Single-domain antibodies (sdAbs), which occur naturally in camelids, are endowed with many characteristics that make them attractive candidates as building blocks to create new antibody-related therapeutic molecules. In this study, we isolated from an immunized llama several high-affinity sdAbs directed against human carcinoembryonic antigen (CEA), a heavily glycosylated tumor-associated molecule expressed in a variety of cancers. These llama sdAbs bind a different epitope from those defined by current murine mAbs, as shown by binding competition experiments using immunofluorescence and surface plasmon resonance. Flow cytometry analysis shows that they bind strongly to CEA-positive tumor cells but show no cross-reaction toward nonspecific cross-reacting antigen, a highly CEA-related molecule expressed on human granulocytes. When injected into mice xenografted with a human CEA-positive tumor, up to 2% of the injected dose of one of these sdAbs was found in the tumor, despite rapid clearance of this 15 kDa protein, demonstrating its high potential as a targeting moiety. The single-domain nature of these new anti-CEA IgG fragments should facilitate the design of new molecules for immunotherapy or diagnosis of CEA-positive tumors.
Cancer immunotherapy, either active, i.e. based on the stimulation of specific anti-tumor responses with tumor-associated antigens/peptides as the immunizing materials, or passive, i.e. based on the injection of mAbs, is now delivering an increasing amount of encouraging data. Notably, several mAbs have been approved for therapeutic use over the last decade. Antibody engineering makes it possible to design new molecules capable of increasing the efficiency of antibody-based therapies. Since the discovery that functional heavy-chain gamma-immunoglobulins lacking light chains occur naturally in the Camelidae , several groups have reported the isolation of single-domain antibodies (sdAbs) consisting of the variable domain of these heavy chain antibodies, also named VHH . These minimal antibody domains are endowed with a large number of properties that make them very attractive for antibody engineering. Despite the reduced size of their antigen-binding surface, VHH domains exhibit affinities similar to those of conventional mAbs and are also capable of binding small molecules as haptens [3,4]. Strikingly, they often use complementarity determining region (CDR) 3 longer than the one of VH domains, which allow them to bind otherwise difficult-to-reach epitopes within the cavities on the antigen surface. Consequently, these fragments can recognize epitopes inaccessible to conventional antibodies and are a good source of enzyme inhibitors . Most importantly, the single-domain nature of VHH permits the amplification and subsequent straightforward cloning of the corresponding genes, without requiring an artificial linker peptide (as for single-chain Fv fragments) or bi-cistronic constructs (as for Fab fragments). This feature allows direct cloning of large sdAb repertoires from immunized animals, without the need to be concerned by the usual disruption of VH/VL pairing faced when generating scFv and Fab fragment libraries. The sdAb format is also likely responsible for the high production yield obtained when these domains or sdAb-based fusion molecules are expressed. A number of sdAb and sdAb-derived molecules has been produced in large amounts in prokaryotic  and eukaryotic [7,8] cell lines, and in plants . Moreover, VHH fragments show exquisite refolding capabilities and amazing physical stability . Last, but not least, the genes encoding VHH show a large degree of homology with the VH3 subset family of human VH genes , which might confer low antigenicity in humans, a very attractive feature for immunotherapeutic approaches. Taken together, these data make VHH excellent candidates to engineer multispecific or multifunctional proteins for immunotherapy .
Carcinoembryonic antigen (CEA or CEACAM5), a member of the immunoglobulin supergene family, is a heavily glycosylated protein involved in cell adhesion and normally produced by fetal gastrointestinal tissues. It was first described by Gold and Freedman in 1965  as a high-molecular mass glycoprotein (∼ 180 kDa) found in colonic tumors and fetal colon, but not in normal adult colon; its expression has since been described in almost all tumors (> 95%) including rectum, breast, lung, liver, pancreas, stomach, thyroid and ovarian tumors. Nonspecific cross-reacting antigen (NCA or CEACAM6) is a highly related member of the same CEACAM family. CEA and NCA polypeptides have extracellular domains, some with cysteine-linked loops, that share extensive amino acid sequence homology (∼ 78% overall) with each other and appear similar to other immunoglobulin superfamily members. A major difference between the two apoproteins is the presence of a single loop-domain in NCA, compared with three tandemly repeated loop-domains in CEA. Comparisons between the extracellular domains of CEA and NCA show that the N-terminal and adjacent loop-domains of each apoprotein have high sequence homology (85–90%). Consequently, many mAbs raised against CEA also bind with high affinity to NCA. CEA has been extensively chosen as the target for directed cancer therapies aimed at selectively destroying cells expressing this tumor antigen but sparing normal cells. Based on a large body of evidence indicating that CEA is associated with the growth and metastasis of cancers , this tumor marker represents an interesting model target with which to monitor the efficacy of sdAb-based multifunctional molecules.
