Ab affinity maturation in vivo is always accompanied by negative selection to maintain Ag specificity. In contrast, in vitro affinity maturation can lead to epitope spread, resulting in loss of specificity. Anti-ganglioside-GD2 mAbs are clinically effective against neuroblastoma; pain and neuropathy are major side effects. We used structural relatives of GD2 to define epitope spread during in vitro affinity maturation of an anti-GD2 single-chain variable fragment (scFv) called 5F11-scFv. Clonal dominance identified by polyclonal sequencing was confirmed by analyzing individual clones. Affinity-matured mutations were introduced into scFv-streptavidin for functional studies. Without a negative selector, 19-fold affinity improvement (clone Q, where Q is the symbol for glutamine) was associated with strong cross-reactivity with GM2 and GD1b and moderate cross-reactivity with GD3, resulting in positive immunohistochemical staining of all 13 non-neural normal human tissues, in contrast to none of 13 tissues with parental clone P. With GM2 as a negative selector, clone Y (where Y is the symbol for tyrosine) was generated with only weak cross-reactivity with GD1b, adrenal and thyroid glands, and no staining of other non-neural normal tissues. Even though there was only a 3-fold affinity improvement, clone Y showed significantly higher tumor uptake over parental clone P (134%, p = 0.04), whereas clone Q was inferior (54% of clone P; p = 0.05) as confirmed by tumor-to-normal tissue ratios across 16 organs (41% of clone P; p < 0.0001). Using the less efficient negative selector GD3, a clone mixture (Q, V, and Y, where V is the symbol for valine) emerged. We conclude that epitope spread during affinity maturation can be reduced by negative selection. Furthermore, efficiency of the negative selector depends on its cross-reactive affinity with the matured scFv.

Antibodies are accepted biologic tools and drugs for an array of human diseases. The affinity of an Ab can have a major impact on its therapeutic potential. When Ag density is low, high-affinity Ab is essential for successful targeting (1). The affinity of Abs generated in vivo is typically no better than 0.1 nM (KD) (2, 3). To overcome this limitation, in vitro affinity maturation is necessary, resulting in a 5,000-fold improvement in affinity (final KD of ∼400 fM) (4). However, for therapeutic purposes, affinity-matured Abs must maintain its high specificity. Loss of specificity will diminish targeting efficiency and cause unforeseen toxicities. During normal in vivo affinity maturation of Ab-producing B cells, binding to foreign Ags initiates their activation and differentiation. In contrast, binding to self-Ags leads to their inactivation or deletion (5, 6), a process known as negative selection (7, 8). In pathologic conditions, e.g., autoimmune diseases, lymphocytes that cross-react with self-Ags can become reactivated. This may be due to the escape of lymphocytes from immune regulation or the emergence of new mutants from somatic hypermutation (9, 10). One may expect that without negative selection, the Ab created during in vitro affinity maturation could become cross-reactive with self-Ags.

Gangliosides provide a unique opportunity to study epitope spreading because of their well-defined structural pathways. Disialoganglioside GD2 (ganglioside names are abbreviated according to Svennerholm; Ref. 11) is a glycolipid Ag highly expressed on tumors of neuroectodermal origin, including melanoma, neuroblastoma (NB),3 sarcoma, and small cell lung cancer. GD2 is an attractive target for Ab-based immunotherapy (12). The anti-GD2 Abs 3F8 (mouse IgG3) and 14.18 have shown clinical potential (13, 14, 15), and single-chain variable fragment (scFv) derived from 5F11 (mouse IgM) has been genetically fused with streptavidin (SA) and used successfully in multistep targeting to GD2-positive human tumors (16). Specific anti-GD2 Abs bind to an epitope formed by the two sialic acids and N-acetylgalactosamine. Among gangliosides there are at least four relatives with structural similarities to GD2 (Fig. 1) (17). They are GM2, GD3, GD1b, and GT2, characterized by one less sialic acid, one less N-acetylgalactosamine, one additional galactose, and one additional sialic acid, respectively. Being the nearest neighbors to GD2 in the synthesis pathway, they are potential cross-reactive Ags or epitopes during affinity maturation of anti-GD2 Abs.

FIGURE 1.

Structural pathways of gangliosides.

FIGURE 1.

Structural pathways of gangliosides.

Close modal

This report is the first demonstration that affinity maturation can yield high-affinity Abs with substantial epitope spread, resulting in poorer in vivo targeting. When the structural or epitope neighbors are known, controlling specificity through negative selection can avoid these undesirable clones. We demonstrated that the efficiency of the negative selector (Ag used in negative selection) was dependent on its binding affinity to these undesirable cross-reactive clones. We further showed that the use of “polyclonal sequencing” could facilitate single clone selection during affinity maturation.

GM3-Sepharose, GM2-Sepharose, GD3-Sepharose, and GD2-Sepharose were synthesized as previously described (18, 19).

Conjugation of GM3OC11H22NH2 (GM3-Sepharose).

N-hydroxysuccinimide (NHS)-activated Sepharose (Pharmacia) (3.4-ml dry volume) was washed with 50 ml of ice-cold HCl (1 mM) and followed with 10 ml of PBS (50 mM; pH 7.0). The wet gel and 4.04 mg of GM3OC11H22NH2 in 3.3 ml of PBS (50 mM; pH 7.0) reacted for 18 h at room temperature with shaking. The gel was filtered off and washed with 2 ml of PBS (50 mM; pH 7.0) and repeated four additional times. The combined filtrate was lyophilized and unreacted GM3OC11H22NH2 was recovered by C18 Sep-Pak for the calculation of coupling efficiency. The recovered GM3OC11H22NH2 was 1.86 mg and the coupling efficiency was ∼54% (2.18 mg coupled; 2.72 μmol in 3.4 ml of dry gel). The functionalized gel was suspended in 3.3 ml of PBS (50 mM; pH 7.0). Ethanolamine (0.2 ml) was added and the mixture was gently shaken for 3 h. The gel was filtered off and successively washed with 3 ml of cold potassium biphthalate buffer (50 mM; pH 4.0), and this step was repeated four additional times; the gel was then washed with 3 ml of PBS (50 mM; pH 8.0), and again this step was repeated four additional times. After three washing cycles the gel was stored in 20% aqueous ethanol. The calculated trisaccharide incorporation density was 0.80 μmol of GM3OC11H22NH2/ml dry gel.

Conjugation of GM2OC11H22NH2 (GM2-Sepharose).

Tetrasaccharide GM2OC11H22NH2 was conjugated to the NHS-activated Sepharose gel as described above. GM2OC11H22NH2 (5.02 mg) and 3.4 ml of gel (dry volume) were coupled. The coupling efficiency was ∼54% (2.70 μmol in 3.4 ml of dry gel). The calculated trisaccharide incorporation density was 0.80 μmol of GM2OC11H22NH2/ml dry gel.

