Immunity against cancer cells may promote their proliferation and metastasis

Significance

We present a concept in cancer where an immune response is detrimental rather than helpful. Two autoantibodies recognize the growth factor receptor which was isolated from cancer patients. One antibody was shown to increase breast cancer cell growth, whereas the other antibody inhibited growth. This study demonstrates that individual autoantibodies may play a role in cancer and that the agonist autoantibody in cancer patients is deleterious and harmful.

Abstract

Herein we present a concept in cancer where an immune response is detrimental rather than helpful. In the cancer setting, the immune system is generally considered to be helpful in curtailing the initiation and progression of tumors. In this work we show that a patient’s immune response to their tumor can, in fact, either enhance or inhibit tumor cell growth. Two closely related autoantibodies to the growth factor receptor TrkB were isolated from cancer patients’ B cells. Although highly similar in sequence, one antibody was an agonist while the other was an antagonist. The agonist antibody was shown to increase breast cancer cell growth both in vitro and in vivo, whereas the antagonist antibody inhibited growth. From a mechanistic point of view, we showed that binding of the agonist antibody to the TrkB receptor was functional in that it initiated downstream signaling identical to its natural growth factor ligand, brain-derived neurotrophic factor (BDNF). Our study shows that individual autoantibodies may play a role in cancer patients.

It is well known that there is often a high degree of mutation in the surface proteins of cancer cells (1–3). From an immunological point of view, we would expect a loss of self-tolerance to these mutated proteins, resulting in their becoming immunogenic and triggering the formation of autoantibodies. When the mutated proteins are growth factor receptors, some of these autoantibodies can be expected to behave as agonists or antagonists, thereby affecting cancer cell proliferation. This situation would be reminiscent of that seen with long-acting thyroid stimulator (LATS), a TSH receptor–binding autoantibody found in patients with Graves’ disease, which stimulates the synthesis and inappropriate release of thyroid hormone (4, 5). In fact, evidence does suggest that analogous autoantibodies are found in the plasma of patients with a number of different cancers. The immune system appears to be able to sense an aberrant structure of cellular components and makes autoantibody responses to the tumor-associated antigens. In some cases, they are used in diagnosis or as predictors of disease progression (6–10).

In this study, we initially used morphology-based selection from combinatorial libraries to identify antibodies that affect tumor cell growth. One of these antibodies was capable of reversing epithelial–mesenchymal transition (EMT), and its target protein was shown to be tropomyosin receptor kinase B (TrkB). We were intrigued by the fact that a randomly selected antibody capable of reversing EMT was directed against a growth factor receptor on the cell surface. These phenotypic transitions between states are not binary, and tumors often exhibit a variety of epithelial and mesenchymal phenotypes.

Later, we identified autoantibodies to TrkB in the plasma of breast cancer patients. In order to study the effect of such antibodies on cancer growth and progression, we selected and purified 2 TrkB autoantibodies using our human combinatorial antibody library containing antibody genes from patients’ peripheral blood lymphocytes. One of these antibodies was shown to be an agonist, while the other was an antagonist. Functional studies showed that the agonist antibody promoted tumor cell growth, while the antagonist antibody inhibited growth. Our work shows that individual autoantibodies may significantly enhance tumor growth in cancer patients.

Results

Scheme for Selection of Antibodies That Induce Mesenchymal–Epithelial Transition in Breast Cancer Cells.