As a first step towards the generation of new CEA-targeted therapeutic molecules, we generated a panel of llama-derived sdAbs capable of binding with high affinity to CEA. Importantly, we were able to select CEA-specific sdAbs showing no cross-reaction with NCA that is expressed on several normal cell types, including granulocytes. Moreover, these sdAbs do not bind to known epitopes recognized by monoclonal murine anti-CEA IgGs. Thus, these sdAbs represent versatile tools to generate potent antibody-based molecules for cancer therapy.
Isolation of sdAbs against CEA
A male llama was immunized subcutaneously five times with 250 μg recombinant purified soluble CEA (sCEA) per injection. A library of 106 clones was obtained by RT-PCR amplification and cloning of VHH genes, using RNA purified from llama peripheral blood cells. A classic issue with anti-CEA IgGs is their high tendency to cross-react with the highly related NCA receptor. To increase the chance of selecting a diverse panel of sdAbs against CEA, two antigen immobilization methods were used. Recombinant human sCEA was either directly immobilized by adsorption onto plastic (immunotubes) (Method A), or indirectly immobilized on magnetic beads via a biotin/streptavidin system (Method B) (see Material and Methods).
After one round of affinity selection, 48 clones randomly picked from each output were assayed by phage-ELISA for binding to biotinylated sCEA immobilized on streptavidin plates. All clones picked from the output of Method A were positive, whereas only 61% from Method B were positive. Consequently, two more rounds were performed for Method B, which ultimately led to 100% of binders by phage ELISA.
Sequence analysis of the 48 clones picked from Method A (round 1) revealed that three highly related antibodies, displaying identical residues in their CDRs, dominated the population. Sequence analysis of the 48 clones picked at random from method B (rounds 1–3) revealed that the output included five sdAbs, namely C3, C17, C25, C43 and C44. C44 was the clone dominating the output of Method A.
Interestingly, the amino acid alignment (Fig. 1) showed that sdAbs C3, C17, C25 and C43 are likely clonally related, despite the presence of a relatively high number of differences scattered all along the gene. CDR3 of C3, C25 and C43 are very similar, suggesting use of the same D gene. C17 shows very similar CDR1 and CDR2, but a rather different CDR3. Interestingly, CDR1 and CDR3 of clone C44 are totally unrelated to CDR1 and CDR3 of the other sdAbs. In all cases, the presence of an arginine at position 50 (Fig. 1) confirmed the Camelidae nature of these sdAbs. All of them belong to subfamily VHH2 .
Affinity determination by surface plasmon resonance
To further characterize these sdAbs, the corresponding cDNA were cloned into the expression vector pPelB55PhoA’ , allowing efficient production and purification of the molecules. SdAbs harboring a hexahistidine tag at the C-terminus were produced in the periplasm of Escherichia coli and purified by immobilized ion metal-affinity chromatography. Final yields were in the range 5–10 mg·L culture−1 for all clones. SDS/PAGE analysis demonstrated a satisfying degree of purity (> 95%, data not shown). Pure sdAbs were then indirectly immobilized on BIAcore sensorchips and their affinity for soluble CEA determined. As shown in Table 1, all sdAbs exhibited a good affinity for sCEA, with a KD ranging from 3 to 32 nm.
Table 1. Kinetic and affinity constants of the binding of soluble carcinoembryonic antigen (sCEA) to CEA-specific sdAbs immobilized via monoclonal anti-(c-myc) 9E10 on CM5 microchips. A Langmuir 1 : 1 model was used to fit six different sCEA concentrations. No binding of sCEA on immobilized 9E10, mass transport or rebinding effect was observed. χ2, statistical value for describing the closeness of the fit. Values of χ2 < 10% of the Rmax are usually acceptable. RU, relative units.
ka × 105 (1·Ms−1)
kd × 10−3 (1·s−1)
8.21 ± 0.040
2.64 ± 0.004
1.78 ± 0.019
1.83 ± 0.002
1.13 ± 0.014
3.60 ± 0.004
1.56 ± 0.014
1.30 ± 0.002
1.24 ± 0.014
1.68 ± 0.002
Specificity analysis by flow cytometry
Flow cytometry was used to determine whether the selected sdAbs specifically bind to CEA+, but not CEA−, cells and to examine if any cross-reaction with NCA could be detected.
sdAbs were assayed by flow cytometry for binding to colon cancer MC38 cells (CEA− NCA−), or to transfected MC38 cells expressing either CEA or NCA (kindly provided by F.J. Primus, Vanderbilt University Medical Center, Nashville, TN, USA). Cells from the CEA+ colon cancer cell line LS174T, as well as from freshly purified human granulocytes that display NCA but not CEA, were also tested.
