Bioinformatic pipeline to identify potential therapeutic targets with subsequent isolation and characterization of novel human anti- DDR1 antibodies Authors: Divya Agrawal, Stephen M. Mahler, Anurag S. Rathore, Martina L. Jones, Jessica C. Mar, Lucía F. Zacchi Published: 2026 ABSTRACT Breast cancer remains a major global health challenge, driving the need for safer and more effective targeted therapies. This study aimed to systematically identify therapeutic targets for breast cancer using an integrated bioinformatics approach and to generate novel human monoclonal antibodies against an identified potential target followed by characterization. A multi-step bioinformatic pipeline was developed to analyse paired tumour and matched non-cancerous breast tissue samples from The Cancer Genome Atlas. Differential expression analysis, target selection criteria, and surface-protein enrichment identified several dysregulated and therapeutically relevant candidates, including established markers such as HER2, supporting the robustness of the approach. From the shortlisted targets, epithelial discoidin domain-containing receptor 1 was selected for antibody discovery. A naïve human phage display library was screened to isolate DDR1-specific single-chain variable fragments that were reformatted into full-length human IgG1. These antibodies were expressed, purified, and characterized through biophysical evaluation and in vitro assays. Several candidates demonstrated high-affinity binding to native DDR1 on breast cancer cells and showed the ability to mediate tumour cell lysis through antibody-dependent cell cytotoxicity. Overall, this study highlights the effectiveness of combining publicly available omics datasets with phage display technology to identify and develop new therapeutic antibody candidates for breast cancer.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-026-56361-4. Introduction Despite significant advances, cancer remains a challenging disease to manage. Breast cancer accounts for the highest number of new cancer cases every year in women compared to any other cancer type1. Monoclonal antibodies (mAbs) are highly effective treatments for cancer because they can precisely target cancer cells while strengthening the immune system’s response against these cancer cells2. Notable examples include rituximab and trastuzumab, which target CD20 and Her2, respectively, and have demonstrated substantial therapeutic efficacy3,4. However, although these antibodies boost a patient’s immune response to target cancer, they do not always guarantee complete remission from the disease, and these treatments are often followed by relapse with high mortality5. A key challenge for developing effective cancer therapies lies in identifying suitable tumor-associated antigens that are both highly specific to cancer cells with minimal or no expression on normal tissues to prevent off-target effects, and sufficiently exposed or accessible on the cell surface to enable effective antibody targeting6. The barriers to identifying antigens that satisfy these properties is that, in tumours, antigen expression can be heterogeneous and variable, thereby complicating the identification of reliable therapeutic targets7. To overcome these hurdles, one strategy is to combine multiple sources of input data from genetics, gene expression, and functional data to identify potential targets. One low-cost approach to identifying new therapeutic targets is to leverage publicly available omics datasets from large cancer patient cohorts. Indeed, several computational pipelines that mine public data have been developed to identify suitable therapeutic targets8–10. However, bioinformatic pipelines may overlook factors like the complexity of disease heterogeneity and the influence of the tumor microenvironment, which are critical to determining whether a target can effectively drive tumor remission11. For instance, tumor-associated macrophages, stromal interactions, and immune checkpoint activity can modulate target expression and therapeutic response, ultimately affecting the success of targeted therapies12. Regardless of the numerous advantages that a computational approach brings, it is important to highlight that bioinformatics-based predictions nevertheless require experimental validation. In the case of RNA sequencing studies, inferences are often derived from bulk tissue samples or single cell studies which lack the ability to associate individual gene expression values with spatial context within a tumor sample. While spatial transcriptomics data exists for cancer tumours, data available for large patient cohorts is only RNA-seq and other bulk-based sequencing assays13,14.It is well-established that there is not always good correlation between mRNA levels and protein expression levels due to post-transcriptional, translational, and post-translational regulation15. As such, it is necessary to validate the functional predictions of genes identified through mRNA-based studies. Gene expression data does not provide information on where a protein is localized inside a cell and, in cancer, proteins can relocate without any change in gene expression levels16,17. Finally, gene expression data does not capture post-translational modifications or quaternary structure of proteins, which provide important context for protein function and stability, both valuable metrics to consider for disease progression. These limitations reinforce the fact that cancer is regulated by multiple levels of the genome, that experimental validation is a critical tool to include when identifying therapeutic targets, and that collaborative, interdisciplinary approaches are required to arrive at robust inferences. In this study, we developed a bioinformatic pipeline capable of handling large cancer mRNA and protein datasets using the breast cancer patient cohorts from The Cancer Genome Atlas Program (TCGA), to identify potential anti-cancer targets, and combined this with an experimental pipeline to isolate and characterise a monoclonal antibody developed against the target that was identified from this computational prediction. Our pipeline leverages the paired design where tumor and non-tumor tissues are sampled from the same donor, a strategy that improves precision of the inferences because gene expression changes in the tumor are assessed relative to non-tumor tissue from the same donor. Here, we combined RNA-sequencing (RNA-Seq) and Reverse Phase Protein Array (RPPA) proteomic datasets from paired normal and tumor tissue samples from the TCGA breast cancer cohort18. RPPA is a high throughput, reproducible, and sensitive assay that quantifies protein levels directly from tissue lysates using antibodies. RPPA enables quantitative profiling of protein expression within the predefined antibody panel and post-translational modifications (PTMs), preserving functional proteomic profiles in tumors19. To identify targets with lower risk of off-target effects on healthy tissues, we focused on proteins with significantly higher abundance on tumor cells compared to normal tissues (preferentially with limited or no expression on normal tissues). Next, we used a variety of resources from the public domain, including data of tumor resistance to current therapies, to support the validity of the identified candidate potential targets. Using this pipeline and a set of stringent selection criteria for targets that would be suitable for antibody therapy, we identified DDR1 as a potential therapeutic target for breast cancer (Fig. 1).Fig. 1Schematics of the workflow followed in the study. (a) Data collection and analysis: TCGA Breast tumor and normal tissue donor ID’s (samples) retrieved to identify a subset of donors with both RNA Seq and RPPA datasets. Differential expression analysis was performed to identify potential common cell surface targets (“n”), and mRNA and proteins were annotated using HUGO gene symbols. “n” targets were narrowed down using a set of selection criteria and were validated through several bioinformatics databases. (b) Monoclonal antibody discovery and development: phage display biopanning using a human naïve library was conducted against a target of interest, and clones were screened using ELISA and flow cytometry. Promising candidates were re-formatted and expressed as full-length monoclonal antibodies (mAbs) and characterized by flow cytometry, SPR, and ADCC assay. GO: Gene Ontology, KEGG: Kyoto Encyclopedia of Genes and Genomes, HPA: Human Protein Atlas, GEPIA: Gene Expression Profiling Interactive Analysis, RPPA: Reverse phase protein array, HUGO: Human Genome Organization, ELISA: Enzyme linked Immuno-sorbent assay, SPR: Surface plasmon resonance, ADCC: Antibody-dependent cellular cytotoxicity (Created in BioRender. Agrawal, D. (2026) https://BioRender.com/bgzshgj). Schematics of the workflow followed in the study. (a) Data collection and analysis: TCGA Breast tumor and normal tissue donor ID’s (samples) retrieved to identify a subset of donors with both RNA Seq and RPPA datasets. Differential expression analysis was performed to identify potential common cell surface targets (“n”), and mRNA and proteins were annotated using HUGO gene symbols. “n” targets were narrowed down using a set of selection criteria and were validated through several bioinformatics databases. (b) Monoclonal antibody discovery and development: phage display biopanning using a human naïve library was conducted against a target of interest, and clones were screened using ELISA and flow cytometry. Promising candidates were re-formatted and expressed as full-length monoclonal antibodies (mAbs) and characterized by flow cytometry, SPR, and ADCC assay. GO: Gene Ontology, KEGG: Kyoto Encyclopedia of Genes and Genomes, HPA: Human Protein Atlas, GEPIA: Gene Expression Profiling Interactive Analysis, RPPA: Reverse phase protein array, HUGO: Human Genome Organization, ELISA: Enzyme linked Immuno-sorbent assay, SPR: Surface plasmon resonance, ADCC: Antibody-dependent cellular