Serological Analysis of Human Anti-Human Antibody Responses in Colon Cancer Patients Treated with Repeated Doses of Humanized Monoclonal Antibody A33

INTRODUCTION

mAb 2 A33 recognizes a cell surface differentiation antigen (A33) of normal human gastrointestinal epithelium that is expressed in 95% of primary and metastatic colon cancer cells but which is absent in most other normal tissues and tumor types (1 , 2) . A33 is a M_r 43,000 transmembrane glycoprotein of the immunoglobulin superfamily (3, 4, 5) and distantly related to a lymphocyte differentiation antigen of Xenopus (6) and a coxsackievirus and adenovirus receptor (7) . A33 consists of two extracellular immunoglobulin domains, a single transmembrane domain, and a short intracellular tail containing four acylation sites proximal to the transmembrane domain (3 , 8) . A33 is not secreted or shed into the blood stream. Some colon cancer cell lines express large amounts of A33, binding up to 800,000 antibody molecules/cell (9) . Upon binding to A33, the A33 antibody is internalized into an incompletely characterized vesicular compartment, and a significant fraction of the internalized antibody is recycled back to the extracellular environment (9) .

The A33 antigenic system is the focus of several clinical studies in patients with colon cancer (10) . Phase I/II clinical trials have shown that the murine mAb A33: (a) localizes with high specificity to colon cancer; (b) is retained for prolonged periods (up to 6 weeks) in the cancer but cleared rapidly from normal colon (5–6 days); and (c) has antitumor activity as a carrier of ¹²⁵I or ¹³¹I (1 , 11 , 12) . However, the use of the murine mAb A33 is limited to very few injections because of the induction of strong HAMA responses in patients. To overcome the immunogenicity problem of the mouse mAb, a humanized version of the A33 antibody (huAb A33) had been constructed by grafting of the murine CDR regions into a human IgG1 framework (13) . The humanized antibody maintained full binding specificity and affinity and was subsequently prepared for clinical testing in Phase I and II studies in patients with colon cancer. We report here that, despite humanization, HAHA responses against huAb A33 could be detected in a large number of patients that had been treated with repeated doses of huAb A33. In this report, we describe the results of a detailed serological characterization of the HAHA responses with the use of a novel, highly sensitive biosensor technique.

PATIENTS AND METHODS

Patients

Sera were obtained from patients with colon cancer enrolled in Phase I and II clinical trials of huAb A33 at MSKCC. Dose and treatment schedules are shown in Table 1 ⇓ (BB-IND-6084, MSKCC Institutional Review Board approved protocols 94-100A, 96-22, 98-056, and 98-011). Blood for HAHA analysis was sampled immediately before antibody administration. Clinical results of the trials will be reported elsewhere.

Antibodies

HuAb33 Production and Purification.

huAb A33, a fully humanized IgG1 mAb was derived from the murine mAb A33 by CDR grafting as described (13) . The antibody was expressed in NSO cells. NSO cell culture supernatant was produced in bioreactors by Celltech Limited, United Kingdom, and huAb A33 was purified from NSO culture supernatant at the New York Branch of the Ludwig Institute for Cancer Research at MSKCC using a three-step column chromatography process: Q-Sepharose anion-exchange, protein A affinity, and S-Sepharose cation-exchange chromatography. Virus inactivation was achieved by using a pH-dependent procedure. The antibody was formulated in PBS (0.05 m sodium phosphate, 0.15 m sodium chloride; pH 7.0), sterile filtered, and stored at −80°C. The purified huAb A33 retained the original binding specificity and binding avidity of the murine mAb A33.

Other Antibodies and Reagents.