Figure 2 shows that C17 sdAb efficiently binds to CEA-expressing cells (MC38–CEA+ and LS174T) but not to NCA+ human granulocytes. C3, C25 and C43 sdAb binding profiles were identical to that of C17 sdAb (data not shown). By contrast, sdAb C44 was also capable of binding to MC38–NCA+ cells and to human granulocytes, whereas all other sdAbs did not show any binding to these cells. C44 sdAb is therefore recognizing an epitope shared by CEA and NCA, in contrast to C3, C17, C25 and C43, which are strictly specific for CEA. Monoclonal antibodies that either bind both CEA and NCA (mAb 192) or only CEA (mAb 35A7) were used as controls (Fig. 2).
Competitive inhibition of antibody binding to LS174T cells
To further characterize the binding properties of sdAbs C3, C17, C25, C43 and to investigate whether different CEA epitopes were recognized by these antibodies, binding competition experiments were performed using cells from the CEA-expressing cell line LS174T. Cells were incubated with trace amounts of 125I-labeled sdAb C17 (0.4 nm) in the presence of increasing concentrations of unlabeled sdAbs. As shown in Fig. 3, sdAbs C25, C3 and C43 were able to compete with C17, indicating that these sdAbs bind to overlapping epitopes or to the same epitope. Moreover, sdAb IC50 is in the nm range for three of the four sdAbs (C3, 1.6 ± 0.4 nm; C17, 7.8 ± 1.3 nm; C43, 5.2 ± 0.3 nm). Only sdAb C25 exhibits a significantly lower apparent affinity (59.1 ± 0.6 nm). These values obtained from cell-binding experiments are in good agreement with surface plasmon resonance (SPR) data (Table 1), except for sdAb C3 (KD = 13.6 nm). This difference might be because of a different conformation and/or glycosylation of the sdAb epitope on cell-displayed CEA and on recombinant sCEA immobilized on sensor chips.
Epitope analysis by surface plasmon resonance
Most CEA-specific mAbs available to date can be classified into five categories (Gold 1-5) according to their epitope . To determine if one or more of these epitopes was recognized by the CEA-specific sdAbs (C3, C17, C25, C43), a qualitative SPR-based sandwich assay was used. sdAbs were first captured via their c-myc tag on the CM5 sensorchip surface coated with the monoclonal anti-c-myc IgG 9E10 to favor a good exposition of the captured sdAbs. Recombinant sCEA was then injected and captured by the sdAbs. Subsequently, one of the five Gold mAbs recognizing one of the five Gold epitopes was injected. Under these conditions, binding of the Gold mAb indicates that its corresponding epitope is not blocked by the capturing sdAb. All Gold mAbs were tested against all CEA-specific sdAbs. Sensorgrams obtained with sdAb C17 are shown in Fig. 4A. All sdAb and Gold mAb combinations led to efficient binding of Gold mAb to the captured sCEA, demonstrating that none of the five Gold epitopes is recognized by the CEA-specific sdAbs. All five anti-CEA Gold IgGs were able to bind captured sCEA, whereas an irrelevant mAb (mouse anti FcγRIII) did not. As a positive of competition, a version of the sdAb used for CEA capture but devoid of the c-myc tag was injected instead of Gold mAbs. No increase in signal was obtained, showing that competition can be efficiently demonstrated between these molecules in our assay. An irrelevant sdAb [anti-(HIV-1 NEF) devoid of c-myc tag] injected at the same concentration was used as a negative control.
Moreover, in a reverse scheme, all sdAbs were able to bind to recombinant sCEA captured on the chip via Gold mAbs covalently immobilized on CM5 sensorchip. Figure 4B shows the results obtained with the mAb 35A7 as an example.
Epitope analysis by flow cytometry
The absence of competition between gold mAbs and sdAbs was confirmed using nonrecombinant cell-surface displayed CEA by flow cytometry. CEA-expressing cells were first incubated with very high concentrations (up to 2190 nm) of sdAb C17 devoid of the c-myc tag. After 1 h of incubation, subsaturating concentrations of either gold mAbs or c-myc-tagged sdAb C17 or C43 (30-70 nm as determined in a previous flow cytometry experiment, data not shown) were added to the wells. After an additional 1 h of incubation, bound gold mAbs and c-myc-tagged sdAbs were revealed. As shown in Fig. 5, the presence of a large excess (more than two orders of magnitude) of untagged sdAb C17 did not interfere with the binding of gold mAbs, but completely inhibited the binding of c-myc-tagged sdAb C17 or sdAb C43 that bind to the same epitope (see above). These results confirmed that these sdAbs do not bind to any of the epitopes recognized by gold mAbs.
In vivo localization
To analyze the behavior of sdAbs against CEA under more physiological conditions, immunocompromised mice were xenografted with LS174T cancer cells. Once tumors were established (i.e., day 7), radiolabeled sdAb C17 (displaying the highest affinity as measured by SPR) was injected and the biodistribution was monitored.