EMT is thought to play a key role in certain instances of malignant transformation (11, 12). EMT is associated with an obvious morphologic change, and in an effort to select antibodies capable of reversing this process (i.e., capable of inducing mesenchymal–epithelial transition [MET]), we decided to build a morphogenic selection system (Fig. 1A). A human combinatorial single-chain variable fragment (scFv) antibody library containing ∼10⁸ different members was incorporated into lentiviral vectors. The lentiviral library was then used at a multiplicity of infection (MOI) of 2 to infect the MDA-MB-231 breast cancer cell line, which is known to exhibit a mesenchymal phenotype. The infected cells were then incorporated into methylcellulose agar for morphogenic screening. The autocrine format used caused the antibodies to localize to the plasma membrane of infected cells (13). We used the scFv-fragment crystallizable (Fc) format, which offers advantages over the phage display–derived scFvs; scFv-Fcs have similar characterization as full-length immunoglobulin G (IgG) antibodies, including bivalent binding, stability, longer half-life, and production. This format allows for rapid characterization of candidate scFvs isolated from phage display libraries or other libraries before conversion into a full-length IgG (14). The construct included an scFv fusion with Fc domains linked via a flexible linker to a platelet-derived growth factor receptor (PDGFR) membrane–spanning domain. This resulted in the antibody molecules being integrated into the plasma membrane, where they were capable of binding to and affecting other receptors. Compared with controls, cells infected with members of the antibody library formed colonies with unique morphologies that included amorphous, scattered, linear, or round phenotypes. After 2 to 3 wk in culture, we harvested colonies that showed unique morphologies (Fig. 1B). Using flow cytometry, we determined that cells in these colonies expressed the epithelial marker E-cadherin (Fig. 1C). All antibody genes were recovered from the colonies by genomic PCR. Based on this phenotypic screening and the epithelial marker assay results, we chose one antibody (641) for further study because it appeared to induce morphologic changes consistent with MET.

Fig. 1.

Scheme for unbiased selection of antibodies that affect cell morphology from combinatorial libraries. (A) Lentiviral antibody libraries were constructed from human combinatorial antibody libraries. Plasmids of lentiviral antibody libraries and virus packaging were used to transiently transfect HEK293 cells. The MDA-MB-231 breast cancer cell line was infected with lentiviral antibody libraries in which antibodies were displayed on the cell surface and therefore could react with plasma membrane–bound receptors. The resulting cells were grown on methylcellulose agar for morphogenic screening. Colonies showing unusual and different morphologies were picked after 2 wk. Genomic DNA was extracted; then antibody genes were recovered by PCR and sequenced. After verification of antibody binding, the specific antibody was incubated with cell lysates and the resulting immunoprecipitates were analyzed by LC-MS/MS. (B) Representative colony morphologies. (C) The morphology change of the cell was determined by the epithelial marker (E-cadherin) using flow cytometry. PE, phycoerythrin.

Identification of the Target Protein.

To identify its target protein, antibody 641 was incubated with MDA-MB-231 cell lysate, and the resulting immune-precipitate was captured on a protein G column and eluted. The eluates were analyzed by Western blot and silver staining. The gel bands were excised and analyzed by liquid chromatography–mass spectrometry and liquid chromatography–tandem mass spectrometry (LC/MS/MS) to identify the target protein. The MS/MS data are shown in SI Appendix, Fig. S1. Only proteins known to be membrane-bound were considered. The identity of the target protein was confirmed by Western blot analysis using TrkB and Vimentin. The data (Fig. 2A) indicated that antibody 641 reacted only with the TRKB protein. The protein sequence of TrkB is 40% homologous with TrkA and TrkC. Therefore, in further studies using ELISA analyses we tested the ability of antibody 641 to bind these homologous proteins, as well as another related protein, p7NTR (Fig. 2B). These results confirmed that antibody 641 is absolutely specific for TrkB. We also found that TrkB antigen expression was reduced by siRNA with antibody 641 (Fig. 2C), further confirming TrkB as the target antigen.

Fig. 2.

Identification of target protein. (A) Western blot confirming TrkB as the target of antibody 641. (B) ELISA showing specific binding of TrkB by antibody 641. (C) Effects of TrkB knockdown in a 293T-overexpressing TrkB cell line. Flow cytometry analysis of cell surface TrkB expression 48 h after treatment with TrkB siRNA (orange) or nontargeting siRNA (blue) by antibody 641. Isotype-matched negative control antibodies are shown in red. PE, phycoerythrin.

Detection of Autoantibody to TrkB Antigen in the Plasma of Patients with Breast Cancer.

