To clone the SF-25 variable region into a human IgG₁ backbone we performed a four-fragment Polymerase Incomplete Primer Extension (PIPE) PCR: two big fragments F2 and F4 (3000–4000 bp) containing at 5’ the constant regions of the light and heavy chain, respectively; and two small fragments F1 and F3 (300–400 bp) composed by SF-25 Vk and VH, respectively.10–12 The PCR reactions contained 0.5 µM of each primer, 25 µL of PhusionTM Flash High-Fidelity PCR Master Mix, 10 ng of template DNA and sterile water up to 50 µL. The cycles for amplification were: 10 s at 98°C, 35 cycles of 1 s at 98°C, 5 s at 62°C, and 10 s at 72°C. PCR products were treated with DpnI, bacteria were transformed with equal amount of the digested PCR products to combine the four fragments and plated in Luria Bertani (LB) agar plates supplemented with 200 µg/mL hygromycin B. Colonies were amplified overnight in LB supplemented with 200 µg/mL hygromycin B and DNA was extracted with a QIAprep spin miniprep kit (Qiagen). Correct assembly of the plasmid was verified by sending the newly generated and extracted plasmid for sequencing (Source Bioscience). The sequencing output was analyzed using FinchTV.

To obtain the pVitro1-SF-25 IgE expression vector, we designed the primers for a three fragment PIPE cloning protocol, amplifying the fragment containing the epsilon heavy chain constant region (PCR1) from an in-house representative pVitro1-IgE vector and the other two fragments (PCR2 and PCR3) from our pVitro1-SF25-IgG₁. The cycles for PCR1 and 2 were: 10 s at 98°C, 35 cycles of 1 s at 98°C, 5 s at 62°C, and 10 s at 72°C. PCR3 was performed with 10 s at 98°C, 35 cycles of 1 s at 98°C, 5 s at 62°C, and 8 s at 72°C. PCR products were separated on 1% agarose gels to discriminate the multiple PCR products by molecular weight. DNA was purified from the gel using the PureLinkTM Quick Gel Extraction Kit (ThermoFisher). PCR products were treated with DpnI to digest the template DNA. One-Shot TOP10 bacteria were transformed with equal amount of the digested PCR products to combine the four fragments (F1-F2-F3-F4) and generate the pVitro1 SF-25 IgG₁ vector. The correct assembly of the plasmid was verified via sequencing and analyzed using FinchTV. Antibodies were produced in Sp2/0 (IgE) and Expi293F (IgG) mammalian cells and purified using previously described methods (see online supplementalry materials and methods).11–13

Cell lines were detached with Trypsin-EDTA treatment, counted and 2.5×10⁵ cells were used per tube. 2 mL FACS buffer [Phosphate Buffer Saline (PBS); 5% Foetal Bovine Serum (FBS); 3 mM Ethylenediaminetetraacetic acid (EDTA)] was added to each tube before a 5 min centrifugation at 400rcf at 4°C. Cells were resuspended in 100 µL FACS buffer and incubated for 20 min at 4°C with a range of concentrations from 0 to 50 ng SF-25 IgG₁ or 0 to 5 µg SF-25 IgE. Cells were washed with 2 mL FACS Buffer and were then incubated with goat antihuman IgG-FITC (Fluorescein isothiocyanate) or antihuman IgE-FITC for 20 min at 4°C. Cells were washed with 2 mL FACS buffer and resuspended in 400 µL FACS buffer for cytometry analysis. Analysis of flow cytometry data was performed by FlowJo (TreeStar Inc) software.

SF-25 antibody binding scores were generated from three experimental datasets, two binding datasets generated for this study by Flow Cytometry and one previously published radioligand binding dataset.2 Transcriptome RNAseq datasets E-MTAB-2706; E-MTAB-2770 and E-MTAB-3983, including several of the human cancer cell lines present in the binding datasets, were downloaded and filtered to only keep the cell lines for which binding data were available. Further filtering and sorting steps were performed to generate matching tables between one binding dataset and the transcription level files. The matched tables were then analyzed, and Spearman correlation scores were calculated for each individual gene in each binding experiment. Average Spearman scores and their variances were generated across the three different combinations made. Further details on the bioinformatic process design and coding are reported in online supplementalry materials and methods .