Humanized 3S193 (IgG1) was prepared as described (14 , 15) . Murine anti-A33 mAbs A33 (IgG2a) and 100.310 (IgG2a) were prepared in this laboratory (1) . Pegylated huAb A33 was prepared as described (16) . SK10B, a humanized IgG1 antibody of the same allotype as huAb, and humanized IgG4 antibody SK10C were provided by Celltech. Human-mouse chimeric G250 (IgG1) was produced by Centocor B.V. (Leiden, the Netherlands; Ref. 17 ). Goat anti-human IgG (γ-chain specific), goat anti-human IgM, and goat anti-mouse IgG were obtained from Sigma Chemical Co. (St. Louis, MO). mAbs against human IgA, IgE, and IgM were from Zymed Laboratories, Inc. (San Francisco, CA), and antibodies against human IgG1, IgG2, IgG3, and IgG4 were from Southern Biotechnologies (Birmingham, AL). Synthetic peptides (for sequences, see Fig. 6 ⇓ ) corresponding to various areas in the V_L and V_H chain of huAb A33 were obtained from Research Genetics, Inc. (Huntsville, AL).

Antibody Fragmentation.

Fab and Fc were prepared from huAbs A33 and 3S193 by digestion with immobilized papain according to the manufacturer’s instructions (Pierce, Rockford, IL). Purity of the fragments was assessed by SDS-PAGE electrophoresis and Coomassie staining. Deglycosylated huAbs A33 and 3S193 were prepared using the GLYCOPRO deglycosylation kit (Prozyme, San Leandro, CA) according to the manufacturer’s instructions. The nondenaturing protocol was followed. Deglycosylation of antibodies was assessed by a mobility shift on SDS-PAGE gels. To ensure that the antibody had not been denatured during the deglycosylation process, the immunoreactivity of humanized A33 was confirmed by BIACORE analysis (BIACORE, Inc., Piscataway, NJ) on immobilized recombinant A33 prepared in a baculovirus expression system. huAbs A33 and 3S193 were reduced overnight at room temperature with 100 mm 1,4-DTT or 5% 2-mercaptoethanol in PBS containing 2% SDS (pH 7.4). After reduction was complete, the reducing agent was removed using a PD-10 desalting column (Pharmacia) equilibrated in PBS. Antibodies were concentrated using Millipore Ultrafree-15 centrifugal concentrators.

Peptide Immunoblots.

Membranes with immobilized peptides were incubated with 1:100 diluted patient serum overnight at 4°C. Bound antibody was visualized by chemiluminescence (Tropix, Bedford, MA).

HAHA BIACORE Analysis

Antibody responses against humanized antibodies (HAHAs) induced after treatment of patients with huAb A33 were analyzed by surface plasmon resonance technology using a BIACORE 2000 instrument. Humanized antibodies (in 5 mm sodium phosphate, pH 5.5) and mouse antibodies (in 6 mm sodium acetate, pH 4.6) were immobilized to CM5 biosensor microchips using N-hydroxy-succinimide and N-ethyl-N′-dimethylaminopropyl carbodiimide. About 10,000 ± 2,000 RUs per flow cell were immobilized for HAHA measurements. Microchip sensor surfaces were conditioned with three cycles of five injections each of 5 μl of 15 mm HCl, followed by one injection of 10 μl of 0.2 m Na₂CO₃ buffer containing 1 m NaCl (pH 10) prior to use for HAHA analysis. Patient serum diluted with HBS buffer (10 mm HEPES, 0.15 m sodium chloride, 3.4 mm EDTA, and 0.05% surfactant P20) containing 1 mg/ml carboxymethyldextran (pH 7.4) was passed over the chip; after we washed this with HBS containing 0.5 M NaCl (pH 7.4), alterations in the refractory index were recorded as relative RUs as described by the manufacturer. Antibody binding is given in RUs over baseline recorded 70 s after injection of the wash buffer. The baseline was recorded 10 s prior to sample injection and set to 0 RU. Microchip sensor surfaces were regenerated with 10 μl of 15 mm hydrochloric acid prior to every new injection cycle. Microchip sensor surfaces were stable for >80 cycles. Patient serum was considered HAHA positive if the RU value at a serum dilution of 1:100 exceeded a cutoff value, defined as the mean inter-patient baseline RU value + 3 × SD of pretreatment sera at a serum dilution of 1:100.

