Development and Preclinical Characterization of a Humanized Antibody Targeting CXCL12

This article is featured in Highlights of This Issue, p. 4293

CXCL12, also known as stromal-derived factor 1 (SDF1), is a CXC chemokine that binds to two G-protein–coupled receptors, CXCR4 and CXCR7 (reviewed in ref. 1). It participates in many developmental and physiologic processes, including hematopoiesis and angiogenesis. Gene deletion of CXCL12 or CXCR4 results in embryonic lethality at E18.5, with severe developmental defects affecting central nervous system, bone marrow, heart, and vascular system (2, 3). CXCR7 inactivation, on the other hand, seems to affect primarily cardiac development, although defects in neuronal migration have also been noted (4). CXCL12α, the major form present in mice and humans, is highly conserved, as the mouse and human forms differ only in one amino acid (1). CXCL12α is produced by numerous cell types including macrophages, lymphocytes, endothelial cells, and fibroblasts. CXCR4 activates multiple downstream targets, including extracellular signal–regulated kinase (ERK)1/2, AKT, and small GTPases of the Rho family (5). CXCR7, on the other hand, seems to function at least in part by sequestering CXCL12 (4, 6).

The CXCL12/CXCR4/CXCR7 pathway has also generated considerable interest as a potential therapeutic target given its role in tumor growth, survival, and angiogenesis (1, 7). CXCL12 levels have been shown to be important in dictating organ-specific metastasis of several cancers (8). It has been previously reported that CXCL12 autocrine signaling loops, together with TGF-β, initiate and maintain the differentiation of fibroblasts to myofibroblasts and the concurrent mammary tumor-promoting phenotype (9, 10). CXCL12 induces angiogenesis, both in vitro and in vivo, synergistically with VEGF (11). VEGF, in turn, can regulate expression of CXCR4 in vascular endothelial cells, hematopoietic stem cells, and promote invasion in an autocrine manner by regulating CXCR4 on cancer cells (12, 13). In recent years, it has become clear that CXCR4⁺ myeloid cells, recruited from the bone marrow into the tumor microenvironment, participate in tumor angiogenesis (14, 15).Growing evidence also suggests that CXCR7 is involved in several aspects of tumorigenesis and could become an important target for new antimetastatic and anticancer drugs (16). Recent preclinical studies suggest that small-molecule inhibitors that target the tyrosine kinase function of VEGF receptors result in increases in serum CXCL12 levels, which might relate at least in part to their ability to promote a more invasive tumor phenotype, when administered at high doses (17, 18). Induction of CXCL12 has also been shown to cause development of resistance and to favor recurrences upon radiation (19).

Here, we describe a high-affinity anti-CXCL12 antibody that shows antitumor effects in multiple mouse and human tumor models, as a single agent or in combination with anti-VEGF antibodies. Our data further validate CXCL12 as a target not only in cancer but also in several inflammation/angiogenesis–related disease models.

Jurkat, EL4, LL/2 (LLC1), 4T1, BxPC3, KMS11, Calu6, A673, and Ovcar-3 tumor cells were from the American Type Culture Collection (ATCC). HM7 cells were originally from Dr. Young Kim (University of California, San Francisco, San Francisco, CA). Jurkat cells were cultured in RPMI media containing 10% FBS (Sigma). 4T1 cells were cultured in Iscove's modified Dulbecco's medium supplemented with 10% FBS. The remaining cell lines were maintained in high glucose Dulbecco's modified Eagle medium (DMEM; Invitrogen) containing 10% FBS, with the exception of EL4 cells which were cultured in DMEM with 10% horse serum.

Female Armenian hamsters were hyperimmunized intraperitoneally, twice per week for a total of 12 to 13 boosts, with 2 μg/animal recombinant human CXCL12α (R&D Systems). Three days after the final boost, splenocytes were fused with cells derived from the murine myeloma cell line P3 × 63AgU.1 (ATCC). Hybridomas were selected using methylcellulose-based hypoxanthine–aminopterin–thymidine (HAT) medium (STEMCELL Technologies). Ten to 14 days after fusion, culture supernatants were collected and screened by direct ELISA as previously described (20). Positive cell lines were subcloned four to six times by limiting dilutions. Those that kept showing consistent activity were scaled up and purified by Protein A affinity chromatography. The eluate was dialyzed against PBS and concentrated. For large-scale antibody purification, HiTrap (Amersham) protein A column chromatography was used.