Conjugation of GD3OC11H22NH2 (GD3-Sepharose).

Tetrasaccharide GD3OC11H22NH2 was conjugated to the NHS-activated Sepharose gel as described above. GD3OC11H22NH2 (3.50 mg) and 3.4 ml of gel (dry volume) were coupled. The coupling efficiency was ∼61% (1.98 μmol in 3.4 ml of dry gel). The calculated trisaccharide incorporation density was 0.58 μmol of GD3OC11H22NH2/ml dry gel.

Conjugation of GD2OC11H22NH2 (GD2-Sepharose).

Pentasaccharide GD2OC11H22NH2 was conjugated to the NHS-activated Sepharose gel as described above. GD2OC11H22NH2 (3.92 mg) and 3.4 ml of gel (dry volume) were coupled. The coupling efficiency was ∼70% (2.12 μmol in 3.4 ml of dry gel). The calculated trisaccharide incorporation density was 0.62 μmol of GD2OC11H22NH2/ml dry gel.

The 5F11-scFv VH fragment was randomly mutated by PCR with a GeneMorph II random mutagenesis kit (Stratagene) according to the manufacturer’s manual. It was designed to introduce a single mutation into the 5F11-scFv VH gene by controlling the quantity of template and the number of PCR cycles. Library size was estimated by counting all variants with a single mutation, i.e., the number of possible deoxynucleosides in each position multiplied by VH gene size, i.e., 4 × 396 = 1584. PCR products of mutated VH fragments were ligated into pHEN-1 containing the 5F11-scFv VL gene. Escherichia coli TG1 was used as the host.

The 5F11-scFv phage library with a randomly mutated VH fragment contained a total of 1.44 × 107 clones. By sequencing 20 random clones from this library, the mutation rate for VH was estimated at 0.35 per clone. The ratio of the parental (P) clone to the mutated clone is 2942 ((1584/0.35) − 1584). Each mutant was estimated to have 3181 ((1.44 × 107) × (0.35/1584)) copies. Because a total of 1013 phagemids was used for selection, 2.2 × 109 copies of each mutant were expected to be present. Even with stringent selection conditions, the probability of missing a promising candidate should be low.

In all affinity maturation experiments, 5F11-scFv-P-SA (P-SA, where P represents the parental clone) (16) or 3F8 (murine IgG3 specific for GD2) (20) was used as the competitor to eliminate low affinity clones. By Biacore affinity analysis, 3F8 could efficiently inhibit the binding of P-SA to GD2, but not vice versa (data not shown).

Selection without GM2-Sepharose as negative selector was done as follows. Approximately 1013 phagemids purified from a 5F11-scFv phage display library were added into 14 ml of 2% BSA. After incubation at room temperature for 1 h, 100 μl of GD2-Sepharose was added and incubated at room temperature for 2 h with shaking and then washed three times with PBS. During the first and second round selection, P-SA was added to a final concentration of 100 μg/ml and 600 μg/ml and incubated for 24 and 72 h with shaking at 4°C, respectively. During the third and fourth round selection, 3F8 was added to a final concentration of 600 μg/ml and 750 μg/ml, respectively, and incubated for 120 h with shaking at 4°C. After thorough washing with PBS for 10 times, phagemids binding to GD2 rescued from each round of selection were amplified by transfection into E. coli TG1. When selection was done with negative selector, 100 μl of GM2-Sepharose was incubated with 1013 phagemids in 14 ml of 2% BSA at room temperature for 1 h to remove cross-reactive clones. This negative selection step was repeated once for both first and second round selection. In separate experiments, GD3-Sepharose, GM2-Sepharose, or GM3-Sepharose was used as a negative selector under the same conditions.

GD2 at 20 ng/well was coated onto a 96-well microtiter plate (Thermo Electron). BSA (0.5%) in PBS was used as blocking reagent. Approximately 2 × 109 phagemids from each round of selection were mixed with P-SA started at 650 μg/ml and added to GD2-coated wells for incubation at 37°C for 2 h. An HRP-conjugated anti-M13 mAb (GE Healthcare) was used to detect GD2 binding phagemids. IB50 (the concentration of competitor required to inhibit 50% binding) was calculated using SigmaPlot 8.0 (Systat Software).

Rescued phagemids from each round of selection were transfected into E. coli TG1. Three-milliliter aliquots were grown overnight without adding helper phage. The unseparated pool of plasmids (polyclonal plasmids) was extracted for polyclonal sequencing.

After transfection of rescued phagemids, a sample was spread on a TYE plate containing 100 μl of ampicillin and 1% glucose to obtain single colonies. Single colonies were grown in 48-well plates (BD Biosciences) with Luria-Bertani medium containing 100 μg/ml ampicillin. Isopropyl-β-D-thiogalactoside was added to a final concentration of 1 mM when OD600 reached 0.8–1.0 to express monoclonal scFv. Soluble scFv supernatants were tested for high-affinity clones in ELISA using 3F8 as the competitor. The scFv supernatants were diluted 4-fold in 0.5% BSA or 0.5% BSA containing a final concentration of 34.2 μg/ml 3F8 and then added to ELISA plates coated with 20 ng/well GD2. After incubation at 4°C for 3 days, bound scFv was detected with biotin-conjugated mouse anti-c-Myc Ab and HRP/SA complex. Affinity of scFv was estimated by its inhibition by 34.2 μg/ml 3F8; the percentage of inhibition was expressed as (1 − (OD450 of scFv + 3F8 well)/(OD450 of scFv-only well)) × 100. Any clone with at least 10% less inhibition by 3F8 than that of the parental clone (∼40%) was selected for gene sequencing (Fig. 2).

FIGURE 2.

cDNA sequences of 5F11-scFv-VH parental clone, P, and its affinity-matured clones Q, V, and K (without negative selection), and clone Y (with negative selection). The underlined sequences are the CDR 1, CDR2, and CDR3 regions, as labeled). Nucleotide sequences for the amino acid mutations in CDR1 (from Glu (designated P for the parental clone) to Gln (Q), to Val (V), and to Lys (K)) and in CDR3 (from Phe (designated P for the parental clone) to Tyr (Y)) are depicted in italics.

FIGURE 2.

cDNA sequences of 5F11-scFv-VH parental clone, P, and its affinity-matured clones Q, V, and K (without negative selection), and clone Y (with negative selection). The underlined sequences are the CDR 1, CDR2, and CDR3 regions, as labeled). Nucleotide sequences for the amino acid mutations in CDR1 (from Glu (designated P for the parental clone) to Gln (Q), to Val (V), and to Lys (K)) and in CDR3 (from Phe (designated P for the parental clone) to Tyr (Y)) are depicted in italics.

