Generation of MANAbodies specific to HLA-restricted epitopes encoded by somatically mutated genes

Significance

Mutations in oncogenes and tumor suppressor genes drive tumorigenesis and their protein products form therapeutic targets that are absent from normal cells. However, nearly all such mutant epitopes lie in the interior of the cells, either in the cytoplasm or nucleus, complicating immunotherapies directed against the mutants. We have developed an approach to identify antibodies selectively targeting complexes containing common HLA types bound to peptide products of commonly mutated oncogenes. Because these peptide-HLA complexes are expected to be exclusively present on the surface of cancer cells, antibodies targeting them could in principle be used for therapeutic purposes.

Abstract

Mutant epitopes encoded by cancer genes are virtually always located in the interior of cells, making them invisible to conventional antibodies. We here describe an approach to identify single-chain variable fragments (scFvs) specific for mutant peptides presented on the cell surface by HLA molecules. We demonstrate that these scFvs can be successfully converted to full-length antibodies, termed MANAbodies, targeting “Mutation-Associated Neo-Antigens” bound to HLA. A phage display library representing a highly diverse array of single-chain variable fragment sequences was first designed and constructed. A competitive selection protocol was then used to identify clones specific for mutant peptides bound to predefined HLA types. In this way, we obtained two scFvs, one specific for a peptide encoded by a common KRAS mutant and the other by a common epidermal growth factor receptor (EGFR) mutant. The scFvs bound to these peptides only when the peptides were complexed with HLA-A2 (KRAS peptide) or HLA-A3 (EGFR peptide). We converted one scFv to a full-length antibody (MANAbody) and demonstrate that the MANAbody specifically reacts with mutant peptide–HLA complex even when the peptide differs by only one amino acid from the normal, WT form.

Cancers are the result of sequential mutations of oncogenes and tumor suppressor genes (1). In theory, somatic mutations are ideal therapeutic targets because they are not found in virtually any normal cell (2). Even though the protein products of these mutations generally only subtly differ from the WT form, often by a single amino acid, this difference is sufficient for effective targeting. When the protein is an enzyme, such as that encoded by BRAF, the resulting structural change can provide a pocket for the binding of specific enzymatic inhibitors (3–5). Antibodies are one of the most successful types of modern pharmaceutical agents and have been shown to be able to specifically recognize proteins that differ only by a single amino acid or by the modification of a single amino acid (5–11). However, all antibodies used in the clinic are directed against cell-surface or secreted proteins rather than intracellular proteins. Intracellular proteins are not accessible to large molecules such as antibodies, but unfortunately the vast majority of abnormal epitopes encoded by mutant genes are not on the cell surface (2).

Intracellular antigens, such as viral components, can be recognized by the immune system, although this is based on recognition of proteolytically processed peptides complexed to HLA molecules on the cell surface (12). Indeed, 10–20% of the epitopes created by mutant genes in cancers (hereinafter referred to as MANAs, for Mutation-Associated Neo-Antigens) are predicted to bind to common HLA types (12). Moreover, examples of T cells that can bind to such peptide–HLA complexes have been found in patients as well as in experimental animals (13–16).

The majority of T-cell responses generated in vivo against MANAs are “private”, that is, directed against mutant eptitopes encoded by passenger mutations that are present in cancers of individual patients or mice but are not commonly found in patients and do not drive neoplastic growth (2). Immunologic agents targeting such antigens are only useful for the treatment of the individual patients harboring the particular MANA (16–20). We wondered whether antibodies that recognize MANA–HLA complexes derived from frequently mutated cancer genes could be developed, because these could theoretically be used in a large number of patients. For this purpose, we attempted to develop antibodies, called MANAbodies, directed against MANAs complexed with HLA proteins. T-cell receptors (TCRs) seem preferable to antibodies because the former naturally bind to peptides complexed with HLA proteins. However, the technologies for generating antibodies reactive against specific epitopes are more advanced than those available for TCRs. Moreover, it was known that antibodies specific for polypeptides that differ by one amino acid, or an amino acid derivative, from other polypeptides can be generated (21). We therefore hypothesized that some MANAs, even when embedded in the pocket of a large HLA protein, might be recognizable by antibodies. The experiments described below were performed to test this hypothesis.