293 cells were cotransfected with CXCR7 in pCMV-XL or CXCR4 in PRK5 with P8vE-Neo vectors. Cells were cultured in F12:DMEM (50:50) media containing 2 mmol/L l-glutamine, 1× GHT (glycine, hypoxanthine, thymidine), and 10% FBS in the presence of 400 μg/mL G418 (Calbiochem) with change of media every 2 to 3 days. Expression of CXCR4 and CXCR7 was confirmed by FACS analysis using specific antibodies (MAB173/FAB170 for CXCR4; MAB4227/MAB42273 for CXCR7; R&D systems). All secondary antibodies were obtained from Jackson ImmunoResearch Laboratories, Inc..

Anti-CXCL12 antibodies were prepared in PBS containing 2% FBS, pH 7.4 in serial 1:3 dilutions and incubated with 1 ng/mL ¹²⁵I-huCXCL12α (PerkinElmer) in 96-well polypropylene plates (Costar) for 1.5 hours. Meanwhile, 293-CXCR4 cells or 293-CXCR7 cells were prepared at 0.8 × 10⁵ cells per well and incubated with preincubated ¹²⁵I-huCXCL12α and anti-CXCL12 antibodies for 2 hours. The incubated cell solutions were transferred to prewet MultiScreen, BV filter plate (Millipore). The cell-bound CXCL12α was determined by counting the cpm captured on individual filters. AMD3100 (Sigma) was included as a control and used in the same way as antibodies.

Duplicate wells of isolated mouse bone marrow or Jurkat cells (5 × 10⁵ cells), resuspended in Hank's balanced salt solution containing 0.2% bovine serum albumin or in 5% FBS, were put into 24-Transwell inserts with 5-μm pore size. Media alone or media with various chemokines were added to the lower chamber. After 3 hours at 37°C/5% CO₂ incubator, cells in the lower chamber were counted. For 4T1 migration, the top of the 24-Transwell plates (8-μm pore size; BD Biosciences) was coated with 10 μg/mL collagen type I from rat tail (Upstate) prepared in PBS for 1 hour at 37°C. A total of 5 × 10⁴ 4T1 cells per well were seeded in 200 μL of serum-free medium on the top and 350 μL of serum-free media containing CXCL12α, with or without various concentrations of 30D8 were added to the bottom well. After 16-hour incubation at 37°C/5% CO₂, cells remaining on the top were removed and those migrated through the filter were counted under inverted microscopy (Zeiss Inc.).

Jurkat cells stimulated with CXCL12α in the absence or presence of anti-CXCL12 antibodies were lysed in RIPA buffer (Sigma) with proteinase (Roche) and phosphatase inhibitor cocktails (Sigma). Lysates were incubated at 4°C for 2 hours with GST-human p21-activated kinase PBD-coupled glutathione-sepharose beads (Cytoskeleton), followed by washing with RIPA buffer. Total and GTP-bound Rac1 were detected by Western blotting with an antibody against Rac1 (Cell Signaling Technology, Inc.).

Cells were incubated for various lengths of time in clear bottom/black wall 96 wells at 37°C in a 5% CO₂ incubator. Cell numbers were determined by the CellTiter-Glo Luminescent Cell Viability Assay Kit (Promega). In some cases, proliferating cells and apoptotic cells were visualized by Ki67 and caspase-3 staining.

Six- to 8-week-old female Balb/c, Balb/c nude mice, and C57BL/6 mice were obtained from Charles River or Jackson Laboratory. Maintenance of animals and experimental protocols were conducted following federal regulations and were approved by Institutional Animal Care and Use Committee at Genentech. For EL4, LLC, A673, HM7, Calu6, BxPC3, and KMS11 tumors, 10⁶ cells were washed with PBS once before resuspending in growth factor–reduced Matrigel (BD Biosciences) at a concentration of 10⁷ cells/mL and injected (100 μL) subcutaneously into the dorsal flank of mice. Anti-Ragweed, gD:2566, and gD:5237 were used as anti-mouse, -hamster, and humanized antibody controls. Primary tumor volumes were measured once a week using the ellipsoid volume formulas (0.5 × L × W², where L is the length and W is width) and verified at terminal point by tumor weight. All antibodies used in animal studies had endotoxin level below 0.02 eu/mg. AMD3100 was administered five times/week at a dose of 5 mg/kg, i.p.