A fast blood clearance was observed, as assessed by the low residual blood radioactivity 6 h after injection [0.30 ± 0.06% of the injected dose per gram % ID·g−1 ± SEM)]. Activity uptake was observed primarily in kidneys (7.4 ± 0.4% ID·g−1 3 h post injection and 4.8 ± 0.6% ID·g−1 at 6 h post injection). Three hours after injection, 1.9 ± 0.1% of the injected dose was localized in the tumor. This was at least twofold higher than the radioactivity found in blood, liver and major organs except kidneys, and was 3.5- and 5-fold higher than the radioactivity found in bones and leg muscles, respectively (Fig. 6). Six hours after injection, the increased clearance resulted in higher tumor-to-normal tissue uptake ratios for all these organs, reaching a ratio of 10 in the case of muscles.
As a first step toward the construction of multispecific and/or multivalent molecules aiming at redirecting immune cells such as T cells or NK cells to tumor cells, we isolated sdAbs able to bind to CEA (or CEACAM5), a tumor marker used in cancer diagnosis and immunotherapy. We used phage display to select binders from one sdAb library derived from peripheral blood mononuclear cells isolated from an immunized llama. Interestingly, two selection methods led to strikingly different outputs. Selection by panning on recombinant sCEA directly adsorbed on plastic allowed the isolation of a single family of highly related sdAbs that dominated the selected population only after a single round of selection. However, the epitope recognized by these antibodies is also present on a highly related molecule, nonspecific cross-reacting antigen (NCA or CEACAM6) that shares the same Ig domain-based structure with CEA and displays a high percentage of sequence homology. By contrast, a selection based on biotinylated sCEA captured on streptavidin-coated magnetic beads led to only 61% binders after a single round and two more rounds of selection were needed to reach 100% binders. Unlike the first method, the output of this selection was more diverse. This method yielded antibodies belonging to the family selected by panning on coated sCEA, but also made it possible to isolate several other clones displaying very similar CDR1 and CDR2, but more diverse CDR3, suggesting use of the same VHH gene but different D genes. Flow cytometry analyses demonstrated that these sdAbs bind specifically to CEA expressed on cancer cells but do not cross-react with NCA. These antibodies were not present in the output of the panning method, despite very similar affinities for the antigen, in the nm range, as determined by SPR. One can hypothesize that the conformational changes resulting from the adsorption of CEA on plastic are either denaturing the epitope recognized by the second family of sdAbs or are favoring the display of the epitope recognized by the first family, leading to a large presence of this latter family during the selection process.
As expected for llama VHHs, the sequences of the CEA-specific sdAbs are homologous to human IGHV3 subgroup genes (C3, 79% homology to IGHV3-23; C17, 68% homology to IGHV3-74; C25, 69% homology to IGHV3-48; C43, 69% homology to IGHV3-13). The most divergent sequences between the four llama sdAbs and human IGHV3 are localized in the CDRs, and in the former VL and CH1 interfaces (residues 40, 42, 49, 50 and residues 15 and 96, respectively). Moreover, the selected llama sdAbs belong to subfamily VHH2 . As described earlier for VHH belonging to this family, CDR3 from these four sdAbs do not contain an additional disulfide bond and do not exceed the mean CDR3 length of classical VH, in contrast to most camel VHH .
Of note is that that none of these antibodies binds to one of the Gold epitopes. These essentially nonoverlapping epitopes have been defined by analyzing the binding specificities of 52 monoclonal anti-CEA IgGs and define five antigenic regions recognized by murine mAbs . In this study, the only four mAbs not binding to these five regions were directed against carbohydrate epitopes, suggesting that the rest of the CEA surface does not elicit antibodies. Epitope analysis performed on recombinant sCEA using SPR or on the surface of living cells demonstrated that the sdAbs target overlapping epitopes and might even share a unique epitope, because it could be anticipated by the high degree of homology of their CDRs. Interestingly, this is not one of the Gold epitopes because no competition was observed between Gold mAbs and the sdAbs, as demonstrated both by SPR on soluble CEA and flow cytometry experiments on cell-displayed CEA. This new epitope, not found on NCA, is therefore not easily detected by murine mAbs. This finding supports a previous study showing that sdAbs have a tendency to bind to epitopes usually invisible to other mAbs, such as cavities, and are a rich source of enzyme inhibitors . In the case of CEA, a heavily glycosylated molecule, one can also hypothesize that the oligosaccharide chains can hinder the access of some regions of the polypeptide to large molecules such as mAbs (150 kDa) but not to very compact sdAb (13 kDa). However, it should be reminded that the sdAbs were selected as phage–sdAbs, which implies that the large phage particle did not prevent binding of the sdAbs to this epitope.