Antibody dependent cell-mediated cytotoxicity/phagocytosis (ADCC)/ADCP assays were performed according to a previously described method.14 The ability of SF-25 IgE to trigger primary human basophil activation was determined using an ex vivo assay in which CD63 expression on the surface of human basophils was used as an early marker of basophil activation, as described previously (see online supplemental materials and methods).15

After 24 hours coculture, the viability of tumor cell monolayers was quantified. T cells were removed from the wells and MTT (3-(4,5-dimethylthiazol-2-yl)−2,5-diphenyltetrazolium bromide, Sigma) was added at 500 µg/mL in 200 µL complete DMEM medium and incubated for 1 hour at 37°C and 5% CO₂. After removal of the supernatant, formazan crystals were resuspended in 200 µL DMSO. Absorbance was measured at 560 nm using a spectrophotometric plate reader (FluoSTAR Omega) and tumor cell viability percentage was calculated as follows: (absorbance of coculture/absorbance of monolayer alone) x 100. Further details of CAR T in vitro and in vivo assays in online supplemental materials and methods.

We cloned the murine variable region sequences of SF-25 into human IgG₁ and IgE antibody scaffolds (figure 1A) using previously established platforms.10–12 High performance liquid chromatography (HPLC) analyses demonstrated high antibody purity and negligible aggregation (<5%) (figure 1B). Production of high-purity intact antibody was demonstrated in different expression systems, culture media, serum content and culture vessel conditions (online supplemental figure S1A, B). Serum-free, serum depleted, or specialist serum-free (ADCF, animal-derived component-free) media all led to high purity antibody with negligible protein aggregates or non-assembled light chain (online supplemental figure S1C,D). Intact antibody can be generated at small and large scale in serum-free conditions demonstrating utility for preclinical process development and clinical testing.

IgE class antibodies are highly glycosylated (12% of molecular weight). We evaluated IgE glycosylation in two SF-25 IgEs produced using mouse (Sp2/0) and human (Expi293F) expression systems.16 17 Lectin blot analyses were performed for the most common sugars known to decorate antibodies expressed in mammalian systems: mannose, fucose, galactose and sialic acid. No significant difference in fucose and mannose was observed between murine and human expression systems. Sp2/0-derived SF-25 IgE contained 2.5 times higher galactose and sialic acid content compared with IgE engineered in human Expi293F (figure 1B).

We confirmed the reactivity of each SF-25 chimeric SF-25 IgG₁ and IgE against a panel of human and non-human tumor cell lines. We showed largely comparable reactivity between IgG and IgE and variable binding across human tumor cell lines, likely reflective of target expression levels. Overall, 18 out of 19 human cancer cell lines from 7 different origins tested were recognized by SF-25 antibodies. Antigen density varied, but with clear population shifts demonstrated for the cell lines tested (figure 1C). No or low binding were detected to the non-human rat colon carcinoma cell line CC531.

Together, these findings confirm the generation of intact, monomeric human Fc IgG₁ and IgE SF-25, and production of functional IgE in serum free-culture mammalian expression systems suitable for future translation. We confirmed binding of the antibodies to human tumor cells of different malignant origins in agreement with reports of the original SF-25 mouse clone.

Three complementary approaches were employed to identify the target of SF-25. We performed immunoprecipitations with the chimeric SF-25 IgG1 on cell lysates from three different human cancer cell lines with differing SF-25 target expression levels (figure 1C): MDA-MB-231, MDA-MB-468 (breast cancer, BRCA) and A2058 (melanoma) (figure 2A-left panel). No specific protein band emerged by comparing these three samples based on the predicted target abundance. However, bands at 27 and 42 kDa were visible across the three cell lines. They were further analyzed by trypsin digestion and mass spectrometry. Peptides belonging to 138 different proteins were identified (online supplemental table S1). A transcriptomic analysis was performed to compare binding levels on panels of human cancer cell lines against transcript levels for each gene in the same cell lines (figure 2A-right panel). Three sets of binding scores were generated reflecting the binding intensity measured for the SF-25 antibody on each cell line, counts per minute for previous radioligand binding data,2 and mean fluorescent intensity (MFI) for the experiments in figure 1C. The relative binding scores for the three binding panels are shown in online supplemental table S2. The binding intensities observed across different cell lines in one binding experiment were compared with transcript levels in aggregate RNA-seq data for the same cell lines. A Spearman correlation score was calculated between SF-25 binding and transcript levels for each gene and average Spearman scores and variance values were calculated from the three comparative studies. The heavy chain of CD98 (CD98hc) encoded by the solute carrier family 3 member 2 (SLC3A2) demonstrated the highest Spearman score (figure 2A-bottom panel). The gene SLC7A5, coding a binding partner of CD98hc, was the second highest hit. Both partners were identified among the candidates immunoprecipitated (figure 2A–bottom panel). SLC3A2 was also identified among 138 different proteins after immunoprecipitation, offering a weak biochemical validation. We, therefore, identified SLC3A2 as the top candidate.