cytotoxicity (Created in BioRender. Agrawal, D. (2026) https://BioRender.com/bgzshgj). DDR1 (also known as CD167a) is a single pass type-I transmembrane protein and a receptor tyrosine kinase (RTK) that is upregulated in highly invasive epithelial-like breast carcinoma cells and other solid tumors20,21. To date, several therapeutic approaches targeting DDR1 have been developed, including an antibody-drug conjugate for colon carcinoma, as well as repurposed FDA-approved tyrosine kinase small molecular inhibitors such as imatinib, dasatinib, and nilotinib22–28. Furthermore, over four commercially available anti-DDR1 antibodies for diagnostics and eight candidate therapeutic antibodies exist, some of which have advanced to clinical trials for the treatment of breast cancer, non-small cell lung carcinoma, colorectal cancer, and gastric cancer29. Despite these developments, current therapeutic strategies face limitations. For example, in the triple-negative breast cancer (TNBC) patient subgroup, small-molecule DDR1 inhibitors have shown limited effectiveness and off-target cytotoxicity30. Some anti-DDR1 mAbs under development primarily act by inhibiting RTK signalling31, which limits tumor growth but cannot induce complete tumor regression. The challenges with current DDR1-associated therapies point to the need for more effective therapies to enhance patient outcomes. In summary, here we present an integrated workflow for target identification that includes a bioinformatic pipeline to identify potential targets from publicly available mRNA and protein omics data with matching same-donor disease-control samples, and selection criteria to focus target selection for antibody-dependent therapeutic/diagnostic applications. Using this pipeline, we generated a list of potential targets for breast cancer therapy, and focused on DDR1/CD167a, a target with appropriate characteristics to facilitate mAb discovery and downstream applications. We then identified new antibodies targeting the extracellular domain of DDR1/CD167a and demonstrated their in vitro cytotoxicity using different breast cancer cell lines. These new anti-DDR1 antibodies illustrate the power of this pipeline to identify targets and generate novel molecules with therapeutic/diagnostic potential. Results Generating a bioinformatics pipeline to identify potential therapeutic targets We sought to generate a bioinformatics pipeline that combines transcriptomic and proteomic data from published sources to generate a list of potential anti-cancer targets. RNA levels do not always match protein levels because cells can regulate protein abundance post-transcriptionally through various mechanisms15,32,33. By combining both RNA and protein data in a single analysis, we can identify biologically important targets that would be missed if we looked at either data type alone. In this work, we applied this pipeline to breast cancer due to its high global incidence and relevance1. The transcriptomics and proteomics data were obtained from tissue samples from the ductal and lobular region of the breast, where most Breast Cancer (BRCA) adenocarcinomas form34. Previous studies such as Huang et al.35 and Hu et al.36 demonstrated the utility of paired tumor-normal samples in uncovering prognostic transcriptional signatures and clustering of expression patterns across cancers, respectively. The paired sample design is especially advantageous as it reduces patient-to-patient variation (the normal tissue samples are derived from non-tumor tissue of the same patient from whom the tumour sample was extracted). Thus, building on these strategies, our pipeline integrates transcriptomic and proteomic profiles from paired samples. Further, to increase the likelihood of finding true targets across a heterogenous population, we cross-referenced our results using external databases including GeneCards, NCBI (Gene/Protein), KEGG, GEPIA, HPA, UniProt, and DRUGBANK, among others (see methods). The pipeline used in this study is depicted in Fig. 1. The data used for the analysis presented here is available from The Cancer Genome Atlas Program (TCGA). The combined transcriptomic (113 patients) and proteomic (33 patients) data was stratified based on patient’s sex and age, and cancer stage and subtype, as shown in Supplementary Table S1T1. The 113 paired transcriptomic samples analyzed (Supplementary Table S1T1 and Fig. 2a) provided a total of 60,660 mRNAs which were first annotated as protein-coding genes using the HGNC (The Human Genome Organisation (HUGO) Gene Nomenclature Committee) identifiers present in transcriptomic metadata, resulting in 19,962 mRNAs (Fig. 2b, computed logCPM value for visualization as a density plot, where CPM is counts per million). Next, we filtered low-abundance genes, where we calculated CPM values for each gene across all samples. Genes were then flagged as TRUE in samples in which their CPM exceeded 0.5. Finally, only genes with CPM > 0.5 in at least 10 from all 226 samples were retained (R=0.0779 before normalization). To reduce noise, we applied Trimmed Mean of M-values (TMM) normalization (R= 0.0774) to correct for compositional bias, leading to 16,404 mRNAs. Using edgeR37, we identified 8,040 mRNAs that were significantly differentially expressed with higher abundance in tumor vs normal tissue, 4,791 mRNAs that were significantly differentially expressed with lower abundance in tumor vs normal tissue, and 3,573 mRNAs with no significant changes between tumor vs normal tissue in their gene expression values (adjusted P-value < 0.05, and Fig. 2c in green, red, and blue, respectively).Fig. 2RNA-Seq analysis of TCGA tumor and normal breast cancer tissue samples. (a) Venn diagrams showing the overlap of patient samples available in TCGA RNA-Seq data, resulting in 113 tumor-normal tissue pairs that were used in the primary analysis. Paired primary tumor (PT), solid tissue normal (STN), and metastatic tumor (MT) donor IDs. (b) Plots showing the effects of filtering and normalization on the gene expression data (see Methods). (c) Volcano plot showing mRNA differential gene expression (based on log2 fold change value), with higher gene expression in green (log2FC > 0.1 when comparing tumor vs normal), lower gene expression in red (log2FC < −0.1 when comparing tumor vs normal), and non-significant changes in blue (where P-value > 0.05, represented by the dotted line). The mRNAs with the highest and lowest expression, based on their fold change values, are labelled in boxes. RNA Seq analysis: Quasi-Likelihood F-test (edge R package); Multiple testing correction to control the False Discovery Rate (FDR) was applied to both analysis: Benjamini-Hochberg (BH) correction. RNA-Seq analysis of TCGA tumor and normal breast cancer tissue samples. (a) Venn diagrams showing the overlap of patient samples available in TCGA RNA-Seq data, resulting in 113 tumor-normal tissue pairs that were used in the primary analysis. Paired primary tumor (PT), solid tissue normal (STN), and metastatic tumor (MT) donor IDs. (b) Plots showing the effects of filtering and normalization on the gene expression data (see Methods). (c) Volcano plot showing mRNA differential gene expression (based on log2 fold change value), with higher gene expression in green (log2FC > 0.1 when comparing tumor vs normal), lower gene expression in red (log2FC < −0.1 when comparing tumor vs normal), and non-significant changes in blue (where P-value > 0.05, represented by the dotted line). The mRNAs with the highest and lowest expression, based on their fold change values, are labelled in boxes. RNA Seq analysis: Quasi-Likelihood F-test (edge R package); Multiple testing correction to control the False Discovery Rate (FDR) was applied to both analysis: Benjamini-Hochberg (BH) correction. For the proteomics analysis, we analyzed 33 paired samples (Supplementary Table S1T1 and Fig. 3a) with a total of 487 proteins. By following a similar processing approach to the transcriptomic analysis described above, ~1700 missing values across samples were omitted (R= −0.77) and quantile normalization was applied (R= −0.59), resulting in a refined set of 457 proteins (Fig. 3b). Differential protein abundance was determined using limma38 linear model, identifying 111 proteins with statistically significant higher abundance in tumor vs normal tissue, 156 proteins with statistically significant lower abundance in tumor vs normal tissue, and 190 proteins with no significant changes in abundance in tumor vs normal tissue (adjusted P-value < 0.05, and Fig. 3c, in green, red, and blue, respectively). Overall, within the filtered set, ~79% of mRNAs and ~58% of proteins exhibited significant differential abundance when comparing cancer to normal tissues.Fig. 3Proteomic analysis of TCGA tumor and normal breast cancer tissue samples. (a) Venn diagrams showing the overlap of patient samples available in TCGA RPPA data, resulting in 33 tumor-normal tissue pairs that were used in the primary analysis. Paired primary tumor (PT), solid tissue normal (STN), and metastatic tumor (MT) donor IDs. (b) Plots showing the effects of filtering and normalization on the proteomic data (see Methods). (c) Volcano plot showing protein differential expression (based on log2 fold change value), with higher protein expression in green (log2FC > 0.1 when comparing tumor vs normal), lower protein expression in red (log2FC < −0.1 when comparing tumor vs normal), and non-significant changes in blue (where P-value > 0.05, represented by the dotted line). The proteins with the highest and lowest expression, based on their fold change values, are labelled in boxes. Proteomic analysis: linear model with an Empirical Bayes moderated t-test (limma package); Multiple testing correction to control the False Discovery Rate (FDR) was applied to both analysis: Benjamini-Hochberg (BH) correction. Proteomic analysis of TCGA tumor and normal breast cancer tissue samples. (a) Venn diagrams showing the overlap of patient samples available in TCGA RPPA data, resulting in 33 