4T1 cells were injected into the right fourth mammary fat pad of female Balb/c mice at the concentration of 2 × 10⁵ cells per 10 μL of PBS using an insulin syringe. Treatment was started 2 days after tumor cell implantation. At the end of the study, lungs were perfused with PBS, fixed in 10% formalin, and visible tumors were counted (21). In some cases, authenticity of tumors within lungs was confirmed by H&E staining and Bioluminescence Imaging (BLI).

Whole tumors were minced into small pieces, followed by sequential digestion with Media I for 60 minutes and Media II for 15 minutes, as previously described (22). Single cells were first incubated at room temperature for 10 minutes with rat anti-mouse CD16/CD32 mAb (Fc block; BD Biosciences), followed by staining with the following fluorophore-conjugated antibodies: anti-CD11b (integrin αM chain, Mac-1, α-chain, M1/70; BD Pharmingen), anti-Ly6G (Gr1, RB6-8C5; eBioscience), anti-CD31 (PECAM-1; eBioscience), anti-human/mouse CXCR7 (8F11-M16; BioLegend), anti-F4/80 (BM8; eBioscience), and anti-CXCR4 (CD184; BioLegend). All isotype-control antibodies were from R&D Systems. Propidium iodide (Sigma) was used to distinguish viable and dead cells. Data were acquired using the FACSCalibur or BD LSR II instruments (BD Biosciences) and analyzed using FlowJo software (TreeStar).

Unless specifically stated, Student t test was used for all the statistical analyses, and P ≤ 0.05 was considered to be significant. Graphs present mean ± SD.

Antibodies against both mouse and human CXCL12α were generated in hamsters. Serum titers were assayed for their ability to bind mouse and human CXCL12α by direct ELISA (Supplementary Methods). Five hamster antibodies showed different degrees of inhibition of CXCL12α-induced chemotaxis of Jurkat cells, with 30D8 being the most potent (Fig. 1A). MAB310 (R&D Systems) served as a positive control. 30D8 at 2 μg/mL completely blocked chemotaxis of Jurkat cells elicited by 10 ng/mL CXCL12α, with an average IC₅₀ of approximately 0.5 μg/mL (∼3 nmol/L). MAB310 was about 50-fold less potent. None of these antibodies had any effect on migration when tested alone. 30D8 could also bind to CXCL12β and CXCL12γ, as assessed by direct ELISA and CXCL12β-induced Jurkat cell migration (data not shown). Next, we compared the abilities of various antibodies to block CXCL12α binding to 293-CXCR4 or 293-CXCR7 cells (Fig. 1B and C). Again, 30D8 was the most potent in blocking binding to CXCR4, whereas all antibodies showed various degrees of inhibitory effect on binding to CXCR7. The inhibition was specific as 30D8 did not affect chemotaxis of isolated mouse bone marrow cells induced by unrelated chemokines such as KC, MIP2, or RANTES (Fig. 1D).

AMD3100 (23), a CXCR4 small-molecule antagonist, showed inhibitory effects in 293-CXCR4 cell-binding assay, with an IC₅₀ of 0.16 nmol/L.

Upon binding to CXCL12α, CXCR4 undergoes dimerization and stabilization leading to activation of a number of downstream signaling pathways including the small GTPase Rac1, which has an essential role in cytoskeleton remodeling (5). Jurkat cells showed a marked increase in GTP-bound Rac1 when treated with CXCL12α at 100 ng/mL for 5 minutes. 30D8, 18E9, or 46H9 inhibited CXCL12α-induced Rac1 activation in a dose-dependent manner. In the same assay, anti-CXCR4 antibody MAB170 from R&D systems at 200 μg/mL and AMD3100 at 40 μg/mL had moderate effects (Fig. 1E).