For the absorption assays, sera were incubated with various test proteins or peptides at 50 μg/ml (final serum dilution, 1:100 in HBS, pH 7.4) at room temperature prior to BIACORE analysis as described above. Protein A and protein G precipitation was performed as follows. Approximately 100 μl of protein A or G beads were added to 500 μl of 1:100 diluted serum and incubated overnight at 4°C on a rotating platform. After removal of the beads by centrifugation, the serum was used in the BIACORE analysis. Caprylic acid precipitation was performed as described (18) . Isotyping of HAHA serum was performed using a BIACORE sandwich approach as follows. CM5 biosensor microchips, to which 3,000 ± 500 RUs of huAb A33 and huAb 3S193 (IgG1 isotype control) were immobilized, were exposed for 5 min to diluted patient serum (as described above), followed by exposure to isotype-specific monoclonal antibody (50 μg/ml) for 5 min. Binding of an isotyping reagent was recorded 2 min after injection of the wash buffer (HBS containing 0.5 m NaCl, pH 7.4).

RESULTS

Serological Monitoring of Colon Cancer Patients Treated with huAb A33: Definition of Response Pattern

Sequential serum samples obtained from 44 patients treated repeatedly with the humanized IgG1 monoclonal antibody A33 were tested for HAHA reactivity against huAb A33 using a BIACORE biosensor. Of the 44 patients, 28 patients were treated with unmodified huAb A33 alone, 13 patients were treated with unmodified huAb A33 plus a chemotherapy regimen (1,3-bis(2-chloroethyl)-1-nitrosourea, Oncorin, fluorouracil, and streptozotocin), and 3 patients were treated with ¹³¹I-labeled huAb A33 (for treatment schedule, see Table 1 ⇓ ). Prior to treatment with huAb A33, none of the 44 patients had measurable HAHA levels (median level, 14 RUs). After treatment with huAb A33, HAHA reactivity against huAb A33 was detected in 67% (20 of 28) of the patients treated with huAb A33 alone (8 of 11 patients in the Phase I trial and 12 of 17 patients in the Phase II trial), in 46% (6 of 13) of the patients receiving huAb A33 plus chemotherapy, and in all 3 patients treated with ¹³¹I-conjugated huAb A33 (Table 2) ⇓ . HAHA levels in the BIACORE assay in individual patients ranged from 30 RUs up to 1500 RUs after treatment with huAb A33 (Table 3) ⇓ . HAHA was detectable in some patients as early as 1 week after the first huAb A33 infusion. There appeared to be no correlation between induction or strength of the HAHA response and the dose of antibody injected (10–60 mg/m² huAb A33). Patients treated with huAb A33 plus chemotherapy appeared to develop HAHA at a somewhat lower frequency (46%) and had lower peak HAHA levels (median peak, 48 RUs; range, 32–831 RUs) than patients treated with huAb A33 alone (frequency, 71%; median peak, 73 RUs; range, 30–1503 RUs). Approximately 37% (15 of 41) of the patients did not develop HAHA or had only a marginal, temporary increase in measurable BIACORE serum reactivity levels (increase in RU less then two times the preserum level and/or ≤15 RUs) during or after treatment with huAb A33. BIACORE reactivity curves of representative patient sera (1:100 serum dilution) are shown in Fig. 1A ⇓ . BIACORE reactivity of a strong HAHA-positive serum (COL-11) after serial dilution is shown in Fig. 1B ⇓ .

In this series of patients, two major HAHA response types were observed. HAHA response types were defined according to the time of onset and the course of an immune response against the humanized antibody A33 (Fig. 2) ⇓ .

HAHA Response Type I.