tumor-normal tissue pairs that were used in the primary analysis. Paired primary tumor (PT), solid tissue normal (STN), and metastatic tumor (MT) donor IDs. (b) Plots showing the effects of filtering and normalization on the proteomic data (see Methods). (c) Volcano plot showing protein differential expression (based on log2 fold change value), with higher protein expression in green (log2FC > 0.1 when comparing tumor vs normal), lower protein expression in red (log2FC < −0.1 when comparing tumor vs normal), and non-significant changes in blue (where P-value > 0.05, represented by the dotted line). The proteins with the highest and lowest expression, based on their fold change values, are labelled in boxes. Proteomic analysis: linear model with an Empirical Bayes moderated t-test (limma package); Multiple testing correction to control the False Discovery Rate (FDR) was applied to both analysis: Benjamini-Hochberg (BH) correction. To compare the data from both transcriptomic and proteomic data sets, we adopted The Human Genome Organisation (HUGO) gene symbols as the common annotation identifier (Supplementary Files S2-4). HUGO gene symbols were selected due to their widespread use and recognition as the official nomenclature for human genes, facilitating linkage of data using different gene/protein identifiers39. In this way, RefSeq IDs, Ensembl IDs, and UniProt IDs were mapped to their corresponding HUGO gene symbol by cross-referencing multiple databases, including HGNC for official gene names, GeneCards (The Human gene database) for comprehensive gene information, and NCBI Gene and Protein databases for verifying gene-protein identifier consistency. To ensure accurate and unambiguous mapping to gene-level identifiers some proteomic targets were excluded from the analysis, including targets identified with antibodies against specific post-translationally modified forms of the targets (such as phosphorylated, acetylated, and methylated isoforms), or if targets were protein complex subunits (due to lack of one-to-one mapping with a single gene identifier or a complex-level regulation of abundance), or if the target names contained non-standard characters or punctuations (e.g., slashes, hyphens, or asterisks) (Fig. 4a). 335 proteins were further reduced to 333 unique HUGO gene symbols, with CASP8 and NOTCH1 each mapped twice by two distinct proteins: duplicates were identified, and only one representative for each was retained in the final set. In parallel, the 16,404 mRNAs were reduced to 16,403 unique HUGO gene symbols after removing one redundant instance of POLR2J3 mRNA. When we compared the 16,403 mRNAs with the 333 proteins (Figs. 2-4), we observed an overlap of 320 potential targets (~96% of the filtered proteins, and ~2% of the filtered mRNAs) (Fig. 4a, Supplementary File S5). Notably, the expression levels of the mRNA or proteins did not always correlate, showing that only 118 mRNAs and proteins (~37% of the 320 potential targets) shared a similar significant increase, decrease, or no-significant change in the tumor vs normal tissue comparison (Fig. 4b, Supplementary File S5). This result was not surprising given the prior observed lack of correlation between transcriptomic and proteomic datasets due to post-transcriptional regulatory mechanism, as discussed above15,32,33. Importantly, transcriptomic analysis defined the initial candidate identification space, while RPPA data was integrated as an orthogonal protein-level validation layer. This integrative strategy enhances confidence in candidate selection while acknowledging that RPPA coverage is restricted to proteins represented within the antibody panel. To identify any common biological themes that were evident in the 320 hits identified, we performed over representation analysis using pathway definitions from the Kyoto Encyclopaedia of Genes and Genomes (KEGG)40 and Gene Ontology (GO)41. We found significant enrichment of key cancer-associated KEGG pathways, including the PI3K-Akt signalling pathway, apoptosis, and MAPK signalling pathway (Supplementary Fig. S1F1a). These pathways are frequently dysregulated in cancer and are key regulators of cell survival, proliferation and apoptosis42. We identified enrichment in multiple GO categories, such as extracellular space, plasma membrane, receptor complex, transmembrane receptor protein tyrosine kinase activity, and collagen binding (Supplementary Fig. S1F1b). These categories highlight proteins that are either embedded in or associated with the cell membrane and are therefore potentially accessible to therapeutic antibodies, as discussed below. In summary, we performed a comparative analysis of transcriptomic and proteomic data from paired, same patient tumor-normal tissue samples, identifying 320 common transcriptomic and proteomic hits.Fig. 4Summary of mapped targets and their transmembrane subtype distribution: (a) Schematic representation of the workflow to identify potential targets from omics data. (b) Stacked bar plot showing the 320 total mapped targets and 55 cell surface potential targets separated according to whether they were single pass type I, II, III or multi-pass transmembrane proteins and according to their abundance in the tumor vs normal tissue comparisons. The graph legend shows abundance (significantly lower, non-significant, or significantly higher) in tumor vs normal tissue of the protein data on the left and the mRNA data on the right (protein | mRNA). For example, Lower | Lower means significantly lower abundance of a particular target in both protein and mRNA data in tumor vs normal tissue. Summary of mapped targets and their transmembrane subtype distribution: (a) Schematic representation of the workflow to identify potential targets from omics data. (b) Stacked bar plot showing the 320 total mapped targets and 55 cell surface potential targets separated according to whether they were single pass type I, II, III or multi-pass transmembrane proteins and according to their abundance in the tumor vs normal tissue comparisons. The graph legend shows abundance (significantly lower, non-significant, or significantly higher) in tumor vs normal tissue of the protein data on the left and the mRNA data on the right (protein | mRNA). For example, Lower | Lower means significantly lower abundance of a particular target in both protein and mRNA data in tumor vs normal tissue. Target selection criteria The selection of suitable targets is a critical part of therapeutic mAb development, aiming to facilitate mAb isolation against the target and decrease chances of adverse reactions due to target expression in non-relevant tissues43. We established the following six selection criteria to select targets for antibody development: target localization to the cell surface, preferential statistically higher expression in tumor tissues vs normal tissues, absence of existing approved therapeutic antibodies, a large extracellular domain, limited post-translational modifications such as glycosylation (which is a source of variability that may impact antibody binding), and minimal or restricted expression in normal tissues. Proteins exposed to the extracellular space can be bound by serum antibodies (criterium 1). Fifty five of the 320 targets identified (17%) were cell surface proteins (Supplementary File S5). The Endoplasmic Reticulum and Mitochondria localized proteins CANX44, MFN145, PERK46, MLANA47, and STING148, (Table 1) were also included because stress-induced membrane remodelling and aberrant trafficking in cancer can result in these proteins’ mislocalization to the plasma membrane49 and exposure of their luminal or cytoplasmic domains to the extracellular environment, thereby enabling recognition by circulating antibodies. We classified the 55 (potential) cell surface proteins based on topology and number of transmembrane domains, categorizing them as single pass Type-I (38 targets), -II (5 targets), and -III (2 targets), and multi-pass (10 targets) transmembrane proteins (Fig. 4b, Table 1).Table 1List of 23 cell surface proteins identified as potential therapeutic targets.Protein | mRNA (Targets)ProteinmRNAProtein | mRNA(UniProt | HGNC ID)Abundance level*log2FC (Tumor/Normal)Adj.-P valuelog2FC (Tumor/Normal)Adj.-P valueNo. of amino acids exposed extracellularlyType of Protein and subcellular localization**No. of glycans exposed extracellularlyB7-H3 | CD276(Q5ZPR3 | 19137)Higher | Higher0.54665\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1.93\times {10}^{-04}$$\end{document}0.85142\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$7.41\times {10}^{-23}$$\end{document}437I6 (N)Calnexin | CANX (P27824 | 1473)0.43628\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1.43\times {10}^{-04}$$\end{document}0.52561\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$9.94\times {10}^{-19}$$\end{document}460^,89#ER, Mit, I0Claudin7 | CLDN7 (O95471 | 2049)0.38921\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$7.61\times {10}^{-03}$$\end{document}1.13041\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$5.80\times {10}^{-14}$$\end{document}52, 21M0EMA | MUC1(P15941 | 7508)2.34703\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1.36\times {10}^{-09}$$\end{document}2.69845\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1.06\times {10}^{-27}$$\end{document}1134I4 (O), 5(N)HER2 | ERBB2(P04626 | 3430)0.33613\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$3.77\times {10}^{-02}$$\end{document}1.85573\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1.67\times {10}^{-15}$$\end{document}629I7 (N)HLA-DQ1 | HLA-DQ1(P01909 | 4942)1.37865\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1.65\times {10}^{-10}$$\end{document}0.88539\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$2.05\times {10}^{-09}$$\end{document}192I2 (N)IR-b | INSR(P06213 | 6091)0.32641\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1.18\times {10}^{-02}$$\end{document}0.17995\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$3.14\times {10}^{-02}$$\end{document}730, 