A database of human CDR3 sequences

We compiled human CDR3 sequences from three main sources: the Kabat database [35], the IMGT database [36] and the RCSB PDB [37]. After removing the duplicates, the database contained 6247 H3, 1332 λ CDR3 and 1495 κ CDR3 unique sequences (Figure 1b). It can be noted that most of the H3 sequences were unique since, for instance, in the Kabat database, 2703 H3 sequences (78%) were found only once among the 3469 complete H3 sequences. The result was comparable in the case of L3 and K3 since, respectively, 74% and 64% of the sequences were unique in the Kabat database. This underlines the very high variability of the human CDR3 sequences.

As noted by several authors, this variability is not evenly distributed in the loop, and the frequency of each amino acid varies from one position to another and for each loop length [18, 38]. In addition, the amino acid distribution depends on the animal origin of the antibody sequence [32]. This bias can be due to a structural constraint as for instance in the case of the antepenultimate residue which is frequently an aspartate (D101 using Kabat numbering scheme) and plays an important role in the switch between the extended and the kinked conformation of the H3 [39]. In other cases, this bias may only be due to the limited number of sequences available for the D and J segments, and amino acids other than those found in natural antibodies may be tolerated [40]. For the construction of the library we decided to make CDR3s that mimic the natural distribution for two main reasons: i) we wanted the scFvs generated to be as human as possible for possible use in human therapy [41]; ii) since the antibodies found in the database are functional, maintaining their amino acid distribution will more likely result in functional antibodies, since we incorporated at each position only the amino acids tolerated in natural antibody CDRs [38].

CDR3 sequences from the database were aligned by length, and the frequency for each of the 20 possible amino acids at each position and for each loop length were calculated. In the case of the light chain CDR3s, sequences were analyzed independently for each class. In the case of the H3 sequences, this resulted in 35 tables, one for each H3 length between 1 and 34 amino acids. For a loop of length n, this table contained 20 n frequencies. This is illustrated in figure 1c with the 5 amino acid long H3 loop case (columns D). The only strongly variable position was the position 2 where the most frequent amino acid (Tyr) was only present in 16% of the loops. This is not surprising since this residue is frequently in contact with the antigen [42]. By contrast, strong biases were observed at position 1 (29% of Gly), 4 (45% of Asp and Gly) and 5 (36% of Tyr). The number of highly variable positions increased with the loop length but, in all the cases, none of the positions was truly random. For the other loop lengths, the frequencies are given in additional file 1 "Amino acid distribution in CDR3s".

Design of the oligonucleotides encoding the CDR3 loops

Eighteen oligonucleotides were designed to follow the amino acid distribution found in the compiled CDR3 database. We used 192 optimized mixes of the four nucleotides at each position of the codon to match as well as possible the desired amino acid distribution (see [31] and references therein). Plots showing the target and the oligonucleotide-encoded distributions are given in additional file 1 "Amino acid distribution in CDR3s". The process is illustrated in figure 1c with the 5 amino acid long H3 loops. D bars represent the frequencies of the 20 amino acids in the human database at the 5 positions. T bars are the expected frequencies with the optimized mixes (after rounding the nucleotide frequencies at the nearest 5%, see Methods). Due to the constraint of the genetic code, some positions were not perfectly optimized. For instance, at position 3, alanine and glycine were under-represented in our mix and we had to introduce a substantial amount of some non naturally found amino acids like cysteine in order to match other amino acid frequencies. There was however a good overall agreement between the database and the oligonucleotide-encoded frequencies since the most frequently found amino acids were represented in the library and the rare ones were most of the time present at a low frequency. The sequences of the 18 degenerate oligonucleotides used to construct the CDR3 loops are given in additional file 2 "Sequences of the spiked oligonucleotides used to introduce the random CDR3 loops".

Construction of the library

We constructed independent libraries for each CDR3 loop length. This was done independently for the heavy and the two classes of light chains. For each library, random CDR3 loops were introduced by PCR and the resulting library cloned back in the scFv13R4 gene fused to the CAT gene in vector pscFvΔCAT (see below). This resulted in libraries of scFv13R4 clones with one and only one randomized CDR3 loop.