This response pattern is characterized by an early onset of HAHA, with HAHA levels peaking after ∼2 weeks of antibody treatment, followed by a decline of measurable HAHA levels over time, although huAb A33 treatment was continued as scheduled (Fig. 2A) ⇓ . Approximately 49% (20 of 41) of patients treated with unmodified huAb A33 showed this pattern of response. Peak HAHA levels after two rounds of infusions ranged from high (an increase >100 RUs; 4 of 41 patients), medium high (an increase 41–100 RUs; 5 of 41 patients), to low (an increase 20–40 RUs; 11 of 41 patients). All three patients treated with ¹³¹I-labeled huAb A33 developed this type of response, with 2 patients having high and 1 patient having medium high HAHA levels.

HAHA Response Type II.

This response pattern is characterized by a delayed onset of HAHA and an increase of the HAHA level with subsequent treatments (Fig. 2B) ⇓ . Approximately 17% (7 of 41) of the patients treated with huAb A33 showed this pattern of response. Peak HAHA levels ranged from high (an increase >100 RUs; 3 of 41 patients) to medium high (an increase 41–100 RUs; 4 of 41 patients). Treatment with huAb A33 was discontinued in those patients because of adverse events. One patient (COL-41), who was treated with huAb A33 plus chemotherapy, developed first a type I response and subsequently a type II response.

Identification of the BIACORE-reactive Serum Component as IgG and IgM

The BIACORE assay measures the direct interaction of serum components with an immobilized antigen without using specific secondary reagents. To verify that the measured reactivity was attributable to immunoglobulin binding to huAb A33, selected serum samples were subjected to a protein G (to remove IgG) or a caprylic acid precipitation (for removal of non-IgG serum proteins) prior to BIACORE analysis. Serum reactivity with huAb A33 was completely depleted after removal of IgG by protein G precipitation in the majority of sera tested (Fig. 3) ⇓ . After treatment of sera with caprylic acid, the reactivity with immobilized huAb A33 was found in the supernatant (IgG fraction), whereas little to no reactivity was detected in the precipitate (data not shown). In sera of patients COL-3 and COL-25 sampled at week 2 of treatment (HAHA type I; containing the peak level reactivity with huAb A33), some residual reactivity with huAb A33 was detected after removal of IgG by protein G precipitation. This protein G nonprecipitable reactivity was identified as IgM binding using anti-human IgM in a BIACORE sandwich assay. The IgM reactivity was relatively weak (an increase <40 RUs), transient (detectable in sera after protein G precipitation at week 2 but nondetectable at week 9 of treatment), and occurred together with IgG in the early weeks of treatment (patient COL-3, week 2; Fig. 3 ⇓ ). IgM reactivity could not be detected by BIACORE in sera that were not IgG depleted or in an ELISA assay. Anti-human IgA and IgE did not react with serum antibodies bound to huAb A33 in a BIACORE sandwich assay. Selected HAHA sera (COL-11 week 2 and week 8, COL-24 week 7, COL-14 week 2, and COL-19 week 2) were typed for IgG subclasses using a BIACORE sandwich assay and IgG subclass-specific monoclonal antibodies. In all patients tested, the antibodies that bound to huAb A33 were identified as IgG1. No IgG2, IgG3, or IgG4 antibodies were detected (data not shown). Taken together, these results indicate that the increase in BIACORE reactivity with immobilized huAb A33 in sera from colon cancer patients treated with huAb A33 was predominantly attributable to induction of IgG1. In addition, transient, weakly reactive IgM against huAb A33 was occasionally induced.