193I18 (N)MCT4 | SLC16A3(O15427 | 10924)0.54870\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$4.42\times {10}^{-04}$$\end{document}1.77911\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1.69\times {10}^{-22}$$\end{document}22, 3, 8, 15, 1, 11M0Mitofusin-1 | MFN1 (Q8IWA4 | 18262)0.55329\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1.52\times {10}^{-06}$$\end{document}0.34365\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$6.55\times {10}^{-08}$$\end{document}583#, 111#Mit, M0MMP14 | MMP14(P50281 | 7160)1.27962\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$5.42\times {10}^{-12}$$\end{document}1.33360\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1.19\times {10}^{-23}$$\end{document}429I0PERK | EIF2AK3 (Q9NZJ5 | 3255)0.64044\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$3.49\times {10}^{-09}$$\end{document}0.59547\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$5.94\times {10}^{-21}$$\end{document}484^, 580#ER, I1 (N)CD44 | CD44(P16070 | 1681)Non-sig | Higher−0.02326\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$8.20\times {10}^{-01}$$\end{document}0.52359\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1.30\times {10}^{-05}$$\end{document}628I1 (O), 9 (N)CD45 | PTPRC(P08575 | 9666)−0.13817\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1.98\times {10}^{-01}$$\end{document}0.62010\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$4.90\times {10}^{-05}$$\end{document}551I17 (N)Connexin-43 | GJA1(P17302 | 4274)0.53996\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$6.20\times {10}^{-02}$$\end{document}1.27968\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1.32\times {10}^{-11}$$\end{document}31, 30M0DDR1 | DDR1(Q08345 | 2730)0.09333\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$3.73\times {10}^{-01}$$\end{document}0.78951\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$6.25\times {10}^{-12}$$\end{document}396I4 (N)E-Cadherin| CDH1(P12830 | 1748)−0.09834\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$7.21\times {10}^{-01}$$\end{document}0.77139\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$3.99\times {10}^{-05}$$\end{document}554I9 (O), 4 (N)FN14 | TNFRSF12A(Q9NP84 | 18152)0.11536\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$6.88\times {10}^{-02}$$\end{document}0.72924\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1.58\times {10}^{-05}$$\end{document}52I0HER3 | ERBB3(P21860 | 3431)−0.14799\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$4.24\times {10}^{-01}$$\end{document}1.07337\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$5.16\times {10}^{-11}$$\end{document}623I10 (N)IGFRb | IGF1R(Q92637 | 5465)0.10223\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$6.61\times {10}^{-01}$$\end{document}1.24603\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1.13\times {10}^{-11}$$\end{document}182I3 (N)MelanA| MLANA(Q16655 | 7124)−0.02987\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$7.83\times {10}^{-01}$$\end{document}0.85818\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$2.82\times {10}^{-07}$$\end{document}70#ER, III0TFRC | TFRC(P02786 | 11763)0.18920\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$4.72\times {10}^{-01}$$\end{document}1.30190\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$6.79\times {10}^{-23}$$\end{document}671II1(O), 3 (N)CD4 | CD4(P01730 | 1678)Higher | Non-sig0.64089\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$5.55\times {10}^{-06}$$\end{document}0.22279\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$8.10\times {10}^{-02}$$\end{document}370I2 (N)STING1 | STING1(Q86WV6 | 27962)0.70509\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$5.81\times {10}^{-06}$$\end{document}−0.09410\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$3.22\times {10}^{-01}$$\end{document}16#,9^,21#,9^,244#ER, G, Mit, M0Adj., adjusted.*Relative abundance of the target in tumor vs normal tissues, based on protein and mRNA data. “Higher | Higher” means a significant increase in protein and mRNA abundance in tumor vs normal tissue. “Higher | Non-sig” or “Non-sig | Higher” refers to significant increase in protein or mRNA abundance, but not both.**Topological classification of the membrane protein and subcellular localization in normal tissue: I, Single pass type-I; II, Single pass type-II; III, Single pass type-III; and M, Multi pass transmembrane protein. ER, Endoplasmic Reticulum; G, Golgi Complex; Mit, Mitochondria.^, #Amino acid number corresponding to the luminal (^) or cytoplasmic (#) domain of the protein, which would be exposed to the extracellular domain if/when the protein traffics to the cell surface.(N) and (O) refer to the number of N-linked and O-linked glycosylation sites exposed to the extracellular domain. Table 1List of 23 cell surface proteins identified as potential therapeutic targets.Protein | mRNA (Targets)ProteinmRNAProtein | mRNA(UniProt | HGNC ID)Abundance level*log2FC (Tumor/Normal)Adj.-P valuelog2FC (Tumor/Normal)Adj.-P valueNo. of amino acids exposed extracellularlyType of Protein and subcellular localization**No. of glycans exposed extracellularlyB7-H3 | CD276(Q5ZPR3 | 19137)Higher | Higher0.54665\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1.93\times {10}^{-04}$$\end{document}0.85142\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$7.41\times {10}^{-23}$$\end{document}437I6 (N)Calnexin | CANX (P27824 | 1473)0.43628\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1.43\times {10}^{-04}$$\end{document}0.52561\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$9.94\times {10}^{-19}$$\end{document}460^,89#ER, Mit, I0Claudin7 | CLDN7 (O95471 | 2049)0.38921\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$7.61\times {10}^{-03}$$\end{document}1.13041\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$5.80\times {10}^{-14}$$\end{document}52, 21M0EMA | MUC1(P15941 | 7508)2.34703\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1.36\times {10}^{-09}$$\end{document}2.69845\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1.06\times {10}^{-27}$$\end{document}1134I4 (O), 5(N)HER2 | ERBB2(P04626 | 3430)0.33613\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$3.77\times {10}^{-02}$$\end{document}1.85573\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1.67\times {10}^{-15}$$\end{document}629I7 (N)HLA-DQ1 | HLA-DQ1(P01909 | 4942)1.37865\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1.65\times {10}^{-10}$$\end{document}0.88539\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$2.05\times {10}^{-09}$$\end{document}192I2 (N)IR-b | INSR(P06213 | 6091)0.32641\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1.18\times {10}^{-02}$$\end{document}0.17995\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$3.14\times {10}^{-02}$$\end{document}730, 193I18 (N)MCT4 | SLC16A3(O15427 | 10924)0.54870\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$4.42\times {10}^{-04}$$\end{document}1.77911\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1.69\times {10}^{-22}$$\end{document}22, 3, 8, 15, 1, 11M0Mitofusin-1 | MFN1 (Q8IWA4 | 18262)0.55329\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1.52\times {10}^{-06}$$\end{document}0.34365\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$6.55\times {10}^{-08}$$\end{document}583#, 111#Mit, M0MMP14 | MMP14(P50281 | 7160)1.27962\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$5.42\times {10}^{-12}$$\end{document}1.33360\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1.19\times {10}^{-23}$$\end{document}429I0PERK | EIF2AK3 (Q9NZJ5 | 3255)0.64044\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$3.49\times {10}^{-09}$$\end{document}0.59547\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$5.94\times {10}^{-21}$$\end{document}484^, 580#ER, I1 (N)CD44 | CD44(P16070 | 1681)Non-sig | Higher−0.02326\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$8.20\times {10}^{-01}$$\end{document}0.52359\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1.30\times {10}^{-05}$$\end{document}628I1 (O), 9 (N)CD45 | PTPRC(P08575 | 9666)−0.13817\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1.98\times {10}^{-01}$$\end{document}0.62010\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$4.90\times {10}^{-05}$$\end{document}551I17 (N)Connexin-43 | GJA1(P17302 | 4274)0.53996\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$6.20\times {10}^{-02}$$\end{document}1.27968\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1.32\times {10}^{-11}$$\end{document}31, 30M0DDR1 | DDR1(Q08345 | 2730)0.09333\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$3.73\times {10}^{-01}$$\end{document}0.78951\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$6.25\times {10}^{-12}$$\end{document}396I4 (N)E-Cadherin| CDH1(P12830 | 1748)−0.09834\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$7.21\times {10}^{-01}$$\end{document}0.77139\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$3.99\times {10}^{-05}$$\end{document}554I9 (O), 4 (N)FN14 | TNFRSF12A(Q9NP84 | 18152)0.11536\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$6.88\times {10}^{-02}$$\end{document}0.72924\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1.58\times {10}^{-05}$$\end{document}52I0HER3 | ERBB3(P21860 | 3431)−0.14799\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$4.24\times {10}^{-01}$$\end{document}1.07337\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$5.16\times {10}^{-11}$$\end{document}623I10 (N)IGFRb | IGF1R(Q92637 | 5465)0.10223\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$6.61\times {10}^{-01}$$\end{document}1.24603\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1.13\times {10}^{-11}$$\end{document}182I3 (N)MelanA| MLANA(Q16655 | 7124)−0.02987\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$7.83\times {10}^{-01}$$\end{document}0.85818\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$2.82\times {10}^{-07}$$\end{document}70#ER, III0TFRC | TFRC(P02786 | 11763)0.18920\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$4.72\times {10}^{-01}$$\end{document}1.30190\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$6.79\times {10}^{-23}$$\end{document}671II1(O), 3 (N)CD4 | CD4(P01730 | 1678)Higher | Non-sig0.64089\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$5.55\times {10}^{-06}$$\end{document}0.22279\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$8.10\times {10}^{-02}$$\end{document}370I2 (N)STING1 | STING1(Q86WV6 | 27962)0.70509\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$5.81\times {10}^{-06}$$\end{document}−0.09410\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$3.22\times {10}^{-01}$$\end{document}16#,9^,21#,9^,244#ER, G, Mit, M0Adj., adjusted.