoFlo Astrios flow cytometer (Beckman Coulter France S.A.S) for their GFP fluorescence intensity. This fluorescence was analyzed in mCherry transfected cells and non transfected cells. Flow cytometry data were analyzed with Kaluza software (Beckman Coulter). 1 µM of proteasome inhibitor MG132 (Sigma-Aldrich) was used in the cell growth medium for 48 hr. Values reported represent median ± standard deviation (SD) of at least three independent experiments. p values were calculated with GraphPad Prism 6 (RRID:SCR_002798) using a Student’s t test. **p<0.01; ***p<0.001; ****p<0.0001.

40All binding studies based on SPR technology were performed on BIAcore T200 optical biosensor instrument (RRID:SCR_008424, GE Healthcare). Capture of single domain Hs2dAb-6xHis was performed on a nitrilotriacetic acid (NTA) sensorchip in HBS-P+ buffer (10 mM Hepes pH 7.4, 150 mM NaCl, and 0.05% surfactant P20) (GE Healthcare). The four flow cells (FC) of the sensorchip were used: one (FC 1) to monitor nonspecific binding and to provide background corrections for analyses and the other three flow cells (FC 2, 3, and 4) containing immobilized Hs2dAb-6xHis for measurement.

41For immobilization strategies, the four flow cells were loaded with nickel solution (10 μL/min for 60 s) in order to saturate the NTA surface with Ni2+ and an extra wash using running buffer containing 3 mM EDTA after the nickel injection. Each His-tagged hs2dAb in running buffer was injected in flow cells at a flow-rate of 10 μL/min. The total amount of immobilized hs2dAb-6xHis was 250–300 resonance units. (RUs; 1 RU corresponds approximately to 1 pg/mm2 of protein on the sensor chip). A Single-Cycle Kinetics (SCK) analysis to determine association (on-rates), dissociation (off-rates) and affinity constants (kon, koff and KD respectively) was carried out. SCK method prevents potential inaccuracy due to sensorchip regeneration between cycles which are necessary in the conventional multiple cycle kinetics (MCK). SCK binding parameters are evaluated for each injection according to the tools and fit models of the BIAevaluation software, giving similar values than MCK. As hs2dAb were smaller proteins than their respective antigens, hs2dAbs were captured on the sensorchip while the recombinant antigens were used as analytes. Analytes were injected sequentially with increased concentrations ranging between 3.125 nM to 50 nM in a single cycle without regeneration of the sensorship between injections. Binding parameters were obtained by fitting the overlaid sensorgrams with the 1:1. Langmuir binding model of the BIAevaluation software version 1.0.