Results

Design and Construction of a Single-Chain Variable Fragment-Based Phage Display Library.

We began these studies with attempts to generate an antibody against a mutant KRAS peptide using conventional approaches to derive monoclonal antibodies after mouse immunization. These efforts failed, as no antibodies specifically reactive with the MANA were identified. We therefore turned to phage display approaches for generating MANAbodies (Fig. 1). The design of the phage display library followed principles used in published studies and included some special features (22). The framework of the library was based on the single-chain variable fragment (scFv) sequence of the humanized 4D5 antibody (trastuzumab), generated against the protein encoded by ERBB2 (23). This framework was chosen by virtue of its stability on phage and its ease of conversion to a soluble scFv, fragment antigen-binding (Fab), or antibody (22, 24). High-resolution crystal structures of the humanized 4D5 have identified the residues within the highly variable complementarity-determining regions (CDRs) that play the most significant role in antigen binding (25). This allowed us to focus variability on the most important residues for antigen binding rather than backbone residues. In our library, amino acid substitutions were limited to defined paratope residues in four CDRs: CDR-L3, CDR-H1, CDR-H2, and CDR-H3 (26). We skewed the library to include specific amino acids within the CDRs that either were previously demonstrated to play a significant role in antigen binding or were enriched in naturally occurring antibodies (SI Appendix, Table S1). One important randomization was at CDR-L3, which contained a mixture of serines and tyrosines, two amino acids previously shown to facilitate a minimalist approach to library design (26, 27). CDRs in the heavy chain have been shown to play a more significant role in antigen-binding diversity (28–30), and we therefore introduced more degeneracy in the heavy-chain CDRs than in the light-chain CDRs (SI Appendix, Table S1). Additionally, at locations wherein increased diversity was sought, the more commonly used IUB nucleotides (designated NNK and NNS degenerate codons) were substituted by DMT codons; this eliminated unwanted residues. In all cases, we used degenerate nucleotide mixtures intended to minimize the incorporation of sequences resulting in cysteine or stop codons. Finally, we introduced length polymorphisms in CDR-H3, allowing for a stem-loop binding length diversity of 7 aa. These changes resulted in a calculated theoretical library complexity of 5 × 10¹³ (SI Appendix, Table S1).

Fig. 1.

Generation of MANAbody. The process of MANAbody generation is outlined with the competitive phage selection highlighted at the center.

The synthesized oligonucleotide library was cloned into a phagemid vector for scFv expression. This scFv carried a myc tag and was fused to the bacteriophage M13 pIII coat protein through a tobacco etch virus (TEV) cleavage site (SI Appendix, Fig. S1). This design facilitated purification of scFvs from the phage particles and provided an alternative elution method, accomplished via TEV cleavage, during the phage-selection process. After library synthesis, 45 clones were sequenced by the Sanger method. The sequencing showed a library success rate of 53%, as defined by the absence of mutations within the framework region and the presence of the expected amino acids within the CDRs. Library diversity was calculated based on transformation efficiencies achieved during library construction, resulting in an estimated diversity of 5.5 × 10⁹. To further assess the quality of the library, DNA purified from the library was subjected to massively parallel sequencing (31). This analysis revealed 3,785,138 unique clones (46.5% of all clones analyzed). The sequenced region included only the CDR-H3 and not the other three CDRs (CDR-L3, CDR-H1, and CDR-H2) that were systematically varied. The fraction (46.5%) of unique clones therefore represents a minimum estimate of the diversity. Translation of a random subset of sequences further showed the expected amino acid distribution (SI Appendix, Table S2) as well as length diversity (SI Appendix, Table S3) in CDR-H3.

Target Selection and Competitive Strategy for Identifying Selectively Reactive Phage Clones.