*Relative abundance of the target in tumor vs normal tissues, based on protein and mRNA data. “Higher | Higher” means a significant increase in protein and mRNA abundance in tumor vs normal tissue. “Higher | Non-sig” or “Non-sig | Higher” refers to significant increase in protein or mRNA abundance, but not both.**Topological classification of the membrane protein and subcellular localization in normal tissue: I, Single pass type-I; II, Single pass type-II; III, Single pass type-III; and M, Multi pass transmembrane protein. ER, Endoplasmic Reticulum; G, Golgi Complex; Mit, Mitochondria.^, #Amino acid number corresponding to the luminal (^) or cytoplasmic (#) domain of the protein, which would be exposed to the extracellular domain if/when the protein traffics to the cell surface.(N) and (O) refer to the number of N-linked and O-linked glycosylation sites exposed to the extracellular domain. List of 23 cell surface proteins identified as potential therapeutic targets. B7-H3 | CD276 (Q5ZPR3 | 19137) EMA | MUC1 (P15941 | 7508) HER2 | ERBB2 (P04626 | 3430) HLA-DQ1 | HLA-DQ1 (P01909 | 4942) IR-b | INSR (P06213 | 6091) MCT4 | SLC16A3 (O15427 | 10924) MMP14 | MMP14 (P50281 | 7160) CD44 | CD44 (P16070 | 1681) CD45 | PTPRC (P08575 | 9666) Connexin-43 | GJA1 (P17302 | 4274) DDR1 | DDR1 (Q08345 | 2730) E-Cadherin| CDH1 (P12830 | 1748) FN14 | TNFRSF12A (Q9NP84 | 18152) HER3 | ERBB3 (P21860 | 3431) IGFRb | IGF1R (Q92637 | 5465) MelanA| MLANA (Q16655 | 7124) TFRC | TFRC (P02786 | 11763) CD4 | CD4 (P01730 | 1678) STING1 | STING1 (Q86WV6 | 27962) Adj., adjusted. *Relative abundance of the target in tumor vs normal tissues, based on protein and mRNA data. “Higher | Higher” means a significant increase in protein and mRNA abundance in tumor vs normal tissue. “Higher | Non-sig” or “Non-sig | Higher” refers to significant increase in protein or mRNA abundance, but not both. **Topological classification of the membrane protein and subcellular localization in normal tissue: I, Single pass type-I; II, Single pass type-II; III, Single pass type-III; and M, Multi pass transmembrane protein. ER, Endoplasmic Reticulum; G, Golgi Complex; Mit, Mitochondria. ^, #Amino acid number corresponding to the luminal (^) or cytoplasmic (#) domain of the protein, which would be exposed to the extracellular domain if/when the protein traffics to the cell surface. (N) and (O) refer to the number of N-linked and O-linked glycosylation sites exposed to the extracellular domain. Within these 55 proteins, we next looked for targets with preferential expression in tumor vs normal tissues in both mRNA and protein analyses (criterium 2) (|log2FC| ≥ 0.1 and an adjusted p-value ≤ 0.05), which included 11 hits (MCT4, CANX, MFN1, B7-H3, CLDN7, PERK, HER2, HLA-DQA1, IR-b, MMP14, and EMA). To increase the number of potential targets, we also included targets showing higher abundance in mRNA but no significant difference in protein (10 hits: TFRC, CD44, CDH1, FN14, DDR1, HER3, GJA1, IGFR-b, MLANA, and CD45), and no significant difference in mRNA but higher abundance in protein (2 hits: CD4 and STING1) (Fig. 4b, Table 1, Supplementary File S6 and Table S1T2). These led to 23 promising candidates (Table 1). To develop new drugs against novel targets or new therapeutic modalities against potential targets, it is critical to understand what targets have already been explored for therapeutic development and at what stage in the pipeline the different therapies are in (criterium 3).The absence of antibodies or therapies under development against targets suggests a low or unexplored target status and supports future freedom to operate (an important consideration for drug development). To identify any existing drugs, ligands, or inhibitors associated with the 23 potential targets we performed searches in the Therapeutic Target Database50 and DRUGBANK51 (Supplementary Table S1T2). We found no existing approved mAbs against 11 targets: GJA1, MLANA, CANX, PERK, MFN1, CDH1, DDR1, STING1, MCT4, CLDN7, and HLA-DQA1, although therapies targeting CDH152 and DDR123,29,53,54 are under development. On the other hand, the other 12 targets have approved mAbs on the market, such as Omburtamab I-131 targeting B7-H3 and XmetA and XmetD targeting IR-b, (Supplementary File S6 and Table S1T2). Consequently, these 12 targets were excluded from the target selection list. These analyses demonstrate that our pipeline identified candidate hits that were already deemed as potential disease targets (indirectly validating the pipeline) as well as currently underexplored targets. Next, we searched the UniProt database to identify membrane proteins with large extracellular domains (criterium 4). Large, well-structured extracellular domains facilitate antibody discovery55,56. Among the 11 targets satisfying criteria 1–4, CLDN7, GJA1, MCT4, STING1, and MLANA, exhibited extracellular domains of less than 100 amino acids long (Table 1) and were therefore excluded from further analysis. In contrast, CDH1, HLA-DQA1, DDR1, CANX, PERK, and MFN1 have extracellular domains greater than 100 amino acids long, making them more suitable for antibody discovery. Criterium 5 focused on glycosylation of the extracellular domain. High levels of glycosylation can mask epitopes or bias antibody selection for glycans that may not be represented in cancer tissue due to glycan heterogeneity57,58. CDH1 contains four N-linked and nine O-linked glycosylation sites, while DDR1 and HLA-DQA1 have 4 and 2 N-linked glycosylation sites, respectively. Notably, CANX59 is glycosylated in tumor cells and can translocate to the plasma membrane, indicating unpredictable cell surface and glycosylation status in tumor. MFN1 is a mitochondrial protein and therefore it should not be glycosylated, while PERK1, an ER resident protein, has 5 sites of potential N-linked glycosylation (predicted by NetNGlyc 1.060). However, no report was found regarding the PERK’s glycosylation in cancer, making this target unreliable for antibody targeting. Given the potential heterogeneity in CANX, PERK, MFN1 cell surface expression and glycosylation, these targets require further study to be used for antibody discovery, which is out of scope for this work. Thus, CANX, PERK, and MFN1 were excluded from the list. Since HLA-DQA1 is a polymorphic protein involved in immune system regulation61, we also excluded it from further analysis. Therefore, we moved forward with CDH1 and DDR1 proteins, which satisfied criteria 1–5. Minimal or restricted expression of cancer targets in normal tissue should decrease side effects due to undesired targeting of non-cancer tissue (criterium 6). We searched online databases such as GEPIA and UniProt for information on normal tissue expression of CDH1 and DDR1. GEPIA was used due to its advanced algorithms and comprehensive coverage of transcriptomic data normalized across TCGA and Genotype-Tissue Expression (GTex) databases62. CDH1 showed high widespread expression in epithelial cells across normal tissues including the breast63, whereas DDR1 expression appeared more restricted, largely confined to breast and gastrointestinal epithelium with minimal or no expression in most other normal tissues23,64. Consistent with these findings, Zhong et al.20 further reported that DDR1 expression levels are significantly higher in breast cancer tissues compared to adjacent normal tissues, and that DDR1 regulated tumor growth by modulating tumor-infiltrating T cells. Thus, DDR1 demonstrates a more limited normal tissue distribution and cancer-specific upregulation, making it the more attractive therapeutic target and biomarker for antibody development. To validate the results of our analysis that identified DDR1 as a potential cancer therapeutic target, we looked at DDR1 expression in cancer tissues using GEPIA and HPA. The results (Supplementary Fig. S1F2) revealed that DDR1 expression was significantly (P-value < 0.05) higher in breast tumor tissues compared to normal tissues, supporting our analysis from TCGA data. Validation through immunohistochemistry (IHC) staining data from the HPA database confirmed high DDR1 expression in breast cancer, with predominant localization at the plasma membrane, nucleoplasm, and cytosol. In contrast, normal breast tissue analysis further revealed medium expression in myoepithelial and glandular cells with no detectable levels in adipocytes (Supplementary Fig. S1F3a-c). All together, we selected DDR1 as a target for development of new antibodies. DDR1 is a type-I transmembrane, moderately glycosylated protein with a large extracellular domain that is overexpressed in breast cancer, while showing limited expression across normal tissues. Notably, current therapies against DDR1, including DDR1 antibody drug conjugate have demonstrated efficacy but also present toxicity profiles particularly in gastrointestinal tissue23 and FDA-approved small molecule treatments lack tissue specificity are associated with off-target effects, opening opportunities for the development of new therapeutic modalities and drugs targeting this protein. Isolation of novel antibody binders to hDDR1/CD167a To identify novel antibodies binding the extracellular domain of DDR1, we performed 3 rounds of phage display biopanning as per standard protocols65, using commercial purified DDR1-His extracellular domain as the target. The phage display library used (Jones-Mahler) was