We validated the approach by first constructing the 5 amino acid long H3 loop library. We sequenced 43 randomly chosen clones and the frequencies of the 20 amino acids at each of the five positions are shown in figure 1c (columns E). Some positions diverged from the expected values as, for instance, the glycine at position 2, which is over-represented in our library. However, on average, the distribution in the library matched the expected distribution. This showed that the quality of the oligonucleotides was good enough for the approach and that the resulting library followed the natural distribution of the amino acid in human H3 loops.

Some of these clones, however, may not be functional as a result of the cloning steps (stop codons present in the oligonucleotides and PCR) or because of a poor expression level in the bacterial cytoplasm. To remove these non functional scFvs, we selected for clones expressed in the bacterial cytoplasm by means of a fusion between the scFv and the CAT enzyme [33]. We tested different CAM concentrations for this selection step ranging from 15 to 200 μg/ml. At the highest concentration, the library was indeed enriched in well-expressed scFvs, but also in clones containing recombined plasmids harboring partial or complete deletions of the scFv gene (data not shown). We thus plated the libraries on a medium CAM concentration (30 μg/ml). This concentration was high enough to remove all the non-expressed or strongly aggregating scFvs but did not result in a detectable amount of plasmids harboring scFv deletion. The size of the libraries of expressed scFv was thus estimated as the product of the original library size (selected on ampicillin) by the frequency of the CAM^R clones. The sizes of the 18 libraries ranged from 2.5 × 10⁶ to 1.9 × 10⁸ (Table 1).

The 18 CAM selected libraries were assembled in amounts proportional to the natural distribution of CDR3 loop lengths in human antibodies to give a final library of more than a billion of clones. Since the theoretical possible diversity is about 10¹⁵ (~13 VH × 10⁷ × 5 VL × 10⁶), it is very unlikely to obtain twice the same clone in the final library. One hundred and eighteen clones were sequenced to determine loop lengths and sequences. As shown in figure 2, all the introduced lengths except 6 for the H3 were found in the library. 11 and 16 amino acid long H3 loops were under-represented in the library. This is presumably due to the poor quality of these oligonucleotides as shown by their profile on an Agilent Bioanalizer (data not shown). H3 loop lengths ranging from 7 to 12 were over-represented in the library but only by a two-fold factor. The H3 loop lengths between 8 and 17 amino acids, which are the most frequently found in human antibodies, were all present in the library. The number of sequenced clones was too small to analyze the frequency of the amino acids found in the CDR3 loops. Except for some contaminations with the original scFv13R4 sequence (4% for H3), no CDR3 sequence was found twice in the sequenced clones. The sequences of the 118 clones are shown in the additional file 3 "Sequence of randomly picked clones".

Cytoplasmic expression of scFvs

The idea underlying the library construction is that the CAM selection step of the VH and VL libraries should result in only expressed scFv proteins. Indeed, since only the CDR3 loops are modified between scFv13R4 and the libraries, most of the interface residues between the two domains are conserved between clones [43]. It is thus likely that any VH will assemble correctly with any VL and that the expression level of the resulting scFv will be close to that of both clones from the VH and VL libraries selected on CAM. We tested this hypothesis by picking random clones of the final library and expressing them in E.coli cytoplasm and in mammalian cytosol.

Twenty clones were tested in E.coli and 19 of them showed some soluble expression in the cytoplasm (Figure 3). One-fourth of the clones (5/20; clones 3, 10, 11, 16, 19) were expressed at very high levels since the scFvs were clearly visible on a Coomassie stained gel. To obtain a more global view of the soluble expression levels in E.coli, the library was cloned in front of the GFPuv gene under the control of the T7 promoter. If the scFv is soluble and expressed in the cytoplasm, this should result in GFP activity that can be directly monitored on an UV transilluminator [44, 45]. About 1000 clones were tested for the presence of detectable GFP activity and approximately 60% exhibited a GFP⁺ phenotype (data not shown) showing again that most of the scFv clones from the library were expressed in E.coli cytoplasm.

