Exploiting Cross-reactivity to Neutralize Two Different Scorpion Venoms with One Single Chain Antibody Fragment*

We report the optimization of a family of human single chain antibody fragments (scFv) for neutralizing two scorpion venoms. The parental scFv 3F recognizes the main toxins of Centruroides noxius Hoffmann (Cn2) and Centruroides suffusus suffusus (Css2), albeit with low affinity. This scFv was subjected to independent processes of directed evolution to improve its recognition toward Cn2 (Riaño-Umbarila, L., Juárez-González, V. R., Olamendi-Portugal, T., Ortíz-León, M., Possani, L. D., and Becerril, B. (2005) FEBS J. 272, 2591–2601) and Css2 (this work). Each evolved variant showed strong cross-reactivity against several toxins, and was capable of neutralizing Cn2 and Css2. Furthermore, each variant neutralized the whole venoms of the above species. As far as we know, this is the first report of antibodies with such characteristics. Maturation processes revealed key residue changes to attain expression, stability, and affinity improvements as compared with the parental scFv. Combination of these changes resulted in the scFv LR, which is capable of rescuing mice from severe envenomation by 3 LD50 of freshly prepared whole venom of C. noxius (7.5 μg/20 g of mouse) and C. suffusus (26.25 μg/20 g of mouse), with surviving rates between 90 and 100%. Our research is leading to the formulation of an antivenom consisting of a discrete number of human scFvs endowed with strong cross-reactivity and low immunogenicity.

Scorpion venoms are essentially neurotoxic and contain diverse constituents, including peptide toxins that cause direct impairment of Na ϩ channel activity (3,4). This effect results in excessive firing of neuronal axons. Neuronal overstimulation is reflected by both autonomic and neuromuscular symptoms, which may range from local pain, paresthesia, and seizures, to encephalopathy, pulmonary edema, and fatal cardiotoxicity (5). Peptide toxins are related by common ancestry (4). Indeed, the most relevant toxins of the aforementioned species share the same size (66 amino acids) and significant sequence identities (Fig. 1). On the other hand, toxins exhibit significant variation of target specificity. For instance, toxins Css2 and Cn2 bind to the hNav1.6 channel with high specificity (6). In contrast, Cll1 and Cll2 interact with various Na ϩ channels. 2 Current treatment against scorpionism includes the use of antivenoms, which consist of polyclonal antibodies obtained from immunized horses. Such polyclonal antibodies are effective neutralizing agents against diverse scorpion venoms (7). However, neutralization depends on non-human molecules that may become immunogenic (8). Thus, an ongoing research goal is the improvement of antivenom therapy by generating human recombinant antibodies against specific toxins. The combined use of protein engineering and directed evolution further allows the generation of antibody fragments endowed with special features (9 -13). Simplified antibodies including Fabs (500 amino acids) and scFvs 3 (250 amino acids) can be generated and evolved experimentally. The resulting antibody variants bind to target antigens with high efficiency and, due to their low sizes, achieve fast diffusion and clearance rates (14), two desirable features in any antivenom against scorpionism.
In a previous work, our group reported the generation of the antibody variant scFv 6009F by directed evolution and phage display for neutralizing the in vivo effects of toxin Cn2 (1). This antibody variant showed increased affinity for the target antigen as compared with its progenitor, the parental scFv 3F. The therapeutic potential of scFv 6009F was further confirmed by neutralizing whole venom of C. noxius (1). Recent tests showed that scFv 6009F has crossreactivity against toxins Css2, Cn3, and Css4 (evidence shown and discussed below). This was somewhat expected, given the high identities shared by these toxins (Fig. 1). It is widely accepted that an antibody may recognize structurally similar antigens, albeit with low affinity as compared with that for its primary antigen (15)(16)(17)(18)(19). An argument of negative trade offs between affinity and cross-reactivity stands out. Affinity is determined by geometric complementarity and direct interactions between the antibody and its primary antigen (20). Secondary antigens cannot reproduce some interactions, giving rise to affinity differences (21). Notwithstanding this, cross-reactivity turned out to be fundamental for the ability of scFv 6009F to neutralize toxin Css2. Furthermore, this antibody variant also neutralized whole venom of C. suffusus.
The above observations suggested that the parental scFv 3F holds structural properties useful to generate antibody variants with neutralizing activity against closely related scorpion toxins. To test this assumption, the parental scFv 3F was matured by directed evolution to attain neutralizing activity against toxin Css2. The generated antibody (scFv 9004G) neutralized toxin Css2 and also whole venom of C. suffusus. The maturation process that led to scFv 9004G also resulted in neutralizing activity against toxin Cn2 and whole venom of C. noxius. Thus, scFv 9004G retained the recognition potential of its progenitor, and exhibited a behavior analogous to that of scFv 6009F. The properties of these antibody variants are similar, but not identical. That is, scFv 6009F is better as neutralizing agent, whereas scFv 9004G shows higher expression yields.