In separate immuno-mass spectrometry experiments, SF-25 IgG₁ reactivity was tested against human cutaneous and tumor antigens. This immuno-mass spectrometry method can be applied to discover proteome-wide targets of antibodies by using complex protein mixtures from human tissues as sources of candidate proteins.18 Tissue lysates were generated from three pooled human skin samples, two human cutaneous metastatic melanoma lesions, and the human ovarian carcinoma cell line IGROV1. In two independent experiments, applying filtering for contaminates and non-specific binding using a non-binding (hapten-specific) antibody (figure 2B), 10 and 7 peptides corresponding to SLC3A2 were identified from the three skin samples. Separately, 18 peptides from two human melanoma lesion samples and 16 peptides from the IGROV1 cell line identified recognition of SLC3A2 by the antibody (figure 2B). The average peak area of binding to melanoma cells and normal skin across two independent experiments was significantly higher in malignant cells for SLC3A2, demonstrating highest expression of the target in tumor samples (figure 2C). In concordance, the identified SLC3A2 peptides (online supplemental figure S2) were demonstrated at significantly higher levels in melanoma vs normal skin (p<0.05, data from three normal skin and two melanoma samples tested twice each in two independent experiments) (figure 2C). Finally, using single guide CRISPR Cas-9 knock out of the SLC3A2 gene in A2058 cells, comparable loss of binding was seen with SF-25 IgG and a commercially available CD98hc specific monoclonal antibody (figure 2D).

These data identify the protein CD98hc, coded for by SLC3A2, as the target of the SF-25 antibody clone both bioinformatically and biochemically.

It was previously reported that the murine SF-25 clone demonstrated normal human tissue binding against a subpopulation of cells in the distal tubules of the kidney.2 Using RNA-sequencing raw data from The Cancer Genome Atlas (TCGA) and the Genome Tissue Expression (GTex) we examined the expression of SLC3A2 in malignant vs normal tissues across a range of human cancers (figure 3A and online supplemental figure S3). Tumors were selected for analysis when adequate equivalent normal tissue data were available. For the majority of tumors assessed, significantly higher levels of SLC3A2 expression were observed in malignant vs normal tissues, most notably in colorectal, breast, genitourinary cancers and cutaneous melanomas (figure 3A). When considering primary vs metastatic tumors, analysis was hampered by the paucity of metastatic tumors covered by TCGA and GTex. Data were analyzed for melanoma (SKCM) and breast cancer (BRCA) where notably only melanoma had a large resource of metastatic disease available (figure 3B). These data showed that SLC3A2 overexpression was maintained in metastatic deposits of melanomas (figure 3B). True TAAs, limited only to tumor tissue, are very hard to come by in solid tumor oncology. Instead, we develop targeted immune therapy directed against antigens where expression is largely higher in malignant vs normal tissue. To explore the levels of SLC3A2 in relation to established TAAs, we compared SLC3A2 expression levels with those recorded for HER-1 and HER-2 in normal skin and breast tissues. HER-1 and HER-2 were selected as key examples of antigens targeted safely and successfully in the oncology clinic with immune therapy strategies including antibody and cell therapy approaches. SLC3A2 expression, in these normal tissues, is similar or lower than HER-1 and HER-2 expression levels (figure 3C).

We next evaluated the reactivity of the antibody clone against human malignant and non-malignant tissues by immunohistochemistry. Directly labeled SF-25 IgG₁-AF488 showed reactivity against melanoma but not against normal skin samples (figure 4A). Tissue microarray immunofluorescence analyses showed no/restricted reactivity in most normal tissues. Consistent with previous data, we confirmed kidney binding and revealed lower-level antibody bindings in testis, and at low levels in human cerebellum (figure 4B). In contrast, robust staining was observed in human malignant tissues, including melanoma, breast, ovarian, testis and soft tissue cancers (figure 4C). Our transcriptomic and immunohistochemical analyses (figures 3 and 4) are consistent with previous reports of SF-25 antibody reactivity against several cancer types versus normal tissues.

To investigate reports of SLC3A2 expression in human peripheral blood mononuclear cells (PBMCs) we stained PMBCs with the MEM-108 anti-CD98hc monoclonal antibody, demonstrating binding that was significantly increased by activation of PBMCs with Phytohemagglutinin (online supplemental figure S4).

