Toward generalizable prediction of antibody thermostability using machine learning on sequence and structure features.

Over the last three decades, the appeal for monoclonal antibodies (mAbs) as therapeutics has been steadily increasing as evident with FDA's recent landmark approval of the 100th mAb. Unlike mAbs that bind to single targets, multispecific biologics (msAbs) have garnered particular interest owing to the advantage of engaging distinct targets. One important modular component of msAbs is the single-chain variable fragment (scFv). Despite the exquisite specificity and affinity of these scFv modules, their relatively poor thermostability often hampers their development as a potential therapeutic drug. In recent years, engineering antibody sequences to enhance their stability by mutations has gained considerable momentum. As experimental methods for antibody engineering are time-intensive, laborious and expensive, computational methods serve as a fast and inexpensive alternative to conventional routes. In this work, we show two machine learning approaches - one with pre-trained language models (PTLM) capturing functional effects of sequence variation, and second, a supervised convolutional neural network (CNN) trained with Rosetta energetic features - to better classify thermostable scFv variants from sequence. Both of these models are trained over temperature-specific data (TS50 measurements) derived from multiple libraries of scFv sequences. On out-of-distribution (refers to the fact that the out-of-distribution sequnes are blind to the algorithm) sequences, we show that a sufficiently simple CNN model performs better than general pre-trained language models trained on diverse protein sequences (average Spearman correlation coefficient, ρ, of 0.4 as opposed to 0.15). On the other hand, an antibody-specific language model performs comparatively better than the CNN model on the same task (ρ= 0.52). Further, we demonstrate that for an independent mAb with available thermal melting temperatures for 20 experimentally characterized thermostable mutations, these models trained on TS50 data could identify 18 residue positions and 5 identical amino-acid mutations showing remarkable generalizability. Our results suggest that such models can be broadly applicable for improving the biological characteristics of antibodies. Further, transferring such models for alternative physicochemical properties of scFvs can have potential applications in optimizing large-scale production and delivery of mAbs or bsAbs.

The PSMA-targeting Half-life Extended BiTE Therapy AMG 160 has Potent Antitumor Activity in Preclinical Models of Metastatic Castration-resistant Prostate Cancer.

PURPOSE: Metastatic castration-resistant prostate cancer (mCRPC) remains a disease with high unmet medical need, as most patients do not achieve durable response with available treatments. Prostate-specific membrane antigen (PSMA) is a compelling target for mCRPC. It is highly expressed by primary and metastatic prostate cancer cells, with increased expression after progression on androgen deprivation therapy. EXPERIMENTAL DESIGN: We developed AMG 160, a half-life extended, bispecific T-cell engager immuno-oncology therapy that binds PSMA on prostate cancer cells and cluster of differentiation 3 on T cells for treatment of mCRPC. AMG 160 was evaluated in vitro and in mCRPC xenograft models. AMG 160 tolerability was assessed in nonhuman primates (NHP). AMG 160 activity as monotherapy and in combination with a PSMA-imaging agent, novel hormonal therapy, and immune checkpoint blockade was evaluated. RESULTS: AMG 160 induces potent, specific killing of PSMA-expressing prostate cancer cell lines in vitro, with half-maximal lysis of 6-42 pmol/L. In vivo, AMG 160 administered weekly at 0.2 mg/kg engages T cells administered systemically and promotes regression of established 22Rv-1 mCRPC xenograft tumors. AMG 160 is compatible with the imaging agent gallium 68-labeled PSMA-11, and shows enhanced cytotoxic activity when combined with enzalutamide or an anti-programmed death-1 antibody. AMG 160 exhibits an extended half-life and has an acceptable safety profile in NHPs. CONCLUSIONS: The preclinical characterization of AMG 160 highlights its potent antitumor activity in vitro and in vivo, and its potential for use with known diagnostic or therapeutic agents in mCRPC. These data support the ongoing clinical evaluation of AMG 160 in patients with mCRPC.See related commentary by Kamat et al., p. 2675.

Regression of human prostate cancer xenografts in mice by AMG 212/BAY2010112, a novel PSMA/CD3-Bispecific BiTE antibody cross-reactive with non-human primate antigens.

