Aglycosylated Igg Variants Expressed in Bacteria That Selectively Bind Fcγri Potentiate Tumor Cell Killing by Monocyte-Dendritic Cells

Aglycosylated IgG variants expressed in bacteria that selectively bind FcγRI potentiate tumor cell killing by monocyte-dendritic cells

Sang Taek Junga, Sai T. Reddya, Tae Hyun Kangb, M. Jack Borroka, Inger Sandliec, Philip W. Tuckerb,d, and George Georgioua,b,d,e,1

aDepartment of Chemical Engineering, bInstitute for Cellular and Molecular Biology, dSection of Molecular Genetics and Microbiology, and eDepartment of Biomedical Engineering, University of Texas, Austin, TX 78712; and cDepartment of Molecular Biosciences and Center for Immune Regulation, University of Oslo, 0316 Oslo, Norway

Edited by Frances H Arnold, California Institute of Technology, Pasadena, CA, and approved November 12, 2009 (received for review July 31, 2009)

The N-linked glycan of immunoglobulin G (IgG) is indispensable The effector FcγRs interact with the upper CH2 and hinge re- for the interaction of the Fc domain with Fcγ receptors on effector gion of the IgG-Fc domain (Fig. 1A) in a manner that is critically cells and the clearance of target cells via antibody dependent dependent on the presence and composition of the single cell-mediated cytotoxicity (ADCC). Escherichia coli expressed, N-linked glycan attached to N297. In contrast, FcRn, which binds aglycosylated Fc domains bind effector FcγRs poorly and cannot to the interface of CH2 and CH3 (5), is not significantly affected elicit ADCC. Using a novel bacterial display/flow cytometric library by the glycosylation status of the antibody constant region. The screening system we isolated Fc variants that bind to FcγRI (CD64) glycosylation state of IgG determines its affinity for effector with nanomolar affinity. Binding was critically dependent on ami- FcγRs and thus, modulates the activation of cytotoxic immune no acid substitutions (E382V, and to a lesser extent, M428I) distal to cells and their ability to participate in antibody-dependent cell- the putative FcγRI binding epitope within the CH3 domain. These mediated cytotoxicity (ADCC) (6, 7). ADCC is important for mutations did not adversely affect its pH-dependent interaction the clinical efficacy and outcomes of many important antibody with FcRn in vitro nor its serum persistence in vivo. Remarkably, therapeutics (8). Glycan composition and glycoform heterogene- the anti-Her2 IgG trastuzumab containing the E382V, M428I substi- ity, which is intrinsic to the expression of antibodies in mamma- tutions and expressed in E. coli exhibited highly selective binding lian cells, has a well-documented effect on the response to widely to FcγRI but not to the other activating receptors (FcγRIIa, FcγRIIIa) used therapeutics such as Rituxan (rituximab, anti-CD20) and nor to the inhibitory receptor, FcγRIIb. In contrast, the glycosylated Herceptin (trastuzumab, anti-Her2) (8, 9). This has led to intense version of trastuzumab (E382V, M428I) purified from HEK293T cells efforts to generate improved antibodies via glycoengineering (7). bound to all Fcγ receptors in a manner similar to that of clinical Wild-type aglycosylated antibodies can be expressed at a high grade trastuzumab. E. coli-purified trastuzumab (E382V, M428I), yield in bacterial cells and exhibit normal serum persistence but not glycosylated trastuzumab (E382V, M428I) or clinical grade and binding to FcRn (10). However, their lack of glycosylation trastuzumab, was capable of potentiating the killing of Her2 renders them completely unable to bind to FcγRs, and therefore overexpressing tumor cells with dendritic cells (DCs) as effectors. they do not induce FcγR-mediated effector functions (9). Re- These results indicate that aglycosylated IgGs can be engineered cently, structure-based mutagenesis and, in a separate study to display unique FcγR selectivity profiles that, in turn, mediate screening of small saturation libraries by yeast display, were used ADCC via mechanisms that are not normally displayed by glycosyl- to isolate aglycosylated IgG1 variants lacking N297 and carrying ated monoclonal antibodies. mutations in residues S298 and T299 that conferred binding to FcγRIIIa or FcγRIIa, respectively (11, 12). However, the respec- antibody engineering ∣ bacterial display ∣ bacterial expression ∣ tive antibody variants, following expression in HEK293 cells, directed evolution ∣ effector function exhibited significant binding to the inhibitory FcγRIIb receptor (12). This result was not surprising given the high degree of γ γ c receptors (FcRs) play a crucial role in linking the adaptive homology between Fc RIIa and Fc RIIb. Fimmune system to the effector functions of innate cells such as We have developed a system that relies on bacterial expression antigen presentation and antibody homeostasis (1). Human and flow cytometric screening of very large libraries of Fc mutants immune cells express six immunoglobulin (IgG) binding FcγRs for the isolation of variants that display a high degree of selec- B (FcγRI, FcγRIIa, FcγRIIb, FcγRIIc, FcγRIIIa, FcγRIIIb) that tivity and receptor affinity (Fig. 1 ). In this study, we focused γ enable various functions, including phagocytosis, cytokine pro- on the isolation of aglycosylated IgG that bound to Fc RI, as duction, superoxide production, release of serotonin, and inhibi- its role in ADCC is far less understood than that of other acti- tion of B cell activation (1, 2). An additional IgG binding vating receptors (1, 2). It has been suggested that because of γ receptor, the neonatal Fc receptor (FcRn), plays a critical role its high affinity, Fc RI on effector cells is occupied by circulating in IgG homeostasis by mediating antibody recycling via intracel- IgG and thus is involved in ADCC only during the early stage of lular trafficking, thus preventing degradation in lysosomes (3). the immune response when antibody concentration is low (13). With the exception of FcγRIIIb, human FcγRs are type I trans- However, the avidity associated with the formation of immune γ membrane glycoproteins. Activating FcγRs such as FcγRI complexes must be capable of displacing soluble IgG from Fc RI (CD64), FcγRIIa (CD32a), and FcγRIIIa (CD16a) have an immunoreceptor tyrosine-based activation motif in their cyto- Author contributions: S.T.J. and G.G. designed research; S.T.J., S.T.R., T.H.K., and M.J.B. plasmic tail and trigger activation of immune responses. The inhib- performed research; I.S. contributed new reagents and analytic tools; S.T.J., S.T.R., itory receptor FcγRIIb (CD32b) processes an immunoreceptor T.H.K., M.J.B., P.W.T., and G.G. analyzed data; and S.T.J. and G.G. wrote the paper. tyrosine-based inhibitory motif in its cytoplasmic domain that The authors declare no conflict of interest. downregulates immune responses (4). The affinity of FcγRII This article is a PNAS Direct Submission. −6 and FcγRIII to monomeric IgG is very low (Kd ¼ ∼10 − 1To whom correspondence should be addressed. E-mail: [email protected]. 10−7 γ M), whereas Fc RI exhibits high affinity to monomeric This article contains supporting information online at www.pnas.org/cgi/content/full/ −8 −9 IgG (Kd ¼ ∼10 − 10 M). 0908590107/DCSupplemental.