Important mutations for toxin recognition and other properties were identified by comparing the maturation processes of scFv 9004G and scFv 6009F. A key mutation appeared in both processes and resulted in improved recognition toward toxins Css2 and Cn2. Another key mutation appeared during the maturation of scFv 6009F (1). This change was introduced into scFv 9004G to merge the best properties of both antibody variants. The resulting scFv, named LR, exhibited outstanding levels of expression and stability. Furthermore, neutralizing activity was also increased against Cn2 and Css2, as well as against whole venom of both C. noxius and C. suffusus. Thus, we are generating a family of antibody variants against different scorpion venoms. Cross-reactivity is a feature that allows each antibody variant to neutralize different toxins without adverse secondary effects. This model system entails an open ended potential for generating new therapeutic agents against scorpionism. Such potential can be harnessed by directed evolution and by combining relevant mutations from different sequence contexts.

EXPERIMENTAL PROCEDURES
Venoms of C. noxius and C. suffusus-Fresh whole venom was obtained from scorpions by electrical stimulation. Samples were dissolved in water and centrifuged at 12,000 ϫ g at 4°C for 10 min. Insoluble material was discarded, whereas the toxin-containing supernatant was recovered and spectrometrically quantified at ϭ 280 nm, assuming that 1 unit of absorbance is equivalent of 1 mg ml Ϫ1 of protein.
Toxin Purification-Toxins Css2 and Css4 were isolated from whole venom of C. suffussus as described elsewhere (6,22). Toxins Cn2 and Cn3 were obtained from whole venom of C. noxius Hoffmann as described in Ref. 23. Other venom constituents used in this work, including toxins Cll1 and Cll2, were isolated from C. limpidus limpidus as described elsewhere (24,25).
Directed Evolution of the Parental scFv 3F-The maturation process consisted of three cycles of directed evolution to obtain antibody variants against toxin Css2 (Fig. 2). Each cycle included the construction of a mutant library by random mutagenesis and four bio-panning rounds by means of phage display. Error-prone PCRs were performed under conditions to obtain different mutation rates (26,27). Coding sequences used as templates corresponded to the parental scFv 3F in cycle 1; the antibody variants 9D and 5C in cycle 2; and the antibody variant 910F in cycle 3. Amplification products were combined to obtain mutant libraries FIGURE 1. Evolutionary conservation of the main toxins of C. noxius and C. suffusus. A multiple sequence alignment was generated with Clustal W2. The amino acid sequences of toxins Cll1, Cll2, Cn2, Css2, Css4, and Cn3 are shown. Symbols indicate: asterisks, absolute conservation; colons, conservation of identity and physicochemical properties; dots, conserved physicochemical properties; blank spaces, strikingly different residues. FIGURE 2. Cross-reactivity of scFv 6009F and scFv 9004G against closely related scorpion toxins. An ELISA was used to test recognition toward Cn2, Css2, Css4, Cn3, Cll1, and Cll2. Concentrations used were: toxin, 3 g ml Ϫ1 ; scFv, 5 g ml Ϫ1 . Bars: white, scFv 6009F; black, scFv 9004G. Error bars of four experiments are shown.
as described previously (1). In cycle 1 of directed evolution, bio-panning rounds were performed as described by Marks et al. (28); the toxin concentration was 3 g ml Ϫ1 in the first and second bio-panning rounds, 1.5 g ml Ϫ1 in the third round, and 300 ng ml Ϫ1 in the fourth round. In cycles 2 and 3 of directed evolution, bio-panning was carried out under stringent conditions as described previously (1). A toxin concentration of 300 ng ml Ϫ1 was used in all biopanning rounds of cycle 2; in the fourth round, phage antibodies were preincubated with 2 M guanidinium chloride at 25°C for 30 min. In cycle 3, toxin concentration was decreased from 350 to 170 ng ml Ϫ1 ; phage antibodies were preincubated at 40 and 50°C for 30 min. Antibody variants were evaluated as phage antibodies and soluble protein by ELISA as described previously (1). The sequence of each variant was determined with the primers: forward, 5Ј-ATACCTATTGC-CTACGGC-3Ј, and reverse, 5Ј-TTTCAACAGTCTA TG-CGG-3Ј in an Applied BioSystems Sequencer model 3100 (Foster City, CA).
Antibody Expression and Purification-Antibody coding sequences were subcloned in vector pSyn1 to transform Escherichia coli TG1 cells by electroporation. Protein expression and purification were carried out as described previously (1). Protein purity and identity were confirmed by mass spectrometry. Protein concentration was determined spectrometrically at ϭ 280 nm, assuming 1 absorbance unit as equivalent to 0.7 mg ml Ϫ1 of protein.