a human scFv phage library prepared from naïve donors55. A mix of negative His-tagged antigens (CD64, CD16a, CD32a) was used to reduce enrichment of binders to the His epitope tag during biopanning (Supplementary Table S1T3). To assess whether the biopanning enriched the library for hDDR1 binders, phage pools from all biopanning rounds were screened in a polyclonal phage ELISA against hDDR1-His in duplicate. Round 2 (R2) and Round 3 (R3) phage pools showed clear enrichment for binders, compared to the naïve library and Round 1 (R1), showing higher absorbance at 450 nm (A450) with increasing 1:10 dilutions of R2 and R3 phage pools (Fig. 5a). To identify individual positive clones, monoclonal phage ELISA was performed using phage expressed into the growth media from random clones picked from both R2 and R3 pools (Fig. 5b,c). PBS only (no phage clones) was used as control to establish baseline A450 readings, with the mean A450 + 3-times SD of controls used as a cut-off to define positive binding clones. In total, 38 positive binders were isolated from 84 phages screened (Fig. 5b,c). All positive clones were sequenced, out of which 22 clones returned good sequence quality. The highly variable complementarity determining region 3 of the heavy chain (CDR3-H) is often cited as the major determinant of antibody-binding specificity66. Within these 22 clones, there were 9 unique sequences, represented by clones A12, B11, C9, C12, D7, E5, E8, F8 and G11, all showing a unique CDR3-H (data not shown).Fig. 5Identification of novel anti-DDR1/CD167a antibodies by phage display biopanning. (a) Polyclonal phage ELISA comparing the three rounds of panning to the naïve Jones-Mahler (J-M) library. Binding was assessed in duplicate by immobilizing hDDR1-His on ELISA plates and incubating with successive 10-fold dilution of phage pools for each round. Shown is the mean ± range from one experiment done in duplicate. (b) Round 2 and (c) Round 3 monoclonal phage ELISA. Phage clones that exceeded an A450 reading (mean+ 3-times SD from No Phage Control (NPC) were considered positive binders (green). Negative binders are shown in grey. Identification of novel anti-DDR1/CD167a antibodies by phage display biopanning. (a) Polyclonal phage ELISA comparing the three rounds of panning to the naïve Jones-Mahler (J-M) library. Binding was assessed in duplicate by immobilizing hDDR1-His on ELISA plates and incubating with successive 10-fold dilution of phage pools for each round. Shown is the mean ± range from one experiment done in duplicate. (b) Round 2 and (c) Round 3 monoclonal phage ELISA. Phage clones that exceeded an A450 reading (mean+ 3-times SD from No Phage Control (NPC) were considered positive binders (green). Negative binders are shown in grey. DDR1/CD167a has 54% homology in the extracellular domain (ECD) to DDR2/CD167b, another RTK. ELISA was used to determine DDR1 target specificity and potential cross reactivity with DDR2 of the identified clones, using purified phage particles (Fig. 6). Interestingly, the purified unique phage clones C9 and G11 did not retain binding to hDDR1, emphasizing the importance of confirming the binding after an initial high-throughput screen. On the other hand, the other 7 clones showed the expected binding to hDDR1 with no cross-reactivity to hDDR2 or the His-tagged controls.Fig. 6Confirmatory ELISA using purified phage clones against hDDR1 (true target), hDDR2 (potential cross-reactive homologue), and His tag. Purified phage preparations from the 9 binders from Rounds 2 and 3 were used to test binding to the positive antigen (hDDR1), a related protein hDDR2-His, and a mixture of negative His-tagged antigens, CD64-His, CD16-His, and CD32a-His (His-Ag). All antigens were coated on ELISA plates at 1 μg/mL. PBS only was used as no phage control (NPC). This experiment was performed once with one replicate per phage sample. Confirmatory ELISA using purified phage clones against hDDR1 (true target), hDDR2 (potential cross-reactive homologue), and His tag. Purified phage preparations from the 9 binders from Rounds 2 and 3 were used to test binding to the positive antigen (hDDR1), a related protein hDDR2-His, and a mixture of negative His-tagged antigens, CD64-His, CD16-His, and CD32a-His (His-Ag). All antigens were coated on ELISA plates at 1 μg/mL. PBS only was used as no phage control (NPC). This experiment was performed once with one replicate per phage sample. To further validate the binding to DDR1, we performed flow cytometry to test the ability of the 7 positive phage clones to bind hDDR1 in its cell-surface conformation. To do this we expressed the hDDR1 extracellular and transmembrane domains fused to intracellular eGFP (hDDR1-(ECD-TM)-eGFP) in ExpiCHO cells. GFP fluorescence confirmed expression of the construct (Fig. 7). Untransfected and transfected cells only (no antibody treatment) and secondary antibody only stained transfected cells were used as controls (Fig. 7a,b and Supplementary Fig. S1F4a,b). Transfected cells were incubated with the 7 positive phage clones and the negative clone G11 (from Fig. 6). hDDR1 specific clones should show high signal in the Q2 (high GFP expression, high anti-phage secondary antibody) quadrant and low signal in the Q1 (low GFP expression, high anti-phage secondary antibody) quadrant, indicating binding to high GFP expressing cells, and low binding to non-GFP expressing cells (Fig. 7). Consistent with the ELISA result (Fig. 6), clone G11 did not bind DDR1 by flow cytometry (Fig. 7c), while the other 7 clones (A12, B11, C12, D7, E5, E8, and F8) demonstrated similar binding to hDDR1 expressed in ExpiCHO cells, with no background binding to untransfected ExpiCHO cells (clone C12 shown as example in Fig. 7d, and other clones can be seen in Supplementary Fig. S1F4c-h).Fig. 7Monoclonal phage flow cytometry demonstrates binding of isolated phage clones to membrane bound hDDR1. ExpiCHO cells transiently expressing hDDR1(ECD-TM)-eGFP were stained with phages and an anti-phage antibody, followed by a secondary antibody (Dylight650). GFP expression (x-axis) and phage binding (y-axis) was monitored by flow cytometry. Controls included: (a) Untransfected and (b) transfected cells stained with secondary antibody only, and transfected cells treated with (c) G11 negative phage clone. (d) Transfected cells stained with C12 positive phage clone shown as an example of a positive binder. Monoclonal phage flow cytometry demonstrates binding of isolated phage clones to membrane bound hDDR1. ExpiCHO cells transiently expressing hDDR1(ECD-TM)-eGFP were stained with phages and an anti-phage antibody, followed by a secondary antibody (Dylight650). GFP expression (x-axis) and phage binding (y-axis) was monitored by flow cytometry. Controls included: (a) Untransfected and (b) transfected cells stained with secondary antibody only, and transfected cells treated with (c) G11 negative phage clone. (d) Transfected cells stained with C12 positive phage clone shown as an example of a positive binder. Expression and purification of full-length IgG1 anti-DDR1 antibodies Due to resource constraints, we moved forward with only four of the seven unique phage-scFvs for subsequent analysis, which were selected on the basis of strong and consistent binding to the purified and native DDR1 extracellular domain by ELISA and flow cytometry, low non-specific binding to DDR2 and negative antigens, and sequence diversity in CDR3-H to minimize redundancy (see results section 3.3). Clones A12, B11, C12, and E8 were reformatted into full-length human IgG1 mAbs (hIgG1), expressed in ExpiCHO cells, and purified via Protein-A affinity chromatography67. Purified antibodies migrated at the expected molecular weight by SDS-PAGE, with one predominant band at ~150 kDa under non-reducing conditions and two separate bands at ~25 kDa and ~50 kDa under reducing conditions, indicating the correct association of light and heavy chains, respectively (Supplementary Fig. S1F5). Protein expression yields of individual clones post-purification varied from 87.5 µg (clone B11) to 177.45 µg (clone C12) per mL of culture transfected, which is within normal range for non-optimized sequences or processes68. The aggregation profile of A12, B11, C12, and E8 mAbs was assessed by size exclusion-HPLC, where a single dominant monomer peak was observed, with less than 2% aggregation and no fragmentation (Supplementary Fig. S1F6). The retention times for all 4 mAbs overlaid with human IgG1 isotype control mAb. These results indicated successful purification of the 4 mAbs with good quality and purity. Verifying binding of purified mAbs to target antigen Retainment of hDDR1 binding after reformatting was confirmed by flow cytometry, using hDDR1-(ECD-TM)-GFP transfected ExpiCHO cells (Fig. 8a). Recombinant hDDR1 expression at the cell surface was first confirmed using a commercial anti-hDDR1 mAb (Fig. 8a-c, and Supplementary Figs. S1F7, 8). All four mAbs showed increased binding to GFP positive transfected cells, compared with non-transfected cells or cells transfected with a non-related membrane protein (VCAM-1, Supplementary Fig. S1F9-10). However, mAb A12 exhibited considerable binding to non-transfected ExpiCHO cells (Supplementary Fig. S1F9d). These results suggest that clones B11, E8, and C12 exhibit greater specificity towards hDDR1 compared to A12, or alternatively A12 has higher affinity for hDDR1 than the other clones and can detect very low levels of endogenous hamster DDR1 on ExpiCHO that the other mAbs (including the positive control) are unable to detect. There is 98% homology between human and hamster DDR1.Fig. 8Flow cytometry analysis of anti-DDR1 mAbs. Flow cytometry analysis of novel anti-DDR1 mAb (5 µg/mL) and a commercial PE-conjugated anti-DDR1 mAb (Comm. mAb; final dilution 1:20) binding to transfected ExpiCHO cells (a), and MCF-7 (b) and MDA-MB-231 human breast cancer cell lines (c). Bar graphs display the mean ± SD of the mean fluorescence intensity (MFI) for novel anti-DDR1 mAb (left y-axis) and