Close modal

Alternatively, 0.5 M GuHCl was used as washing solution to select for scFv clones with slow off rate. ScFv supernatants were diluted 4-fold in 0.5% BSA and added to ELISA plates coated with 20 ng/well GD2. After 2 h of incubation at 37°C, wells were washed with 0.5 M GuHCl three times followed by PBS or washed with PBS alone. ScFv binding was detected by mouse anti-c-Myc and HRP/goat anti-mouse Ab complex. Inhibition of scFv binding by 0.5 M GuHCl was expressed as (1 − (OD450 of scFv + 0.5M GuHCl washed well)/(OD450 of scFv + PBS washed well)) × 100. Any clone with at least 10% less inhibition by GuHCl than that of the parental clone (∼20%) was selected for gene sequencing (Fig. 2).

Because anti-GD2 scFv proteins were relatively unstable, we expressed the newly identified clones as scFv-SA fusion proteins for functional studies. ScFv fusion with SA (e.g., 5F11-scFv-SA or P-SA) was previously used to test the functional properties of 5F11-scFv (16). A QuikChange II-E site-directed mutagenesis kit (Stratagene) was used to introduce identified mutations into a plasmid containing the P-SA gene. E. coli XL-1 blue was used as host strain. The method of fermentation of P-SA in a BioFlow 3000 fermentator (New Brunswick Scientific) was as previously described (16, 21), except that pH was controlled at 6.5 for both Y-SA and Q-SA (where Y and Q are the symbols for tyrosine and glutamine, respectively). The purification of scFv-SA was based on the method described previously (16, 21) with a minor modification. Briefly, scFv-SA was captured by 2-iminobiotin-Sepharose 4 fast flow (Affiland) and was eluted with 0.2 M sodium acetate buffer (pH 4.6) containing 0.1 M NaCl. Eluates were directly neutralized with 0.5 M Tris buffer (pH 8.0), and then the product was concentrated and changed to its storage buffer (30 mM Tris (pH 7.5) containing 1 mM EDTA, 150 mM NaCl, and 5% sorbitol) by ultrafiltration with a 100,000 m.w. cutoff membrane at 4°C. Protein concentration was calculated by A280/2.05. The A280 (1 mg/ml = 2.05 arbitrary units) was calculated by Vector NTI Suit7 (Invitrogen). A TSK-GEL G3000SWxl size exclusion column (30 cm × 7.8 mm; 5 μ) (Tosoh Bioscience) was used for HPLC analysis with 0.4M NaClO4 and 0.05 M NaH2PO4 (pH 6.0) buffer as the mobile phase at a flow rate of 0.5 ml/min and analyzed at 215 nM.

Protein yield of P-SA was 200 mg/L; for Q-SA and Y-SA, it was 30–50 mg/L. After purification, the purified proteins were concentrated by ultrafiltration to ∼6 mg/ml in 30 mM Tris (pH 7.5) containing 1 mM EDTA, 150 mM NaCl, and 5% sorbitol. HPLC analysis exhibited one major peak at a retention time of 15.271 min with purity of 82, 63, and 65% for P-SA, Y-SA, and Q-SA, respectively.

The a-series and b-series of gangliosides (GD3, GD2, GD1b, GT1b, GM3, GM2, GM1, and GD1a) were coated on ELISA plates at 20 ng/well. Gelatin (0.01%) was used as the blocking reagent. P-SA, Y-SA, and Q-SA at the highest concentration of 3 μg/ml was diluted 1/3 in 0.5% BSA. HRP/biotin was used as a detection agent. Wells without Ag were used as nonspecific binding controls, and wells with Ag but no scFv-SA were used for background subtraction. Cross-reactivity was expressed as the relative binding of other gangliosides to GD2, where cross reactivity = (OD450(gangliosides) − background)/(OD450(GD2) − background) × 100%.

The comparative affinity of 5F11-scFv-SAs was determined by BIACORE T100 (GE Healthcare). Gangliosides can be directly immobilized on sensor chip CM5 by hydrophobic interaction (22). The reference surface was immobilized with GM1. GM1 at 50 μg/ml in 20% ethanol was passed through the reference surface with a flow rate of 5 μl/min for 15 min and followed by five washes of 25 μl of 20 mM NaOH at a flow rate of 50 μl/min. To reduce mass transport as the limiting factor in Biacore measurements or multivalency during binding, the minimal amount of GD2 was used for immobilization on the Biacore sensor chip. Moreover, to ensure comparability of the reference channel (coated with GM1), the active channels were coated with equal amount of GM1 mixed with GD2 at a ratio of 1:1 (final concentration of 50 μg/ml in 20% ethanol for GD2 and GM1, respectively) to ensure an even distribution of the Ags. Immobilization was conducted for 15 min with a flow rate of 5 μl/min and followed by five washes of 25 μl of 20 mM NaOH at a flow rate of 50 μl/min. The GD2 coating concentration was chosen to allow the analyte concentration (10 × KD) to be ∼100–200 resonance units. Affinity analysis was performed at 25°C and a 30 μl/min flow rate in HBS-E buffer (0.01M HEPES (pH 7.4), 3 mM EDTA, and 300 mM NaCl). Purified scFv-SA was diluted in HBS-E buffer at increasing concentrations (12.5, 25, 50, 100, and 200 nM). An association phase was run for 2 min followed by 3 min of dissociation. At the end of each cycle, the surface was regenerated using 20 mM NaOH at a flow rate of 50 μl/min over 1 min. BIACORE T-100 evaluation software (Version 2.0) was used for analysis. The biosensor curves obtained following injection of the samples over the active surface were subtracted with the control curves obtained with the samples injected over the reference surface before kinetics analysis. Based on the association rate constant (kon, ka1) and the dissociation rate constant (koff, kd1), the equilibrium constant (KD = kd1/ka1) could be calculated.

Stage 4 NB tumors and normal tissues were obtained at Memorial Sloan-Kettering Cancer Center with institutional review board approval. Five- to seven-micrometer sections of snap-frozen tissues were fixed in acetone for 30 min at −20°C. Endogenous biotin-binding activity was blocked by sequential treatment with avidin and biotin (Vector avidin-biotin blocking kit; Invitrogen) for 20 min each. Sections were incubated with 2 μg/ml scFv-SA at room temperature for 1 h. Following washing, sections were incubated with HRP/biotin for 30 min at room temperature and subsequent incubation with 3,3′-diaminobenzidine for 5 min. H&E staining was also performed.