Construction of the Antibody Variant scFv LR by Site-directed Mutagenesis-The coding sequence of scFv 9004G was modified to generate a V101F change. A megaprimer was generated by PCR with the oligonucleotides V101F 5Ј-CAAAAC-TTCCGAACCCCCCTC-3Ј and forward. The megaprimer was purified from agarose gel. The scFv 9004G sequence was re-amplified using the megaprimer and the oligonucleotide Myc 5Ј-TCAGATCCTCTTCTGAGATG-3Ј. The product was gel purified, digested with restriction endonucleases SfiI and NotI, gel purified again, and ligated to appropriately digested vector pSyn1. The coding sequence of the new antibody variant scFv LR was verified by sequencing.
Surface Plasmon Resonance Measurements-Kinetic constants of antibody binding to immobilized toxin were determined in a Biacore biosensor system (BIACORE X, Uppsala, Sweden). 90 ng of toxin dissolved in 10 mM MES (pH 6) were bound to a CM5 sensor chip using a solution of 0.05 M N-hydroxysuccinimide and 0.2 M N-ethyl-N-(dimethylaminopropyl)carbodiimide. Approximately 70 to 100 resonance units were coupled. scFv proteins were serially diluted in HBS-EP buffer (Biacore); 100-l samples were injected over immobilized toxin at a flow rate of 50 l min Ϫ1 . Delay between injections of the parental scFv 3F was 400 s. Delay for scFv 6009F, scFv 9004G, and scFv LR was 1000 s. Biosensor measurements were performed at 25°C. Protein concentrations ranged from 0.5 to 50 nM for scFv 6009F, scFv 9004G, and scFv LR. For the parental scFv 3F, protein concentrations ranged between 1 and 100 nM. Constants were calculated with the Langmuir (1:1) model of the BIAEVALUATION software version 3.1.
Functional Stability Measurements-To test recognition of toxin Cn2 in the presence of strong denaturing conditions, samples of each antibody variant were prepared at 1.5 g ml Ϫ1 with different concentrations of guanidinium chloride (0, 0.9, 1.1, 1.3, and 1.5 M). Samples were incubated overnight at 37°C and then evaluated by ELISA in a plate with immobilized toxin Cn2 at 1.5 g ml Ϫ1 . The ELISA test was followed as described previously (1). Average values (n ϭ 4) were calculated from readings at ϭ 492 nm, allowing observation of stability differences among antibody variants.
In Vivo Neutralization Tests-LD 50 values used in this work were ϳ0.25 g/20 g of mouse and ϳ0.7 g/20 g of mouse for toxins Cn2 and Css2, respectively (22,29). Unless otherwise noted, 2 LD 50 of toxin was used. Groups of 8 to 10 female CD4 mice weighting ϳ20 g were injected with toxin by the intraperitoneal passage, thus providing a control population. Intoxication symptoms were followed until death or symptom remission was observed. In the experimental population, the neutralizing activity of antibody variants was tested by a common procedure. Toxins were mixed with each antibody variant to different toxin-to-antibody molar ratios (1:10, 1:2, or 1:1). Mixtures were preincubated for 30 min at room temperature (ϳ25°C) prior to their injection into mice by the intraperitoneal passage. To test the neutralization activity against freshly prepared whole venom, two different tests were performed. In test 1, an amount of venom equivalent to 2 or 3 LD 50 was mixed with antibody to a final 1:3 toxin-to-antibody molar ratio. (Ratios were calculated relative to the main toxin of the venom.) A LD 50 of ϳ8.75 g/20 g of mouse was determined for the venom of C. suffusus suffusus, because ϳ2.8% of the total venom corresponds to toxin Css2. 4 The LD 50 of the venom of C. noxius was calculated at ϳ2.5 g/20 g of mouse, because toxin Cn2 represents ϳ6.8% of the total content (29). Mixtures were preincubated at room temperature for 30 min prior to their injection into mice. Test 2 was aimed at determining the ability of the antibody to rescue mice from developing envenomation. An amount of freshly prepared whole venom equivalent to 3 LD 50 was injected into mice. A time lapse of 5-10 min was allowed to pass before injecting antibody in a 1:3 toxin-to-antibody molar ratio relative to the main toxin of venom.

RESULTS
The parental scFv 3F was previously isolated against toxin Cn2 from a human non-immune antibody library. However, this scFv was unable to neutralize the in vivo effects of the toxin. The scFv 3F was subjected to a maturation process to increase its affinity to Cn2. The process culminated in variant scFv 6009F, which showed neutralizing activity against Cn2 and whole venom of C. noxius (1). scFv 6009F did not show recognition toward toxins Cll1 and Cll2 (1). More recently, ELISA tests revealed that scFv 6009F cross-reacts with other toxins, including Cn3, Css2, and Css4 (Fig. 2). Cross-reactivity against Css2 was relevant, because this toxin is the major con-stituent of the venom of C. suffusus. Recognition of scFv 6009F toward Css2 and Cn2 was similar (Fig. 2).