604–609 ∣ PNAS ∣ January 12, 2010 ∣ vol. 107 ∣ no. 2 www.pnas.org/cgi/doi/10.1073/pnas.0908590107 Downloaded by guest on September 24, 2021 Fig. 1. Isolation of high-affinity aglycosylated IgG1-Fc to FcγR. (A) Binding epitopes of Fc receptors and C1q on IgG-Fc (8). (B) Schematic showing the display system for library screening. (C) Effect of trehalose on the periplasmic display of Fc. Cells expressing either IgG-Fc domain or the M18 scFv as a control were grown with or without 0.5 M trehalose, spheroplasted and labeled with 2.5 μg∕ml of Protein A-FITC for detection. (D) Fluorescence histogram showing spheroplasted cells expressing either wt IgG-Fc or highly fluorescent variants labeled with 30 nM of FcγRI-FITC (P < 0.001 using unpaired Student’s t-test relative to wt IgG-Fc) isolated from the fourth round. M: Mean fluorescence intensity.

and consistent with this hypothesis there is increasing evidence protein retention following spheroplasting, even in the presence of that the activation of this receptor plays an important role in 0.5 M trehalose (Fig. S1C). the resolution of certain infections and in cancer (14–16). We report here the engineering of aglycosylated IgG anti- High Throughput Screening of Aglycosylated Fc Libraries for bodies that contain amino acid substitutions in the CH3 domain Increased FcγRI Binding. The retention of ligand-accessible Fc in that are distant from the FcγR-binding epitope located in the spheroplasted cells was exploited for the flow cytometric screen- CH2 domain yet confer low nanomolar affinity to FcγRI but ing of large libraries for FcγR binding. We initially constructed not to the inhibitory FcγRIIb nor to other effector FcγRs. libraries composed of (i) random mutants by error-prone PCR Engineered, bacterially expressed trastuzumab IgGs potentiated (∼9.2 × 108 transformants) containing 0.5% nt substitutions the lysis of Her2 overexpressing SkBr3 breast cancer cells by per gene (as determined by the sequencing of 20 random effector monocyte-derived dendritic cells (mDCs). Strikingly, clones) and (ii) 10-a.a. random insertions encoded by NNS glycosylated IgGs, clinical grade trastuzumab (Herceptin) or randomization scheme (where N ¼ equimolar A; T; C; G; mammalian cell-expressed trastuzumab that contained the same S ¼ equimolar C; G) (∼2.8 × 107 transformants representing a CH3 domain mutations failed to activate the cytotoxic ability of miniscule fraction of the theoretical diversity) between N297 mDCs. Our results indicate that aglycosylated IgGs engineered to and S298 with the intent of replacing the glycan that is normally display unique FcγR-binding characteristics may be useful both attached to N297 with a peptide epitope. Transformants from the for elucidating the function of FcγRI and for the selective recruit- two libraries were pooled and, following induction of protein ment of mDCs in cancer therapy. synthesis, spheroplasts were prepared and labeled with the appro- γ B

priate Fc R conjugated to FITC (Fig. 1 ). Highly fluorescent SCIENCES Results clones were isolated by FACS. The respective genes were rescued