Specificity Analysis of the Human Anti-huAb A33 Response and Epitope Mapping

The specificity analysis of the HAHA response was carried out using two complementary types of BIACORE-based assays. In one type of assay, sera from patients treated with huAb A33 were probed for reactivity against unrelated antibodies or huAb A33 antibody fragments that were immobilized on the biosensor chip surface. Sera that reacted with huAb A33 showed no reactivity with human IgG1 isotype control antibodies (including humanized mAb 3S193 and mouse-human chimeric mAb chG250) or a huAb A33 unrelated humanized IgG4 antibody SK10C (Fig. 4) ⇓ . In the second type of assay, serum reactivity against immobilized huAb A33 was measured using patient serum absorbed with various humanized, mouse-human chimeric, or mouse mAbs prior to BIACORE analysis. Of the antibodies tested in the absorption assay, only huAb A33 and mouse mAb A33 (from which the grafted CDR regions in huAb A33 were derived) completely absorbed serum reactivity against huAb A33 in all immune sera. Human IgG1-matched control mAbs and other murine mAbs did not absorb reactivity against huAb A33, including mAb 100.310, a murine antibody that recognizes an epitope on the A33 antigen that is different from the epitope recognized by murine mAb A33 (Fig. 5) ⇓ . huAb A33 has a non-Caucasian IgG1 allotype (non-2, a-1, z-allotype). Potential allotypic reactivity was investigated using a humanized antibody construct (SK10B) of an identical allotype but different CDR regions than huAb A33 and with no A33 antigen binding activity. No immunoglobulin allotype binding was observed in the BIACORE adsorption assay or with immobilized allotype-matched IgG1 (Fig. 4) ⇓ . These results indicate that the serum reactivity was specific for epitopes contained in huAb A33 and mAb A33 and that the immunodominant epitopes are located in the variable region of the huAb A33.

Representative HAHA sera were selected for a more detailed mapping of the antigenic determinant. The following results suggest that the primary HAHA reactivity is directed against antigenic determinants in the V_L and V_H regions in huAb A33 that include the CDRs: (a) HAHA sera reacted with immobilized huAb A33-derived Fab and Fab′ but not IgG with different V regions (Fig. 4) ⇓ or huAb A33-derived Fc (data not shown). In addition, these sera did not react with a Fab of a new generation of humanized A33 antibodies derived from rabbits using phage display technology and CDR grafting (19) ; and (b) in serum absorption assays, reactivity of immune patient sera against huAb A33 was absorbed by huAb A33 Fab′ (Fig. 5) ⇓ and also by a pegylated derivative of huAb A33, which had retained full A33 antigen binding activity after chemical conjugation of polyethyleneglycol to the antibody.

After localizing the immunodominant epitopes to the V_L and V_H region, we attempted to map the epitope recognized by individual patient immune sera to a defined sequence of amino acids in the V_L and V_H region as follows:

(a) Multiple synthetic peptides (n = 24) corresponding to overlapping huAb A33 V_L and V_H sequences (Fig. 6) ⇓ were immobilized on polyvinylidene difluoride membranes and used to probe preimmune and immune sera from two high-titered HAHA patients (COL-11 and COL-24) for reactivity with those peptides. The patients’ preimmune and immune sera showed indistinguishable reactivity pattern. Thus, this method did not allow mapping of specific amino acid sequences recognized by the induced IgG antibodies against huAb A33.

(b) Eight peptides corresponding to all huAb A33 V_L and V_H sequences that contained murine amino acid residues (Fig. 6) ⇓ were synthesized and used in BIACORE absorption assays. None of the peptides tested blocked binding of huAb A33-positive immune sera to huAb A33 (COL-3, COL-5, COL-10, COL-11, COL-20, and COL-24).

(c) To test whether the epitopes recognized by the patients’ immune sera involved glycosylated sites, the huAb A33 was N- and O-deglycosylated and used in the absorption assay. Reactivity of immune sera against huAb A33 was absorbed by deglycosylated huAb A33 (which had retained A33 antigen activity), indicating that posttranslational carbohydrate modifications in huAb A33 do not contribute to the antigenicity of the antibody (Fig. 7) ⇓ .

(d) To test whether the epitopes recognized by the patients’ immune sera require association of V_L and V_H chain, huAb A33 was reduced with 2-mercaptoethanol or 1,4-DTT prior use in the absorption assay. huAb A33 reduced by 2-mercaptoethanol or 1,4-DTT lost the ability to block reactivity of immune sera with huAb A33 (Fig. 7) ⇓ .