Maturation of scFv Antibody Variants
The parental scFv 3F was used again as starting point of directed evolution, but this time to achieve neutralizing activity against toxin Css2. This new maturation process was expected to reveal mutations fundamental to cross-reactivity (30). Conceivably, however, it could have also resulted in modified specificity and even loss of recognition toward the toxin for which the parental antibody was first isolated, as it has occurred in other cases (18). The maturation process consisted of three cycles of directed evolution and multiple rounds of bio-panning. Fig. 3a compares the maturation processes from the parental scFv 3F to improve its recognition toward toxins Cn2 and Css2.
Cycle 1 of Directed Evolution-A mutant library was constructed using the parental scFv 3F as template for errorprone PCR. The resulting library exhibited 1.3 ϫ 10 7 sequences and a mutagenic rate of 1.6%. Four bio-panning rounds against Css2 were performed. Two antibody variants, 9D and 5C, were isolated and further characterized: 5C exhib-ited one amino acid change (N74D), whereas 9D showed two changes (A23T and N74D) (Fig. 3b). Interestingly, the N74D change was present in both variants and emerged independently in the maturation process of scFv 6009F (1). This mutation was sufficient to increase recognition of both antibody variants to Css2, Cn2, and Cn3, as demonstrated by phageantibody ELISA tests (data not shown). The sufficiency of a single amino acid change to achieve significant binding activities has been observed before with a mutant Fab fragment. Such mutant recognizes digoxin and three of its analogs (30).
Cycle 2 of Directed Evolution-Antibody variants 5C and 9D were used as templates to construct a second mutant library (2.5 ϫ 10 7 variants; mutation rate, 0.4%). Four bio-panning rounds against Css2 were performed under stringent conditions. The antibody variant 910F showed the best recognition to Css2 in soluble-protein ELISA tests (A 492 nm ϳ2). This antibody variant has amino acid changes in the VH domain (Fig. 3b). The G59D change was drastic and mapped in proximity of CDR2, whereas the I37L change was located in framework region 2.
Cycle 3 of Directed Evolution-A third mutant library was constructed using antibody variant 910F as template (1.8 ϫ 10 7 variants; mutation rate, 0.45%). Bio-panning was carried out under stringent conditions. Twelve antibody variants showed good recognition to Css2 (A 492 nm Ͼ2). Sequence analyses revealed that an antibody variant, named scFv 9004G, was isolated in three independent occasions. The coding sequence of scFv 9004G was cloned into plasmid pSyn1 to obtain soluble protein. Cross-reactivity was evaluated in an ELISA test that included scFv 6009F as control (Fig. 2). Both antibody variants showed similar recognition to Cn2 and Css2, as well as to Cn3 and Css4. Neither scFv showed recognition to toxins Cll1 and Cll2 of C. limpidus.

Characterization of scFv 9004G
The neutralizing activity of scFv 9004G was evaluated with CD1 female mice. Table 1 indicates that 1 LD 50 of Css2 killed 60% of individuals in the control population. By mixing Css2 and scFv 9004G in a 1:10 toxin-to-antibody molar ratio, the fraction of surviving individuals in the experimental population was 100%. No symptoms of intoxication were observed throughout the experiment. The neutralizing ability of scFv 9004G was further tested with 2 LD 50 of freshly prepared whole venom of C. suffusus. A LD 50 of ϳ8.75 g/20 g of  mouse was determined by the "up and down" method (31). A high molar ratio (1:20 with respect to the main toxin of venom) was used because of the observed strong cross-reactivity against Css4 (Fig. 2). No individuals survived in the control population. Impressively, scFv 9004G allowed survival of 100% of the individuals in the experimental population (Table  1). Throughout the test, surviving mice did not manifest intoxication symptoms and behaved normally. The ability of scFv 9004G to neutralize toxin Cn2 was also tested. Experiments included 2 LD 50 of Cn2 in a 1:10 molar ratio. Table 1 indicates that all individuals survived in the experimental population. scFv 9004G prevented the onset of symptoms associated with intoxication by Cn2. The neutralization of 2 LD 50 of freshly prepared whole venom of C. noxius resulted in a survival rate of 90%. Overall, scFv 9004G neutralized toxins Css2 and Cn2, as well as the whole venom of either C. suffusus or C. noxius.