For treatment of patients with prostate cancer (PCa), we developed a novel T cell-engaging (BiTE) antibody designated AMG 212 or BAY2010112 that is bispecific for prostate-specific membrane antigen (PSMA) and the CD3 epsilon subunit of the T cell receptor complex. AMG 212/BAY2010112 induced target cell-dependent activation and cytokine release of T cells, and efficiently redirected T cells for lysis of target cells. In addition to Chinese hamster ovary cells stably expressing human or cynomolgus monkey PSMA, T cells redirected by AMG 212/BAY2010112 also lysed human PCa cell lines VCaP, 22Rv1, MDA PCa 2b, C4-2, PC-3-huPSMA, and LnCaP at half maximal BiTE concentrations between 0.1 and 4 ng/mL (1.8-72 pmol/L). No lysis of PSMA-negative human PCa cell lines PC-3 and DU145 was observed. The subcutaneous (s.c.) formation of tumors from PC-3-huPSMA cells in NOD/SCID mice was significantly prevented by once daily intravenous (i.v.) injection of AMG 212/BAY2010112 at a dose level as low as 0.005 mg/kg/d. Rapid tumor shrinkage with complete remissions were observed in NOD/SCID mice bearing established s.c. 22Rv1 xenografts after repeated daily treatment with AMG 212/BAY2010112 by either the i.v. or s.c. route. Of note, 22Rv1 tumors were grown in the absence of human T cells followed by intraperitoneal injection of T cells 3 days before BiTE treatment. No effects on tumor growth were observed in the absence of human T cells or AMG 212/BAY2010112. On the basis of these preclinical results, AMG 212/BAY2010112 appears as a promising new BiTE antibody for the treatment of patients with PSMA-expressing PCa.

Side-by-side analysis of five clinically tested anti-EpCAM monoclonal antibodies.

BACKGROUND: Epithelial cell adhesion molecule (EpCAM) is frequently and highly expressed on human carcinomas. The emerging role of EpCAM as a signalling receptor and activator of the wnt pathway, and its expression on tumor-initiating cells, further add to its attractiveness as target for immunotherapy of cancer. Thus far, five conventional monoclonal IgG antibodies have been tested in cancer patients. These are murine IgG2a edrecolomab and its murine/human chimeric IgG1 antibody version, and humanized, human-engineered and fully human IgG1 antibodies 3622W94, ING-1, and adecatumumab (MT201), respectively. Here we compared all anti-EpCAM antibodies in an attempt to explain differences in clinical activity and safety. METHODS: We recombinantly produced all antibodies but murine edrecolomab and investigated them for binding affinity, EpCAM epitope recognition, ADCC and CDC, and inhibition of breast cancer cell proliferation. RESULTS: ING-1 and 3622W94 bound to EpCAM with much higher affinity than adecatumumab and edrecolomab. Edrecolomab, ING-1, and 3622W94 all recognized epitopes in the exon 2-encoded N-terminal domain of EpCAM, while adecatumumab recognized a more membrane proximal epitope encoded by exon 5. All antibodies induced lysis of EpCAM-expressing cancer cell lines by both ADCC and CDC with potencies that correlated with their binding affinities. The chimeric version of edrecolomab with a human Fcγ1 domain was much more potent in ADCC than the murine IgG2a version. Only adecatumumab showed a significant inhibition of MCF-7 breast cancer cell proliferation in the absence of complement and immune cells. CONCLUSION: A moderate binding affinity and recognition of a distinct domain of EpCAM may best explain why adecatumumab showed a larger therapeutic window in cancer patients than the two high-affinity IgG1 antibodies ING-1 and 3622W94, both of which caused acute pancreatitis.

T cell-engaging BiTE antibodies specific for EGFR potently eliminate KRAS- and BRAF-mutated colorectal cancer cells.

Epidermal growth factor receptor (EGFR)-specific monoclonal antibodies predominantly inhibit colorectal cancer (CRC) growth by interfering with receptor signaling. Recent analyses have shown that patients with CRC with mutated KRAS and BRAF oncogenes do not profit from treatment with such antibodies. Here we have used the binding domains of cetuximab and pantitumumab for constructing T cell-engaging BiTE antibodies. Both EGFR-specific BiTE antibodies mediated potent redirected lysis of KRAS- and BRAF-mutated CRC lines by human T cells at subpicomolar concentrations. The cetuximab-based BiTE antibody also prevented at very low doses growth of tumors from KRAS- and BRAF-mutated human CRC xenografts, whereas cetuximab was not effective. In nonhuman primates, no significant adverse events were observed during treatment for 3 wk at BiTE serum concentrations inducing, within 1 d, complete lysis of EGFR-overexpressing cancer cells. EGFR-specific BiTE antibodies may have potential to treat CRC that does not respond to conventional antibodies.