We next tested the expression of the library in mammalian cells. Fifteen scFvs were cloned in a mammalian expression vector as fusions with the EGFP gene and under the control of the SV40 early promoter, then transfected in HeLa cells. Typical results are shown in figure 4. Three clones were expressed at a high soluble level, comparable to that of the parental scFv13R4 (clone 15), 10 scFvs were found to be mainly soluble but some aggregated material was still present in the cell (clones 33 and 36) and 2 clones accumulated essentially as cytoplasmic aggregates (clone 24), as observed with the hybridoma-derived anti-oncoprotein E6 scFv 1F4 (Figure 4) [28]. In conclusion, thirteen out of the fifteen scFvs tested were expressed as soluble proteins that could be easily detected in the cytoplasm and in the nucleus of the transfected cells.

Altogether these results showed that more than 85% of the clones from the final library expressed soluble scFv in E.coli (16/20) and mammalian cytoplasm (13/15), while about 20% of them expressed scFv at a very high level (5/20 in E. coli and 3/15 in eucaryotic cells). This is a great improvement over the results obtained by other authors with non optimized scFv libraries [12].

Selection of binders

As shown above, the library contains a high proportion of expressed clones but it remains to be shown that antibodies against particular proteins can be selected. We thus used the phage displayed library to select for binders against five different antigens using purified proteins adsorbed on microtiter plates. We performed three rounds of selection and the eluted phages were tested by ELISA against the immobilized antigens. As shown in figure 5a, in all the cases a positive signal was obtained after a single round of selection. This signal increased strongly after two rounds and did not increase further during the third round of selection. This very fast selection process was presumably due both to the focused library itself, which contains only expressed scFvs resulting in a low background, and to the use of a trypsin-sensitive helper phage that further decreased the background level [46, 47].

We next tested whether the library contained clones expressing soluble scFvs in the periplasm. The non-suppressive HB2151 strain was infected with the phages eluted after the third round of selection against tubulin, GST:Syk and the core histones. Periplasmic extracts were prepared and tested for binding activity by ELISA (Figure 5b). In the three cases, 12–20% of the clones gave a strong signal with absorbance values higher than 0.5 (10 times the background) and about 30% were clearly positive with an absorbance value higher than 0.1 (twice the background). These results compared favorably with those reported with other scFv libraries, underlining again the high proportion of well-expressed clones present in the library. In addition, this showed that our CAT fusion approach selected efficiently for constructs without stop codons present in the oligonucleotides. This is indeed of premium importance to isolate soluble scFvs from phage-displayed libraries since amber stop codons in CDRs are frequently selected during panning of synthetic and semi-synthetic libraries [48].

In both the previous characterizations, the scFvs were expressed under oxidizing conditions in E.coli periplasm, either as scFv-pIII fusion or as soluble protein. In addition, panning was done on phage, again under oxidizing conditions. We could not exclude that the selection and the screen introduced a bias towards clones expressed in the periplasm at the expense of those expressed in the cytoplasm. To test that the selected scFvs were indeed also expressed in the cytoplasm, we subcloned the same pool of clones (round 3) in a cytoplasmic expression vector under the control of the strong T7 promoter. For each antigen, ninety-five clones were tested by ELISA for binding to their respective antigen. In each case, the number of positive clones was comparable or even better than in the periplasmic screen (Figure 5b). For instance, in the case of GST:Syk, 80% of the tested clones were positive after three rounds of selection. This demonstrated that the periplasmic selection step did not decrease the proportion of soluble scFvs in the cytoplasm and that by using our library it is not necessary to directly select within the cytoplasm to avoid introducing a bias during the selection [13].