Rescue Experiments
Preliminary rescue tests were made using scFv 9004G. In these experiments, animals were envenomed by intraperitoneal injection of an amount equivalent to 2 LD 50 of freshly prepared whole venom C. suffusus. Seven to 10 min were allowed to pass before administering scFv 9004G up to a 1:20 toxin-to-antibody molar ratio with respect to the main toxin. Five of six individuals were rescued. Surviving animals showed rapid recovery and did not present dramatic symptoms. When time lapses between the injections of venom and antibody were reduced to 5 min, a survival rate of 100% was attained.

Detailed Characterization of scFv 9004G and scFv 6009F
Expression Yield and Binding Kinetics-At the level of expression, scFv 9004G yielded 2 times more protein than scFv 6009F. Kinetic constants of binding to Css2 and Cn2 were also determined for both antibody variants and the parental scFv 3F (Table 2 and supplemental Fig. S1). scFv 3F showed slightly higher affinity for Cn2 than for Css2: K a values were similar in both cases, whereas the K d for Cn2 was 1.5 times lower than for Css2. That is to say, the parental scFv 3F binds both toxins, but has higher affinity toward the antigen for which it was isolated. With Cn2, binding kinetics measurements were strikingly similar for scFv 6009F and scFv 9004G. The situation was to some extent different with Css2. scFv 6009F showed a slightly lower dissociation constant, and consequently an affinity increase relative to the values observed for scFv 9004G. Interestingly, K D values for scFv 9004G and scFv 6009F were increased by 2 orders of magnitude relative to the K D value determined for the parental scFv 3F with both toxins. This indicates that independent maturation processes can result in antibody variants with similar binding properties.
Neutralization-The neutralizing activity was challenged against 2 LD 50 of Css2 or Cn2 in different toxin-to-antibody molar ratios (Table 3). scFv 6009F and scFv 9004G provided full protection in a 1:10 molar ratio. However, scFv 6009F showed superior neutralizing activity when antibody concentration was decreased to a 1:2 molar ratio. The surviving rate was 100%, although mice manifested mild intoxication symptoms. In contrast, scFv 9004G failed to protect half of the experimental population, and survivors suffered severe symptoms. Apparently, the maturation process of scFv 9004G resulted in an antibody variant with minor neutralizing activity as compared with that of scFv 6009F. Nevertheless, scFv 9004G exhibited a comparatively superior expression yield (Table 3), an aspect that would be important for its production as therapeutic agent.

Optimization of a scFv by Site-directed Mutagenesis
scFv 6009F and scFv 9004G have similar affinity constants and cross-reactivity, we attempted to generate an improved scFv by combining their mutations. The V101F change was inserted by site-directed mutagenesis in the sequence context of scFv 9004G. This change appeared previously in the maturation process of scFv 6009F, and it has been shown to be a determinant for binding to Cn2 (1). The resulting antibody variant, scFv LR, confirmed that the properties of antibodies can be dramatically improved by simply combining key muta-

TABLE 3 Neutralization test with 2 LD 50 of toxins Cn2 and Css2
Mice were intraperitoneally injected with toxin or with a mixture of toxin and antibody. The presence and intensity of intoxication symptoms in surviving individuals are indicated as follows: a, severe symptoms; b, mild symptoms; c, ND, not determined. tions from separate maturation processes. ELISA determined that scFv LR kept the strong cross-reactivity of its progenitors (data not shown). However, this antibody variant has better neutralizing activity than scFv 9004G and even scFv 6009F (Table 3). Neutralization tests were performed using 2 LD 50 of toxin and low toxin-to-antibody molar ratios. scFv LR protected all individuals and prevented the onset of symptoms when used in a 1:2 molar ratio. In a 1:1 molar ratio, the antibody variant protected half of the individuals in the experimental population. These survivors showed mild intoxication symptoms. A similar survival rate was attained with scFv 6009F against Cn2, but severe symptoms were observed; no protection was observed against Css2. In summary, the protection range of the new variant extends to toxins Css2 and Cn2 in the lowest toxin-to-antibody molar ratios we have ever tested. In addition to its improved neutralizing activity, scFv LR showed an expression yield of 2.4 mg liter Ϫ1 , the highest value among all antibody variants including scFv 9004G ( Table 2). Interestingly, affinity constants for Css2 and Cn2 were significantly increased in scFv LR. K D values were in the picomolar range for both toxins; accordingly, k d values were the lowest.
Binding Measurements under Strong Denaturing Conditions-Stability of the antibody variants was compared by measuring recognition to Cn2 upon treatment with increasing concentrations of guanidinium chloride. Recognition levels were determined by ELISA. In Fig. 4, scFv LR exhibited the best recognition, followed by scFv 6009F, and finally by scFv 9004G. The most drastic differences of functional stability were observed at 1.3 M guanidinium chloride. scFv 9004G and scFv 6009F retained 10 and 50% of binding to Cn2, respectively. However, scFv LR retained 80% of binding. The above results clearly show the effect of a single residue change over several properties including affinity, stability, expression yield, and neutralizing activity.