In vivo biodistribution was performed using a multistep targeting technique as previously described (16). Briefly, athymic nude mice with established human LAN-1 xenografts were fed biotin-free diet for 1 wk. Mice with measurable tumors (0.5- to 0.8-cm diameter at the time of Ab injection) were injected i.v. with 900 μg of 5F11-SA. Twenty-four hours later, 450 μg of synthetic clearing agent (biotin-LC-NM-(Gal-NAc)16, where LC- NM is amidocaproyl-N-methyl (16), was injected i.v. and followed 4 h later by 100 μCi of 111In-DOTA-biotin i.v. (where DOTA is 1,4,7,10-tetraazacyclododecane-1,4,7,10-tetraacetic acid). Mice were sacrificed 24 h after the injection of 111In-DOTA-biotin for biodistribution studies. The percentage of an injected dose (μCi) of radiolabeled ligand per gram of weight (%ID/g) of tissue was calculated for each mouse. The %ID/g of a tumor measures the uptake intensity in the tumor and the ratio of %ID/g of a tumor vs the %ID/g of each normal tissue in each mouse represents targeting specificity in vivo.

ScFvs were selected on GD2-Sepharose in the presence of anti-GD2 competing Abs (competitors) either with negative selection (competitive selection in the presence of a negative selector (CSN)) or without negative selection (competitive selection (CS)). The first negative selector used was GM2-Sepharose.

5F11-SA (P-SA) or 3F8 (both specific for GD2) were used as competitors. CS in the presence of free Ag was not feasible because the glycolipid GD2 was not water soluble. We chose P-SA as the competing ligand for selection rounds no.1 and no.2 and 3F8 for the subsequent rounds. The number of GD2-binding phagemids recovered increased from the first through the fourth round (1.6 × 108, 4.4 × 108, 1.46 × 109, and 9.23 × 109 for CS-1, CS-2, CS-3, and CS-4, respectively). With GM2-Sepharose as a negative selector, the number of GD2-binding phagemids recovered increased only slightly from the first through the fourth round (2.1 × 108, 1.3 × 108, 3.1 × 108, and 7.1 × 108 for CSN-1, CSN-2, CSN-3, and CSN-4, respectively).

The IB50 of pooled phagemids to GD2 was used to estimate the average phagemid affinity. As expected, after each round of selection the enhancement of IB50 was consistent with an affinity gain. For CS, IB50 for pooled phagemids for each subsequent round was 18, 589, >650, and >650 μg/ml 3F8, respectively. As for CSN, IB50 of four rounds was 9, 20, 158, and 218 μg/ml 3F8, respectively. Using an IB50 of 6 μg per milliliter of the parental clone as the benchmark, CS-4 had a >108-fold affinity increase, whereas CSN-4 had a 36-fold improvement.

With each round of CS, certain clones became increasingly dominant in the population. Polyclonal sequencing of these phagemid pools identified specific mutants (Fig. 3, A and B). These clones were later shown (see below) to have high affinity for GD2. By the fourth round, mutant clones (Table I) became fully dominant. Without a negative selector, CS yielded clone Q with a Glu to Gln (Q) mutation in the sixth amino acid of CDR1 (see Fig. 2, CS-4 graph in Fig. 3,A, and Table I). In contrast, clone K, with a glutamic acid to lysine (K) mutation in CDR1, did not substantially expand from CS-3 to CS-4.

FIGURE 3.

Polyclonal sequencing to track the dominance of scFv populations. The color code for nucleotides is as follows: black, G; green, A; blue, C; and red, T. The size of each nucleotide peak reflects the abundance of the mutant scFv. A, Affinity maturation without negative selector; scFv with Glu (parental clone)→Gln (Q) (GAA→CAA in CDR1 (bold italics indicate nucleotide substitution) mutation became the dominant population starting from the third round (CS-3) through the fourth round of selection (CS-4). B, With GM2 as negative selector, scFv with Phe (parental clone)→Tyr (Y) (TTT→TAT in CDR3) became the dominant population from the third round (CSN-3) through the fourth round of selection (CSN-4). C–F, Affinity maturation was conducted with 20-fold fewer phagemids. C, With GM3 as negative selector, scFv with Glu (parental clone)→Gln (Q) (GAA→CAA) mutation became increasingly dominant from the third round (CSGM3-3) to the fourth round (CSGM3-4) to the sixth round of selection (CSGM3-6). D, With GM2 as negative selector, scFv with Phe (parental clone)→Tyr (Y) (TTT→TAT in CDR3) mutation became increasingly dominant from the fourth (GSGM2-4) to the fifth (CSGM2-5) to the sixth round of selection (CSGM2-6). E and F, With GD3 as the negative selector, scFv with mutations in CDR1 (CAA (Q) and GTA (V) in E) and CDR3 (TAT (Y) in F) were all enriched. However, clone Q with Glu→Gln mutation (GAA→CAA in CDR1) rapidly became dominant from the fourth (GSGD3-4) to the fifth (CSGD3-5) to the sixth round of selection (CSGD3-6).

FIGURE 3.

Polyclonal sequencing to track the dominance of scFv populations. The color code for nucleotides is as follows: black, G; green, A; blue, C; and red, T. The size of each nucleotide peak reflects the abundance of the mutant scFv. A, Affinity maturation without negative selector; scFv with Glu (parental clone)→Gln (Q) (GAA→CAA in CDR1 (bold italics indicate nucleotide substitution) mutation became the dominant population starting from the third round (CS-3) through the fourth round of selection (CS-4). B, With GM2 as negative selector, scFv with Phe (parental clone)→Tyr (Y) (TTT→TAT in CDR3) became the dominant population from the third round (CSN-3) through the fourth round of selection (CSN-4). C–F, Affinity maturation was conducted with 20-fold fewer phagemids. C, With GM3 as negative selector, scFv with Glu (parental clone)→Gln (Q) (GAA→CAA) mutation became increasingly dominant from the third round (CSGM3-3) to the fourth round (CSGM3-4) to the sixth round of selection (CSGM3-6). D, With GM2 as negative selector, scFv with Phe (parental clone)→Tyr (Y) (TTT→TAT in CDR3) mutation became increasingly dominant from the fourth (GSGM2-4) to the fifth (CSGM2-5) to the sixth round of selection (CSGM2-6). E and F, With GD3 as the negative selector, scFv with mutations in CDR1 (CAA (Q) and GTA (V) in E) and CDR3 (TAT (Y) in F) were all enriched. However, clone Q with Glu→Gln mutation (GAA→CAA in CDR1) rapidly became dominant from the fourth (GSGD3-4) to the fifth (CSGD3-5) to the sixth round of selection (CSGD3-6).

Close modal
Table I.