IC50 values calculated from cell-surface competition experiments and KD values measured by SPR are in the nm range for sdAbs C17, C3 and C43, demonstrating that these sdAbs, selected against a recombinant ectodomain of CEA can efficiently bind their antigen when displayed at the cell surface of human tumor cells. The high affinities of the selected sdAbs, which compares favorably with conventional mAbs despite their monovalency, should allow an efficient in vivo targeting of tumor cells expressing CEA. To verify this hypothesis, we conducted an in vivo localization experiment in LS174T-xenografted nude mice with sdAb C17, which binds to CEA with the highest affinity as measured by SPR. Blood clearance of this sdAb was fast because low blood radioactivity was observed as early as 6 h after injection. This result is in agreement with the sdAb blood half-life that has been estimated to be 20–40 min in mice [20,21]. Despite this rapid blood clearance and the monovalent nature of the sdAb excluding an avidity effect, almost 2% of the injected dose accumulates in tumor tissues (fivefold higher than in muscle tissues), which compares well with the results of Cortez-Retamozo et al. . These authors injected LS174T-xenografted mice with a CEA-specific sdAb fused to beta-lactamase and 2.8% of the total injected dose was found in the tumor 6 h after injection, despite a blood half-life expected to be significantly higher for this 45 kDa construct than for sdAb C17 (13 kDa). In this study, 3 h after injection, 7% ID·g−1 were found in kidneys, which decreased to 5% at 6 h post injection. This renal accumulation is expected for very small molecules such as sdAbs (13-15 kDa). Ultrafiltration of low molecular mass proteins and subsequent uptake by proximal tubular cells followed by lysosomal degradation leads to the intracellular accumulation of radioactivity. It is expected that systemic administration of basic amino acids may reduce renal retention of radioiodinated sdAbs because it is efficient in lowering kidney uptake of antibody fragments [22a]. The low activity accretion observed in other organs led to tumor-to-organ radioactivity uptake ratios of at least 2 (range 2–5) 3 h after injection and 3 (range 3–9) at 6 h after injection. Overall, sdAb C17 showed an expected biodistribution profile, demonstrating its utility as a CEA+ tumor-targeting molecule.
In conclusion, the specificity, affinity and single-domain structure of the new CEA-specific sdAbs isolated here make them very attractive candidates to build, together with sdAbs targeting receptors such as CD16 (FcγRIII)  or other activating receptors or radiolabeled haptens , new multivalent and/or multispecific molecules with superior characteristics for immunotherapy or radioimmunotherapy.
Material and methods
A young adult male llama (Lama glama) was immunized subcutaneously at days 1, 30, 60, 90 and 120 with 250 μg recombinant human soluble CEA extracellular domain (sCEA) produced as previously described . Sera were collected 15 days prior to each injection to follow the immune response against the immunogen.
VHH library construction
Blood samples (100 mL) were taken 15 days after each of the three latest immunizations and peripheral blood mononuclear cells were isolated by Ficoll-Histopaque-1077 (Sigma-Aldrich, St. Louis, MO, USA) discontinuous gradient centrifugation. Total RNA was isolated by acid guanidinium thiocyanate/phenol/chloroform extraction  and synthesis of the cDNA was performed with Superscript II reverse transcriptase (GibcoBRL, Gaithersburg, MD, USA) using primer CH2FORTA4 . A first PCR was performed using an equimolar mixture of four backward primers originally designed to anneal on human VH genes (5′ VH1–Sfi: 5′-CATGCCATGACTCGCGGCCCAGCCGGCCATGGCCCAGGTGCAGCTGGTGCAGTCTGG-3′; 5′ VH2–Sfi: 5′-CATGCCATGACTCGCGGCCCAGCCGGCCATGGCCCAGGTCACCTTGAAGGAGTCTGG-3′; 5′ VH3–Sfi: 5′-CATGCCATGACTCGCGGCCCAGCCGGCCATGGCCGAGGTGCAGCTGGTGGAGTCTGG-3′; 5′ VH4–Sfi: 5′-CATGCCATGACTCGCGGCCCAGCCGGCCATGGCCCAGGTGCAGCTGCAGGAGTCGGG-3′) and one forward primer (CH2FORTA4). These primers allow the amplification of two bands corresponding two the VH + CH1 + hinge + part of CH2 gene fragment of traditional antibodies or the VHH + hinge + part of CH2 gene fragment of HcAbs. Using the gel-purified (Qiaquick gel extraction kit; Qiagen, Hilden, Germany) lower band as the template, VHH genes were re-amplified using an equimolar mixture of the four backward primers (5′ VH1 to 4-Sfi) and 3′ VHH–Not primer (5′-CCACGATTCTGCGGCCGCTGAGGAGACRGTGACCTGGGTCC-3′) containing SfiI and NotI restriction enzyme sites. Resulting VHH fragments were purified from 1% agarose gel, digested with SfiI and NotI and ligated into pHEN1 phagemid  digested with SfiI and NotI. The ligated material was transformed into TG1 E. coli electroporation-competent cells (Stratagen, Miami, FL, USA). Cells were plated on 2YT/ampicillin (100 μg·mL−1)/glucose (2%) agar plates. Colonies (106) were scraped from the plates with 2YT/ampicillin (100 μg·mL−1)/glucose (2%), and stored at −80 °C in the presence of 20% glycerol. Because llamas were hyperimmunized, a library containing a million of different clone can be considered as representative.