We chose MANAbody targets based on the frequency of particular mutations and the strength of their predicted binding to HLA alleles. KRAS is one of the most commonly mutated genes in human cancers, with mutations particularly prevalent in pancreatic, colorectal, and lung adenocarcinomas. We chose the G12V mutation as the target because a relevant peptide containing the mutation was predicted to bind to HLA-A2, which is the most common HLA allele in many ethnic groups (32). This in silico prediction was made using the NetMHC v3.4 algorithm (33–35). Additionally, the critical mutant residue (V at codon 12) was expected to be exposed on the surface of the HLA protein based on structural studies of other peptide–HLA complexes (36). The peptide KLVVVGAVGV, in which the valine residue (V) at position 8 represents the G12V mutation, was chemically synthesized by conventional means. Peptides corresponding to the product of a mutant allele will henceforth be termed “mutant peptides,” and peptides representing the product of a WT allele will be referred to as “WT peptides.” The mutant KRAS peptide was then folded into a complex (monomer) of HLA-A2 and beta-2-microglobulin [KRAS(G12V)-HLA-A2]. Two peptides corresponding to WT KRAS sequences were also synthesized and folded with HLA-A2 or HLA-A3 to form KRAS(WT)-HLA-A2 and KRAS(WT)-HLA-A3 monomers, respectively. Additional mutant KRAS monomers corresponding to other codon 12 mutations were also assembled. In most cases, monomers were biotinylated to facilitate purification and subsequent experimentation (SI Appendix, SI Materials and Methods).

The phage display selection process consisted of 10 rounds of selection and amplification, which were divided into three distinct phases: an enrichment phase (rounds 1–3), a competitive phase (rounds 4–8), and a final selection phase (rounds 9–10) (SI Appendix, Fig. S2). The overall objective of these phases was to maximize recovery of clones that bound KRAS(G12V)-HLA-A2 better than KRAS(WT)-HLA-A2 or HLA alone. In each round of the enrichment phase, negative selection with heat-denatured biotinylated HLA-A2 monomers was followed by positive selection with KRAS(G12V)-HLA-A2 (SI Appendix, Fig. S2A and SI Materials and Methods). In each successive round (rounds 2 and 3), the amount of KRAS(G12V)-HLA-A2 monomer was reduced to enrich for stronger binders.

The competitive phase described in this study was intended to enrich for the rare KRAS(G12V)-HLA-A2 binders over KRAS(WT)-HLA-A2 binders and the much more frequent pan-HLA binders that we expected to be present in the library following the enrichment phase. Each round of the competitive phase began with negative selection using denatured HLA-A2 and native HLA-A3 monomers (SI Appendix, Fig. S2B). Then, the phage were simultaneously incubated with KRAS(G12V)-HLA-A2 bound to streptavidin magnetic beads and KRAS(WT)-HLA-A2 bound to streptavidin agarose beads. KRAS(WT)-HLA-A2 served as a competitor, because phage bound to it would not be recovered in the magnetic bead trapping step (SI Appendix, Fig. S2B). Moreover, in each round of the competitive phase, decreasing amounts of KRAS(G12V)-HLA-A2, but the same amount of KRAS(WT)-HLA-A2, were used in an attempt to enrich for high-affinity binders. In the final selection phase, each round started with stringent negative selection using denatured and native KRAS(WT)-HLA-A2 monomer and proceeded with positive selection with KRAS(G12V)-HLA-A2 monomer (SI Appendix, Fig. S2C).

Evaluation of the Selected Phage Clones.

We used an ELISA to evaluate the binding of phage to peptide–HLA complexes. After the enrichment phase (SI Appendix, Fig. S2A), the selected phage (en masse) did not show preference for mutant over WT KRAS peptides complexed to HLA-A2, or preference for KRAS peptides bound to HLA-A2 over KRAS peptides bound to HLA-A3 (SI Appendix, Fig. S3A). Only after the final selection phase (SI Appendix, Fig. S2C) did specificity for mutant KRAS bound to HLA-A2 emerge. In particular, these phage bound to KRAS(G12V)-HLA-A2 better than to KRAS(WT)-HLA-A2 or to KRAS(WT)-HLA-A3 (SI Appendix, Fig. S3B). The phage were cloned by limiting dilution and expanded in a 96-well plate format. One clone (D10) showed substantial binding to the KRAS(G12V)-HLA-A2 monomer (Fig. 2A). The D10 clone was highly specific for KRAS(G12V)-HLA-A2, as it failed to bind all other monomers tested (Fig. 2B).