The mouse lymphoma model EL4 has been shown to be relatively refractory to anti-VEGF antibody therapy in vivo (15, 24, 25). Single-agent 30D8 significantly inhibited EL4 tumor growth in C57BL6 mice compared with the control antibody at a dose as low as 10 mg/kg. A 50% growth inhibition was achieved at the dose of 30 mg/kg, three times a week. The magnitude of inhibition was comparable to that obtained with the anti-VEGF antibody B20.4.1.1 (26) at 10 mg/kg twice weekly (Fig. 2A). A combination of B20.4.1.1 and 30D8 monoclonal antibodies (mAb) achieved a greater inhibition (∼70%). There was a significant difference in tumor size between mice treated solely with B20.4.1.1 versus those treated with a combination of 30D8 and B20.4.1.1 (P < 0.01).

A significant inhibition of proliferation and induction of apoptosis was noted in tumor sections from animals treated with 30D8, as assessed by caspase-3 and Ki67 staining (Supplementary Fig. S1). There was a 53% and 40% reduction in F4/80⁺ and CD11b⁺Gr1⁺ cell infiltration, respectively (P < 0.01) in 30D8-treated mice compared with control antibody–treated mice. Only a 21% reduction in F4/80⁺ cell infiltration was observed in B20.4.1.1-treated mice (Fig. 2B). In addition, an increase in CD11b⁺Gr1⁺ cells within the tumors was observed following treatment with B20.4.1.1. Blood vessel density as assessed by CD31⁺ cells was reduced in both 30D8- and B20.4.1.1-treated groups.

Next, we evaluated the effects of 30D8 on growth of Lewis lung carcinoma (LLC) cells implanted subcutaneously in C57BL6 mice (ref. 25; Fig. 2C). 30D8, administered on the same day of tumor cell implantation at the dose of 30 mg/kg, resulted in about 50% inhibition at day 14, which was comparable with that obtained with B20.4.1.1 at 10 mg/kg. Single-agent activity was also observed when 30D8 was given to animals bearing established tumors of approximately 400 mm³ (data not shown). An additive effect (P < 0.01) was observed when 30D8 was combined with B20.4.1.1 (Fig. 2C). Interestingly, CXCL12α had a significant effect on LLC cell survival in vitro under serum-free conditions. As shown in Fig. 2D, about 50% of the cells in the control group began to die after 24 hours in culture, whereas cells incubated with as low as 1.6 ng/mL CXCL12α survived as well as those in 10% serum and continued to proliferate (Fig. 2D). Such effects were completely neutralized in the presence of 100 μg/mL 30D8.

To investigate the effects of 30D8 on cancer metastasis, 4T1 breast cancer cells transfected with the luciferase gene were orthotopically inoculated into Balb/c mice (21, 27). Although treatment with 30D8 had no effect on primary tumor growth (Fig. 2E), it resulted in approximately 60% reduction in the number of lung cancer nodules at the dose of 10 mg/kg (Fig. 2F). All lungs in the control group developed metastasis, with an average of 14 nodules, whereas one third of the mice treated with 30D8 had no visible lung metastasis and those that did develop lung metastasis had an average of 5 nodules. Combination of 30D8 at 30 mg/kg with B20.4.1.1 at 10 mg/kg further enhanced the inhibitory effects compared with B20.4.1.1 alone (P < 0.05). The result was confirmed on 7-μm thick lung paraffin-embedded sections stained with hematoxylin and eosin (H&E) to reveal cancer nodules not only on the surface of lung but also within lung tissues (data not shown).

Consistent with the in vivo observations, CXCL12α had no effect on 4T1 cell proliferation. However, it significantly stimulated 4T1 cell migration at 10 ng/mL (∼4.5-fold increase compared with control) in vitro. Such stimulation could be completely blocked by 50 μg/mL 30D8 (Supplementary Fig. S2).

CXCL12α is highly expressed in the synovium of patients with rheumatoid arthritis and has been implicated in the recruitment and accumulation of CD4⁺ memory T cells, B cells to the synovium during the induction phase as well as in the recruitment of myeloid cells and synovial fibroblasts during disease onset and development (28). To investigate whether 30D8 had any effect in an experimental mouse collagen-induced arthritis (CIA) model (Supplementary Methods), we first injected 30D8 into DBA-1J mice at day 24 following bovine collagen type II injection (preventive phase). Mouse/hamster chimera 30D8 (ch30D8) was used to avoid immune responses. muTNFRII-IgG2a was used as a positive control (29). Compared with muTNFRII-IgG2a, ch30D8 at 30 mg/kg was not effective during onset of the disease (day 24–35) but could significantly slow down disease progression at day 35 to 73 (Fig. 3A).