The protein derivative of SLC3A2 is an established heterodimerization partner for multiple solute carriers. SLC3A2/CD98hc and two L-type amino acid transport binding partners, LAT1 and ascAT1 (derived from SLC7A5 and SLC7A10, respectively), were analyzed independently as prognostic markers in bladder, breast, cervical, lung, renal and head and neck cancers. Survival over 10 years from diagnosis was assessed by the Kaplan-Meier method for high and low expressing tumors (online supplemental figure S5). Diversity in prognostic value of SLC3A2 and two of its many binding partners highlights the breadth of impact potentially associated with targeting this key heterodimerization TAA.

These findings, at the transcriptomic and cell surface proteomic levels, support SLC3A2 as a TAA and confirm reactivity of the human Fc engineered antibody to human tumor tissues of different origins.

Since IgE antibodies may offer an alternative immunotherapy approach for solid tumors, we investigated the antitumor functions of the engineered SF-25 IgE antibody in vitro.

To determine direct effects of SF-25 IgE on target cancer cells, colony formation assays with A2058, IGROV1 and PaTu8988t cells were performed. SF-25 IgE did not impair the ability of cancer cells to form colonies (figure 5A).

We next investigated whether SF-25 IgE could trigger Fc-mediated effector functions through the high affinity Fc receptor FcεRI. Mast cells and basophils are known to express FcεRI and to participate in parasite clearance through IgE.19 The rat basophilic leukemia RBL SX-38 in vitro mast cell model expressing human FcεRI, was used to examine the Fc-mediated biological activities of SF-25 IgE.20 We investigated the Fc-mediated functions of SF-25 IgE to trigger mast cell degranulation when cross-linked by multiple copies of its target antigen expressed on the surface of cancer cell lines A2058, IGROV1, SKBR-3 and LS-180. Degranulation, measured by β-hexosaminidase release, in the presence of non-specific control IgE was minimal, while SF-25 IgE triggered significant mast cell degranulation in the presence of different target expressing cancer cells (figure 5B). ADCC was measured in coculture experiments using human effector cell and target cell fluorescence reporters (online supplemental figure S6A,B). In the presence of healthy volunteer or melanoma patient PBMCs, SF-25 IgE engendered significant A2058 melanoma cell killing above non-specific isotype IgE or no antibody controls (figure 5C).

In vitro, SF-25 IgE did not affect proliferation or the clonogenic ability of cancer cells. The antibody exerted Fc-mediated effector functions via the high affinity FcεRI receptor. When cross-linked by multiple copies of an antigen expressed on the target cell surface, SF-25 IgE mediated specific mast cell degranulation and triggered cytotoxicity of cancer cells by both healthy volunteer and cancer patient immune cells.

An IgE antibody introduced in the human circulation could bind to FcεRI-expressing basophils in the blood. If basophil bound IgE is cross-linked by signals such as multivalent soluble circulating antigen or by antibodies recognizing circulating multivalent antigen, this could trigger basophil degranulation, potential hypersensitivity and the onset of systemic anaphylaxis. We therefore evaluated the potential of SF-25 IgE to trigger basophil degranulation in human cancer patient blood using the basophil activation test (BAT). The BAT assay is used to monitor for hypersensitivity to the first-in-class IgE immunotherapeutic as part of its clinical development.9 15 21 Whole blood samples from ovarian cancer patients were incubated with either stimulation controls: anti-FcεRI antibody to crosslink the IgE receptor, fMLP (N-Formylmethionyl-leucyl-phenylalanine), a polyclonal activator of human basophils, or polyclonal anti-human IgE to cross-link endogenous IgEs already bound to the surface of human blood basophils. Basophil populations were identified by expression of CCR3 (online supplemental figure S5A). Basophil activation was evaluated by detection of CD63 cell surface expression, a marker normally absent in resting basophils (online supplemental figure S6C,D). Marked levels of basophil activation were observed when ovarian cancer whole blood samples were incubated with the three positive stimulation controls. These data confirmed propensity for patient basophil activation by both IgE and non-IgE-mediated mechanisms (figure 5D). In contrast, the incubation of blood with SF-25 IgE did not lead to basophil activation above a threefold threshold (figure 5D inset) in any of the 23 cancer patient samples tested.

These findings suggest that SF-25 IgE could not trigger basophil activation in a functional ex vivo assay, demonstrating no hypersensitivity reaction in cancer patient whole blood. These findings provide early evidence to support safe administration of this antibody to patients.