Fig. 2.

Selective binding of phage and purified scFv to mutant monomers. Monomers folded with the indicated peptides, beta-2 microglobulin, and HLA molecules were incubated with phage clones or purified scFv at different dilutions, followed by ELISA with anti-M13 (for phage) or anti-Flag tag (for scFv) antibody. (A) Selective binding of phage clones collected and expanded after the final selection phase for KRAS(G12V)-HLA-A2 binders. Clone D10 is highlighted by the red arrow. (B) Selective binding of phage clone D10 to different monomers. ****P < 0.0001, comparing KRAS(G12V)-HLA-A2 against every other monomer at 1:80 dilution. (C) Selective binding of the purified D10 scFv to different monomers. ****P < 0.0001 comparing KRAS(G12V) HLA-A2 against every other monomer at 1 µg/mL dilution. (D) Selective binding of phage clone C9 to different monomers. ****P < 0.0001, comparing EGFR(L858R)-HLA-A3 against every other monomer at 1:900 dilution. (E) Selective binding of the purified C9 scFv to different monomers. ****P < 0.0001, comparing EGFR(L858R)-HLA-A3 against every other monomer at 1 µg/mL dilution. In B–E, monomers folded with WT or specified mutant peptides and HLA molecules are shown on the x axis. ELA, negative control peptide; No Monomer, well coated with streptavidin without monomer attached.

To produce D10 scFv uncoupled from M13 pIII, ssDNA from the D10 phage was purified. The scFv portion was amplified by PCR, sequenced (SI Appendix, Table S4), and cloned into a prokaryotic expression vector containing either a Flag or a V5 epitope tag in addition to a 6X His tag. This facilitated high-level expression and affinity purification of D10 scFv. Similar to the phage expressing D10 scFv:pIII fusion protein, purified D10 scFv interacted with KRAS(G12V)-HLA-A2 in a highly specific fashion (Fig. 2C). Importantly, the D10 scFv did not show any binding above background to KRAS(WT)-HLA-A2, KRAS(WT)-HLA-A3, or to other KRAS mutants [KRAS(G12C) or KRAS(G12D)] bound to HLA-A2. Additionally, D10 scFv did not bind to KRAS peptides when not complexed with HLA proteins (SI Appendix, Fig. S4). These results demonstrate successful selection for scFv bound to peptides in the context of HLA.

Affinity Maturation.

The affinity of the D10 scFv for KRAS(G12V)-HLA-A2 was estimated to be 49 nM (SI Appendix, Table S5), using the AlphaScreen method for affinity measurement (37). We next proceeded to affinity-mature D10. Briefly, a phage library of D10 scFv mutagenized through error-prone PCR was generated from the original D10 scFv sequence and subject to three rounds of selection against the KRAS(G12V)-HLA-A2 and KRAS(WT)-HLA-A2 monomers. Evaluation of the clones yielded a candidate, D10-7, that showed a newly acquired capacity to bind to KRAS(G12C)-HLA-A2 while still retaining the ability to differentiate between mutant and WT KRAS epitopes (SI Appendix, Fig. S5). To compare D10-7 and D10 for their relative binding to KRAS(G12V)-HLA-A2, we used off-rate-based assays to measure the k_off value. Unlike affinity measurements, these assays allow for rapid comparison of multiple scFvs within the same test, thus providing a more direct comparison of the relative binding of multiple clones (38). The off-rate measurements showed an almost twofold decrease in the disassociation rate constant for the affinity-matured D10-7 scFv (3.2 × 10⁻⁶ s⁻¹) compared with the D10 scFv (5.7 × 10⁻⁶ s⁻¹) (SI Appendix, Table S5). No measurable binding of the KRAS(WT)-HLA-A2 monomer to these scFvs occurred, documenting that the k_off for KRAS(G12V)-HLA-A2 was at least 200-fold lower than for KRAS(WT)-HLA-A2 (SI Appendix, Table S5). The large differential binding of these scFvs to mutant vs. WT peptides complexed with HLA-A2 was also evident in the other assays described below.

Identification of Phage That Can Bind to a Different MANA.