Selection of phage–sdAbs
Selections were performed as described previously . Briefly, 10 μL of the library was grown in 50 mL of 2YT/ampicillin (100 μg·mL−1)/glucose (2%) at 37 °C to an D600 of 0.5. Five milliliters of the culture were then infected with 2 × 1010 M13KO7 helper phage for 30 min at 37 °C without shaking. The culture was centrifuged for 10 min at 3000 g. The bacterial pellet was resuspended in 25 mL of 2YT/ampicillin (100 μg·mL−1)/kanamycine (25 μg·mL−1), and incubated for 16 h at 30 °C with shaking (270 rpm). The culture was then centrifuged for 20 min at 3000 g and one-fifth of the volume of 20% PEG 6000, 2.5 m NaCl was added to the supernatant and incubated for 1 h on ice to precipitate phage particles. The solution was then centrifuged for 15 min at 3000 g at 4°C and the phage-containing pellet was re-suspended with 1 mL of NaCl/Pi.
Phage were selected using either immunotubes coated with recombinant sCEA  (10 μg·mL−1 in NaCl/Pi, overnight 4 °C) or biotinylated sCEA and streptavidin-coated paramagnetic beads (Dynabeads M-280; Dynal Biotech, Oslo, Norway). Recombinant sCEA was biotinylated using a biotin protein-labeling kit according to the manufacturer’s instructions (Roche, Basel, Switzerland). Two hundreds microliters of beads were mixed with 1 mL NaCl/Pi containing 2% skimmed milk powder (NaCl/Pi/2% milk) for 45 min at room temperature in a siliconized Eppendorf tube. Beads were washed with NaCl/Pi/2% milk using a magnetic particle concentrator and resuspended with 250 μL NaCl/Pi/2% milk. We added 200 μL of biotinylated sCEA and the solution was gently rotated for 30 min at room temperature; 150, 75 and 25 nm of biotinylated sCEA were used for the first, second and third rounds of selection, respectively. We then added 450 μL of the phage preparation (1012 pfu), preincubated for 1 h in 500 μL NaCl/Pi/2% milk. The mixture was rotated for 3 h at room temperature and washed five times with 800 μL NaCl/Pi/4% milk, five times with 800 μL NaCl/Pi containing 0.1% Tween and five times with 800 μL NaCl/Pi. Every five washes, the mixture was transferred to a new siliconized tube. Phage fixed on sCEA-coated beads were resuspended with 200 μL NaCl/Pi and incubated without shaking with 1 mL of log-phase TG1 cells and plated on 2YT/ampicillin (100 μg·mL−1)/glucose (2%) in 243 × 243 mm dishes (Nalgene Nunc, Roskilde, Denmark). Some isolated colonies were grown overnight in microtiter plate containing 200 μL 2YT/ampicillin (100 μg·mL−1)/glucose (2%) and stored at −80 °C after the addition of 15% glycerol (masterplates). The remaining colonies were harvested from the plates, suspended in 2 mL 2YT/ampicillin (100 μg·mL−1)/glucose (2%) and used for phage production for the next round of selection.
ELISA screening of phage–sdAb
A 96-well plate replicator was used to replicate the masterplates in 120 μL of fresh broth. Colonies were grown for 2 h at 37 °C under shaking (400 rpm) and 35 μL 2YT/ampicillin (100 μg·mL−1)/glucose (2%) containing 2 × 109 M13KO7 helper phage were added to each well and incubated for 30 min at 37 °C without shaking. The plate was centrifuged for 10 min at 1200 g and the bacterial pellet was suspended in 150 μL 2YT/ampicillin (100 μg·mL−1)/kanamycine (25 μg·mL−1) and grown for 16 h at 30 °C under shaking (400 rpm). Phage-containing supernatants were tested for binding to sCEA by ELISA. Briefly, biotinylated sCEA (5 μg·mL−1) was coated on streptavidin 96-well microplates (BioBind assembly streptavidin coated; Thermo Fischer Scientific, Waltham, MA, USA) saturated with NaCl/Pi/2% milk. Fifty microliters of phage supernatant were added to 50 μL NaCl/Pi/4% milk and incubated for 2 h at room temperature in the ELISA microplate. Bound phage were detected with a peroxidase-conjugated monoclonal anti-M13 mouse IgG (GE Healthcare, Munich, Germany). Reading was performed at A405. DNA of positive phage (A405 three times above the blank) was sequenced using abi prism®bigdye™ Terminators (Applied Biosystems, Foster City, CA, USA).