Affinity maturation of anti-GD2 5F11-scFv

Clone NameSelection ProcessCompeting Anti-GD2 Ligand PresentGM2 as Negative SelectorLocation of Mutation in VHAmino Acid change in VH
− − − − − 
CS − CDR1 Glu→Gln (Q) 
CS − CDR1 Glu→Val (V) 
CS − CDR1 Glu→Lys (K) 
CSN CDR3 Phe→Tyr (Y) 
Clone NameSelection ProcessCompeting Anti-GD2 Ligand PresentGM2 as Negative SelectorLocation of Mutation in VHAmino Acid change in VH
− − − − − 
CS − CDR1 Glu→Gln (Q) 
CS − CDR1 Glu→Val (V) 
CS − CDR1 Glu→Lys (K) 
CSN CDR3 Phe→Tyr (Y) 

To be certain that these polyclonal sequences were truly representative of the clonal population, individual clones were isolated after selection with either 3F8 as the competitive inhibitor or 0.5 M GuHCl as the washing buffer. We first chose clones with better affinity than that of the parental clone in the presence of a competing ligand (i.e., less inhibition by 3F8). At 38.4 μg/ml, 3F8 inhibited 30–40% binding of 5F11-scFv (P-scFv). Of 1500 clones, only 60 had <30% inhibition by 3F8. Upon sequencing these 60 clones, we consistently found mutations of the sixth amino acid, i.e., Glu, in the CDR1 H chain to Gln (Q), Lys (K), Val (V), and Gly (G) with frequencies of 81.6, 13.3, 3.3, and 1.7%, respectively (Fig. 2). In a subsequent experiment, 0.5 M GuHCl was used to eliminate low affinity clones, which had reduced binding on ELISA. Washing thrice with 0.5 M GuHCl lowered ELISA binding of P-scFv by 20%. Fifty of the 2400 clones with a <10% drop following 0.5M GuHCl wash were sequenced, and they all carried the GAA (Glu)→CAA (Gln) Q mutation. Thus, both polyclonal sequencing and monoclonal selection pointed to a dominant candidate (i.e., clone Q) when selection was done without negative selection.

In contrast, with GM2-Sepharose as a negative selector, the Y clone with a TTT (Phe)→TAT (Tyr) mutation in CDR3 assumed dominance after CSN-4 (Fig. 3,B and Table I). In this case, no Q clone was detected. Moreover, the Y clone was not found in the sequencing data from CS-4, suggesting that they were competed out by the high-affinity Q clone.

5F11-scFv was genetically fused to SA (e.g., 5F11-scFv-SA (P-SA)) to make stable proteins to test their functional properties (16). These purified scFv-SAs were analyzed by Biacore T100 for their relative affinities to GD2 at concentrations of 12.5–200 nM. The affinities of Y-SA and Q-SA were superior by 3- and 19-fold, respectively, over P-SA (Table II). When tested against the structural neighbors of GD2, GM2 and GD1b, Q-SA showed higher affinity to GM2 than GD1b. The relatively low resonance unit values per Biacore analysis precluded reliable determination of the cross-reactive affinity of Q-SA on GD3.

Table II.

Affinity and cross-reactivity of scFv-SA on gangliosides

ClonesP-SAY-SAQ-SA
Affinity by SPR (GD2)a Kon (M−1 s−17.811 × 104 1.437 × 105 5.675 × 105 
 Koff (s−11.550 × 10−3 8.654 × 10−4 5.923 × 10−4 
 KD (nM) 19.8 6.02 1.04 
Affinity by SPR (GM2) Kon (M−1 s−1ND ND 2.013 × 105 
 Koff (s−1ND ND 1.638 × 10−3 
 KD (nM) ND ND 8.14 
Affinity by SPR (GD1b) Kon (M−1 s−1ND ND 1.595 × 105 
 Koff (s−1ND ND 1.807 × 10−3 
 KD (nM) ND ND 11.3 
Cross-reactivity by ELISAb GD3/GD2 0.00% 0.31% 13.78% 
 GD1b/GD2 0.36% 5.05% 25.74% 
 GT1b/GD2 0.00% 0.14% 2.47% 
 GM3/GD2 0.00% 0.14% 2.42% 
 GM2/GD2 0.55% 0.71% 43.44% 
 GM1/GD2 0.08% 0.00% 2.60% 
 GD1a/GD2 0.00% 0.06% 4.45% 
 Gelatin/GD2 0.32% 0.19% 1.09% 
ClonesP-SAY-SAQ-SA
Affinity by SPR (GD2)a Kon (M−1 s−17.811 × 104 1.437 × 105 5.675 × 105 
 Koff (s−11.550 × 10−3 8.654 × 10−4 5.923 × 10−4 
 KD (nM) 19.8 6.02 1.04 
Affinity by SPR (GM2) Kon (M−1 s−1ND ND 2.013 × 105 
 Koff (s−1ND ND 1.638 × 10−3 
 KD (nM) ND ND 8.14 
Affinity by SPR (GD1b) Kon (M−1 s−1ND ND 1.595 × 105 
 Koff (s−1ND ND 1.807 × 10−3 
 KD (nM) ND ND 11.3 
Cross-reactivity by ELISAb GD3/GD2 0.00% 0.31% 13.78% 
 GD1b/GD2 0.36% 5.05% 25.74% 
 GT1b/GD2 0.00% 0.14% 2.47% 
 GM3/GD2 0.00% 0.14% 2.42% 
 GM2/GD2 0.55% 0.71% 43.44% 
 GM1/GD2 0.08% 0.00% 2.60% 
 GD1a/GD2 0.00% 0.06% 4.45% 
 Gelatin/GD2 0.32% 0.19% 1.09% 
a

Relative affinity was compared by surface plasmon resonance (Biacore T-100) using ganglioside (GD2, GM2, or GD1b) coated on CM5 chips.

b

Values of ≥1% were defined as cross-reactive; ≥15% was defined as strongly cross-reactive.

In vitro cross-reactivity was tested using both the a-series and b-series of gangliosides (Fig. 1), and results were summarized in Table II. Cross-reactivity with other gangliosides was defined as significant if relative binding was >1% of binding to GD2. In contrast to P-SA, which had no cross-reactivity, Q-SA had strong cross-reactivity with GM2 and GD1b and weak cross-reactivity with GD3 and GD1a, whereas Y-SA had only weak cross-reactivity with GD1b. Cross-reactivity of Q-SA determined by ELISA was in agreement with the affinity results obtained by Biacore (Table II).

P-SA, Y-SA, and Q-SA were tested for tissue specificity by IHC on human NB tumors and normal human tissues. NB tumors are known to express high levels of GD2. All 11 tumors were stained positive by P-SA, Y-SA, and Q-SA (Table III). Seventeen normal tissues were also tested (Table III). Frontal lobe, pons, cerebellum, and spinal cord all stained positive with both affinity-matured and parental clones as expected, because GD2 is known to be present on neuronal tissues. The remaining 13 normal tissues were all positive with Q-SA, in sharp contrast to complete negativity with P-SA. As for Y-SA, it showed only weak heterogeneous staining in thyroid and adrenal tissues, whereas the remaining 11 normal tissues were negative.