To determine whether this approach was applicable to other MANAs, we sought to identify scFvs specific for a mutant epidermal growth factor receptor (EGFR) peptide complexed to different HLA allele. The EGFR L858R mutation is found in ∼10% of lung adenocarcinomas and accounts for ∼40% of all EGFR mutations in this cancer type (39). Codon 858 is in the cytoplasmic domain, rather than the extracellular or membrane domains, of the EGFR protein and normally should not be visible on the cell surface (40). A peptide (KITDFGRAK) containing this mutation was predicted to bind to the HLA-A3 allele when analyzed by the NetMHC v3.4 algorithm. To identify scFvs specific to this peptide–HLA complex, we adopted a modified scheme of selection and amplification in which isopropyl β-D-1-thiogalactopyranoside was used to induce scFv expression. Additionally, the number of rounds in each selection phase was adjusted to favor the enrichment of desired phage rather than the elimination of undesired ones (compare SI Appendix, Fig. S6 with Fig. S2; also see SI Appendix, SI Materials and Methods). With these modifications, we were able to obtain a phage clone (C9) that showed selective binding to mutant EGFR peptide complexed to HLA-A3 [EGFR(L858R)-HLA-A3] compared to a variety of control monomers, including WT EGFR bound to HLA-A3 (Fig. 2D). The C9 scFv generated from this clone showed similar selective binding to EGFR(L858R)-HLA-A3 (Fig. 2E). The estimated k_off of the mutant EGFR peptide bound to HLA-A3 was an order of magnitude lower than the k_off of the WT peptide (value of 2.6 × 10⁻⁶ s⁻¹ vs. 3.0 × 10⁻⁵ s⁻¹, respectively; SI Appendix, Table S5).

Selective Binding to Cells Displaying Mutant Peptides on the Cell Surface.

We next attempted to determine whether the D10 and C9 scFvs would bind to mutant KRAS and EGFR peptide–HLA complexes on the surface of cells. The T2 cell line was derived from an Epstein–Barr virus-transformed human lymphoblast line defective in presentation of endogenous HLA-associated peptide antigens owing to a deletion that involves genes for TAP1 and TAP2 peptide transporters (41). T2A3 is a modified version of T2 with stable expression of the HLA-A3 transgene (42, 43). T2 and T2A3 cells express low levels of HLA that can be stabilized by the addition of exogenous HLA-binding peptides, and thus can serve as a platform for assaying interactions with specific HLA-binding peptides (44, 45). We first pulsed T2 cells with KRAS(G12V), KRAS(WT), or a negative control peptide. To assess loading efficiency, we used the W6/32 antibody that targets HLA molecules stabilized by any HLA-binding peptides. The efficiency of peptide loading between WT and mutant peptides were comparable as suggested by W6/32 staining (SI Appendix, Fig. S7). Analysis of the pulsed cells by flow cytometry after incubation with D10 phage showed that binding of the D10 phage to KRAS(G12V) peptide-pulsed cells was evident, whereas only background levels of staining to the KRAS(WT) or control peptide-pulsed cells was observed (Fig. 3A and SI Appendix, Fig. S8). A similar experiment with purified D10 scFv rather than D10 phage confirmed the selective binding to KRAS(G12V) presented on the cell surface (Fig. 3B). We also pulsed T2A3 cells with mutant EGFR(L858R), EGFR(WT), or a negative control peptide and assessed C9 phage binding by flow cytometry. Again, binding of the C9 phage to the EGFR(L858R) peptide-pulsed cells was evident, whereas no binding to the EGFR(WT) or control peptide-pulsed cells was observed (Fig. 3C and SI Appendix, Fig. S9). Only background fluorescence was observed when phage or scFvs were not included in the reaction or cells were not loaded with the peptides (Fig. 3 A–C and SI Appendix, Figs. S8 and S9).

Fig. 3.