Bacterial Display of Aglycosylated Fc Domains. Simmons (10) pre- by PCR, subcloned, and transformants were subjected to addi- APPLIED BIOLOGICAL viously reported that aglycosylated, correctly folded IgG1 anti- tional rounds of FACS sorting. Positive populations were enriched bodies purified from E. coli exhibit negligible binding to the following iterative screening with fluorescently labeled extracellu- extracellular domain of FcγRI or C1q. Toidentify amino acid sub- lar domain of FcγRI (CD64) (Fig. S1D). After the fourth round, stitutions that enable aglycosylated IgG to bind to the extracel- 50 clones were selected at random and their individual fluores- lular domain of FcγRs, it was first necessary to develop an cence was determined. The six clones displaying the highest fluo- appropriate system for E. coli library screening. The e-clonal rescence (>5x above background, Fig. 1D) upon labeling with technology we developed previously (17) is well suited for the iso- FcγRI-FITC were isolated and sequenced (Fig. S1E). All six clones lation of IgGs with desired antigen specificities by fluorescence- were derived from the error-prone PCR Fc library (Fig. S1E) and activated cell sorting (FACS). However, in the e-clonal system, contained two consensus amino acid substitutions, E382V and the Fc domain of IgG secreted into the periplasm is captured M428I that fall within the CH3 C and F β-strands (19) (Fig. 2A by an inner-membrane anchored ZZ domain from Protein A and B), respectively. This finding was surprising because these and, therefore, is not suitable for engineering Fc domains. We mutations are distal from the FcγRI-binding epitope that is noticed that, under certain conditions, soluble Fc secreted into purportedly comprised of the lower hinge and the upper part of the periplasm via a posttranslational signal peptide (such as the CH2 domain (20). The clone exhibiting the highest fluores- the widely used PelB leader) (Fig. S1A) is retained on the surface cence, Fc5, contained only those two substitutions. of spheroplasted cells following treatment with Tris-EDTA- Aglycosylated (wt) Fc and the mutants exhibiting the highest lysozyme (Fig. 1B). Thus, spheroplasts grown in terrific broth FcγRI binding via FACS (Fc5, E382V and M428I; Fc11, E382V; media and expressing secreted, soluble Fc exhibited 2-fold higher and Fc49, C229R, E382V and M428I) were expressed and puri- fluorescence upon labeling with Protein A-fluorecein isothiocya- fied to near homogeneity. Optimal soluble expression of nate (-FITC) relative to spheroplasts expressing an unrelated homodimeric Fc fragments was obtained at 30 °C in E. coli scFv (Fig. 1C). The retention of secreted Fc by spheroplasted Jude-1 cells where secretion into the periplasm was mediated cells was highly dependent on growth conditions. Optimal labeling by the cotranslational ssDsbA signal peptide (18) (Fig. S2A); this with Protein A-FITC was obtained with cells grown at 25 ºC in the led to a substantially higher amount of correctly assembled Fc presence of 0.5 M trehalose but not with other nonmetabolizable both in the periplasmic fraction and in the growth medium com- sugars such as sucrose or sorbitol. Under these conditions sphe- pared to expression with the PelB signal peptide (Fig. S2B). Fol- roplasts expressing the Fc and labeled with Protein A-FITC exhib- lowing protein A and gel filtration chromatography, ∼0.8 mg∕L ited 10-fold higher signal than controls (Fig. 1C). Western blot of dimeric Fc protein (Fig. 2C) was routinely obtained in analysis revealed that a significant fraction of the Fc protein shake flask cultures (Fig. S2C). The affinity of the purified was retained in spheroplasted cells after growth with 0.5 M treha- homodimeric Fc fragments for FcγRI measured by ELISA lose (Fig. S1B). Secretion of Fc by the signal recognition particle (Fig. 2D) correlated with the FACS analysis of Fc fragments dis- dependent DsbA signal sequence (18) did not result in significant played on spheroplasts (Fig. 1D). Surface plasmon resonance