Table III.

Tissue cross-reactivity by immunohistochemistrya

OrganP-SAY-SAQ-SAb
Heart − − 
Ileum − − ++ 
kidney − − 
Liver − − 
Lung − − 
Pancreas − − ++ 
Spleen − − ++ 
Stomach − − ++ 
Testes − − ++ 
Thyroid − +/− ++ 
Adrenal − +/− ++ 
Sigmoid Colon − − +/− 
Skeletal Muscle − − 
Pons ++ +++ 
Cerebellum +/− ++ +++ 
Spinal Cord ++ 
Frontal Lobe ++ +++ 
Stage 4 NB tumors (n = 11) ++ ++ ++/+++ 
OrganP-SAY-SAQ-SAb
Heart − − 
Ileum − − ++ 
kidney − − 
Liver − − 
Lung − − 
Pancreas − − ++ 
Spleen − − ++ 
Stomach − − ++ 
Testes − − ++ 
Thyroid − +/− ++ 
Adrenal − +/− ++ 
Sigmoid Colon − − +/− 
Skeletal Muscle − − 
Pons ++ +++ 
Cerebellum +/− ++ +++ 
Spinal Cord ++ 
Frontal Lobe ++ +++ 
Stage 4 NB tumors (n = 11) ++ ++ ++/+++ 
a

The minus sign (−) indicates negative. Positive was expressed as +, ++, and +++ according to intensity and homogeneity. Plus/minus (+/−) heterogeneously positive, i.e. some area was negative and some area was 1+ positive. The percentage positive and intensity of staining consistently ranked the clones as follows: Q-SA > Y-SA ≈ P-SA.

b

With clone Q, there was consistent diffuse staining of nuclei and cytoplasm in all tissues tested.

Athymic nude mice xenografted with human NB LAN-1 tumors were injected i.v. with either P-SA (n = 10), Y-SA (n = 5), or Q-SA (n = 5). %ID/g of each scFv-SA for each organ, as well as tumor to normal organ ratios, was compared in Fig. 4. Q-SA had the lowest tumor uptake (2.93 ± 0.50%ID/g) despite its 19-fold increase in affinity over the parental scFv-SA (P-SA, 3.99 ± 0.34% ID/g, p = 0.05). It also had inferior tumor-to-normal organ ratios across the 16 organs tested (41 ± 18% of clone P; p < 0.0001). In contrast, despite a mere 3-fold increase in affinity, Y-SA had significantly higher tumor uptake (5.29 ± 0.54%ID/g; p = 0.04).

FIGURE 4.

Efficiency of tumor targeting; %ID/g and tumor to organ ratio using clones P, Y and Q genetically linked to SA in multistep targeting. Athymic mice (n = 5–10 per group) xenografted with NB LAN-1 were injected with a standard dose of 900 μg of scFv-SA, 450 μg of synthetic clearing agent, and 100μCi 111In DOTA-biotin as described previously (16 ). Mice were sacrificed and their organ radioactivity was measured to calculate the %ID/g in tumor and various organs. Tumor-to-organ ratio was calculated as follows: (%ID/g in tumor)/(%ID/g in organ).

FIGURE 4.

Efficiency of tumor targeting; %ID/g and tumor to organ ratio using clones P, Y and Q genetically linked to SA in multistep targeting. Athymic mice (n = 5–10 per group) xenografted with NB LAN-1 were injected with a standard dose of 900 μg of scFv-SA, 450 μg of synthetic clearing agent, and 100μCi 111In DOTA-biotin as described previously (16 ). Mice were sacrificed and their organ radioactivity was measured to calculate the %ID/g in tumor and various organs. Tumor-to-organ ratio was calculated as follows: (%ID/g in tumor)/(%ID/g in organ).

Close modal

We next tested the relationship between the efficiency of a negative selector and its cross-reactive affinity with affinity-matured clones. GM2 had the highest cross-reactive affinity toward Q-SA (KD = 8.14 nM by Biacore; 43.44% cross-reactivity by ELISA). GD3 had moderate cross-reactivity toward Q-SA (13.78% cross-reactivity by ELISA), whereas GM3 had minimal to no detectable cross-reactivity. We tested the individual efficiency of GM2-Sepharose, GD3-Sepharose, and GM3-Sepharose as negative selectors in affinity maturation experiments labeled as CSGM2, CSGD3, and CSGM3, respectively (Fig. 3). In these experiments, we used 20-fold fewer starting phagemids. Polyclonal sequencing was used to follow the clonal evolution during each of the consecutive selections. With GM2-Sepharose as a negative selector (CSGM2), clone Y again became enriched by the sixth round of selection (CSGM2-6), and Q or K clones were never detected (Fig. 3,D). With GM3-Sepharose as a negative selector (CSGM3), clone Q became dominant after the sixth round (CSGM3-6), whereas clone K was soon competed out (Fig. 3,C) as expected, because it cross-reacted strongly with GM3 (data not shown). With GD3-Sepharose as negative selector (CSGD3), clone V was first detected after the fourth round (CSGD3-4) (Fig. 3,E), whereas clones Q and Y appeared after CSGD3-5 (Fig. 3, E and F). By CSGD3-6, ∼40% population was represented by clone Q, ∼30% by clone Y, ∼20% by clone P, and ∼10% by clone V (Fig. 3, E and F). The reason why clone V appeared early on was because of its lower cross-reactivity with GD3 compared with that of clone Q; it was finally overtaken by clone Q, which had higher affinity to GD2. We conclude that the lower affinity negative selectors (e.g., GD3) could not efficiently remove all of the cross-reactive clones. The cross-reactive affinity of the negative selector appeared critical in determining the efficiency of the negative selection step. In addition, when affinity maturation was conducted using a 20-fold lower starting number of phagemids, clonal dominance was delayed by one to two rounds.

For healthy individuals, affinity improvement in vivo is part of the maturation of immune response and is always accompanied by negative clonal selection. High-affinity clones mature whereas autoreactive clones are deleted. The Abs eventually achieve both high affinity and Ag specificity. In contrast, most if not all in vitro affinity maturation strategies imitate adaptive immunity without constraints. Very few reports incorporate steps to avoid cross-reactive clones. This is generally not due to the lack of foresight, but rather the difficulty in defining cross-reactive epitopes. Our study using gangliosides provides a working model to demonstrate this breakdown in specificity during affinity maturation with subsequent epitope spreading. Knowing the epitope neighborhood, strategies could therefore be tested to reduce this undesirable epitope spread, thereby improving the quality of affinity-matured anti-ganglioside Abs.