Taken together, these results suggest that the main antibody response in patients treated with huAb A33 is directed against one or multiple conformational antigenic determinants located within the V_H and V_L chain regions in huAb A33 and requires association of V_L and V_H chains.

HAHA and Adverse Events

Colon cancer patients developing a HAHA response of type I did not show significant infusion-related symptoms. Although most patients with a type II HAHA response did not clearly have higher titers or RU values in the BIACORE assay when compared with type I responders, virtually all infusion-related adverse events occurred in the HAHA type II group (Table 4) ⇓ . Infusion-related symptoms in those patients included fever, chills, nausea, vomiting, rash, and myalgias. Although patients with HAHA type II responses had increasing RU values in the BIACORE measurements from week to week with continued treatment of huAb A33, symptoms did not occur until after several weeks of progressively increasing HAHA levels (Fig. 8) ⇓ . Thus, after observing this pattern of increasing HAHA levels and induction of symptoms, we were able to predict which patients would develop infusion- related reactions, and we discontinued huAb A33 treatment in subsequent patients (e.g., COL-33 and COL-41; Table 3 ⇓ ) exhibiting this pattern on BIACORE measurements.

DISCUSSION

Because of their potential to target and mediate selective destruction of cancer cells, mAbs have generated much interest as diagnostic and therapeutic agents. For therapeutic application, the immunogenicity of the injected antibody is of paramount importance. For example, mouse-derived mAbs are highly immunogenic in humans and generally can be given only for a limited period of time before onset of a HAMA response, which rapidly clears the murine antibody from the blood and may trigger anaphylactic reactions (1 , 11 , 12) . To attenuate or abolish immunogenicity and to allow long-term treatment of patients without the interference of an anti-immunoglobulin response, various molecular technologies have been used that reduce the amount of foreign immunoglobulin residues in a therapeutic antibody, a factor hypothesized to inversely correlate with the immunogenicity of an antibody. Antibodies obtained by molecular engineering include human-mouse chimeric versions, humanized versions, and fully human antibodies produced in human immunoglobulin transgenic animals and from human immunoglobulin gene libraries (20, 21, 22, 23, 24, 25, 26) . However, the CDR regions of all these antibodies are unique and may contain inherently antigenic determinants that potentially can become immunogenic no matter how an antibody was generated.

To date, limited information is available on the immunogenicity of humanized therapeutic antibodies in cancer patients. Only a few clinical studies thus far have assessed the immunogenicity of repeated doses of different chimeric or humanized antibodies, and the methodological challenge of measuring unambiguously and with high sensitivity HAHA responses in patients. Clinical trials reporting on the HAHA response to humanized antibodies in cancer patients generally have used ELISA-based techniques for assessment of HAHA (27, 28, 29, 30) . ELISA, however, may not be the ideal methodology to assess quantitatively and with high sensitivity the total polyclonal HAHA response generated in a patient against a therapeutic humanized antibody because of several inherent problems. These include: (a) secondary reagents have been used for sandwiching (bridging) and visualizing a positive reactivity that depend on the multivalency of an antigen; (b) when both antigen (administered antibody) and the polyclonal antibodies to be assessed (induced antibodies) are human immunoglobulins, applicable ELISA formats are limited to those that are based on multivalency; and (c) assays have been generally designed and optimized to measure a specific anti-idiotypic response and would therefore not detect responses to other antigenic determinants. Thus, even when HAHA had been assessed for a particular antibody, the results cannot be compared to studies with other antibodies, and little has been learned that is applicable to other antigenic systems. Immune responses in cancer patients after treatment with human-mouse chimeric antibodies have also been analyzed using a “double antigen” assay (31 , 32) . Although this design overcomes some of the disadvantages of the ELISA, it requires radioisotope-labeled antibody.