To determine whether ch30D8 can inhibit established inflammation, we initiated treatments in cohorts of CIA mice at day 42. ch30D8 monotherapy significantly ameliorated arthritis compared with the control antibody (Fig. 3B). Although treatment with muTNFRII-IgG2a or ch30D8 alone could reduce the clinical score on day 83, the combination was significantly better than each monotherapy in preventing the progression of arthritis in mice with mild disease at enrollment (Fig. 3C). Such combination substantially reduced bone-erosive changes in CIA mice, resulting in a considerably higher joint cortical bone volume (JCBV) compared with mice treated with either inhibitor alone (ref. 30; Fig. 3D).

Laser-induced rupture of Bruch's membrane results in choroidal neovascularization (CNV) in mice (31). Bone marrow–derived hematopoietic stem cells are known to be incorporated into sites of retinal and CNV (32). Treatment with 30D8 at 30 mg/kg intraperitoneally (i.p.) resulted in a 50% reduction of neovascularization not only in prevention (Fig. 3E) but also in intervention studies, in which 30D8 was given 48 hours post-laser (Supplementary Methods and Supplementary Fig. S3A). Representative images of CNV in animals treated with various antibodies are shown in Supplementary Fig. S3B. In both circumstances, 10 mg/kg B20.4.1.1 resulted in more than 90% inhibition.

Infiltration of macrophages/dendritic cells expressing CX3CR1 is known to occur as early as 8 hours after laser injury, with a peak at day 2 (33). We used a transgenic mouse model in which one copy of the CX3CR1 gene is replaced by the GFP reporter gene (34). Compared with normal eyes, there was an approximately 20-fold increase in macrophage infiltration in eyes with laser injury. When animals were pretreated with 30D8 24 hours before laser injury, macrophage/dendritic cell infiltration into the injured sites was significantly reduced compared with the control antibody group (Fig. 3F). Representative images of infiltrating GFP⁺ cells are shown in Supplementary Fig. S3C.

30D8 was selected for humanization (Supplementary Methods) based on its in vitro and in vivo activities. The alignment of light chain κ between hamster 30D8 and humanized 30D8 (hu30D8) is shown in Supplementary Fig. S4 (35–38). hu30D8 bound not only to CXCL12α, but also to CXCL12β and CXCL12γ in direct ELISA. It specifically blocked CXCL12α- and CXCL12β-induced Jurkat cell migration, with an IC₅₀ of 0.16 μg/mL (∼1 nmol/L), which is comparable with both hamster and mouse/hamster chimeric counterparts. Because of a high apparent on-rate and nonspecific binding of CXCL12α to the dextran-coated sensor chip at high concentration, the affinity for binding to hu30D8 could not be reliably determined using Biacore technology. K_D was later measured by Biolayer Interferometry (Supplementary Methods) as 0.923 and 2.39 nmol/L for human and mouse CXCL12, respectively.

We also tested the activity of hu30D8 as a single agent in other human xenografts, including the A673, BxPC3, HM7, KMS11, and Ovcar-3 models. In general, hu30D8 was less effective in these models compared with Calu6, with 15% to 30% tumor inhibition achieved at the dose of 30 mg/kg. On the basis of TaqMan analysis, expression levels of CXCL12α and its receptors in human tumor cells in vitro did not correlate with efficacy of hu30D8 in vivo.

We initially sought to identify the epitope in several hamster antibodies, including 30D8, by alanine scanning of CXCL12α (Supplementary Methods). This analysis indicated that Asn⁴⁴ in the middle region of the molecule is the key binding epitope for 30D8, whereas Val¹⁸, Leu⁴², and Asn⁴⁵ were also important for 30D8 binding. Other hamster antibodies and the commercial antibody MAB310 seemed to cover a wider region. On the basis of epitope mapping data, CXCL12α processing at N- and C-termini in blood should not affect its binding to 30D8.