the commercial mAb (right y-axis) from one experiment performed in triplicate (n=3, shown as dots). For novel mAbs, statistical comparisons are to the secondary antibody only (Sec. ab only, black) and unstained controls (Unst., red). A One-way ANOVA with a Tukey’s multiple comparisons test was performed to determine significance (*** p < 0.001; ** p < 0.01; * p < 0.05); n.s.: non-significant. For the commercial mAb, an unpaired t-test was used to determine significance. (**** p < 0.0001), ns: non-significant. Flow cytometry analysis of anti-DDR1 mAbs. Flow cytometry analysis of novel anti-DDR1 mAb (5 µg/mL) and a commercial PE-conjugated anti-DDR1 mAb (Comm. mAb; final dilution 1:20) binding to transfected ExpiCHO cells (a), and MCF-7 (b) and MDA-MB-231 human breast cancer cell lines (c). Bar graphs display the mean ± SD of the mean fluorescence intensity (MFI) for novel anti-DDR1 mAb (left y-axis) and the commercial mAb (right y-axis) from one experiment performed in triplicate (n=3, shown as dots). For novel mAbs, statistical comparisons are to the secondary antibody only (Sec. ab only, black) and unstained controls (Unst., red). A One-way ANOVA with a Tukey’s multiple comparisons test was performed to determine significance (*** p < 0.001; ** p < 0.01; * p < 0.05); n.s.: non-significant. For the commercial mAb, an unpaired t-test was used to determine significance. (**** p < 0.0001), ns: non-significant. To confirm binding of the mAbs to native hDDR1, we used human breast cancer cell lines MCF-7 and MDA-MB-231, which are known to constitutively express hDDR1 receptor on their surface69. First, we verified hDDR1 expression using flow cytometry with the commercial PE-conjugated antibody, and observed significantly higher expression of hDDR1 in MCF-7 cells compared with MDA-MD-231 cells (P-value < 0.0001) (Fig. 8a-c, and Supplementary Figs. S1F7, 8), consistent with previous reports69. Next, we tested our four novel mAbs on the same two cancer cell lines (Fig. 8b,c), observing significantly higher binding to both breast cancer cell lines compared to secondary antibody only and unstained controls, suggesting binding to native DDR1 (Fig. 8 and Supplementary Fig. S1F8c,d). Interestingly, we observed differences in the binding of the different mAbs to the different cell lines. MCF-7 has the highest DDR1 expression, and all 4 mAbs bound well to this cell line (Fig. 8b). However, in the case of MDA-MB-231 cell line (Fig. 8c), the B11 clone showed significantly lower binding to this cancer cell line compared to the clone A12 and C12 (P < 0.05), and E8 (P < 0.01). Together, our data shows that our antibodies bind both hDDR1 expressing breast cancer cell lines and hDDR1-transfected ExpiCHO cells with different ability. Among them, B11 displayed the weakest binding across all cell types. To further characterize our anti-DDR1 mAbs, we performed a surface plasmon resonance (SPR) assay (Table 2, Supplementary Fig. S1F11). Multicycle kinetic SPR experiment allows for the determination of association and dissociation kinetics of the mAbs (analytes) to the target antigens, in this case soluble hDDR1 ECD-His and soluble hDDR2 ECD-His (ligands). To do this, we first captured the soluble hDDR1 ECD-His and hDDR2 ECD-His on a CM5 chip with immobilised anti-His antibody and probed the functionalized surface with anti-DDR1 mAbs (A12, B11, C12, and E8) as analyte. The sensorgram data was analyzed using the 1:1 kinetic binding model. As expected, no binding with the human IgG1 isotype control was detected, nor was any binding between hDDR2 and the anti-DDR1 mAbs detected, consistent with the ELISA results (Supplementary Fig. S1F11a, b, d, f, h, j). The A12 mAb exhibited a binding affinity of 27.3 pM to hDDR1, surpassing the respective affinities of other mAbs (E8, C12, B11), which were 0.558 nM, 0.251 nM and 2.54 nM, respectively (Table 2). This result supports our hypothesis that the increased binding of A12 to non-transfected ExpiCHO cells is due to a higher affinity for the target (Supplementary Fig. S1F9d,k), and the decreased binding of B11 to all cell types (Fig. 8) is due to lower affinity for the target (Table 2). Together, the results show that the new mAbs, isolated from a naïve library without any subsequent affinity maturation, show good affinity for DDR1 with no cross-reactivity to DDR2.Table 2Kinetic parameters for the interaction (1:1 binding model) between anti-hDDR1 IgG and immobilized hDDR1-His (recombinant soluble extracellular domain) receptor.mAbska (1/Ms)kd(1/s)KD (M)Rmax (RU)A121.49 × 1064.07 × 10−52.73 × 10−1182B111.13 × 10 62.88 × 10−32.54 × 10−932.8C121.41 × 10 63.54 × 10−42.51 × 10−1072.1E88.63 × 10 54.82 × 10−45.58 × 10−1040.3 Table 2Kinetic parameters for the interaction (1:1 binding model) between anti-hDDR1 IgG and immobilized hDDR1-His (recombinant soluble extracellular domain) receptor.mAbska (1/Ms)kd(1/s)KD (M)Rmax (RU)A121.49 × 1064.07 × 10−52.73 × 10−1182B111.13 × 10 62.88 × 10−32.54 × 10−932.8C121.41 × 10 63.54 × 10−42.51 × 10−1072.1E88.63 × 10 54.82 × 10−45.58 × 10−1040.3 Kinetic parameters for the interaction (1:1 binding model) between anti-hDDR1 IgG and immobilized hDDR1-His (recombinant soluble extracellular domain) receptor. In-vitro functional assessment of novel anti-DDR1 mAbs To determine the functional activity of the anti-DDR1 antibodies, we performed two different cell-based assays: AlamarBlue cell proliferation and antibody dependent cell-mediated cytotoxicity (ADCC) assays. The AlamarBlue cell proliferation assay was used to evaluate the cytotoxic effects of the mAbs on breast cancer cells by measuring changes in metabolic activity. In this assay cell viability was measured by the reduction of AlamarBlue reagent (resazurin) into a fluorescent product (rosorufin) by metabolically active cells. Lower fluorescence indicates decreased cell viability, reflecting cytotoxic effects or loss of metabolically active cells. Several controls were included in this experiment. First, rituximab, a well-characterized B-cell CD20-targeting mAb3 served as an assay positive control (Fig. 9a) as it induces cytotoxicity in Raji cells (a human B lymphoblastoid cell line) via ADCC mechanisms. Rituximab treatment led to a reduction in cell viability below 50% across the 1.25–20 µg/mL concentration range, in cell viability assay, ensuring the proper functioning of assay reagents and conditions, and consistent with previous observations4. Next, ethanol, a cytotoxic agent, was used as a positive control for the assay in the relevant cell lines (Fig. 9b-e). Ethanol induced cell death in non-DDR1 expressing controls (HEK293-T and CHO-S) and breast cancer (MCF-7 and MDA-MB-231) cell lines at 1.25–20 % v/v (Fig. 9b-e), as expected. Finally, the human IgG1 isotype antibody was used as negative control and showed the expected limited effect on cell viability across the tested concentrations (0–20 µg/mL), with over 80% viability observed at the highest antibody concentration for all cell lines in the cell viability assay, with the exception of CHO-S (Figs. 9b-e). Overall, all controls behaved as expected.Fig. 9Novel anti-DDR1 antibodies lower breast cancer cell line viability, measured by AlamarBlue assay. (a) Raji cell line treated with Rituximab(assay positive control); (b-e) Cell Viability in the presence of four novel anti-DDR1 mAbs, human isotype control (hIgG1), or Ethanol treatment control (viability positive control)for non-DDR1 expressing control cell lines CHO-S (b) and HEK293-T (c) or DDR1 expressing breast cancer cell lines MCF-7 (d) and MDA-MD-231 (e). Each error bar on points represents the quality range of upper and lower limits shown as mean±range. The experiment was performed once with 2 replicates (n=2). Data are expressed as percent cell viability relative to cells only control (without antibody treatment). Novel anti-DDR1 antibodies lower breast cancer cell line viability, measured by AlamarBlue assay. (a) Raji cell line treated with Rituximab(assay positive control); (b-e) Cell Viability in the presence of four novel anti-DDR1 mAbs, human isotype control (hIgG1), or Ethanol treatment control (viability positive control)for non-DDR1 expressing control cell lines CHO-S (b) and HEK293-T (c) or DDR1 expressing breast cancer cell lines MCF-7 (d) and MDA-MD-231 (e). Each error bar on points represents the quality range of upper and lower limits shown as mean±range. The experiment was performed once with 2 replicates (n=2). Data are expressed as percent cell viability relative to cells only control (without antibody treatment). Upon anti-DDR1 treatment, HEK293-T and CHO-S cell lines exhibited minimal reduction in cell viability, staying above 50% relative viability, similar to the IgG1 control (Figs. 9b-c). These results are consistent with the reported low/null expression of DDR1 in these cell lines (Supplementary Fig. S1F7a,b and70,71). In contrast, the breast cancer cell lines MCF-7 and MDA-MB-231 showed increased susceptibility to anti-DDR1 mAb-induced cytotoxicity, while the human IgG1 isotype control treatment remained above 80% viability throughout the entire mAb concentration range (Fig. 9d-e). For example, the E8 mAb reduced MCF-7 and MDA-MB-231 cells viability below 50% at concentrations as low as 5 µg/ml and 2.5 µg/ml, respectively(Fig. 9d-e).. Generally, the antibodies induced comparable reductions in breast cancer cell viability (Fig. 9). This phenomenon has been previously observed for other antibodies72. The observed reduction in cell viability may therefore be mediated through mechanisms such as direct induction of apoptosis, or growth inhibition. Further studies will be required to experimentally confirm the specific cell-death pathways engaged by these anti-DDR1 antibodies. To determine IC50 value with a high confidence, future work will require the experiment to be repeated at a refined concentration range with more technical replicates. Together, the data supports that the anti-DDR1 antibodies selectively decrease the viability of DDR1 expressing breast cancer cell