To overcome these pitfalls, we have developed a HAHA detection assay that uses biosensor technology (33) . In this technique, the humanized antibody and control reagents are chemically immobilized to the surface of a biosensor chip. The chip is then exposed to patient sera, and antibody-antibody binding is quantitated in real time by surface plasmon resonance, a methodology that measures refractive index changes of a light beam. Secondary serological reagents are therefore not required. We have found that the assay is specific for the test antibody and highly sensitive, able to detect picomolar concentrations of anti-human immunoglobulin antibodies in the sera of patients. Using this technique, we were able to carry out a detailed monitoring of HAHA reactivities in 44 patients with colon cancer enrolled in Phase I and Phase II clinical studies with the humanized antibody A33. Patients in these studies were injected repeatedly with huAb A33 alone, with huAb A33 in combination with chemotherapy, or with radioisotope-labeled huAb A33 (¹³¹I-labeled huAb A33).

Before treatment with huAb A33, none of the patients with colon cancer had any measurable HAHA levels against huAb A33 or against other humanized control antibodies. Pretreatment sera were indistinguishable from sera obtained from normal individuals, indicating the absence of preexisting allotypic, idiotypic, and heterophilic antibodies against the humanized antibodies. After treatment with huAb A33, we detected HAHA in 71% (20 of 28) of the patients receiving unmodified huAb A33 alone, in 46% (6 of 13) of the patients receiving unmodified huAb A33 in combination with chemotherapy, and in all 3 (3 of 3) patients receiving ¹³¹I-labeled huAb A33. We observed two distinct types of HAHA responses. The two types differed with regard to the time point of HAHA onset, the course of HAHA after subsequent treatment with huAb A33, and in their impact on continued administration of huAb A33. In both types, the induced antibodies were predominantly of the IgG isotype and specifically reactive with epitopes located within the variable regions of the assembled H and L chains of huAb A33.

A HAHA response of type I was characterized by an early onset of HAHA, with peak serum levels detectable after ∼2 weeks of treatment with huAb A33. With subsequent antibody injections, detectable HAHA levels declined and in some patients fell to baseline values. Patients developing this type of HAHA response did not have any significant treatment-limiting, infusion-related reactions. HAHA of type I was found in 49% (20 of 41) of patients treated with huAb A33. A HAHA response of this type was unexpected, because we have not observed this response pattern in our trials with other humanized or chimeric antibodies nor, to our knowledge, has this type of HAHA response been reported before. Possibly, the response is unique for the A33 antigenic system. On the basis of the fast onset and the predominantly IgG response, the type I response pattern has the characteristics of a recall reaction (memory response). However, the sensitizing immunogen is unknown, because none of the patients had prior exposure to huAb A33. Because the antigenic determinant recognized by the HAHA antibodies has been located within the assembled V_H and V_L region of huAb A33 and includes the CDR regions, it can be speculated that the unknown immunogen might mimic this epitope. Candidate molecules include antibodies against an anti-idiotypic huAb A33 antibody (Ab3; whereby huAb A33 is Ab1, and the huAb A33 anti-idiotypic antibody is Ab2). According to the Jerne network theory, Ab2 is required for the induction of Ab3. However, we did not detect antibodies against huAb A33 (Ab2) in patients prior to treatment, against the A33 antigen (Ab1 or Ab3) in patients with colon cancer, or in normal individuals. Other candidate immunogens may include A33 antigen ligands that bind to the same epitope on the A33 antigen as huAb A33 and thus may sterically mimic the antigenic determinant on huAb A33. The natural A33 ligand has not been defined as yet. Other immunogenic candidates include intestinal microflora or alimentary-derived molecules that bind to the A33 antigen that is expressed abundantly in normal intestinal epithelial cells. Alternatively, this type of HAHA response may be interpreted as an immediate primary IgG immune response that is attenuated and down-regulated by subsequent treatments with antibody.