To further elucidate the molecular recognition of CXCL12α by 30D8 and the mechanism of inhibition, we determined the crystal structure of hu30D8 Fab in complex with human CXCL12α (Supplementary Methods and Supplementary Fig. S4). The crystallographic asymmetric unit contains a CXCL12α dimer and 2 Fab molecules each binding to one CXCL12α monomer (Fig. 4A). This is consistent with concentration-dependent dimerization of CXCL12α (40) as the crystallization media contains 160 μmol/L CXCL12α (Supplementary Methods; ref. 41). It also indicates that the antibody does not inhibit CXCL12α dimerization.

hu30D8 induced dose-dependent inhibition of EL4 tumor growth in C57BL6 mice at 10 to 50 mg/kg dose range. This effect was comparable with that of its hamster counterpart (Fig. 5A). When hu30D8 was given in combination with B20.4.1.1, an additive effect was seen (Fig. 5B). hu30D8 also elicited inhibitory effects in the human Calu6 lung carcinoma, a tumor model highly responsive to VEGF blockade (39). Of note, 10 and 30 mg/kg hu30D8 were equally efficacious in this model and resulted in approximately 50% tumor growth inhibition, which was statistically significant compared with the control antibody group (P < 0.05; Fig. 5C). There was no tumor shrinkage when we combined 30D8 with B20.4.1.1. CD11b⁺Gr1⁺, F4/80⁺, and CD31⁺ cells inside the tumor were examined at week 4, 24 hours after the last dosing. Treatment with hu30D8 at 30 mg/kg led to average 45%, 31%, and 36% decreases in tumor F4/80⁺, CD11b⁺Gr1⁺, and CD31⁺ cells, respectively, as assessed by fluorescence-activated cell sorting (FACS) analysis (P < 0.05; Fig. 5D).

The hu30D8 paratope involves all six complementarity-determining regions (CDR; Fig. 4B). The specific interactions at the two Fab-CXCL12α interfaces are essentially identical. Remarkably, each interface is composed of residues from only one CXCL12α monomer, suggesting that the antibody can bind to the monomeric form of CXCL12α as well. Figure 4C depicts the specific interactions at Asn⁴⁴ and Asn⁴⁵ of CXCL12α that occupy a groove generated by the three heavy-chain CDRs of hu30D8, and engage polar interactions with CDR-H3. The network of interactions concentrated in this region seems to contribute to the majority of binding free energy, therefore constitutes a binding “hot spot”. The presence of such a hot spot correlates well with the mutagenesis results, where alanine substitution at Asn⁴⁴ causes drastic reduction in binding, whereas mutations at other sites had moderate to no effects.

The crystal structure revealed key interactions contributing to hu30D8′s inhibitory activity. The RFFESH fragment near N-terminus of CXCL12α is known to be responsible for receptor binding (42). As shown in Fig. 4D, residues Glu¹⁵ and His¹⁷ interact with hu30D8 through multiple hydrogen bonds. Antibody binding is likely to interfere with CXCL12α activity by blocking RFFESH binding to CXCR4, and likely to CXCR7 as well, because both receptors share similar ligand-binding surfaces for the binding of the synthetic ligands (43).

Pharmacokinetic characteristics of 30D8 were first studied in mice and subsequently in rats and cynomolgus monkeys (Supplementary Methods). Fast clearances were observed following a single dose of 30 mg/kg in Balb c/nude mice for both 30D8 (intraperitoneally; Fig. 6A) and hu30D8 (intravenously; Fig. 6E) and in cynomolgus monkeys for hu30D8 (intravenously; Fig. 6C). A significant decrease in serum CXCL12α was observed 3 hours after ch30D8 administration in NCr nude mice. This correlated inversely with the serum 30D8 levels (Fig. 6B). Similarly, serum CXCL12α level was undetectable in HM7-implanted nude mice 3 hours after 30D8 treatment, whereas no significant change was observed after AMD3100 or the control antibody. In the same study, serum VEGF was not affected significantly by any of the treatments (Supplementary Fig. S5A). Furthermore, a pharmacokinetic study was conducted using multiple dosing regimens as the animals received 30D8, three times/week, in all in vivo models. When 30D8 was given intraperitoneally for 24 consecutive days at 30 mg/kg, three times/week in HM7-implanted nude mice, no significant accumulation of the antibody was observed (Supplementary Fig. S5B).