Selective binding of candidate phage clones or purified D10 scFv to cells displaying mutant peptides on the cell surface. T2 or T2A3 cells were pulsed with the indicated peptides and then incubated with D10 phage (A), purified D10 scFv (B), or C9 phage (C) before analysis of the stained cells by flow cytometry. ELA and LLG, negative control peptides; for C9 phage, KRAS(WT) was used as a negative control peptide. (D) scFv-mediated, complement-dependent cell killing. CDC assay was performed by incubating T2 cells with 10% rabbit complement and D10 scFv or D10-7 scFv preconjugated to anti-V5 antibody, after T2 cells were pulsed with the indicated peptides. CellTiter-Glo was used to assess the viability of cells. ***P < 0.001, comparing KRAS(G12V):D10-7 to all other points at the 0.66 nM (−0.18 on x axis) antibody concentration; ns, not significant (P = 0.488), comparing KRAS(WT):D10-7 to Unpulsed:D10-7 at the 0.66 nM antibody concentration.

We next sought to determine whether T2 cells pulsed with the mutant KRAS(G12V) peptide could be targeted by D10 scFv and killed in a complement-dependent cytotoxicity (CDC) assay. As a positive control, we pulsed T2 cells with KRAS(G12V) or KRAS(WT) peptides and performed a CDC assay with the W6/32 antibody. As expected, the antibody killed peptide-pulsed T2 cells efficiently in the presence of complement (SI Appendix, Fig. S10). We then tested D10 scFv and the affinity-matured D10-7 scFv, both conjugated to an anti-V5 epitope tag antibody containing the complement-fixing Fc region, in the CDC assay. Both scFvs killed the KRAS(G12V)-pulsed T2 cells in a dose-dependent fashion and the affinity-matured D10-7 scFv showed a remarkable improvement in killing efficiency (EC₅₀ of 0.79 nM for D10-7 vs. EC₅₀ of 11.2 nM for D10, Fig. 3D). In contrast, cells pulsed with KRAS(WT) or not pulsed with exogenous peptides showed only marginal cell death.

Generation of a Full-Length MANAbody.

Clinical applications of immunotherapeutic reagents generally employ complete antibodies, including the Fc domain, rather than just the scFv component. Another attribute of complete antibodies is the higher avidity achieved as a result of bivalency of the initially monovalent scFvs. To generate a complete MANAbody from the D10 scFv, we grafted the D10 scFv sequence onto the constant region of the clinically used humanized 4D5 antibody trastuzumab. High levels of expression of the D10 MANAbody were achieved in mammalian cells (139 mg of protein per liter of Expi293 cell culture). Similar to the D10 scFv, the D10 MANAbody interacted with KRAS(G12V)-HLA-A2, as assessed by ELISA (Fig. 4A). No binding was observed to the KRAS(WT)-HLA-A2 monomer or to any other monomer tested. The D10 MANAbody also showed relatively stronger staining of T2 cells pulsed with the mutant KRAS(G12V) peptide, compared with those pulsed with KRAS(WT) or a negative control peptide (Fig. 4B and SI Appendix, Fig. S11). Finally, the observed half-life of the full-length D10 MANAbody was similar to that of its scFv derivative when assessed for its monovalent dissassociation (SI Appendix, Table S5). Thus, D10 MANAbody retained the high specificity and low dissociation rate constant observed with the D10 scFv.

Fig. 4.

Selective binding of D10 MANAbody. (A) Selective binding of D10 MANAbody to KRAS(G12V)-HLA-A2. Monomers folded with WT or the specified mutant peptides and HLA molecules as indicated on the x axis were incubated with D10 MANAbody at different dilutions, followed by ELISA with anti-human IgG antibody. ****P < 0.0001 comparing KRAS(G12V)-HLA-A2 against every other monomer at 1 µg/mL dilution. (B) Selective binding of D10 MANAbody to cells displaying mutant peptides on the cell surface. T2 cells were unpulsed or pulsed with indicated peptides and then incubated with D10 MANAbody or with an isotype control antibody, before analysis of the stained cells by flow cytometry.

Discussion

We have established a procedure for generating scFvs that selectively bind to mutant peptides embedded within HLA–beta-2 microglobulin complexes. Using this procedure, we obtained scFvs against the products of two commonly mutated oncogenes (KRAS and EGFR) when complexed with two common HLA types (A2 and A3, respectively). These scFvs bind to the peptide–HLA complexes on the surface of cells and can kill those cells when complement is present. Converting an scFv into a complete, bivalent antibody containing the Fc region sometimes results in loss of affinity (46, 47). However, we successfully generated a complete antibody using the D10 scFv sequence, and this MANAbody retained a dissociation rate constant and specificity comparable to those of the scFv (SI Appendix, Table S5 and Fig. 4B and SI Appendix, Fig. S11). We have not yet attempted to generate a MANAbody using the C9 scFv directed to the mutant EGFR peptide complexed with HLA-A3.