To generate an SLC3A2-specific CAR, we first sequenced the variable regions of the heavy and light chain of the SF-25 clone and performed sequence modifications to the framework 1 region of both chains. The resultant single-chain variable fragments (scFv) were fused to a modified CD28 hinge containing a myc-tag, CD28 transmembrane and costimulatory intracellular domains and CD3ζ stimulatory domain (4SFm28ζ).22 A control CAR, truncated at the intracellular domain of CD28, was also engineered (4SFm28Tr). Retroviral vector cassettes were generated with the 4αβ chimeric cytokine receptor upstream of the CARs separated by a T2A sequence (figure 6A). The 4αβ (interleukin 4, IL4/2R) enables selective expansion and enrichment of CAR-positive T cells after transduction through delivery of an IL-2 binding signal in response to IL-4.23 After 10 days in IL-4 supplemented culture, robust CAR expression was seen (figure 6B).

High (melanoma—A2058), middle (prostate—PC3-LN3, breast—MDA-MB-468) and low (breast—MDA-MB-231) SF-25 expressing cell lines (figure 1C) were used for in vitro cytotoxicity analysis at effector to target (E:T) ratios of 5:1, 1:1 and 1:5. Significant CAR specific cytotoxicity was demonstrated at 24 hours at all coculture densities for the A2058 and MDA-MB-468 cell lines. For PC3-LN3 4SFm28ζ cytotoxicity was significant compared with 4SFm28Tr at the 5:1 E:T ratio and against untransduced T cells at 5:1 and 1:1. For the low density MDA-MB-231 cell line significantly superior 4SFm28ζ cytotoxicity was seen vs 4SFm28Tr at E:T 5:1 and untransduced T cells at E:T 1:1 (figure 6C). Culture supernatants probed for and IL-2 demonstrated donor variability. At an E:T of 1:1 at 24 hours Il-2 levels were significantly higher for 4SFm28ζ compared with both controls against PC3-LN3, MDA-MB-231 and A2058. Interferon gamma was raised for 4SFm28ζ compared with controls in all cocultures (figure 6D). These data demonstrate signaling CAR T restricted antigen specific in vitro efficacy of SF-25 derived CAR T cells.

Using coculture at an E:T ration of 1:2 on the prostate cancer monolayer PC3-LN3 4SFm28ζ, 4SFm28Tr and untransduced T cells were probed for phenotype and markers of activation (CD69 and PD-1) prior to and after exposure to monolayers. MFI of CD69 significantly increased on 4SFm28ζ CAR T cells after exposure to antigen compared with controls. Accordingly, PD-1 percentage increased post antigen exposure for the signaling CAR (figure 6E). On exposure to PC3-LN3, activation marker levels increased for the 4SFm28ζ signaling CAR only, commensurate with antigen specific activation (figure 6E). Viral transduction alone (for 4SFm28ζ and 4SFm28Tr) resulted in loss of naïve populations and expansion of both T effector memory (TEM) and T effector memory-RA (TEMRA) populations compared with the activated but untransduced controls (figure 6F). Allowing for the natural variability in donors, the impact of engagement of antigen by CAR T cells resulted in further enrichment of TEMRA (figure 6E). These data demonstrate robust, antigen specific, activation of 4SFm28ζ CAR T cells.

Immunodeficient NSG mice were intravenously injected with human LS-180 colorectal cancer cells which led to the development of cancer lesions in animal lungs. Freshly isolated human PBMCs were adoptively transferred in the presence or absence of SF-25 IgE. Antibody treatment alone was readministered twice more (figure 7A). After 3 weeks, tumor load in the lungs of mice was measured (figure 7B). Analyses demonstrated significantly restricted tumor growth in SF-25 IgE-treated mice, both in relation to the number of metastatic foci (p=0.0028) and with regard to the area of tumor occupancy in animal lungs (p=0.0061) compared with PBMC treated controls (figure 7C,D).

For CAR T in vivo efficacy testing, PC3-LN3-luciferase subcutaneous xenografts were established in NSG mice. Tumor engraftment was confirmed by bioluminescence imaging. Tumor growth was monitored by caliper measurements as central tumor necrosis limits bioluminescence accuracy. Tail vein injections of 4SFm28ζ and 4SFm28Tr CAR T cells were undertaken 3 days after tumor inoculation (figure 7E). A significant increase in overall survival was seen in the 4SFm28ζ group (p=0.0208) (figure 7F). Of five, two treated mice demonstrated tumor eradication after initial growth, with one showing initial response followed by tumor escape (figure 7G). The animals displayed no signs of weight loss or manifest any symptoms of cytokine release syndrome or graft-versus-host disease (online supplemental figure S7).

These findings collectively support the functional capability of novel immunotherapies targeting SLC3A2 to restrict the growth of tumors and improve survival in vivo in the absence of overt toxic effects.