Neutralization Challenge with 3 LD 50 of Whole Venom
The best antibody variants were challenged against 3 LD 50 of whole venom in two ways. First, we followed the common protocol by incubating toxin/antibody mixtures prior to their injection into mice (Table 4A). The second way was injecting venom alone, waiting 5 to 10 min to allow the onset of severe envenomation symptoms, and finally attempting rescue by antibody injection (Table 4B).
Neutralization Test 1-After preincubating venom with either scFv 6009F or scFv LR, the mixture was injected by the intraperitoneal passage. In the case of the C. suffusus venom, a survival rate of 100% was obtained, even at a molecular ratio as low as 1:3 with respect to the main toxin of this venom (Table 4A). No intoxication symptoms were observed with both antibody variants. In the case of C. noxius venom, differences between scFv 6009F and scFv LR were observed. In a 1:3 molecular ratio, scFv LR protected all individuals. scFv 6009F failed to protect half of the experimental population, but nevertheless, delayed death for more than 3 h as compared with the control population.
Neutralization Test 2-In this challenge, mice were rescued from actual envenomation by injection of 3 LD 50 of whole venom. Antibody was injected after the onset of severe symptoms. With the venom of C. suffusus, rescue was carried out with antibody in a 1:10 molar ratio. scFv 6009F did not allow a significant survival rate, because 7 of 8 individuals died. By contrast, a significant reduction of symptom intensity was observed, together with a survival rate of 100%, when scFv LR was injected in a 1:10 molar ratio. After injecting C. noxius venom into mice, survival rates of 50 and 60% were observed with scFv 6009F and scFv LR, respectively (Table 4B). These results suggested that increased amounts of antibody could be necessary to rescue more animals from this demanding situation. We tested a 1:18 ratio and found that scFv 6009F was able to rescue 80% of envenomed mice, whereas scFv LR rescued 90% of mice. This difference was slight, but it was found repeatedly in independent experiments.  Neutralization tests: (A) Test 1: neutralization challenge with 3 LD 50 of whole soluble venom of C. suffusus and C. noxius. An amount of freshly prepared whole venom equivalent to 3 LD 50 was injected into mice or as a preincubated mixture with antibody at the indicated toxin-to-antibody molecular ratio. (B) Test 2: rescue from a challenge with 3 LD 50 of whole soluble venom of C. suffusus and C. noxius. A time lapse of 5-10 min was allowed to pass before injecting antibody in a several toxin-to-antibody molecular ratios, relative to major component from venoms.

DISCUSSION
The optimization of antibodies for therapeutic purposes can be obtained by diverse methodologies. Interesting proposals have been developed to neutralize the effect of scorpion venoms. Such proposals share two common aspects. First, they were based on the rational approach of neutralizing the major toxic and abundant constituents of the whole venom of a particular species. Second, animals previously immunized with specific toxins were employed as source of antibodies. Several monoclonal antibodies have been generated by immunization of mice (23,32,33). By means of recombinant DNA technology, different simplified antibody formats (scFv, Fab, dimer, and tandem repeats) have been constructed (34 -38). Another antibody source has been the dromedary immune system. Recent papers showed the ability of antibody fragments, known as nanobodies, to neutralize the most abundant toxins of the scorpion Androctonus australis Hector (39,40). The whole venom has been neutralized by a bi-specific antibody fragment (41). Exceptional work has reported the possibility of neutralizing whole venom with a single antibody. However, in the majority of instances, whole venom neutralization has depended on at least two antibodies, or on equivalent bi-specific molecular constructions. We propose the in vitro maturation of human scFv antibodies, by directed evolution and phage display, for generating variants capable of neutralizing the main toxins and venoms of different species.
Several reports have shown that an antibody may recognize structurally related antigens, but with significant affinity variations. It has also been demonstrated that affinity and crossreactivity are readily modulated by mutations in the variable domains (16,18,30,(42)(43)(44)(45). scFv 6009F is an interesting case, because its maturation was aimed at increasing recognition toward toxin Cn2. However, scFv 6009F showed cross-reactivity against 3 scorpion toxins (Cn3, Css2, and Css4) and the ability to neutralize two of them (Cn2 and Css2). Such behavior with diverse toxins was an interesting byproduct of the maturation process, because the increase of affinity would normally be expected to trade-off with cross-reactivity. Interestingly, directed evolution resulted in an antibody variant with increased potential of recognition toward toxins of different scorpion species.