Jung et al. PNAS ∣ January 12, 2010 ∣ vol. 107 ∣ no. 2 ∣ 605 Downloaded by guest on September 24, 2021 specific for Her2/neu (ErbB2) that is overexpressed in ∼30% of breast carcinomas (8). Extensive evidence indicates that recruit- ment of innate immune cells via interactions with Fcγ receptors plays an important role in the therapeutic action of Herceptin (8). For preparative production of aglycosylated trastuzumab (AglycoT), aglycosylated trastuzumab-Fc5 (AglycoT-Fc5) and aglycosylated trastuzumab-Fc601 (AglycoT-Fc601) in E. coli, heavy and light chains were fused to the PelB signal peptide and placed downstream from the lac promoter in a dicistronic operon (Fig. S5A). E. coli BL21(DE3) cells were grown at 30 ° CtoanOD600 of ∼100 by fed-batch fermentation in pH-stat control mode, protein synthesis was induced, the fermentor was cooled to 25 °C, and cells were harvested 7 hr later at an OD600 of ∼130–140 (Fig. S5B). Under these conditions, the yield of aglycosylated tetrameric IgG1 was 40–50 mg∕L with >70% retained in the periplasmic space. IgG1 protein was released from the cell pellet by incubation with lysozyme and EDTA for 16 hr at 30 °C and was purified to >80% homogeneity by protein A and size exclusion chromatography (Fig. 3A and Fig. S5C). SPR analysis (Fig. 3B and Fig. S5D–G) revealed that the E. coli expressed AglycoT-Fc5 and AglycoT-Fc601 IgG1s showed dramatically improved affinity toward FcγRI relative to AglycoT. For example, AglycoT-Fc601 bound to FcγRI with an equilibrium dissociation K constant ( D) nearly identical to that of commercial-grade glycosylated trastuzumab (Herceptin) and >130 fold greater than wild-type bacterially expressed AglycoT. Both AglycoT and γ Fig. 2. Characterization of isolated aglycosylated Fc variants. (A) Amino acid AglycoT-Fc5 displayed higher affinity for Fc RI than the isolated substitutions in Fc5 (a.a. 382 and 428) represented on the 3D structure of Fc domains. Notably and unlike Herceptin, which binds to all glycosylated IgG1 Fc (PBD Code: 1FC1). (B) Two beta sheets including 382E effector human Fcγ receptors with physiologically relevant in β-sheet C and 428M in β-sheet F of CH3 domain represented on the crystal affinities, AglycoT-Fc5, and AglycoT-Fc601 (as well as AglycoT) structure of wild-type full glycosylated IgG. (C) Nonreducing SDS-PAGE show- exhibited poor binding to FcγRIIa, FcγRIIb (EC50 ≥ 1000-fold ing the purified wt Fc and engineered Fc variants. Lane 1: M.W. standards; and 100-fold higher for GST fused FcγRIIa and FcγRIIb, respec- Lane 2: wt Fc; Lane 3: Fc5; Lane 4: Fc11; Lane 5: Fc49. (D) Binding of wt Fc, Fc5, tively, Fig. 3C and D) and for FcγRIIIa (Fig. 3E). Fc11, and Fc49 to FcγRI extracellular domain detected by ELISA. Errors bars indicate the standard deviations from duplicate run on the same sample. Because the mutations that impart the specific interaction of AglycoT-Fc5 to FcγRI are distal to the biochemically defined binding epitope that encompasses the glycosylated N297 residue, (SPR) analysis revealed that wild-type aglycosylated Fc does not we sought to examine whether glycosylation might have an effect γ K > 30 μ bind to Fc RI ( D M), whereas Fc5 and Fc49 exhibited on binding affinity and specificity. Glycosylated antibodies K high affinities with D values of 31 and 92 nM, respec- against Her2/neu were produced in HEK293T cells, and served k ¼ 4 5 × 104 −1 −1 k ¼ 1 4 × 10−3 −1 tively ( on;Fc5 . M sec , off;Fc5 . sec ; as glycosylated counterparts to AglycoT-Fc5 and AglycoT, which k ¼ 2 5 × 104 −1 −1 k ¼ 2 3 × 10−3 −1 on;Fc49 . M sec , off;Fc49 . sec ). we labeled as glycosylated trastuzumab-Fc5 (GlycoT-Fc5) and We next performed additional saturation mutagenesis of resi- glycosylated trastuzumab (GlycoT) (Fig. S6A). The affinities of dues 380, 382, and 384 that lie in the F-contacting C β-sheet of GlycoT-Fc5 and GlycoT for the extracellular domains of effector CH3 and of residues 426, 428, and 430 within the F β-sheet. FcγRs were determined by ELISA. GlycoT-Fc5 showed almost We also carried out insertional mutagenesis between residues identical affinity for FcγRI as wild-type GlycoTand clinical grade 428–429 (Fig. S3A), and the three libraries were screened for Herceptin (trastuzumab produced in CHO rather than HEK293T FcγRI-FITC binding (Fig. S3B). Of 14 individual clones isolated cells) (Fig. 3F). Surprisingly, and in contrast to its aglycosylated from the fourth round of sorting, we again observed the consensus counterpart, GlycoT-Fc5 did not exhibit selectivity for FcγRI and mutations E382Vand M428I, further indicating that these substi- showed binding to FcγRI as well as FcγRIIa (Fig. S6B), FcγRIIb tutions, and particularly the introduction of V residue at position (Fig. S6C), and FcγRIIIa (Fig. S6D) similar to GlycoTand typical 382, are critical for binding to FcγRI (Fig. S3C). Subsequently, an of other IgG1 human antibodies. These results reveal that glyco- error-prone PCR library containing 0.3% nt per gene substitutions sylation at N297 has a dominant effect relative to that of E382V from Fc5 (as determined by the sequencing of 20 random and M428I mutations. clones) was constructed (∼7 × 108 transformants). Over 4 × 108 The mutations in AglycoT-Fc5 and AglycoT-Fc601 reside close spheroloplasts were labeled with FcγRI-FITC and sorted in the to the epitope recognized by the neonatal Fc receptor (FcRn) and first round. After the fifth round of FACS, we isolated 20 clones therefore might affect their ability to escape endosomal degrada- with higher fluorescence than Fc5 (Fig. S4A–C). These substitu- tion a consequence that could be detrimental for human thera- tions localized to one of three distinct regions (i) the upper CH2 peutic purposes (21). AglycoT-Fc5 showed pH-dependent region that is likely to directly contact FcγRI, (ii) the interdomain binding to human FcRn at pH 5.5 and pH 6.0 and significantly region that connects the Fand B β-sheets of CH2 and CH3 and (iii) reduced binding at pH 7.4 typical of human IgG, including the region of the CH3 (Fig. S4D). Fc601 showed the highest affin- Herceptin. On the other hand, AglycoT-Fc601 showed reduced ity and contained two additional mutations: K338R and G341Von affinity to FcRn binding at pH 6.0, indicating that the additional the G and A strands of CH2 and CH3, respectively (Fig. S4E). two substitutions in the CH2-CH3 interface of Fc601 interfere with binding to the human FcRn (Fig. 3G). Engineered, E. coli-Expressed Aglycosylated Trastuzumab Exhibits For a preliminary evaluation of the effect of the E382V and Highly Selective FcγRI Binding and pH-Dependent FcRn Binding. M428I mutations on IgG homeostasis, the serum persistences Herceptin (glycosylated trastuzumab produced in Chinese ham- of AglycoTand AglycoT-Fc5 in Balb/c mice were compared to that ster ovary (CHO) cells) is a highly effective therapeutic antibody of Herceptin. Antibodies were administered by intraperitoneal