Antitumor activity of an EpCAM/CD3-bispecific BiTE antibody during long-term treatment of mice in the absence of T-cell anergy and sustained cytokine release.

muS110 is a BiTE antibody bispecific for murine EpCAM (CD326) and murine CD3. MT110, its human-specific analog, is in a clinical phase 1 trial for treatment of patients with adenocarcinoma of the lung or gastrointestinal tract. Recent studies have shown a therapeutic window for muS110, have explored single-dose toxicity of muS110, and have found that a 1-week low-dose treatment dramatically increased the tolerability of mice to very high doses of muS110 (Cancer Immunol. Immunother. 2009;58:95-109). Here we analyzed the impact of long-term, high-dose treatment of mice with muS110 on antitumor activity and functionality of T cells. After an initial self-limiting cytokine release, the 1-week adaptation period effectively blunted further cytokine production in response to a subsequent high-dose treatment with muS110. The much-increased tolerability of mice adapted to muS110 was not because of anergy of T cells. T cells isolated from chronically muS110-treated mice fully retained their cytotoxic potential, proliferative capacity, and responsiveness to stimulation by either muS110 or anti-CD3/anti-CD28/interleukin-2 when compared with T cells from control mice. Unimpaired T-cell performance was also evident from the effective prevention of orthotopic 4T1 breast tumor outgrowth in mice treated long term with escalating doses of muS110. Finally, we show that muS110 and MT110 recognize orthologous epitopes on mouse and human EpCAM proteins, suggesting that the target-related safety profile of muS110 in mice may be predictive for MT110 in humans.

Potent control of tumor growth by CEA/CD3-bispecific single-chain antibody constructs that are not competitively inhibited by soluble CEA.

Carcinoembryonic antigen (CEA, CD66e) is a well-characterized tumor-associated antigen that is frequently overexpressed in tumors. Phospholipases release CEA from tumor cells resulting in high circulating serum levels of soluble CEA (sCEA) that has been validated as marker for progression of colorectal, breast, and lung cancers. sCEA also acts as a competitive inhibitor for anticancer strategies targeting membrane-bound CEA. As a novel therapeutic approach for treatment of tumors expressing CEA on their cell surface, we constructed a series of bispecific single-chain antibodies (bscAb) combining various single-chain variable fragments recognizing human CEA with a deimmunized single-chain variable fragments recognizing human CD3. CEA/CD3-bscAbs redirected human T cells to lyse CEA-expressing tumor cells in vitro and in vivo. Efficient tumor cell lysis was achieved in vitro at bscAb concentrations from 1 pg/mL (19 fM) to 8.9 pg/mL with preactivated CD8 T cells, and 200 to 500 pg/mL with unstimulated peripheral blood mononuclear cell. The cytotoxic activity of a subset of CEA/CD3-bscAbs was not competitively inhibited by sCEA at concentrations that exceeded levels found in the serum of most cancer patients. Treatment with CEA/CD3-bscAbs prevented the growth of human colorectal cancer lines in a severe combined immunodeficiency mouse model modified to show human T cell killing of tumors. A murine surrogate CEA/CD3-bscAb capable of recruiting murine T cells for redirected tumor lysis in immunocompetent mice prevented the growth of lung tumors expressing human CEA. Together, our results reveal a unique opportunity for targeting cytotoxic T cells toward CEA-expressing tumors without being competitively inhibited by sCEA and establish CEA/CD3-bscAb as a promising and potent therapeutic approach.

Human IgG1 antibodies antagonizing activating receptor NKG2D on natural killer cells.

NKG2D is a surface receptor expressed on NK cells but also on CD8(+) T cells, gammadelta T cells, and auto-reactive CD4(+)/CD28(-) T cells of patients with rheumatoid arthritis. Various studies suggested that NKG2D plays a critical role in autoimmune diseases, e.g., in diabetes, celiac disease and rheumatoid arthritis (RA), rendering the activating receptor a potential target for antibody-based therapies. Here, we describe the generation and characteristics of a panel of human, high-affinity anti-NKG2D IgG1 monoclonal antibodies (mAbs) derived by phage display. The lead molecule mAb E4 bound with an affinity (KD) of 2.7 +/- 1.4 x 10(-11) M to soluble and membrane-bound human NKG2D, and cross-reacted with NKG2D from cynomolgus macaque, indicating potential suitability for studies in a relevant primate model. MAb E4 potently antagonized the cytolytic activity of NKL cells against BaF/3-MICA cells expressing NKG2D ligand, and blocked the NKG2D ligand-induced secretion of TNFalpha, IFNgamma and GM-CSF, as well as surface expression of CRTAM by NK cells cultured on immobilized MICA or ULBP-1 ligands. The antibody did not show a detectable loss of binding to NKG2D after seven days in human serum at 37 degrees C, and resisted thermal inactivation up to 70 degrees C. Based on these results, anti-human NKG2D mAb E4 provides an ideal candidate for development of a novel therapeutic agent antagonizing a key receptor of NK and cytotoxic T cells with implications in autoimmune diseases.