Functionality of isolated scFvs as intrabodies

To determine if the isolated scFvs were able to bind to their target in vivo we characterized the anti-histones scFvs expressed in human cells. The third round of selection was cloned in vector p513-EGFP and ten randomly chosen clones were transfected in HeLa cells. Typical results of the cells expressing the scFv-EGFP fusions and observed by fluorescence microscopy are shown in figure 6. One scFv was expressed as cytoplasmic aggregates. Four scFvs were expressed as soluble cytoplasmic proteins, as judged by the homogeneous staining of the cells, at a level comparable to that of the scFv13R4. Finally, three scFvs gave rise to a strong staining of the nucleus (clones 2, 6 and 9) and, two scFvs were exclusively localized in the nucleus (clones 5 and 10). Since these scFvs fused to the EGFP were expressed in the cytoplasm of the cell and did not contain a nuclear localization signal, this suggested that they were able to interact in vivo with the histones and were thus active inside the cell [5]. This analysis showed that about half of the clones present after the third round of selection against core histones were able to bind to their nuclear target in vivo. This was confirmed in vitro by western and dot blot with purified scFv (additional file 4 "In vitro characterization of some anti-histones scFvs"). In addition, sequencing of the five positive clones showed that they contained different heavy and light chain CDR3 regions (additional file 4 "In vitro characterization of some anti-histones scFvs").

In vitrocharacterization of anti-tubulin scFvs

To demonstrate the activity of the anti-tubulin scFv under the reducing conditions that pertain in the cell cytoplasm, we extracted the scFvs in presence of a reducing agent and we compared the ELISA signal with that obtained with the scFvs extracted under oxidizing conditions. As shown in figure 7, the five scFvs gave the same ELISA signal in both conditions, demonstrating the full activity of the scFvs under reducing conditions and thus in absence of disulfide bond formation.

The five scFvs were able to recognize unfolded tubulin by western blot in brain extracts and the native protein in a competition ELISA (data not shown). Finally, we tested their ability to interact with microtubules in cells by IF. Clones 2F12C and 2G4C, but no the three other scFvs, were able to reveal the microtubule network in cells. In figure 8 is illustrated the utility of this new library as a source for both in vivo and in vivo proteomic studies: HeLa cells were transfected with the anti-histones clone 5 fused to a Red-GFP then the microtubule network was revealed by IF using the 2F12C scFv.

Finally, we characterized clone 2F12C by Biacore since it was the best binder in ELISA and IF. Data were fitted to a two-state reaction model giving a Kd of 50 nM (additional file 5 "Biacore analysis of clone 2F12C"). This value is comparable to those obtained for the best scFvs obtained with other single framework libraries with variability restrained to the CDR3 loops [20, 22].

Altogether, our results show that the library described in this report is highly diverse and functional and allows fast and easy isolation of in vivo and in vitro active fully human intrabodies.

Bacterial strains, chemicals and enzymes

LB and 2xYT media were previously described [55]. Strain C-Max5αF' (Bio-rad laboratories) is E.coli K-12, [F' lacI^q Tn10] φ80dlacZΔM15 Δ(lacZYA-argF)U169 recA1 endA1 hsdR17(r_k^- m_k⁺) phoA supE44 λ- thi-1 gyrA96 relA1. MC1061 (ATCC #53338) is E.coli K-12, F- λ- hsdR2 hsdM+ hsdS+ mcrA mcrB1 araD139 Δ(ara-leu)7696 Δ(lacIPOZY)X74 galE15 galU galK16 rpsL thi. Non-suppressor strain HB2151 is E. coli K-12 [F'proA+B+lacIq lacZΔM15] ara Δ(lac-pro) thi. Chemicals were purchased from Sigma. Restriction enzymes and cloned Taq polymerase were from Fermentas. ProofStart and Pfu DNA polymerases were respectively purchased from Qiagen and Promega. Plasmid DNA, PCR and agarose separated DNA were purified using Macherey-Nagel Nucleospin kits.