When an Ab clone evolves in affinity and spreads its epitope specificity, the cross-reacting peptides or oligosaccharides are usually structurally or conformationally related to the original Ag. Among gangliosides, a hierarchy of cross-reactivity can be derived from their structural similarities. The four closest structural neighbors of GD2, namely GM2, GD3, GD1b, and GT2, differ from GD2 by a single residue, representing four different structural directions. Using GM2, GD3, and GM3, respectively, as negative selectors we found that their cross-reactive affinity highly correlated with their ability to reduce clones with epitope spread. Although one could increase the concentration of a negative selector to compensate for low affinity, the cost is prohibitive at high concentrations.

GM2 was the ideal negative selector because its high cross-reactivity was consistently detected among the high-affinity clones to GD2. We chose a novel ganglioside-oligosaccharide-Sepharose scaffold as a proof of principle for in vitro selection. This platform should also have general applicability to most other glycolipids. We introduced random single base pair mutations to the entire scFv-VH region, including both CDR and the framework regions (23). The efficiency of affinity maturation can be greatly improved by increasing the library size and by focusing mutations onto hot spots (24). This focusing on CDR regions is particularly important if simultaneous mutations in the scFv are desired. Although “polyclonal sequencing” is highly efficient for tracking clonal dominance for single mutations, the presence of simultaneous mutations in a bigger library may make the analysis more complex.

Epitope spread has clinical implications for ganglioside-specific Abs. Anti-GD2 Abs are known to induce significantly painful side effects (25, 26, 27), with occasional reports of cranial or peripheral neuropathy. Anti-GD1a Abs preferentially immunostained motor fibers thought to cause motor neuropathies such as acute motor axonal neuropathy (28). GD1b Abs preferentially immunostained the large dorsal root ganglion neurons, potentially responsible for sensory ataxic neuropathies (28). High-affinity anti-GD2 Abs with promiscuous specificity would potentially cause significant motor and sensory neuropathies if they were to be given to patients. In contrast, with GM2 negative selection during affinity maturation the high-affinity clones isolated were much more restricted in their cross-reactivity and would likely to be safe for clinical use.

Our in vitro findings also raise the possibility that autoimmune Abs can develop when there is no negative selection during somatic hypermutation. B cell tolerance, whether central or peripheral, is a complex process for preventing the development of autoimmunity during an immune response to an epitope on a foreign pathogen. In germinal center follicles of peripheral lymphoid tissues, self-reactive BCRs are continually generated through somatic hypermutation (29, 30). With affinity maturation, autoimmunity, if not prevented, can be severe, especially because follicular B-cells differentiate into long-lived memory cells and plasma cells that in turn will produce autoantibodies indefinitely. It is generally accepted that whereas B-cells with BCRs that recognize both self-Ags and foreign Ags would survive, those that recognize only the foreign Ag will prevail because their BCR engagement is less chronic and specific T cell help is present. Self-reactive B cells that lack appropriate T cell help are rapidly censored or depleted.

In light of this B cell tolerance model, our findings can be interpreted as follows. If GD2 is the foreign epitope and GM2 is the self-epitope, affinity maturation during hypermutation will produce clones like Q and Y. The self-epitope should induce rapid B cell inactivation, i.e., negative selection. Thus, as immunity matures in the presence of self-As, high-affinity clones (e.g., clone Y) emerge while retaining Ag specificity. However, if the censoring/deletion pathway malfunctions (i.e., absence of negative selection), high-affinity clones (e.g., clone Q) will become dominant. The mere absence or depression of the self-epitope (e.g., GM2) during a specific time period or tissue space (e.g., during an infection) can be the reason for this malfunction. The ensuing cross-reactive Ab clones (e.g., clone Q) will be free to propagate and become pathologic when the self-epitope GM2 re-emerges.

We conclude that this report can provide insight into the affinity maturation of carbohydrate (and ganglioside) specific Abs in vitro and in vivo. More importantly, our findings may have broader implications for other epitopes of amino acid, nucleotide, or lipid origin, many of which are implicated in human diseases and some as therapeutic targets. We believe that epitope spread during affinity maturation occurs both in vitro and in vivo. Thus, negative selectors may potentially be necessary during high throughput screens of peptide, aptamer, and nucleotide libraries for target-specific ligands. However, a major challenge remains, i.e., how to define the epitope neighborhood. Moreover, even with sophisticated modeling, the ultimate proof of specificity of novel Abs and ligands will be in vivo testing in appropriate animal models before being used in patients.

We thank Dr. Irene Cheung (Memorial Sloan-Kettering Cancer Center, New York, NY) for critically reviewing the manuscript and Dr. P. Zhang for expertise in ganglioside conjugation to Sepharose.

The authors have no financial conflict of interest.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

1

This work was supported in part by National Institutes of Health Grant CA106450, Robert Steel Foundation, Hope Street Kids, William H. Goodwin, Alice Goodwin, and the Commonwealth Foundation for Cancer Research, and the Experimental Research Center of Memorial Sloan-Kettering Cancer Center.

3

Abbreviations used in this paper: NB, neuroblastoma; CS, competitive selection; CSN, competitive selection in the presence of a negative selector; DOTA, 1,4,7,10-tetraazacyclododecane-1,4,7,10-tetraacetic acid; GuHCl, guanidine chloride; IB50, concentration of competitor required to inhibit 50% binding; %ID/g, percentage of injected dose of radiolabeled ligand per gram; IHC, immunohistochemistry; P, parental (clone); SA, streptavidin; scFv, single chain variable fragment; NHS, N-hydroxysuccinimide.