In contrast, hu30D8 had a normal clearance in nude rats at the same dose level (intravenously; Fig. 6D). K_D for hu30D8 against rat CXCL12α was 154 nmol/L, 80-fold higher than CXCL12α from mouse/human CXCL12α measured by Biolayer Interferometry. Lower affinity was expected as rat CXCL12α differs from mouse/human CXCL12α at amino acids Asn⁴⁴ and Leu⁶³, with the former being the most critical residue for hu30D8 binding.

To investigate whether epitope contributed to fast clearance, several antibodies raised in hamsters that target different binding epitopes in CXCL12α were tested. Although epitopes were different, all these antibodies had similar binding affinity toward CXCL12α and showed similar clearance as 30D8 (Supplementary Fig. S6).

Finally, we generated a series of single point mutations in the heavy chain of hu30D8 around amino acid Asp⁹⁵, Gln⁹⁶, or Tyr¹⁰⁰ (Supplementary Methods). Among them, one mutation, hu30D8^D95A, which removes a key interaction with Asn ⁴⁴ of CXCL12α, showed 1,000-fold lower affinity to human CXCL12α and much slower clearance in vivo than parental hu30D8 (Fig. 6E). Other mutants such as hu30D8^Y100A had similar affinity toward CXCL12α and similar pharmacokinetic profile as the parental antibody.

The CXCL12/CXCR4 signaling axis has been actively explored as a potential drug target (44). So far, most of the studies focused on the CXCL12 receptor CXCR4 and on small-molecule inhibitors targeting CXCR4 (45), although efforts using neutralizing anti-CXCL12 antibodies have been also reported (46, 47).

Here, we show efficacy of the anti-CXCL12 antibody 30D8 in several tumors as well as inflammatory models as a single agent, despite a relatively fast clearance. In all models tested, the efficacious doses of 30D8 were higher than those of anti-VEGF mAb B20.4.1.1, but in the same range as those of DC101, a widely used mAb directed against VEGFR2, which is typically administered at 40 mg/kg every 3 days in mouse models (27).

mRNA levels of CXCL12α and its receptor in human tumor cells were not useful to predict antitumor activity of hu30D8, possibly because they were neither representative of the levels of protein on the cell surface nor indicative of the levels of proteins with specific functions. In fact, a number of CXCR4 isoforms, mutants with varying levels of glycosylation, have been found on the cell surface, with various functions (48, 49).

Humanization of 30D8 allowed us to investigate pharmacokinetics in cynomolgus monkeys. Such antibodies will also enable safety studies to assess the consequences of long-term CXCL12 blockade. hu30D8 bound to circulating CXCL12α with very high affinity and subsequently blocked CXCL12α binding (free or heparin-bound forms) to both CXCR4 and CXCR7. The contribution of CXCR7 in tumor growth/metastasis in each model could not be addressed in the current study, due to lack of specific inhibitors. AMD3100 was less efficacious in some of our in vivo models, which might be due to its activity as a CXCR7 ligand with allosteric agonist properties (50).

When 30D8 was used in combination with a TNFRII-IgG2a fusion protein, efficacy could be further improved, suggesting that CXCL12 could become a novel target at least in subsets of arthritis patients who fail to respond to anti-TNF-α therapy (28). In the CIA and CNV models, blocking CXCL12 function prevented macrophage and/or lymphocyte infiltration at an early phase, although the current study could not rule out the possibility that 30D8 inhibited angiogenesis by directly inhibiting endothelial cell migration and survival as well.

Preliminary data indicate a greater efficacy of hu30D8 in syngeneic immunocompetent mice compared with immunodeficient mice bearing EL4 tumor (data not shown). If CD8⁺ T lymphocyte–mediated immune responses were important for antitumor activity, then immunodeficient mice would not be expected to fully reveal the efficacy of hu30D8.