Oligonucleotides

The sequences of the 18 spiked oligonucleotides used to introduce degenerate CDR3 loops and listed in the additional file 2 "Sequences of the spiked oligonucleotides used to introduce the random CDR3 loops" have been synthesized and HPLC purified by IBA GmbH (Goettingen, Germany). Other oligonucleotides were synthesized by MWG:

T7.back CCGGATATAGTTCCTCCTTT;

T7.for CTGCTAACCAGTAAGGCAAC;

M13rev-49 GAGCGGATAACAATTTCACACAGG;

M13uni-43 AGGGTTTTCCCAGTCACGACGTT;

scFvCAT.rev AACGGTGGTATATCCAGTGA;

scFvCAT2.rev CGGTGGTATATCCAGTGATTTTT;

PliaisonH3 TGGGGCAGAGGCACCCT;

PliaisonH3.back AGGGTGCCTCTGCCCCA;

PliaisonL3 GCAGTAATAATCAGCCTCGTCC

Plasmids

Phagemid vector pCANTAB6 [56] was used for N-terminal fusion of NcoI/NotI-scFv fragments to the minor coat protein pIII of filamentous phage M13. This phagemid is derived from pUC119 and contains in the following order: a lac promoter, the pelB leader sequence, NcoI and NotI sites for scFv cloning, a His6 and a c-myc tag recognized by the 9E10 monoclonal antibody [57], an amber codon and the pIII gene sequence.

For cytoplasmic expression of the scFvs in E.coli we used plasmid pET23NN [58]. This plasmid is derived from pET23d(+) (Novagen) and contains a T7 promoter, followed by a NcoI site containing the ATG initiator, a NotI site followed by a c-myc and a His6 tag.

Plasmid pscFvΔCAT is derived from pTrc99A and contains a tac promoter, followed by a NcoI site containing the ATG initiator of an out-of-frame scFv, a NotI site followed by the CAT gene. When a scFv is inserted within the NcoI-NotI sites, the scFv is expressed as a fusion with the CAT protein. The construction was done as following. First, the unique BstEII site of pTrc99A (A13038) was removed by digestion followed by 5' overhang fill-in, to form blunt ends, and ligation. The resulting plasmid was digested with NcoI and NotI, and the 4210 bp fragment purified (fragment I). Second, the unique NcoI site of plasmid pACYC184 (X06403) located within the CAT gene was removed by site directed mutagenesis by changing the Thr172 codon from ACC to ACG. Then the CAT gene was amplified by PCR using CAT-NotI.for (TAAGGCGGCCGCAATGGAGAAAAAAATCACTG)and CAT-HindIII.back (ACTGCCTTAAAAAGC TT ACGCC). In the oligonucleotide sequences, the introduced restriction sites are underlined and the beginning and the end of the CAT gene are in bold-italic. The 660 bp PCR fragment was digested by NotI and HindIII, and purified (fragment II). Third, a 750 bp NcoI-NotI scFv13E6 fragment, a grafted version of the scFv13R4 containing the CDR loops of an anti-E6 monoclonal antibody (Philibert et al., to be published), was purified (fragment III). Four, the three fragments I, II and III were ligated to give plasmid pscFvCAT. Finally, an internal deletion of 165 bp was introduced in the scFv by removing the fragment between the two PstI sites of the gene. The resulting plasmid, called pscFvΔCAT, is Amp^R and CAM^S since the deletion of the PstI fragment resulted in a frameshift in the scFv.

Plasmid p513-EGFP is a derivative of pSG5 [59] and harbors the EGFP coding region (Clontech) under the control of the SV40 early promoter. The p513-scFv-EGFP constructs correspond to in frame fusions of the scFvs and the EGFP coding region with a linker of 10 residues. The scFv coding regions were amplified with oligonucleotide primers 5'-ACTGATAAGCTTGCCACCATGGCCGAGGTGC and 5'-TTGATTACTAGTGAGTTTTTGTTCTGCGGCC and inserted into the HindIII-SpeI digested p513-EGFP vector.

Database of CDR3 sequences

We used the release 5 (August, 1992) of the Kabat database [35]. This dataset contained 44990 sequences. We extracted the 4643 human VH sequences which were not a pseudogene and were not humanized. H3 sequences were then extracted from this dataset taking first into account the nucleotide sequence when present then the amino acid sequence. Finally we kept the 3469 complete H3 sequences that contained only the 20 regular amino acids, among which 2703 were unique. The same procedure was followed for λ and κ light chains, respectively, resulting in 1044 and 1291 sequences from which 775 and 828 were unique.