Other antibodies, termed TCRmimics, have been generated against peptide–HLA complexes in the past (48, 49). The novel aspect of our study is the design and construction of antibody-based reagents that differentially recognize HLA complexes containing peptides varying only by a single amino acid. A second important component of our study is that the variant peptides are commonly found in human cancers.

Certain limitations of our study should be emphasized. First, we cannot be certain that MANAbodies can be developed to all MANAs. Although we were successful in generating scFvs in two cases (KRAS G12V and EGFR L858R), subsequent (preliminary) attempts to generate scFvs were successful with only two of four other peptides. These failures were not due to an inability to develop scFvs to the mutant peptides in complex with HLA/beta-2 microglobulin monomers, which could always be obtained. The failed attempts were the result of the scFvs’ inabilities to discriminate between the WT and mutant forms of the antigenic peptides. It is possible that more complex scFv libraries, or libraries of different design, could be used to select phage with higher specificity. However, it is also possible that the mutant residues within some MANAs will be shielded by the HLA proteins that bind them, precluding the development of any MANAbodies specific to these peptides. This point is related to another limitation of this approach: Potential clinical uses of MANAbodies are confined to patients whose HLA type matches that of the MANAbody target. Although HLA-A2 and HLA-A3 are common HLA types, not all individuals carry them. Moreover, tumor cells can lose expression of beta-2 microglobulin or other components of antigen processing, thereby making them invisible to any immunotherapy that depends on MHC class I presentation (50).

Another important limitation is that the phage, scFvs, and MANAbody described in our study were shown only to bind to peptide–HLA–beta-2 microglobulin complexes in artificial situations. These included in vitro assays (ELISA) as well as cell-based assays using T2 cells pulsed with peptides. We have not yet shown that any of these reagents can bind to the much lower amounts of mutant peptides likely to be found on the surface of naturally occurring tumor cells. We have so far been unable to demonstrate such binding with the D10, D10-7, or C9 scFvs using standard flow cytometric methods. However, flow cytometry is known to require much higher densities of cell surface antigen (51, 52) than are likely to be present on cells with any specific peptide–HLA complex (53–55). Fortunately, suitably engineered antibody-based reagents can be used to kill cells when their cognate antigens are sparsely present on the surface, numbering as low as 300 per cell (56).

The greatest challenge in both cancer diagnosis and therapy is specificity—developing reagents that recognize or kill cancer cells but not normal cells. The relative lack of specificity currently presents a major obstacle for the wider implementation of powerful immunotherapeutic agents such as chimeric antigen receptors and bispecific antibodies (57–60). In this context, specific somatic mutations that alter the encoded proteins of cancer driver genes represent biochemical features that distinguish cancer cells from normal cells in unparalleled fashion. The strength of the work described here is that it demonstrates the feasibility of generating highly specific reagents that recognize these altered proteins in a context that is clinically relevant (cell surface). This sets the stage for further exploration of such reagents and their incorporation into suitable diagnostic and therapeutic vehicles.

Materials and Methods

Methods for realization of the three key procedures used in this study (phage library construction, phage selection, immunological assays for binding specificity) are described in detail in SI Materials and Methods. In brief, a phage library displaying scFvs with humanized 4D5 antibody sequence as the framework was designed based on ref. 22, with modifications to enhance diversity and affinity. The phage library was constructed and screened for scFvs specific for mutant KRAS and EGFR peptides bound to HLA molecules using a series of sequential positive and negative selections based on ref. 26 with additional steps of competitive binding to enhance specificity. Free scFvs and a full-length antibody were then generated based on sequences from the candidate phage-displayed scFvs (23). Candidate phage, free scFvs, and the full-length antibody were analyzed by ELISA, flow cytometry after cell staining, and complement-dependent cytotoxicity (26, 44).