606 ∣ www.pnas.org/cgi/doi/10.1073/pnas.0908590107 Jung et al. Downloaded by guest on September 24, 2021 Fig. 3. Expression, purification, FcγR binding of aglycosylated or glycosylated trastuzumab antibodies. (A) SDS-PAGE showing the purified aglycosylated trastuzumab (AglycoT) and aglycosylated trastuzumab-Fc5 (AglycoT-Fc5), Lane 1: nonreduced AglycoT; Lane 2: nonreduced AglycoT-Fc5, Lane 3: reduced AglycoT; Lane 4: reduced AglycoT-Fc5. (B) Kinetic rates and equilibrium dissociation constants of aglycosylated trastuzumabs determined by SPR analysis for binding to FcγRI. (C–G) ELISA assays for (C) binding of aglycosylated trastuzumabs to FcγRIIa, (D) binding of aglycosylated trastuzumabs to FcγRIIb, (E) binding of aglycosylated trastuzumab to FcγRIIIa, (F) binding of glycosylated trastuzumabs to FcγRI, and (G) pH-dependent binding of trastuzumabs to FcRn at pH 7.4 and at acidic pH 6.0. Error bars represent standard deviation for two replicates. (H) In vivo serum persistence of trastuzumabs. Balb/c mice were injected intraperitoneally with Herceptin, AglycoT, and AglycoT-Fc5 antibodies at a dosage of 2 μg∕g of mouse. Antibody concentrations present in serum were measured at 132 h postinjection by ELISA. Error bars indicate the standard deviation for the serum concentration of trastuzumab antibodies in Balb/c mice (n ¼ 5 to 7 for each group). **: P < 0.01;*:P < 0.05: unpaired Student’s t-test relative to AglycoT and Herceptin, respectively. SCIENCES injections at dosages of 2.0 mg∕kg and serum antibody concentra- (Fig. S7B). In contrast, AglycoT-Fc5 induced ∼40% (25∶1; APPLIED BIOLOGICAL tion was examined by ELISA 132 hr later. As shown in Fig. 3H, the E∶T ratio) and ∼70% (100∶1;E∶T ratio) tumor cell killing. A concentration of AglycoT-Fc5in the serum was 16.6 μg∕ml, a value comparable degree of cell lysis was observed with AglycoT- that is ∼80% of the injected dose and slightly higher than that of Fc601 (Fig. 4A). Incubation of AglycoT-Fc5 with Herceptin Herceptin (13.1 μg∕ml) and significantly higher than AglycoT markedly reduced ADCC. Similar to Herceptin and to GlycoT, (8.7 μg∕ml). This was presumably due to the documented favor- only background lysis of SkBr3 cells was observed with GlycoT- able effect of the M428I mutation on pH-dependent FcRn binding Fc5 that binds nonselectively to all effector FcγRs (Fig. S7B). In (22). Although the mouse FcRn binds to human IgG1 with higher contrast to these results and in agreement with earlier studies (28, affinity relative to its murine counterpart IgG2a (23), the mouse 29), when PBMCs were used as effector cells, Herceptin poten- B model provides useful information for predicting the rank-order of tiated efficient lysis of SkBr3 cells (Fig. 4 ). PBMCs are com- the t1∕2 of IgG variants in humans (24). posed of several cell types that express various activating FcγRs. Most notably, NK cells that express only FcγRIIIa (1), E. coli Expressed, Aglycosylated Trastuzumab Potentiates the Killing constitute the largest innate immune cell population in PBMCs. γ of Her-2 Overexpressing Cells by mDCs. Macrophages and immature Therefore, glycosylated Herceptin, which displays Fc RIIIa bind- E DCs greatly outnumber the classical killer cells (NK and T cells) ing (Fig. 3 ), was able to induce significant ADCC when PBMCs in tumors (25). In recent years, the cytotoxic properties of various were used as effector cells. In contrast, AglycoT-Fc5 and AglycoT- subpopulations of DCs toward cancer cells has attracted signifi- Fc601 were unable to induce statistically significant tumor cell γ lysis above background levels with PBMC cells, consistent with cant interest (26). Human circulating DCs express Fc RI, γ E FcγRIIa, and FcγRIIb, but not FcγRIIIa (27). the fact that they do not bind Fc RIIIa (Fig. 3 ). 11 þ Human CD c mDCs were differentiated from peripheral Discussion blood mononuclear cells (PBMCs) by incubation for 7 d in the We developed a facile system for the screening of large combi- presence of interleukin-4 (IL-4) and granulocyte macrophage natorial libraries of aglycosylated Fc domains expressed in E. coli colony stimulating factor. Those mDCs activated by lipopolysac- by FACS selection for binding to Fcγ receptors. In the absence of γ A charide (LPS) expressed high levels of Fc RI (Fig. S7 ). SkBr3 the single glycan at N297, IgG antibodies display low to negligible breast carcinoma epithelial cells overexpressing Her-2 were incu- binding to receptors on effector cells or to C1q and thus bated either with Herceptin or with AglycoT-Fc5 or AglycoT- cannot mediate ADCC and complement dependent cytotoxicity Fc601 at 10 μg∕ml. AglycoT and media without effector cells (CDC). In this report we engineered aglycosylated IgG1 mutants were used as negative controls. Activated mDCs were then added that selectively bind to FcγRI with a similar affinity to that of at either 25∶1 or 100∶1 ratios and the lysis of the SkBr3 target CHO-derived, glycosylated antibodies (Herceptin). cells was determined by monitoring the release of radioactive Two amino acid substitutions in CH3 conferred highly selective isotope chromium 51 (51Cr) 24 h later. No cell lysis above back- binding to FcγRI and not to other effector FcγRs. The high- ground was detected with Herceptin (Fig. 4A) or with GlycoT binding selectivity of aglycosylated IgGs containing the E382V