SdAb production and purification
Selected clones were sequenced and amplified by PCR using primers 5′ pJF–VH3–Sfi (CTTTACTATTCTCACGGCCATGGCGGCCGAGGTGCAGCTGGTGG) and 3′ c-myc–6His/HindIII (CCGCGCGCGC CAAGACCCAAGCTTGGGCTARTGRTGRTGRTGRTGRTGTGCGGCCCCATTCAGATC) to add the HindIII site for further cloning, a hexahistidine tag for purification and the c-myc tag for detection. For production of clones without the c-myc tag, the PCR amplification was performed using primers 5′ pJF–VH3–Sfi and 3′ 6His/HindIII (CCGCGCGCGCCAAGACCCAAGCTTGGGCTACTAGCTCCCGTGGTGATGGTGGTGATGTGAGGAGACAGTGACCTG).
PCR fragments were cloned into the pPelB55PhoA′  vector between the SfiI and HindIII sites. E. coli K12 strain TG1 was used to produce the sdAb-tagged fragments. An inoculum was grown overnight at 30 °C in 2YT medium supplemented with 100 μg·mL−1 ampicillin and 2% glucose. Four hundred milliliters of fresh medium were inoculated to obtain an D600 of 0.1, and bacteria were grown at 30 °C to an D600 of 0.5–0.7 and induced with 100 μm isopropyl thio-β-d-galactoside for 16 h. The cells were harvested by centrifugation at 4200 g for 10 min at 4 °C. The cell pellet was suspended in 4 mL of cold TES buffer (0.2 m Tris/HCl, pH 8.0; 0.5 mm EDTA; 0.5 m sucrose), and 160 μL lysozyme (10 mg·mL−1 in TES buffer) was added. The cells were then subjected to osmotic shock by the addition of 16 mL of cold TES diluted 1 : 2 with cold H2O. After incubation for 30 min on ice, the suspension was centrifuged at 4200 g for 40 min at 4 °C. The supernatant was incubated with 150 μL DNase I (10 mg·mL−1) and MgCl2 (5 mm final) for 30 min at room temperature. The solution was dialyzed against 50 mm sodium acetate pH 7.0, 0.1 m NaCl, for 16 h at 4 °C. sdAbs were purified by TALON metal-affinity chromatography (Clontech, Mountain View, CA, USA) and concentrated by ultrafiltration with Amicon Ultra 5000 MWCO (Millipore, Billerica, MA, USA). The protein concentration was determined spectrophotometrically using a protein assay kit (Bio-Rad Laboratories, Hercules, CA, USA).
Kinetic parameters were determined by real-time SPR using a BIACORE 3000 apparatus. Monoclonal anti-c-myc IgG 9E10 was covalently immobilized (3300 RU) on a flow cell of CM5 sensor chip (Biacore AB, Uppsala, Sweden) with EDC/NHS activation according to the manufacturer’s instructions. A control flow cell surface was prepared with the same treatment but without antibody. All analyses were performed at 25 °C, at a flow rate of 30 μg·mL−1 and using HBS-EP (Biacore AB; 10 mm Hepes pH 7.4, 150 mm NaCl, 3.4 mm EDTA and 0.005% Biacore™ surfactant) as running buffer. Each sdAb was injected (90 μL) at a concentration of 50 μg·mL−1 in HBS-EP over 3 min and followed by a 90 μL injection of sCEA at six different concentrations (0.19–6.2 μg·mL−1). A 400 s dissociation step was applied before a pulse of 5 mm HCl to regenerate the flow cell surfaces between each run. The absence of direct sCEA binding to 9E10 was assessed. The control sensorgram obtained by injection of sdAb only on the 9E10 flow cell was subtracted from all other sensorgrams to compensate for sdAb dissociation from 9E10 mAb. Resulting sensograms were fitted to a Langmuir 1 : 1 binding isotherm model and errors on ka and kd were calculated using biaevaluation 3.2 software.
The CEA-positive human colon carcinoma LS174T cell line was obtained from the American Type Culture Collection (Rockville, MD, USA). The murine colon carcinoma MC38 cells either transfected with human CEA (MC38–CEA cell line) or with human NCA (MC38–NCA cell line) were kindly provided by F.J. Primus (Vanderbilt University Medical Center, Nashville, TN, USA) . These cells were cultured in Dulbecco’s modified Eagle’s medium (Gibco Laboratories, Lyon, France) supplemented with 10% heat-inactivated fetal bovine serum (Gibco Laboratories), l-glutamine (300 μg·mL−1), fungizone (0.25 μg·mL−1), streptomycin (100 μg·mL−1), penicillin G (100 Units·mL−1) and geneticin (0.5 mg·mL−1). These cells are adherent and grow as monolayers at 37 °C in a humidified 5% CO2 incubator. Immunofluorescence assays were performed by incubating 5 × 105 indicator cells with sdAbs (10 μg·mL−1) for 30 min on ice. sdAbs binding to MC38, MC38–CEA, MC38–NCA, LS174T cell lines and human granulocytes were then revealed by incubation with the monoclonal anti-c-myc 9E10 IgG (10 μg·mL−1) followed by incubation with fluorescein isothiocyanate (FITC)-labeled F(ab)′2 goat anti-mouse IgG (H+L) (FITC-GAM) antibodies (Jackson ImmunoResearch Laboratories, West Grove, PA, USA). Human granulocytes were purified as described previously .
sdAb C17 (0.5 nmol in 50 μL NaCl/Pi) was iodinated with Na125I (18.5 MBq) using iodogen  for 20 min at 4 °C. Ten microliters of 1 mm dl-tyrosine pH 7.4 was added to the solution and the mix was incubated for a further 5 min. The iodinated antibody was purified by gel-permeation chromatography on a PD 10 column (Sephadex G-25, GE Healthcare, Waukesha, WI, USA).