Cross-reactivity was pre-existent in the progenitor of scFv 6009F. The parental scFv 3F showed recognition against Cn2 and Css2 in ELISA tests (data not shown). Consequently, this antibody family recognizes a key epitope for neutralizing Cn2, Css2, and probably other related toxins. A second relevant aspect of this family is that its structural scaffold derives from the germ-line of domains VH3 and V3, which has been shown to be the domain combination with the best thermodynamic stability (46). This suggested that scFv 3F could be used as a plastic template to generate new neutralizing variants and for identifying residues associated with important properties, as affinity and stability.
Optimization of the antibody family started with the decision of exploring again the sequence space of the parental scFv 3F. Mutant libraries were generated by random mu-tagenesis, and multiple bio-panning rounds were performed to isolate antibody variants with increased binding to Css2. The antibody variant scFv 9004G was isolated. This variant did not loose the binding activity of its progenitor, and thus recognized Cn2 (Fig. 2). Furthermore, scFv 9004G showed the ability to neutralize Css2 (the antigen targeted by directed evolution) and Cn2 (the antigen for which the parental antibody was isolated) ( Table 1). The result of this new maturation process confirmed that the recognition potential toward Cn2 can be preserved, despite 10 mutations that separate the sequences of scFv 9004G and scFv 6009F (Fig. 3b). It is likely that important contact surfaces were not significantly altered and that conserved structural geometry account for the high level of cross-reactivity of scFv 6009F and scFv 9004G.
Noteworthy is the observation that all amino acid changes of scFv 9004G were located in the framework regions (Fig. 3b, maturation against toxin Css2). Of particular importance is the N74D mutation, which appeared recurrently during the first cycle of directed evolution. This change appeared independently in the evolutionary line of scFv 6009F (Fig. 3a, maturation against toxin Cn2). The N74D change affected the recognition for both toxins in a significant manner. Position 74 was subjected to saturation mutagenesis to reveal amino acid residues that could be important to increase recognition to Css2 (data not shown). Invariably, the best residue was aspartic acid, so it can be concluded that this residue plays an important role for a good interaction with toxins Css2 and Cn2, in both scFv 6009F and scFv 9004G.
Mutations located near the CDRs may be relevant for its conformation and interaction with amino acid residues of the antigen (47). This is the case of a mutation that was isolated in the second cycle of directed evolution. The antibody variant 910F exhibited a G59D change that increases recognition to Css2 in a significant manner. On the other hand, mutations located in frameworks may contribute to increase affinity and specificity (48), and therefore may lead to scFv variants that recognize related toxins without loosing binding to Cn2. This was observed after using scFv 9004G in recognition tests that included toxic fractions from venoms of Centruroides sculpturatus and Centruroides elegans (supplemental Fig. S2). These species have medical importance, and the tested fractions include the most abundant and potent toxins of their venoms.
Further characterization of scFv 6009F and scFv 9004G showed that recognition and affinity toward Css2 and Cn2 were similar, and that significant cross-reactivity emerged repeatedly ( Fig. 2 and Table 2). For both antibody variants, K D values were in the range of 10 Ϫ10 M. Indeed, sensorgrams of scFv 6009F and scFv 9004G can be hardly distinguished from each other when superimposed (supplemental Fig. S1). For this reason, protein identity was confirmed by mass spectrometry (supplemental Fig. S3). The antibody variants did not show different neutralizing activities when tested against 2 LD 50 of either toxin in a 1:10 toxin-to-antibody molar ratio (Table 3). However, conspicuous differences were found in more challenging tests with 1:2 molar ratios. On the one hand, scFv 6009F provided full protection, reflected by a survival rate of 100% and the prevention of intoxication symptoms. On the other hand, scFv 9004G provided partial protec- FEBRUARY 25, 2011 • VOLUME 286 • NUMBER 8 tion, reflected by a survival rate of 50% and the onset of mild intoxication symptoms.

Neutralizing Antibody Fragments against Scorpion Venoms
By analyzing the maturation process of scFv 6009F, it was inferred that the V101F change has a positive effect on affinity toward Cn2 (Cycle 2, variant 610A), without being sufficient to provide neutralizing activity (1). This mutation was introduced in the sequence context of scFv 9004G, which shows a comparatively higher expression yield. The resulting scFv LR was notorious in four aspects. First, affinity toward Css2 and Cn2 was significantly increased. This was reflected by K D values in the picomolar range, and by low dissociation constants in the range of 10 Ϫ5 s Ϫ1 for both toxins (Table 2). In our experience, a K D value does not reflect the real potency of an antibody in the absence of an analysis of both k a and k d rate constants. Good neutralizing antibodies are those with off rates in the range 10 Ϫ4 s Ϫ1 to 10 Ϫ5 s Ϫ1 . This recognition level is superior to that observed in antibodies obtained by immunization processes (49). Second, scFv LR showed the highest protein yield (2.4 mg liter Ϫ1 ). Conceivably, expression and purification protocols can be improved still further. Third, scFv LR prevented intoxication of 50% of individuals within a population injected with 2 DL 50 of either toxin in 1:1 toxinto-antibody molar ratio. Fourth, recognition of Cn2 was tested after treatment with the denaturing agent guanidinium chloride, revealing significant stability differences between antibody variants. scFv 9004G was less stable than scFv 6009F. scFv LR showed a higher stability as compared with scFv 6009F (Fig. 4). This observation further confirmed the importance of the V101F change in the sequence context of scFv 9004G.