In a remarkable coincidence, in an effort to select agonist antibodies for therapeutic purposes, we had already established the presence of TrkB-binding autoantibodies in the plasma of breast cancer patients. First, enzyme-linked immunosorbent assays (ELISAs) using microtiter plates coated with the recombinant extracellular domains (ECDs) of TrkB (100 ng/well) and 100× diluted de-identified plasma of breast cancer patients (n = 6) detected the presence of TrkB-binding antibodies in these patients (Fig. 3A). Next, in order to remove their natural ligand, brain-derived neurotrophic factor (BDNF), and any other factors affecting binding to the TrkB receptor, we purified antibodies from plasma samples using protein-A/G beads, then analyzed these purified antibodies to determine their binding to TrkB antigen. The results showed that some of these purified antibodies (10 μg/mL) did specifically bind to TrkB antigen (Fig. 3B), confirming the presence of anti-TrkB autoantibodies in the plasma of some cancer patients. Since most immune responses are polyclonal, we reasoned that if TrkB had become immunogenic there could well be a variety of anti-TrkB autoantibodies, perhaps with different functions. This proved to be the case.

Fig. 3.

Detection of autoantibody to TrkB antigen in the plasma of patients with breast cancer. (A) Patient’s plasma (n = 6), (B) purified antibodies from patient’s plasma were individually analyzed by ELISA.

Different Anti-TrkB Autoantibodies Have Different Effects on Breast Cancer Cell Growth In Vitro and In Vivo.

Since it is difficult to elucidate the role of individual antibodies in a polyclonal mixture, functional studies were done using 2 purified human monoclonal anti-TrkB autoantibodies. One was agonist antibody 93, which we had discovered in previous study in our laboratory We exploited a phage display library derived from the blood of cancer patients to select and isolate high-affinity scFv antibodies specific for TrkB (15). The other was antibody 641, which had been isolated from a human combinatorial antibody library using morphologic selection as described above. To study the effect of anti-TrkB autoantibodies on breast cancer cells, we initially evaluated these 2 antibodies as scFv-Fcs in terms of their effect on the MDA-MB-231 breast cancer cell line in vitro. MDA-MB-231 cells were incubated with purified anti-TrkB antibody 93 or 641 at concentrations of 20 μg/mL. The proliferative response was analyzed at 72 h. As shown in Fig. 4A, incubation with antibody 93 increased cancer cell proliferation, while incubation with antibody 641 inhibited proliferation compared to phosphate-buffered saline (PBS) control.

Fig. 4.

Different effects of anti-TrkB autoantibodies on breast cancer cell growth in vitro and in vivo. (A) Effects of the TrkB 641 antibody on breast cancer cell growth in vitro. MDA-MB-231 cells were incubated with purified antibody 641 or 93 for 72 h. Cell proliferation was measured (n = 6) as described in Materials and Methods. (B) Growth effect of breast cancer cells in vivo by the anti-TrkB 641 and 93 antibodies. CB17/SCID female mice (n = 5) were inoculated orthotopically into the mammary fat pad with MDA-MB-231 human breast cancer cells. Purified antibody 641 or 93 at a dose of 10 mg/kg or 30 mg/kg per injection was given intraperitoneally when the tumor was palpable (60–100 mm³) and twice a week for 4 wk. (C) Tumor volume and body weight were measured twice a week, and results were compared to normal saline controls. Data represented as mean ± SD. An unpaired t test (2-tailed) was used to compare between treatment groups. All statistical evaluations of data were performed using GraphPad Prism 8. Statistical significance was achieved at P value (*P ≤ 0.05). Error bars indicate SD of measurements from independent experiments.

We then studied whether these antibodies could influence breast cancer cell growth in vivo. MDA-MB-231 cells were injected orthotopically into the mammary fat pad of female severe combined immunodeficient (SCID) mice. When the tumor mass was palpable (60–100 mm³), purified anti-TrkB antibody 93 or 641 was injected intraperitoneally twice a week at doses of 10 mg/kg or 30 mg/kg per injection. MDA-MB-231 tumor growth was increased in mice treated with antibody 93 and suppressed in mice treated with antibody 641, as shown in Fig. 4B. There was no effect on body weight (Fig. 4C). In summary, these experiments revealed that one anti-TrkB autoantibody (antibody 93) is an agonist, increasing cancer cell growth both in vitro and in vivo, while another anti-TrkB autoantibody (antibody 641) is an antagonist, inhibiting cancer cell growth both in vitro and in vivo. Thus, different anti-TrkB autoantibodies can have very different functions, with agonist antibodies potentially playing a major, deleterious role in the clinical setting.