Cell-binding experiments were performed on cells from the LS174T colon carcinoma cell line (ATCC). We incubated 150 μL of 125I-labeled sdAb C17 (4 × 10−10m final concentration, specific activity: 5 × 1017 cpm·mol−1) with 100 μL of cell suspension (5 × 106 cells·mL−1 final) in binding medium [modified Eagle’s medium with Earle’s salts (GIBCO-Invitrogen-France), 0.2% BSA] in the presence of increasing concentrations of unlabeled sdAb (100 μL in binding medium). After 2.5 h under shaking, 100 μL of the suspensions were centrifuged in triplicate for 30 s through a phthalate mixture . An aliquot of supernatant and the cell pellet from each tube were counted (three experiments, each in triplicate). The non-specific binding was evaluated in the presence of an excess of unlabelled sdAb C17 (2 × 10−7m).
IC50 values and statistics were calculated with graphpad prism® (GraphPad Software, Inc. San Diego, CA, USA) using a one-site competition nonlinear regression analysis.
Epitope mapping by SPR and flow cytometry
In a first set of experiments, epitope mapping of sdAbs and Gold mAbs (B17, CE25, 35A7, B93, 192)  was carried out by SPR at a flow rate of 20 μg·mL−1. First, each sdAb (50 μg·mL−1) was injected on 9E10 mAb immobilized (11 000 RU) on a CM5 sensorchip. Second, 60 μL of sCEA antigen (25 μg·mL−1) were injected and, third, one of the Gold mAbs was injected (10 μg·mL−1). An irrelevant sdAb devoid of c-myc tag [anti-(HIV-1 NEF)] was also used as negative control. As a competition control, the sdAb used for CEA capture was produced without a c-myc tag and tested for its ability to bind to the captured CEA. The 9E10-coupled surface was regenerated with 10 μL of 5 mm HCl and the process was repeated to test the ability of each Gold mAb to bind sCEA once this molecule had been bound to a given sdAb. The absence of binding of the different Gold mAbs to the 9E10 mAb alone or to the sdAbs in absence of sCEA, as well as the absence of binding of an irrelevant antibody to the captured sCEA (mouse anti-FcγRIII) was verified.
The absence of competition between the sdAbs and Gold mAbs for binding to CEA in its normal environment was also assessed by flow cytometry. Briefly, MC38–CEA cells (5 × 105·well−1) were preincubated with various concentrations (from 2 μm to 3 nm) of sdAb C17 (devoid of c-myc tag) for 1 h on ice in NaCl/Pi + 1% BSA. Gold mAbs or sdAb C17 or sdAb C43 were then directly added in the wells at subsaturating concentrations (30–70 nm, determined in a previous experiment) and cells were incubated on ice for an additional hour. After washing, bound Gold mAbs were stained using FITC–GAM (10 μg·mL−1) and sdAb C17 (with c-myc tag) was stained using mAb 9E10 (10 μg·mL−1) followed by FITC–GAM.
In vivo localization
All in vivo experiments were performed in compliance with the French guidelines for experimental animal studies. Female HSD athymic nude-Foxn1nu 8–9 weeks old (Harlan, Gannat, France) were engrafted by subcutaneous injection of 2 × 106 LS174T human colorectal carcinoma cells in the flank. Biodistribution studies were performed 13–15 days later. Mice were injected intravenously in the tail vein with 125I-labeled sdAb C17 (10 pmol in 100 μL NaCl/Pi 0.2% BSA) and killed at 3 and 6 h post injection. Blood, organs and tumors were collected, the two latter were weighted and radioactivity in the samples was determined. Injected doses were corrected for losses by subtraction of non-injected and subcutaneously injected material (remaining in the animal tail) from the total dose. All studies were performed with groups of three mice. Results were expressed as the mean percentage of injected dose per gram of tissue ± SEM.
We would like to thank Martine Chartier, Jallane Abdelhak and Sandra Mendes for their excellent technical assistance, Sophie Sibéril and Agnès Groulet for preliminary work. We are grateful to Dr J. Barbet for fruitful discussions. We also thank Christiane and Bernard Guidicelli for generously providing a llama for immunization. This work was supported by CNRS, INSERM, the Association pour la Recherche sur le Cancer (ARC), the Cancéropole Ile-de-France and by the GDR N°2352 CNRS ‘Tumor immuno-targeting’.