Jung et al. PNAS ∣ January 12, 2010 ∣ vol. 107 ∣ no. 2 ∣ 607 Downloaded by guest on September 24, 2021 FcγRI are located in the CH3, they do not adversely affect pH- dependent binding to human FcRn. In fact mutations in M428 have been shown to favorably affect FcRn affinity at pH 6.0 (22). AglycoT-Fc5 exhibited slightly higher serum persistence in mice relative to glycosylated Herceptin. Mutations that further increase the affinity of aglycosylated antibodies for FcγRI can interfere with FcRn binding as was the case of AglycoT-Fc601. However, this is not a general phenomenon and we have isolated other AglycoT-Fc5 variants that display increased affinity for FcγRI without affecting proper interactions with FcRn. FcγRI is mainly expressed on monocytes, macrophages, DCs, and at lower levels on neutrophils (31). DCs, which are profes- sional antigen presentation cells, are also able to mediate tumor cell killing directly by a variety of apoptotic and necrotic pathways (26) or indirectly by stimulating tumor cell specific cytotoxic T cells (32) and NK cells (33). We showed that mature, LPS- activated mDCs elicit potent cytotoxicity of Her2 overexpressing cells in the presence of AglycoT-Fc5 or AglycoT-Fc601. By contrast, Herceptin, and significantly glycosylated antibodies containing the E382V and M428I mutations were not able to stimulate the lysis of the target cells. Unlike AglycoT-Fc5 and AglycoT-Fc601, immune complexes formed by Herceptin can trigger signaling via the inhibitory FcγRIIb. Earlier studies have shown that selective engagement of FcγRIIa or FcγRIIb mediate opposing effects on mDC maturation and function (34). Addi- tionally, antitumor effect in mice was demonstrated by abrogation of inhibitory FcγRIIb signaling on DCs (35). Thus, the most straightforward explanation for our findings is that immune complexes formed by AglycoT-Fc5 or AglycoT-Fc601 selectively engage FcγRI to mediate cytotoxicity by mDCs, the interaction of Herceptin (and also GlycoTor GlycoT-Fc5) with FcγRIIb may be Fig. 4. Antibody-dependent cellular cytotoxicity (ADCC) toward SkBr3 with the reason for the attenuation of this function. With PBMCs that monocyte-derived dendritic cells (mDCs) as effector cells. (A) Target (T) Her2 are comprised by a large fraction of FcγRIIIa expressing NK cells overexpressing tumor cells, SkBr3, were incubated with either E. coli-purified but relatively low numbers of circulating immune cells expressing AglycoT, AglycoT-Fc5, AglycoT-Fc601, or Herceptin at 10 μg∕ml for 1 hr prior FcγRI, the situation is reversed and Herceptin exhibited strong to incubation with LPS-activated mDCs. Cell lysis was measured by 51Cr ADCC in contrast to AglycoT-Fc5 or AglycoT-Fc601. Activated release at 24 hr. (B) SkBr3 cells were incubated with IgGs as in (A), peripheral mDCs sensitized with the engineered aglycosylated antibodies blood mononuclear cells (PBMCs) were added 1 hr later and cell lysis was de- described above were found to also phagocytose the target cells. termined as above. Standard deviations calculated from triplicate samples are represented by error bars. **: P < 0.01;*:P < 0.05; unpaired Student’s It is possible that in addition to their cytotoxic activity the mDCs t-test relative to controls with AglycoT antibody and with no IgG, respec- cells may also be able to effectively present antigen and activate tively. tumor specific cytotoxic Tcells that in turn could further enhance tumor destruction. Future studies will explore if aglycosylated FcγRI-binding antibodies are able to enhance other relevant and M428I mutation was abolished when this antibody was pro- DC-mediated functions (i.e., antigen presentation and activation duced as a glycoprotein in HEK293T cells. We are not aware of of tumor specific T cells). any other instances where the presence of a naturally occurring Aglycosylated antibodies engineered for ADCC bypass the glycan in a protein abolishes ligand selectivity. The human FcγRs need for glycoengineering, mammalian expression, and problems for which crystal structures are available show that IgG binding related to glycan heterogeneity. Together with technologies for occurs at an epitope that comprises residues in the hinge and IgG isolation (17) and commercial level expression (10, 34), it CH2 regions. Biochemical data indicate that FcγRI also binds is feasible to carry out the complete development of therapeutic to the same region (30). The differences in the binding specificity antibodies in bacteria. The therapeutic utility of aglycosylated of the GlycoT-Fc5 and AglycoT-Fc5 and also the fact that the antibodies will further be determined by pharmacological par- E382V and M428I mutations are distal to the FcγRI binding epi- ameters, including stability, biodistribution, and immunogenicity tope suggest that these amino acid substitutions likely induce a in humans. conformational change in the protein. Conceivably, this confor- Materials and Methods mational change might cause the CH2 domain to slightly bulge (For additional materials and methods see SI Materials and Methods) out, thus mediating a transition from the “closed” form of IgG “ ” with truncated glycans to the open structure of fully glycosyl- Library Screening. For library screening, the extracellular domain of glycosylated IgG (13). This effect is not manifest in GlycoT-Fc5 in which ated FcγRI (R&D Systems) was labeled with FITC according to the manufac- the two heavy chain polypeptides assume the open conformation turer’s instruction (Invitrogen). Binding of FITC-labeled soluble FcγRI to as a result of the glycan appended to N297. Unfortunately, a human IgG-Fc was confirmed by fluorescent ELISA on plates coated with detailed structural interpretation is not possible at the moment, glycosylated human IgG-Fc (Bethyl Laboratories). Prepared spheroplasts were as neither the structure of aglycosylated IgGs nor that of FcγRI is sorted on a MoFlo droplet deflection flow cytometry (Dako Cytomation) using a 488 nm Argon laser for excitation and detection through a yet available. IgG homeostasis is regulated by pH-dependent 530∕40 band pass filter. Sorted cells were resorted immediately after initial binding to FcRn that recognizes an epitope comprised of a.a. sorting. The Fc mutant genes in the resorted spheroplasts were cloned into in the CH3 and the CH3-CH2 hinge (5). Although the two critical SfiI digested pPelBFLAG vector (Table S1) after PCR amplification using mutations for high affinity binding of aglycosylated antibodies to two specific primers STJ#16 and STJ#220 (Table S2), and transformed in