The neutralization activities of scFv 9004G and scFv 6009F were challenged (Table 1) (1) against 2 LD 50 of freshly prepared whole venom of either C. noxius or C. suffusus. Although mild envenomation symptoms were observed, both antibody variants conferred full protection. Symptoms could be associated to the presence of toxins that were not fully neutralized, but that are nevertheless, recognized to a significant extent as a result of cross-reactivity. This effect was more evident in neutralization tests with C. noxius venom, which is more potent. For this reason, more drastic conditions were imposed to the best antibody variants, scFv 6009F and scFv LR. Neutralization was attempted against 3 LD 50 of freshly prepared whole venom. Whole venoms were collected between 1 and 2 h prior to starting the neutralization test. This requirement is explained by the heterogeneous toxicities that have been observed with stored venoms. Thus, we wanted to guarantee that the venom was fully active when injected. In neutralization test 1, scFv 6009F allowed a survival rate of 100% against C. suffusus venom. The same antibody allowed a survival rate of only 50% against C. noxius venom. The observed differences may be attributed to the potent nature of the venom of C. noxius. In stark contrast, neutralization with scFv LR was absolutely successful. Full protection was observed in tests with both venoms at 1:10 and 1:3 molar ratios.
Additional experiments were performed by injecting mice with 3 LD 50 of whole venom, allowing the onset of symptoms, and then rescuing mice by injecting antibody (neutralization test 2). This gave valuable information, because venom was injected by the intraperitoneal passage. In this way, toxins are expected to reach all tissues faster than they could possibly do when venom is subcutaneously injected by a scorpion sting. Rescue was attempted after toxins were already bound to target channels and symptoms were developing. Antibody was administered by the intraperitoneal passage, and not by the ideal route to treat envenomation, which is by endovenous injection. Put in other words, venom diffusion was faster than it could have possibly happened after a scorpion sting, whereas antibody diffusion was slower than it would be wanted to counteract envenomation.
Differences were clearly observed between neutralizing activities of scFv 6009F and scFv LR after mice were injected with whole venom of C. suffusus. Only one individual survived when rescue was attempted with scFv 6009F. In stark contrast, a survival rate of 100% was observed with scFv LR. This antibody variant caused a decrease of symptoms as soon as it was injected; full recovery was observed after 2 to 3 h. With the venom of C. noxius, scFv 6009F and scFv LR rescued between 50 and 60% of the experimental population when used in 1:10 molar ratio. Because this venom is more potent, the molar ratio was increased to 1:18. In this condition, scFv LR allowed a survival rate of 90%; full recovery was observed after 6 to 7 h. It is worth pointing out that the calculated LD 50 for the venom of C. noxius is 3.5 times higher than that for the venom of C. suffusus. Accordingly, the LD 50 for toxin Cn2 is twice as high as that for Css2. Envenomation associated with C. noxius is definitely more potent, and this situation is explained at least in part by the activity of the main toxin Cn2. Therefore, rescue tests were carried out with different toxinto-antibody molar ratios to find the amount of circulating antibody that could inhibit the ongoing interaction of toxins with target channels. In contrast, preincubating venom (or toxin) with antibody allows the capture of toxic constituents even when low molar ratios are used.
Independent maturation processes were directed to isolate antibody variants with neutralizing activity against toxins Cn2 or Css2. Each isolated antibody variant exhibited neutralizing activity against both toxins and two different venoms. This increased potential is associated with few sequence changes, each one conceivably exerting a structural effect from the frameworks or the CDRs. This discrete number of mutations allows dissection of individual effects. Furthermore, by segregating relevant mutations in different sequence contexts we can obtain new antibody variants with improved properties, as has been shown for scFv LR. Thus, the rationale of our research is to exploit the strong cross-reactivity of antibody variants for neutralizing different venoms. The case in point is not unique, because the generation of a second antibody family is on the way. This second family exhibits a different pattern of cross-reactivity. Its members recognize a different epitope of Cn2 (supplemental Fig. S4), and neutralize toxins Cll1 and Cll2 (the main toxins of C. limpidus limpidus). Our ultimate goal is the generation of a new antivenom, equivalent or superior to the commercially available. We are sure to do so by optimizing human antibody families that recognize key epitopes of toxins from the venoms of different Centruroides spp.