Structural Comparison of the Two Anti-TrkB Autoantibodies.

Structural analysis revealed that antibodies 93 and 641 have very similar amino acid sequences in both the heavy-chain third complementarity–determining region (HCDR3) and the light-chain third complementarity–determining region (LCDR3). However, they have distinctly different amino acid sequences in HCDR1 and HCDR2 (Fig. 5).

Fig. 5.

Structural comparison of the 2 anti-TrkB autoantibodies. Antibodies 93 and 641 have very similar amino acid sequences in both HCDR3 and LCDR3 but different amino acid sequences in HCDR1 and HCDR2.

Comparison of Target-Binding Characteristics of the Two Anti-TrkB Autoantibodies.

We used flow cytometry to further characterize binding to cancer cells by antibodies 93 and 641. Both anti-TrkB autoantibodies (10 μg/mL) showed strong binding to the breast cancer line MDA-MB-231 (Fig. 6A). Fig. 6B shows that both antibodies 93 and 641 specifically bind to TrkB, not to p75NTR, TrkA, and TrkC. To determine the binding affinity, antibodies 93 and 641 were analyzed using TrkB-ECD in a biolayer interferometry binding assay. The results showed that the 2 antibodies are similar in terms of their binding affinity to TrkB (5.6 nM for antibody 93 versus 5.7 nM for antibody 641). The data are shown in Fig. 6C. We also used epitope binning analysis by the Octet platform to study the TrkB binding site of BDNF and antibody 641 or 93. The result (SI Appendix, Fig. S2) showed that the binding region may be similar between BDNF and antibodies 641 and 93 to TrkB.

Fig. 6.

Comparison of target-binding characteristics of the 2 anti-TrkB autoantibodies. (A) Comparison of binding with antibodies 93 and 641 (10 μg/mL) in MDA-MB-231 cells by analysis of flow cytometry. PE, phycoerythrin. (B) Binding specifically to the Trk family using Western blot. (C) To determine binding affinity, antibodies 93 and 641 were analyzed using TrkB-ECD in a biolayer interferometry binding assay. His-tagged TrkB protein was loaded onto nickel-nitrilotriacetic acid (Ni-NTA) biosensors. Antibodies 641 and 93 were prepared at 7 concentrations from 450 nM to 7 nM, half dilution.

Functional Comparison of the Two Anti-TrkB Autoantibodies.

TrkB receptors are known to propagate their signal through activation of multiple signaling pathways (e.g., Ras-ERK, PI3K, PLCγ) (16). To determine if binding by antibodies 93 and 641 activate the TrkB receptor, and to analyze any downstream effects, we used 2 reporter cell lines (TkB-CRE-bla 293 and TkB-NFAT-bla CHO-K1). A HEK293 and a CHO cell line were separately infected with a lentiviral vector encoding the full-length human TrkB gene under the EF1a promoter. Next, the HEK293 cells were infected with a lentiviral vector encoding the β-lactamase (bla) protein regulated by a cAMP-response element (CRE), whereas the CHO cells were infected with the β-lactamase gene regulated by an nuclear factor of activated T cells (NFAT) response element. CCF-4 was used as a fluorescence resonance energy transfer (FRET)-based substrate for β-lactamase as a sensitive reporter of mammalian gene expression. SI Appendix, Fig. S3A shows the reporter mechanism when the TrkB receptor is activated and induces subsequent activation of downstream signaling cascades. Initially, we tested the purified antibodies from plasma of breast cancer patients and normal donors. We used these reporter cell lines to observe the effect of purified antibodies (10 μg/mL) in cell CRE and NFAT signaling. The result is shown in SI Appendix, Fig. S3B. Since it is difficult to elucidate the role of autoantibodies in a polyclonal mixture, we determined if antibody 93 or 641 inactivates the TrkB receptor. The 2 reporter cell lines were exposed to antibody 93 or 641 (10 μg/mL) for 5 h; then a β-lactamase assay was performed by flow cytometry (SI Appendix, Fig. S4). As with the natural ligand, BDNF, antibody 93 stimulated signaling in both cell lines whereas antibody 641 did not. Thus, antibody 93 both binds to and activates the TrkB growth factor receptor. On the other hand, antibody 641, despite its close structural relationship to antibody 93 and equivalent binding strength, had no effect (Fig. 7 A and B). As the positive control, BDNF showed an agonist effect in both reporter cell lines.