HAHA response type II was characterized by a generally delayed onset of HAHA and by a progressive increase in HAHA levels as a consequence of each subsequent antibody administration. In contrast to the lack of toxicity associated with retreatment with huAb A33 in the presence of a type I HAHA response, continued re-injection with huAb A33 of patients with type II HAHA responses ultimately led to infusion-related toxicities necessitating termination of huAb A33 treatment. HAHA type II responses were found in 17% (7 of 41) of patients treated with huAb A33. They were less frequent than HAHA type I. HAHA type II has the characteristics of a newly induced immune response against huAb A33. The detectable onset of the immune response varied from patient to patient, ranging from week 2 to week 21 after the first administration of huAb A33. This variable time interval might reflect the immune competence of individual patients or a different degree of reactivity against the huAb A33 in different patients. In some patients, repeated injections of the antibody was required to induce an immune response (e.g., patient COL-24). In other cases, additional immune stimulatory signals might have been required to trigger induction of HAHA (e.g., patient COL-10 with a HAHA induction only after 21 weeks of treatment). Those signals may have been derived from events such as concurrent infections or tumor lysis.

Detailed serological analysis showed that the antigenic determinants in both types of HAHA responses were located solely within the variable regions of the assembled H and L chains of huAb A33. Although the CH and CL were derived from a rather uncommon IgG1 allotype (non-2, a-1, z-allotype), no allotypic responses were detected, suggesting that this allotype does not contribute to the immunogenicity of huAb A33. Terminal α-Gal residues on carbohydrate side chains of an antibody (particularly if derived from a mouse hybridoma or immunoglobulin transgenic mouse) have been considered a potential antigenic determinant for circulating or induced α-Gal antibodies in human serum (34) . In the case of huAb A33, originally derived from a mouse hybridoma (NSO cells), α-Gal antibodies did not contribute to HAHA because the reactivity with deglycosylated huAb A33 was indistinguishable from glycosylated huAb A33. Furthermore, no α-Gal reactivity had been observed in sera from colon cancer patients against mAb 3S193, a humanized IgG1 antibody produced in NSO cells, which was used as isotype control antibody in our studies.

Many factors will influence the immunogenicity of a therapeutic antibody (35) . Some are inherent to the antibody construct, whereas others are related to the way the antibody is administrated or the way the patient responds. To date, there is no adequate “in vitro” or “in silico” method that can predict the “in vivo” immunogenicity of an antibody construct, and clinical trials are still the only way to assess whether an antibody is immunogenic. As we report here, HAHA responses in patients against an administered humanized antibody can come in different forms, and it would be desirable to develop a fast, specific, and sensitive method for HAHA assessment that allows discrimination between response types that are harmful to a patient and require termination of treatment with antibody from those response types that appear tolerable and allow continued treatment with antibody. The BIACORE method used in this study fulfills these criteria and is now being used in all our studies with humanized antibodies. Another approach to assessing immunogenicity include changes in the pharmacokinetics of an administered antibody, e.g., more rapid clearance from the blood may indicate induction of a neutralizing HAHA response. In all three patients treated with ¹³¹I-labeled huAb A33, HAHA data, as measured in the BIACORE, correlated with a more rapid serum clearance of the injected antibody. 3 However, radiolabeling techniques chemically alter an antibody, and thus immunogenicity results obtained with labeled antibodies might not be fully representative of what would be found with unmodified “naked” antibodies. It is obviously important to determine the effect these two types of HAHA responses have on antibody-serum pharmacokinetics. Clearly, in the case of type II responses, the early termination of antibody treatment compromises the ability to obtain the best anticancer response for the patient. However, the type I response may also be associated with accelerated antibody clearance from blood, thereby providing a window for continued tumor growth and leading to incorrect assessment of ineffectiveness of the antibody. The BIACORE technology described here may offer important additional information for the evaluation of the therapeutic potential of antibody-based immunotherapy.