We also extracted CDR3 sequences from the IMGT/LIGM-DB database the 27 Nov. 2003 [36]. We considered only the "productive/regular/human/cDNA+RNA/Rearranged" genes. We obtained 5179 H3, 1432 K3 and 1131 L3 sequences from which 4323, 974 and 812 were unique.

We collected 127 additional human antibody sequences from the Protein data bank [37]. For this we used the file of 510 sequences already compiled by Andrew Martin the 19 August 2003 [60]. The complete list of CDR3 sequences is available on request to PM.

Spiked oligonucleotide design

In biasing the representations of the amino acids, optimized mixtures of the nucleotides at each of the three codon positions were calculated as described previously [31, 61]. Premature termination of protein sequences was limited by imposing and upper bound of 0.05 on the probability of realizing a stop codon. For the 34 positions which did not satisfactorily recover the desired probabilities of the amino acids, a second optimization was done with the same method but with no constraint on the stop codon frequency. For oligonucleotide synthesis, the calculated frequencies were rounded in increments of 5% as follows: all the frequencies between 0% and 5% were rounded to 5%; other frequencies were rounded to the nearest 5%; if the resulting sum was higher than 100%, 5% was removed from the rounded amino acid frequency larger than 5% for which the difference between the rounded and the target frequency was maximal and the process iterated until the sum was 100%; if the sum was lower than 100%, 5% was added to the rounded frequency lower than 95%, for which the difference between the rounded and the target frequency was maximal and the process iterated until the sum was 100%. The sequences of the oligonucleotides are given in the additional file 2 "Sequences of the spiked oligonucleotides used to introduce the random CDR3 loops".

Construction of VH and VL libraries

Variable CDR3 sequences were introduced in scFv13R4 by PCR assembly using a hot-start proofreading polymerase (ProofStart, Qiagen) using as template plasmid pAB1-scFv13R4p [10]. To introduce random H3 loops, the 5' of the gene with the random H3 sequence was obtained with oligonucleotides M13rev-49 and one of the 13 degenerate oligonucleotides (H3_n, see additional file 2: Sequences of the spiked oligonucleotides used to introduce the random CDR3 loops) and the 3' with PliaisonH3 and M13uni-43 (both for 20 cycles at 55°C). The two purified bands were thus assembled by PCR (30 cycles, 55°C) using M13rev-49 and M13uni-43. The resulting PCR was purified using a commercial kit (Nucleospin, Macherey-Nagel), digested for 16 h at 37°C with NcoI and NotI enzymes, then purified on gel. The same procedure was followed to introduce random L3 and K3 loops except that the pairs of primers were M13rev-49/PliaisonL3 for the 5' and one of the degenerate oligonucleotide encoding the L3/K3 loop (K3_n or L3_n, see oligo.txt) with M13uni-43 for the 3' part of the gene.

Each digested band was ligated for 16 h at 16°C with 1 μg of NcoI-NotI digested and purified pscFvΔCAT in 100 μl using 10 Weiss units of T4 DNA ligase. The ligation was heat inactivated and purified using a commercial kit (Nucleospin). The ligation was then electroporated in 300 μl of MC1061 competent cells [62] and plated on a 600 cm² square plate of LB containing 100 μg/ml of ampicillin and incubated for 16 h at 37°C. The 21 libraries (13 VH and 5 VL) were scrapped in 10 ml of LB with 10% glycerol and 10⁹ bacteria were immediately plated on a 600 cm² square plate of LB containing 100 μg/ml of ampicillin, 1 mM IPTG and 30 μg/ml of CAM and incubated for 16 h at 37°C. The 21 libraries were scrapped in 10 ml of LB with 10% glycerol and frozen at -80°C. An aliquot was used to prepare DNA for the library assembly.