In the present study, we observed fast clearance of 30D8 in mice and cynomolgus monkeys. On the basis of data using hu30D8 mutants, fast clearance seemed to correlate well with affinity of the antibody toward CXCL12α. A preliminary screen using a library of 1,000 human secreted proteins indicated that hu30D8 does not nonspecifically bind to other secreted proteins (Unpublished Data), arguing against the possibility that nonspecific binding is a reason for the fast clearance. Likewise, hu30D8 was classified as “low risk” for fast clearance in an assay of nonspecific binding shown to be useful for identifying antibodies likely to show fast clearance in cynomolgus monkeys (51). Notably, the crystal structure of hu30D8/human CXCL12α complex revealed that hu30D8 only partially blocks heparin-binding sites on CXCL12α, raising the possibility that target-mediated clearance could be related to heparin binding by CXCL12α. Therefore, balancing pharmacokinetic properties and binding affinity seems to be a requirement for optimal efficacy of the antibody. Further studies are needed to unravel the mechanism(s) of fast clearance.

In conclusion, hu30D8 is a suitable tool to test the hypothesis that targeting CXCL12 is a valid strategy to treat cancer and inflammatory diseases in humans. The availability of predictive biomarkers to select patients most likely responsive to anti-CXCL12 therapy would be a major step toward the initiation of such human studies.

J. Wang is an employee and shareholder of Genentech/Roche. H. Xiang is employed as Scientist in Roche and has ownership interest (including patents) in the same. L.A. Damico-Beyer is employed as Senior Scientist in Genentech and has ownership interest (including patents) in Roche. R.D. Carano and R.F. Kelley have ownership interest (including patents) in Roche. N. Ferrara is a former employee of Genentech/Roche. No potential conflicts of interest were disclosed by the other authors.

Conception and design: C. Zhong, J. Wang, B. Li, H. Xiang, R. Corpuz, R. Takkar, W.P. Lee, R.F. Kelley, N. Ferrara

Development of methodology: C. Zhong, B. Li, H. Xiang, R. Corpuz, J. Yao, R. Takkar, W.P. Lee, L.A. Damico-Beyer, R.D. Carano, C. Adams, W. Wang

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): C. Zhong, M. Ultsch, M. Coons, T. Wong, S. Clark, R. Clark, L. Quintana, P. Gribling, E. Suto, J. Yao, W.P. Lee, W. Wang

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): C. Zhong, B. Li, H. Xiang, L. Quintana, K. Barck, J. Yao, R. Takkar, W.P. Lee, L.A. Damico-Beyer, R.D. Carano, R.F. Kelley, W. Wang

Writing, review, and/or revision of the manuscript: C. Zhong, J. Wang, B. Li, H. Xiang, M. Ultsch, T. Wong, S. Clark, R. Corpuz, L.A. Damico-Beyer, R.F. Kelley, W. Wang, N. Ferrara

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): C. Zhong, B. Li, H. Xiang, M. Ultsch, T. Wong, J. Yao, W. Wang

Study supervision: C. Zhong, H. Xiang, W.P. Lee, N. Ferrara

Cloning the variable heavy and light chain sequence from the hybridoma cell line using 5′ RACE: N.Y. Chiang

Confirmation of the correct sequence cloned and reformatting into expression vectors for mammalian cells: N.Y. Chiang

The authors thank A. Bruce for graphic artwork and I. Hotzel for sequence alignment; M. Kowanetz for help with the 4T1 model; and M. Dennis and X. Qu for reading the article. The authors also thank several people who provided experimental support or participated in scientific discussions: S. William, G. Zhuang, J. Marik, D. Kallop, Y. Xia, G. Meng, W.L. Wong, K. Katschke, F. Peale, P. Luan, C. Quan, J. Lee, Y. Gan, C. Brown, C. Spiess, L. Gonzales Jr, S. Ranjani Ramani, A. Chuntharapai, R. Vandlen, P. Hass, M. Nagel, M. Schweiger, J. Zavala-Solorio, F. Chu, R. Liu, S. Krycia. We acknowledge the use of synchrotron X-ray sources at Stanford Synchrotron Radiation Lightsource, supported by the Department of Energy's Office of Science and Office of Biological and Environmental Research, and by the National Institutes of Health. The crystal structure of hu30D8 Fab/CXCL12a complex has been deposited into Protein Data Bank, accession code 4LMQ.

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.