608 ∣ www.pnas.org/cgi/doi/10.1073/pnas.0908590107 Jung et al. Downloaded by guest on September 24, 2021 51 electrocompetent E. coli Jude-1 cells. The resulting transformants were sub- ADCC Assays. Target SkBr3 tumor cells were labeled with Na CrO4 (Perkin jected to the next round sorting and resorting. Elmer Life Sciences) at 100 uCi∕106 cells for 1 h at 37 °C. Cells were then washed twice with PBS and resuspended in RPMI and added to a 96-well Preparative Expression of Aglycosylated IgG. Preparative expression was per- plate at 104 cells∕well. Engineered Fc antibodies with trastuzumab and rele- formed by fed-batch fermentation using a 3.3 L jar fermentor (New Bruns- vant controls were added to the target cells in triplicate wells and incubated ∕2 wick Scientific) with 1.2 L working volume. Cells were grown at 30 °C in R at 37 °C for 1 h. The plate was then centrifuged at 2000 rpm for 1 min and ð Þ medium (36) consisting of: 2 g of NH4 2HPO4, 6.75 g of KH2PO4,0.93gof washed twice with PBS. Effector cells, either fully differentiated mDCs (day 7) citric acid-H2O, 0.34 g of MgSO4, 20 g of glucose, 0.05 g of ampicillin and 5 ml or freshly isolated PBMCs, were resuspended in RPMI 1640, 2% low IgG FBS of trace metal solution dissolved in 2 N HCl (10 g of FeSO4-7H2O, 2.25 g (Invitrogen), LPS (250 ng∕106 cells) and added to the wells at various ratios. ZnSO4-7H2O, 1 g of CuSO4-5H2O, 0.35 g of MnSO4-H2O, 0.23 g of Target cells and mDCs were incubated at 37 °C for 24 h. The isotope levels Na2B4O7-10H2O, 1.5 g of CaCl2, and 0.1 g of ðNH4Þ6Mo7O24 per L). E. coli BL21(DE3) (EMD Chemicals) harboring pSTJ4-AglycoT, pSTJ4-AglycoT-Fc5, present in cell media were then measured in a liquid scintillation counter. or pSTJ4-AglycoT-Fc601 were cultured in 500 ml baffled-flask with 120 m∕l Incubation of target cells with 4% SDS was used as a positive control for R∕2 media at 30 °C at 250 rpm for 8 h and used to inoculate the fermentor. maximum lysis and incubation with no effector cells was used as background The dissolved oxygen concentration was maintained at 40% of the air satura- lysis. tion using automatic cascaded control by increasing agitation speed from – – 100 1000 rpm, air flow rate from 1 3 SLPM (standard liquid per minute) ACKNOWLEDGMENTS. We thank Chhaya Das for her invaluable technical and pure oxygen flow rate from 0–1.5 SLPM when required. Initial pH assistance with ADCC assays; Dr. Yong Tang, Dr. Louis M. Weiner, Dr. Gregory was adjusted to 6.8 and controlled by the addition of 30% (v/v) ammonium P. Adams, and Calvin C Shaller for their helpful suggestions and assistance hydroxide when pH decreased to <6.75, and by the supply of either growth with preliminary experiments; Jamye O’Neal for her valuable assistance with feeding solution (700 g∕L of glucose and 10 g∕L of MgSO47H2O) before animal experiments; Dr. Michael Rosenblum, MD (Anderson) for his gift of induction or production feeding solution (500 g∕L glucose, 10 g∕Lof glycosylated trastuzumab (Herceptin); and Dr. Sally Ward (UT Southwestern) MgSO47H2O, and 100 g∕L of yeast extract) after induction when the pH for advice on PK experiments in mice. We also thank Dr. Ki Jun Jeong and Dr. increased above 6.9. When OD600 reached 100, the culture temperature Yariv Mazor for helpful discussions. This work was supported by grants from was reduced to 25 °C and 30 min later, protein expression was induced with the Clayton Foundation and by the Advanced Research Program of the State 1 mM of IPTG. The culture broth was harvested 7 h after induction. of Texas.