Fig. 7.

Functional comparison of the 2 anti-TrkB autoantibodies. (A) TkB-CRE-bla 293 or (B)TkB-NFAT-bla CHO-K1 reporter cell lines were treated with BDNF (100 ng/mL), antibody 93 or 641 at 10 μg/mL or isotype antibody for 5 h. Cells were incubated with the CCF4-AM substrate and subjected to FACS based on the FRET signal. Data were normalized to the PBS control group. (C) TkB-overexpressed 293 cells were incubated with antibody 93 or 641 or PBS. After a 10-min incubation at 37 °C, cell lysates were analyzed for the total and phosphorylated PLCγ, AKT, and ERK using Western blot. Error bars are represented as mean ± SD. An unpaired t test (2-tailed) was used to compare treatment groups. All statistical evaluations of data were performed using GraphPad Prism 8. Statistical significance was achieved at P value (***P ≤ 0.001).

To gain mechanistic insight into the signal transduction, we analyzed the induction of downstream transcription factors triggered by receptor binding by antibody 93 or 641. TkB-CRE-bla 293 cells were incubated with antibody 93 or 641 or PBS. After 10 min incubation at 37 °C, cell lysates were prepared and analyzed for total and phosphorylated PLCγ, AKT, and ERK using Western blot. The results (Fig. 7C) showed that the cell-signaling effect of antibody 93 was similar to that of the natural growth factor ligand, BDNF. By comparison, the effect of antibody 641 on ERK and PLCγ was markedly reduced relative to that of antibody 93. The fact that the binding of these 2 autoantibodies results in significantly different downstream signaling likely accounts for their differential effects on tumor cell growth.

Discussion

Conceptually, one can imagine 2 outcomes following an immune response to cancer. The first is generally recognized and involves a surveillance role that inhibits the initiation and progression of cancer. The second, more obscure possibility is that under certain conditions the immune response may stimulate rather than curtail tumor growth. It is well known that certain proteins associated with cancer cells are often mutated, leading them to escape immune tolerance and become immunogenic. This situation can be especially problematic when the mutated proteins are growth factor receptors. It is reasonable to expect a polyclonal response, and some of the many different antibodies formed may be agonists, with the potential to enhance tumor proliferation. Such a situation is analogous to long-acting thyroid stimulator (LATS), where an agonist autoantibody stimulates cells by binding to its receptor.

Here we give evidence for the second possible outcome following an immune response to cancer. We confirmed the existence of an antitumor response in breast cancer patients which contained autoantibodies specific for the TrkB growth factor receptor. Importantly, selected monoclonal antibodies derived from this population were functional. Although structurally similar, antibodies from 2 representative clones were shown to have markedly different effects on the TrkB growth factor receptor target. Antibody 93 was an agonist capable of stimulating tumor growth, while antibody 641, an antagonist selected on basis of an EMT–MET switch in vitro, was shown to inhibit tumor cell growth. Since EMT is a phenomenon observed in metastasis of many tumors, this phenomenon involved several signaling pathways, such as the TGF-β and Wnt/β-catenin signaling pathways. Other factors such as NF-κB and miRNA are also involved in EMTs. Likewise, autoantibodies that bind with different concentration, affinity, and epitope to self-antigens may present a different biology role in tumor cells. Therefore, it would be helpful to study the effect of coexisting antibodies in the EMT–MET switch or other signaling pathway in cancer. In other words, given the polyclonal nature of the autoantibodies in a patient’s plasma, and given the fact that any single antibody may have a positive or negative effect on tumor growth, the net effect may depend on the relative representation of individual autoantibody clones. This work suggests that a preponderance of agonist autoantibodies to a growth factor receptor would be deleterious and harmful to the cancer patient. Given this possibility, we suggest that antibodies to surface proteins on cancer cells are another parameter that should be measured in cancer patients.