Library assembly

The 13 VH libraries were amplified using primers M13rev-49/PliaisonH3.back using Pfu polymerase and the 5 VL libraries using scFvCAT.rev/H3_Liaison (30 cycles, 55°C). The 21 PCR bands were purified then carefully quantified on gel using ImageJ software [63]. The 13 VH bands were pooled in amounts proportional to their frequency in human H3. This mix is called VHpool. The 2 VL κ bands were pooled in order to obtain 75% of 9 amino acids and 25% of 10 amino acids loops. The VL λ bands were pooled to obtain 30% of 9, 30% of 10 and 40% of 11 amino acids loops. Finally the κ and λ mixes were pooled in order to get 50% of each class in the final mix called VLpool.

VHpool and VLpool were assembled by PCR using Taq DNA polymerase and primers M13rev-49/scFvCAT2.rev in 500 μl (30 cycles, 55°C). The PCR was successively digested with 20 units of NcoI and NotI for at least 6 h each, purified then quantified on gel. 50 μg of vector pCANTAB6 was successively digested with 80 units of NcoI and NotI for at least 6 h each, purified then quantified on gel. 5 μg of linearized pCANTAB6 was ligated with an equal molar amount of insert (0.84 μg) in 500 μl at 16°C using 50 Weiss units of T4 DNA ligase. The ligation was heat inactivated and purified using a commercial kit (Nucleospin). The purified ligation was then electroporated in 10 × 300 μl of C-Max5αF' competent cells [62], plated on ten 600 cm² square plate of LB containing 1% of glucose and 100 μg/ml of ampicillin. After incubation for 16 h at 37°C, cells were scrapped in 2xYT containing 10% of glycerol and kept frozen at -80°C in aliquots corresponding to twenty times the diversity.

Antigens

Aurora-A is an His-tagged protein produced in E.coli. GST:Syk was expressed in E.coli and was a gift from P. Dariavach [64]. E6 protein from papillomavirus HPV16 was expressed in cyanobacterium Anabaena (Desplancq et al., to be published). Histones (a mix of H2a, H2b, H3 and H4) were purchased from Sigma (type II-AS. #H7755). Tubulin was purified from pig brain [65].

Library rescue and selection

Library rescue was done essentially as previously described using a trypsin sensitive helper phage [66]. Briefly, an aliquot of the library corresponding to a 10 to 20 fold excess over the diversity (2–3 × 10¹⁰ bacteria) was inoculated in 1000 ml of 2xYT containing 100 μg/ml ampicillin and 1% glucose and grown with shaking at 37°C until OD_600nmwas 0.7. 200 ml (~3 × 10¹⁰ cells) were infected with 5 × 10¹¹ helper phage KM13 [66] and incubated without shaking for 30 min at 37°C. Cells were pelleted, resuspended in 1000 ml of 2xYT containing 100 μg/ml ampicillin and 25 μg/ml kanamycin and incubated overnight with vigorous shaking at 30°C. The supernatant containing phages was precipitated twice by adding 1/5^th of the volume of PEG-8000 20%, NaCl 2.5 M, resuspended in PBS supplemented with 15% of glycerol and aliquots containing 10¹¹-10¹² phages were stored at -80°C.

To select for binders, 100 μl of purified antigens were coated in a Nunc Maxisorp 96-well plate. For the first round we used an antigen concentration of 10–100 μg/ml and 1–10 μg/ml for the subsequent rounds. Plate was washed 3 times with PBS containing 0.1% of Tween20 (PBST) and saturated 2 h at RT with PBS containing 2% of non fat milk (MBPS). 10¹¹-10¹² phages were added per well in 2% MPBS and incubated for 2 h at RT. The plate was washed 20 times (first round) or 40 times (2^nd and 3^rd rounds) with PBST then 3 times with PBS. Excess PBS was removed and the phages were eluted by adding 100 μl of 100 mM triethylamine for 10 min at RT. The eluted phage suspension was neutralized with 50 μl of 1 M Tris-HCl pH 7.4, then digested 15 min at RT with trypsin by adding 1.5 μl of 0.1 M CaCl₂ and 15 μl of 10 mg/ml TPCK-treated trypsin (Sigma). 1 ml of a 37°C exponentially growing Cmax5αF' strain in 2xYT was infected with 40 μl of trypsin-treated phages, incubated 30 min at 37°C without shaking, then plated on a 15 cm round Petri dish (LB, 100 μg/ml ampicillin, 1% glucose). After overnight incubation at 37°C, bacteria were recovered from the plate and used to prepare a new stock of phages using KM13 helper phage. 10¹¹-10¹² phages of this stock were used for the next round of selection.