1. Nimmerjahn F, Ravetch JV (2008) Fcγ receptors as regulators of immune responses. 20. Jefferis R, Lund J, Pound JD (1998) IgG-Fc-mediated effector functions: molecular de- Nat Rev Immunol, 8(1):34–47. finition of interaction sites for effector ligands and the role of glycosylation. Immunol 2. Cohen-Solal JF, Cassard L, Fridman WH, Sautes-Fridman C (2004) Fc gamma receptors. Rev, 163(1):59–76. – Immunol Lett, 92(3):199 205. 21. Martin WL, West AP, Gan L, Bjorkman PJ (2001) Crystal Structure at 2.8 A of an 3. Ghetie V, Ward ES (2000) Multiple roles for the major histocompatibility complex class FcRn/Heterodimeric Fc Complex: Mechanism of pH-Dependent Binding. Mol Cell, – I- related receptor FcRn. Annu Rev Immunol, 18(1):739 766. 7(4):867–877. 4. Ravetch JV, Bolland S (2001) IgG Fc receptors. Annu Rev Immunol, 19:275–290. 22. Hinton PR, et al. (2004) Engineered human IgG antibodies with longer serum half-lives 5. Roopenian DC, Akilesh S (2007) FcRn: the neonatal Fc receptor comes of age. Nat Rev in primates. J Biol Chem, 279(8):6213–6216. Immunol, 7(9):715–725. 23. Ober RJ, Radu CG, Ghetie V, Ward ES (2001) Differences in promiscuity for antibody- 6. Desjarlais JR, Lazar GA, Zhukovsky EA, Chu SY (2007) Optimizing engagement of the immune system by anti-tumor antibodies: an engineer's perspective. Drug Discov FcRn interactions across species: implications for therapeutic antibodies. Int Immunol, – Today, 12(21-22):898–910. 13(12):1551 1559. 7. Jefferis R (2009) Recombinant antibody therapeutics: the impact of glycosylation on 24. Dall'Acqua WF, et al. (2002) Increasing the Affinity of a Human IgG1 for the Neonatal SCIENCES mechanisms of action. Trends Pharmacol Sci, 30(7):356–362. Fc Receptor: Biological Consequences. J Immunol, 169(9):5171–5180. 8. Liu X-y, Pop LM, Vitetta ES (2008) Engineering therapeutic monoclonal antibodies. 25. Bonmort M, et al. (2008) Killer dendritic cells: IKDC and the others. Curr Opin Immunol, APPLIED BIOLOGICAL Immunol Rev, 222(1):9–27. 20(5):558–565. 9. Jefferis R (2005) Glycosylation of recombinant antibody therapeutics. Biotechnol Prog, 26. Wesa AK, Storkus WJ (2007) Killer dendritic cells: mechanisms of action and therapeu- 21(1):11–16. tic implications for cancer. Cell Death Differ, 15(1):51–57. 10. Simmons LC, et al. (2002) Expression of full-length immunoglobulins in Escherichia coli: 27. Fanger NA, Wardwell K, Shen L, Tedder TF, Guyre PM (1996) Type I (CD64) and type II rapid and efficient production of aglycosylated antibodies. J Immunol Methods, (CD32) Fc gamma receptor-mediated phagocytosis by human blood dendritic cells. – – 263(1 2):133 147. J Immunol, 157(2):541–548. 11. Lazar GA, et al. (2009) Optimized Fc variants and methods for their generation. US 28. Suzuki E, et al. (2007) A Nonfucosylated Anti-HER2 Antibody Augments Antibody- Patent 2009/0092599 A1. Dependent Cellular Cytotoxicity in Breast Cancer Patients. Clin Cancer Res, 13(6): 12. Sazinsky SL, et al. (2008) Aglycosylated immunoglobulin G1 variants productively 1875–1882. engage activating Fc receptors. Proc Natl Acad Sci USA, 105(51):20167–20172. 13. Krapp S, Mimura Y, Jefferis R, Huber R, Sondermann P (2003) Structural analysis of 29. Lazar GA, et al. (2006) Engineered antibody Fc variants with enhanced effector func- – human IgG-Fc glycoforms reveals a correlation between glycosylation and structural tion. Proc Natl Acad Sci USA, 103(11):4005 4010. integrity. J Mol Biol, 325(5):979–989. 30. Shields RL, et al. (2001) High resolution mapping of the binding site on human IgG1 for 14. Bevaart L, et al. (2006) Direct targeting of genetically modified tumour cells to FcγRI Fc gamma RI, Fc gamma RII, Fc gamma RIII, and FcRn and design of IgG1 variants with triggers potent tumour cytotoxicity. Brit J Haematol, 132(3):317–325. improved binding to the Fc gamma R. J Biol Chem, 276(9):6591–6604. 15. Pfefferkorn LC, Fanger MW (1989) Cross-linking of the high affinity Fc receptor for 31. Schmidt RE, Gessner JE (2005) Fc receptors and their interaction with complement in human immunoglobulin G1 triggers transient activation of NADPH oxidase activity. autoimmunity. Immunol Lett, 100(1):56–67. Continuous oxidase activation requires continuous de novo receptor cross-linking. 32. Ashley DM, et al. (1997) Bone marrow-generated dendritic cells pulsed with tumor – J Biol Chem, 264(24):14112 14120. extracts or tumor RNA induce antitumor immunity against central nervous system 16. Tillinger W, et al. (2009) Expression of the High-Affinity IgG Receptor FcRI (CD64) in tumors. J Exp Med, 186(7):1177–1182. Patients With Inflammatory Bowel Disease: A New Biomarker for Gastroenterologic 33. Schmitz M, et al. (2005) Tumoricidal potential of native blood dendritic cells: Diagnostics. Am J Gastroenterol, 104(1):102–109. direct tumor cell killing and activation of NK cell-mediated cytotoxicity. J Immunol, 17. Mazor Y, Blarcom TV, Mabry R, Iverson BL, Georgiou G (2007) Isolation of engineered, 174(7):4127–4134. full-length antibodies from libraries expressed in Escherichia coli. Nat Biotechnol, 34. Boruchov AM, et al. (2005) Activating and inhibitory IgG Fc receptors on human DCs 25(5):563–565. – 18. Schierle CF, et al. (2003) The DsbA signal sequence directs efficient, cotranslational mediate opposing functions. J Clin Invest, 115(10):2914 2923. export of passenger proteins to the Escherichia coli periplasm via the signal recogni- 35. Kalergis AM, Ravetch JV (2002) Inducing Tumor Immunity through the Selective γ tion particle pathway. J Bacteriol, 185(19):5706–5713. Engagement of Activating Fc Receptors on Dendritic Cells. J Exp Med, 195(12): 19. Lefranc M-P, et al. (2005) IMGT unique numbering for immunoglobulin and T cell 1653–1659. receptor constant domains and Ig superfamily C-like domains. Dev Comp Immunol, 36. Jeong KJ, Lee SY (2003) Enhanced production of recombinant proteins in Escherichia 29(3):185–203. coli by filamentation suppression. Appl Environ Microbiol, 69(2):1295–1298.

Jung et al. PNAS ∣ January 12, 2010 ∣ vol. 107 ∣ no. 2 ∣ 609 Downloaded by guest on September 24, 2021