TrkB has been found mutated in human breast, lung, colon, and melanoma cancers (17–19). In addition, the TrkB gene rearrangements and transcript that encode a truncated isoform were also reported in cancer cells. Recently, it was reported that TrkB plays a role in many cancer cells, including breast, lung, colon, and neuroblastoma tumor. Activation of TrkB receptors leads to a series of downstream signaling, including the PI3K/Akt, ERK, and PLCγ pathways. As a result, these pathways showed oncogenic effects by increasing cancer cell growth, proliferation, migration, epithelial–mesenchymal transition, and resistance to anoikis. These findings indicate that TrkB may be a useful target for therapeutic interventions for cancer cells (20–23).

One of the interesting features of our study is how closely related the agonist and antagonist antibodies are to each other. It seems that it is simply a matter of selection of a different germline and the same VDJ recombination. Taken together, our findings provide a perspective that agonist autoantibodies in cancer are deleterious and harmful. This is a + unrecognized driver of cancer cell growth, opening the possibility that removal of their autoantibodies may have a therapeutic benefit.

Materials and Methods

See SI Appendix, Materials and Methods for a detailed description of cell culture, construction of the lentivirus antibody library, lentivirus preparation, immunoprecipitation, purification, and characterization of the scFv-Fc antibody, Western blot, ELISA, siRNA transfection, biolayer interferometry assay, and cell proliferation assay.

Morphogenic Selection by Colony-Forming Cell Assay Using Methylcellulose-Based Media.

MDA-MB-231 cells were transduced with the lentiviral antibody library at an MOI of 2. The MDA-MB-231 cells transduced with the antibody library were added to methylcellulose media such that the methylcellulose final concentration was 1.27% and the cell concentration was ∼3 × 10⁴ cells/mL A total of 1.5 mL cell suspension was added to 35-mm-diameter dishes. The cells in soft agar were cultured for 2 to 3 wk. The colonies that showed unique morphologies were harvested with the aid of a micromanipulator (Sutter Instruments) and lysed with lysis buffer containing protease K for 1 h at 50 °C. The antibody genes from each colony were amplified by PCR with primer pairs customized for our lentiviral vector. The amplified antibody genes were analyzed by electrophoresis and recovered. After digestion with SfiI, the genes were ligated into the lentiviral vector and transformed into XL1-1 Blue competent cells. Bacteria colonies selected from each bacterial transformation were sequenced.

Xenograft Mouse Models.

The tumor orthotopic model was generated by injection of 1 × 10⁷ MDA-MB-231 human breast cells into the mammary fat pad in CB17/SCID female mice (n = 5). When tumors were palpable (60–100 mm³) around day 21, antibody 641 was administered at a dose of 10 mg/kg or 30 mg/kg twice weekly through intraperitoneal injection for 1 mo. PBS was injected in parallel for comparison. Tumor volume and body weight were measured twice a week. Tumor size was determined by Vernier caliper measuring length (L) and width (W), and the tumor volume (mm³) was calculated as 1/2 × LW². All of the animal studies were approved by the Institutional Animal Care and Use Committee (IACUC) (protocol no. 19–0005) of The Scripps Research Institute.

TrkB Reporter Cell Assay.

TkB-CRE-bla 293 or TkB-NFAT-bla CHO-K1 reporter cell lines were treated with BDNF or antibodies for 5–6 h. The adherent cells were loaded with CCF4-AM dye, treated with Accutase detachment solution, and subjected to fluorescence-activated cell sorting (FACS). Upon activation, the reporter cells expressed beta-lactamase, cleaved CCF4, and disrupted the FRET; excitation at 405 nm produced a florescent signal at 450 nm, while the green florescent signal at 520 nm was diminished.

Statistical Analysis.

Sample sizes, statistical tests, and definitions of error bars are indicated in the figure legends and were calculated using GraphPad Prism 8. Differences with P < 0.05 were considered statistically significant.

Data Availability.

The data that support the findings of this study are available in Figshare, https://figshare.com/ (reference number 10.6084/m9.figshare.10302227).