Periplasmic and cytoplasmic screening

For periplasmic screening, phages from round 3 were used to infect the non-suppressive strain HB2151. Individual clones were tested for scFv expression by ELISA on antigen-coated 96-well microtiter plates as described [67]. For cytoplasmic screening, plasmid was prepared from the pool of bacteria of the 2^nd or 3^rd selection round, digested with NcoI and NotI enzymes, and the 750 bp band was cloned in NcoI-NotI digested and dephosphorylated plasmid pET23NN. Ligation was transformed in C-Max5αF', plated on LB with 100 μg/ml ampicillin and incubated for 16 h at 37°C. Cells were scrapped and the plasmid DNA prepared and used to transform chemically competent BL21(DE3)pLysS. Individual clones were grown in a 96-well microtiter plate containing 100 μl of 2xYT, 100 μg/ml of ampicillin with vigorous shaking at 37°C until OD_600nmreached 0.6. IPTG was added to 0.4 mM final and the microtiter plate incubated for 16 h at 24–30°C with vigorous shaking in a humidified atmosphere. After centrifugation, cells were resuspended in 100 μl of 50 mM Tris-HCl pH7.5, 5 mM EDTA, freeze/thawed and incubated 1 h on ice. MgCl₂ was added up to 10 mM and the DNA was digested with 10 μg/ml of DNAseI. 5–20 μl were used in an ELISA on an antigen-coated 96-well microtiter plate (Nunc Maxisorp). Revelation was done using 9E10 monoclonal antibody followed by an HRP conjugated anti-mouse IgG antibody.

Purification of scFv

scFvs cloned in plasmid pET23NN were purified from the cytoplasm of BL21(DE3)pLysS and purified on a Ni-NTA column as described for the parental scFv13R4 [26].

Cell transfection and immunofluorescence

HeLa cells were maintained in Dulbecco's modified Eagle's tissue culture medium (DMEM; Invitrogen) supplemented with L-glutamine (2 mM), penicillin (100 IU/ml), streptomycin (25 μg/ml) and 10% heat-inactivated fetal calf serum at 37°C in a humidified 5% CO₂ atmosphere. Transient transfection was carried out with the TransFectin lipid reagent (Bio-Rad, Hercules, CA, USA) according to the manufacturer's instructions. Cells were seeded on coverslips in 6-well plates at 2.5 × 10⁵ cells/well the day before transfection. 1 μg DNA and 2 μl of reagent diluted in 100 μl of DMEM were mixed and left at room temperature for 20 min. Cells were grown at 37°C for 24 h after addition of the mixture. The expressed GFP-tagged proteins were visualized after fixation of the transfected cells with 4% paraformaldehyde in PBS during 45 min at room temperature. After extensive wash with PBS, cells were dried and mounted with Fluoromount-G (SouthernBiotech, Birmingham, UK). The processed cells were examined with a Zeiss Axioplan fluorescence microscope equipped with an Olympus DP50 camera. Images were collected with a Zeiss 40× plan-neofluar objective and processed using Adobe Photoshop 5.5. For figure 8, HeLa were transfected with anti-histones clone 5 fused to the dsRed-monomer GFP, fixed as above and permeabilized with Triton ×-100 (0.2%, 5 min). The microtubule network was revealed with the 2F12C scFv (Table 2) at 3 μg/ml using the 9E10 anti-myc and an Alexa Fluor 488 anti-mouse IgG antibody. Cells were observed by confocal microscopy (×63).