1
Zuckier, L. S., E. Z. Berkowitz, R. J. Sattenberg, Q. H. Zhao, H. F. Deng, M. D. Scharff.
2000
. Influence of affinity and antigen density on antibody localization in a modifiable tumor targeting model.
Cancer Res.
60
:
7008
-7013.
2
Batista, F. D., M. S. Neuberger.
1998
. Affinity dependence of the B cell response to antigen: a threshold, a ceiling, and the importance of off-rate.
Immunity
8
:
751
-759.
3
Rathanaswami, P., S. Roalstad, L. Roskos, Q. J. Su, S. Lackie, J. Babcook.
2005
. Demonstration of an in vivo generated sub-picomolar affinity fully human monoclonal antibody to interleukin-8.
Biochem. Biophys. Res. Commun.
334
:
1004
-1013.
4
Steidl, S., O. Ratsch, B. Brocks, M. Durr, E. Thomassen-Wolf.
2008
. In vitro affinity maturation of human GM-CSF antibodies by targeted CDR-diversification.
Mol. Immunol.
46
:
135
-144.
5
Carsetti, R., G. Kohler, M. C. Lamers.
1995
. Transitional B cells are the target of negative selection in the B cell compartment.
J. Exp. Med.
181
:
2129
-2140.
6
Strasser, A., H. Puthalakath, L. A. O'Reilly, P. Bouillet.
2008
. What do we know about the mechanisms of elimination of autoreactive T and B cells and what challenges remain.
Immunol. Cell Biol.
86
:
57
-66.
7
Nossal, G. J..
1994
. Negative selection of lymphocytes.
Cell
76
:
229
-239.
8
Palmer, E..
2003
. Negative selection–clearing out the bad apples from the T-cell repertoire.
Nat. Rev.
3
:
383
-391.
9
Vanderlugt, C. L., S. D. Miller.
2002
. Epitope spreading in immune-mediated diseases: implications for immunotherapy.
Nat. Rev.
2
:
85
-95.
10
Deshmukh, U. S., H. Bagavant, J. Lewis, F. Gaskin, S. M. Fu.
2005
. Epitope spreading within lupus-associated ribonucleoprotein antigens.
Clin. Immunol.
117
:
112
-120.
11
Svennerholm, L..
1964
. The gangliosides.
J. Lipid Res.
5
:
145
-155.
12
Modak, S., N. K. Cheung.
2007
. Disialoganglioside directed immunotherapy of neuroblastoma.
Cancer Invest.
25
:
67
-77.
13
Kushner, B., K. Kramer, S. Modak, N. K. Cheung.
2007
. Anti-GD2 anitbody 3F8 plus granulocyte-macrophage colony stimulating factor (GM-CSF) for primary refractory neuroblastoma (NB) in the bone marrow (BM).
Proc. Amer. Soc. Clin. Oncol.
25
:
526s
(Abstr.).
14
Yu, A. L., A. L. Gilman, M. F. Ozkaynak, W. B. London, S. Kreissman, H. X. Chen, K. K. Matthay, S. L. Cohn, J. M. Maris, P. Sondel.
2009
. A phase III randomized trial of the chimeric anti-GD2 antibody ch14.18 with GM-CSF and IL2 as immunotherapy following dose intensive chemotherapy for high-risk neuroblastoma: Children’s Oncology Group (COG) study ANBL0032.
J Clin Oncol.
27
: (Suppl. 15s):
10067z
(Abstr.).
15
Modak, S, and N. K. Cheung. Neuroblastoma: therapeutic strategies for a clinical enigma. Cancer Treat. Rev. In press.
16
Cheung, N. K., S. Modak, Y. Lin, H. Guo, P. Zanzonico, J. Chung, Y. Zuo, J. Sanderson, S. Wilbert, L. J. Theodore, D. B. Axworthy, S. M. Larson.
2004
. Single-chain Fv-streptavidin substantially improved therapeutic index in multistep targeting directed at disialoganglioside GD2.
J. Nucl. Med.
45
:
867
-877.
17
Yu, R. K., E. Bieberich, T. Xia, G. Zeng.
2004
. Regulation of ganglioside biosynthesis in the nervous system.
J. Lipid Res.
45
:
783
-793.
18
Willison, H. J., K. Townson, J. Veitch, J. Boffey, N. Isaacs, S. M. Andersen, P. Zhang, C. C. Ling, D. R. Bundle.
2004
. Synthetic disialylgalactose immunoadsorbents deplete anti-GQ1b antibodies from autoimmune neuropathy sera.
Brain
127
:
680
-691.
19
Andersen, S. M., C. C. Ling, P. Zhang, K. Townson, H. J. Willison, D. R. Bundle.
2004
. Synthesis of ganglioside epitopes for oligosaccharide specific immunoadsorption therapy of Guillian-Barre syndrome.
Org. Biomol. Chem.
2
:
1199
-1212.
20
Saito, M., R. K. Yu, N. K. Cheung.
1985
. Ganglioside GD2 specificity of monoclonal antibodies to human neuroblastoma cell.
Biochem. Biophys. Res. Commun.
127
:
1
-7.
21
Schultz, J., Y. Lin, J. Sanderson, Y. Zuo, D. Stone, R. Mallett, S. Wilbert, D. Axworthy.
2000
. A tetravalent single-chain antibody-streptavidin fusion protein for pretargeted lymphoma therapy.
Cancer Res.
60
:
6663
-6669.
22
Catimel, B., A. M. Scott, F. T. Lee, N. Hanai, G. Ritter, S. Welt, L. J. Old, A. W. Burgess, E. C. Nice.
1998
. Direct immobilization of gangliosides onto gold-carboxymethyldextran sensor surfaces by hydrophobic interaction: applications to antibody characterization.
Glycobiology
8
:
927
-938.
23
Li, Z., C. J. Woo, M. D. Iglesias-Ussel, D. Ronai, M. D. Scharff.
2004
. The generation of antibody diversity through somatic hypermutation and class switch recombination.
Genes Dev.
18
:
1
-11.
24
Rogozin, I. B., M. Diaz.
2004
. Cutting edge: DGYW/WRCH is a better predictor of mutability at G:C bases in Ig hypermutation than the widely accepted RGYW/WRCY motif and probably reflects a two-step activation-induced cytidine deaminase-triggered process.
J. Immunol.
172
:
3382
-3384.
25
Handgretinger, R., P. Baader, R. Dopfer, T. Klingebiel, P. Reuland, J. Treuner, R. A. Reisfeld, D. Niethammer.
1992
. A phase I study of neuroblastoma with the anti-ganglioside GD2 antibody 14.G2a.
Cancer Immunol. Immunother.
35
:
199
-204.
26
Yu, A. L., S. D. Gillies, R. A. Reisfeld.
1991
. Phase I clinical trial of ch14.18 in patients with refractory neuroblastoma.
Proc. Am. Soc. Clin. Oncol.
10
:
318
27
Cheung, N. K., H. Lazarus, F. D. Miraldi, C. R. Abramowsky, S. Kallick, U. M. Saarinen, T. Spitzer, S. E. Strandjord, P. F. Coccia, N. A. Berger.
1987
. Ganglioside GD2 specific monoclonal antibody 3F8: a phase I study in patients with neuroblastoma and malignant melanoma.
J. Clin. Oncol.
5
:
1430
-1440.
28
Gong, Y., Y. Tagawa, M. P. Lunn, W. Laroy, M. Heffer-Lauc, C. Y. Li, J. W. Griffin, R. L. Schnaar, K. A. Sheikh.
2002
. Localization of major gangliosides in the PNS: implications for immune neuropathies.
Brain
125
:
2491
-2506.
29
Goodnow, C. C., J. Sprent, B. Fazekas de St Groth, C. G. Vinuesa.
2005
. Cellular and genetic mechanisms of self tolerance and autoimmunity.
Nature
435
:
590
-597.
30
Manjarrez-Orduno, N., T. D. Quach, I. Sanz.
2009
. B cells and immunological tolerance.
J. Invest. Dermatol.
129
:
278
-288.