CD19-/CD3-bispecific antibody of the BiTE class is far superior to tandem diabody with respect to redirected tumor cell lysis.

Many kinds of bispecific antibodies recruiting T cells for cancer therapy have been developed. Side-by-side comparison has shown that CD19-/CD3-bispecific antibodies of the diabody, tandem diabody (Tandab) and quadroma format had similar cytotoxic activity, with Tandab being the most active format. Tandab has also been claimed to be superior to single-chain (sc) Fv-based bispecific constructs although data from a side-by-side comparison are not available. In this study, we compared side-by-side MT103 (bscCD19xCD3), a single-chain bispecific antibody of the BiTE class, with a CD19-/CD3-bispecific representative of the Tandab class. Based on literature data, we have constructed, produced and characterized the LL linker version of Tandab, which was reported to be the most active version of Tandab proteins. A dimeric protein of 114kDa was obtained that showed proper bispecific binding to CD3- and CD19-positive cells and could redirect both pre-stimulated and unstimulated human T cells for lysis of human B lymphoma lines Raji, MEC-1 and Nalm-6. Raji cells were lysed at a half-maximal concentration (EC50) of 10 nM Tandab using pre-stimulated T cells, which closely matched the published activity of LL-Tandab with this particular cell line. MT103 had between 700- and 8000-fold higher efficacy than Tandab for redirected lysis of the three human B lymphoma lines. These data demonstrate that under identical experimental conditions, the BiTE format has far superior activity compared to the Tandab format and is also superior to conventional diabody and quadroma formats. The extraordinary potency of the BiTE class and its representative MT103 may translate into improved anti-tumor activity, lower dosing and lower costs of production compared to other bispecific antibody formats.

Single-chain Fv antibody-alkaline phosphatase fusion proteins produced by one-step cloning as rapid detection tools for ELISA.

A system was constructed for the production of alkaline phosphatase (aP)-labeled antibody single-chain Fv (scFv) fragments in Escherichia coli. The expression vector pASK75 was modified by sequentially inserting the E. coli aP coding region and the scFv cloning cassette. Engineering the cloning sites SfiI and NotI located at the 5' and 3' end of the scFv gene provides an easy means to insert scFv fragments. These cloning sites are widely used in recombinant antibody technology and, thus, enable the one-step cloning of scFv fragments derived from corresponding antibody phage libraries into the expression vector. An expressed herbicide-specific scFv aP fusion protein retained both, analyte binding and enzymatic activity, as determined by ELISA. Therefore, this system permits the production of scFv-aP conjugates in E. coli, which can replace conventionally prepared aP-labeled antibodies in immunoassays. These fusion proteins are designed to accelerate the immunochemical detection of analytes, since the assay duration is essentially reduced by omitting the use of enzyme labeled secondary antibodies.

Cloning, functional expression and kinetic characterization of pesticide-selective Fab fragment variants derived by molecular evolution of variable antibody genes.

Fab antibody fragments were constructed by subcloning single chain Fv variable regions from the phagemid vector pCANTAB 5E into the expression vector pASK99. The vector was designed for bacterial secretion of Fab fragments and bears coding sequences for murine constant domains including the Strep-tag II at the carboxyl-terminal end of the constant heavy chain domain. The cloning procedure was carried out with the scFv antibodies IPR-7, IPR-53 and IPR-23. The second and third clone originated from the molecular evolution of the s-triazine selective antibody IPR-7. The Fab fragments were expressed under the transcriptional control of the tetA promoter system. Large-scale production benefits from the anhydrotetracycline-inducible system because of the lower costs for the inducer compared to IPTG and the tightly regulated expression of the recombinant antibody fragments. The Strep-tag purification technology facilitates the isolation of Fab fragments from the E. coli periplasm. The characteristics of functionally expressed Fab fragments were determined by employing a BIAcore 2000 system. The KD of the Fab variant IPR-23 (K(D)= 1.1 2 x 10(-9) M) optimized by molecular evolution was improved by a factor of 24 compared to the Fab IPR-7 (K(D) = 2.73 x 10(-8) M), which was derived from the template scFv antibody IPR-7. The affinity alteration was also reflected in the 22-fold reduction of the IC50 values of the variants Fab IPR-7 (IC50 = 60.5 microg/L) and Fab IPR-23 (IC50=2.7 microg/L) in the corresponding atrazine ELISA.