lines.As the AlamarBlue assay measures metabolic activity rather than direct membrane disruption, the observed reduction in viability reflects cytotoxic effects but does not distinguish between apoptosis, necrosis, or other mechanisms of cell death. Next, we tested our antibodies in an Antibody Dependent Cell-mediated Cytotoxicity (ADCC) assay using the same cell lines. ADCC requires binding of the antibody Fab domain to the target cell with binding of the Fc domain to Fc receptors on immune effector cells, triggering immune cell activation and death of the target cell73. Before conducting ADCC assays, we confirmed by SPR that all the novel mAbs were able to bind to two variants (Phe176 and Val176) of the FcƔRIIIa receptor, which is required for ADCC activity (Supplementary Table S1T4 and Fig. S1F12). ADCC can be measured in vitro by monitoring the enzyme activity of lactate dehydrogenase (LDH), a cytosolic enzyme released into the media upon damage to the cell membrane, and an indicator of cytotoxicity. The levels of LDH can then be measured by a coupled enzymatic reaction in which LDH catalyzes the conversion of lactate to pyruvate leading to red formazan product, where lysis is represented by the LDH activity in the treated cells (minus spontaneous LDH activity) divided by the maximum LDH activity in control cells (minus spontaneous LDH activity), expressed as a percentage. A negative lysis value indicates that the experimental conditions gave less cell lysis than spontaneous lysis within each cell line, reflecting a lack of cytotoxic activity under those conditions. Rituximab was used as an assay positive control on Raji cells, where treatment with Peripheral Blood Mononuclear Cells (PBMC, the effector cells) showed significant decrease in cell viability when comparing between doses: 36.5±1.9% cell lysis at a high dose (5 µg/mL), 21.3±1.7% lysis at a medium dose (2.5 µg/mL), and 16.0±1.2% at a low dose (0.1 µg/mL) (P < 0.0001, Fig. 10a), consistent with previous reported findings3,4. Ethanol (%v/v), included as a positive lysis control for the cancer cell lines, also significantly induced cell lysis in a decreasing manner across high to low dose range (Fig. 10b) for all cell lines except for CHO-S at medium-low dose and MCF-7 cells at low dose, where negative and non-significant cell lysis was observed (Fig. 10b). Therefore, overall, all controls again behaved as expected.Fig. 10In-vitro cell killing activity of novel anti-hDDR1 mAbs via ADCC. A 10:1 ratio of Effector (PBMCs):Target (E:T) cells/well were plated in 96-wells plates incubated at 37 ºC with 5% CO2 and treated with 3 different concentrations of mAbs. (a) Raji cell line treated with Rituximab at 5 µg/ml, 2.5 µg/ml, and 0.1 µg/ml doses. One way ANOVA with Tukey’s multiple comparison test was performed to compare among doses (assay positive control). (b) Ethanol (%v/v) treatment on cell lines CHO-S (blue), HEK-293T (red), MCF-7 (green), and MDA-MB-231 (violet) (lysis positive control). (c-f) ADCC activity of (c) A12, (d) B11, (e) C12, and (f) E8 anti-DDR1 mAbs on DDR1 negative (HEK-293-T and CHO-S) control cells, and DDR1 positive (MCF-7 and MDA-MB-231) breast cancer cell lines using human PBMCs as effector cells at E:T cells ratio of 10:1. Graphs show % cell lysis calculated as [(experimental release – spontaneous release)/(maximum release – spontaneous release)] × 100 ± SD. The experiment was performed once, with n=3 for all cell lines and controls, except n=2 for MDA-MB-231 cell line. A two-way ANOVA with Tukey’s multiple comparison test was performed to assess dose and cell line dependent effects on ADCC, with specific comparisons made between DDR1 negative and DDR1 positive cell lines at each dose levels. Significance was represented as * p < 0.05, ** p < 0.01, *** p < 0.001, **** p < 0.0001, “n.s.” non-significant. mAb: Monoclonal antibody, ADCC: Antibody dependent cell-mediated cytotoxicity. In-vitro cell killing activity of novel anti-hDDR1 mAbs via ADCC. A 10:1 ratio of Effector (PBMCs):Target (E:T) cells/well were plated in 96-wells plates incubated at 37 ºC with 5% CO2 and treated with 3 different concentrations of mAbs. (a) Raji cell line treated with Rituximab at 5 µg/ml, 2.5 µg/ml, and 0.1 µg/ml doses. One way ANOVA with Tukey’s multiple comparison test was performed to compare among doses (assay positive control). (b) Ethanol (%v/v) treatment on cell lines CHO-S (blue), HEK-293T (red), MCF-7 (green), and MDA-MB-231 (violet) (lysis positive control). (c-f) ADCC activity of (c) A12, (d) B11, (e) C12, and (f) E8 anti-DDR1 mAbs on DDR1 negative (HEK-293-T and CHO-S) control cells, and DDR1 positive (MCF-7 and MDA-MB-231) breast cancer cell lines using human PBMCs as effector cells at E:T cells ratio of 10:1. Graphs show % cell lysis calculated as [(experimental release – spontaneous release)/(maximum release – spontaneous release)] × 100 ± SD. The experiment was performed once, with n=3 for all cell lines and controls, except n=2 for MDA-MB-231 cell line. A two-way ANOVA with Tukey’s multiple comparison test was performed to assess dose and cell line dependent effects on ADCC, with specific comparisons made between DDR1 negative and DDR1 positive cell lines at each dose levels. Significance was represented as * p < 0.05, ** p < 0.01, *** p < 0.001, **** p < 0.0001, “n.s.” non-significant. mAb: Monoclonal antibody, ADCC: Antibody dependent cell-mediated cytotoxicity. All anti-DDR1 antibodies showed negative cell lysis for the DDR1 negative cell lines (CHO-S and HEK293), indicating no cytotoxicity in the absence of the DDR1 target (Fig. 10). Importantly, ADCC activity was observed specifically in DDR1-expressing cancer cell lines and not in DDR1-negative control cell lines, confirming target-dependent cytotoxicity (Figure 10). Additionally, antibodies demonstrated increasing cytotoxic activity at higher concentrations (Figure 10), consistent with antibody-mediated effector engagement. These findings, together with confirmed FcγRIIIa binding by SPR (Supplementary Table S1T4 and Fig. S1F12), support functional ADCC activity mediated by anti-DDR1 antibodies. A12 mAb induced significant ADCC-mediated lysis in both MCF-7 and MDA-MB-231 breast cancer cell lines, with the highest activity observed in MCF-7 cells at the high dose (34.8±1.41%), and lower yet significant activity in MDA-MB-231 cells (26.6 ± 0.3%). Notably, A12 was the only clone that demonstrated statistically significant dose-dependent increase in cytotoxicity in MCF-7 cells even at the low dose (6.41±0.88%) to high dose compared to DDR1 negative cells (Fig. 10c). Both B11 and C12 showed significant lysis activity against MDA-MB-231 cells across all doses (Fig. 10d,e) but showed only significant lysis of MCF-7 cells at the high dose (8.03±1.82% lysis by B11, and 25.0±1.80% lysis by C12, Fig. 10d,e). Finally, E8 behaved more similarly to clone A12, showing significantly higher lysis in MDA-MB-231 cells across all doses (high: 26.57±0.30%, medium: 13.96±0.39%, low:12.28±1.04%) and in MCF-7 cells at the high (42.32±1.54%) and medium (15.66±1.65%) doses, but not at the low dose, compared to control cells (Fig. 10f). Therefore, in this assay A12 consistently outperformed the other three antibodies, followed closely by E8, while B11 and C12, underperformed. Importantly, the ADCC responses observed for the four antibodies appear to correlate with their binding affinities as reported in Table 2, where A12, with the highest affinity, induced significant cytotoxicity even at the low dose in both cancer cell lines; while B11 with the weakest affinity, elicited only low lysis at the highest dose. E8, and C12 to a lower extent, exhibited intermediate profiles in both affinity and ADCC activity. These findings underscore the potential use of mAbs targeting DDR1-positive breast cancer cells to promote ADCC via engagement of effector cells like neutrophils and NK cells73. Altogether, the results show that targeting DDR1 with the novel antibodies can lead to cell lysis in DDR1 expressing cancer cells via a variety of mechanisms, but not on cells that do not express DDR1, supporting their potential as candidate therapeutic antibodies for further preclinical evaluation. Discussion The development of novel therapies against breast cancer is a fundamental unmet need. In the era of personalized medicine, target identification and validation remain a major bottleneck74. To address this bottleneck, in this study we designed a bioinformatics pipeline to select targets, isolated novel mAbs against the target, and characterized the mAbs’ target binding ability and cytotoxic activity using cell-based models. Our work generated a list of potential targets and a panel of new potential therapeutic antibodies against one selected breast cancer target, the cell surface receptor and tyrosine kinase DDR1. Therapeutic target identification using increasingly available large-scale omics data requires development of comprehensive bioinformatics pipelines. Bulk transcriptomic and proteomic datasets do not resolve gene or protein expression at single-cell resolution and therefore cannot guarantee expression specificity within individual tumor cell populations75. However, our pipeline leverages paired tumor-matched normal samples from the same patient, which provides a powerful approach to identify targets enriched in tumor tissue relative to the patient’s healthy tissue. This design significantly reduces inter-individual variability and strengthens biological relevance of the identified targets. Importantly, the goal of this pipeline was to identify candidate therapeutic targets using robust patient cohort datasets rather than to establish definitive cell-type specificity within individual tumor cell subsets. Single-cell and spatial profiling approaches provide valuable insights into tumor heterogeneity and microevolution, such analysis represent downstream refinement step rather than pre-requisites for cohort-level therapeutic target identification. Our study therefore establishes a robust integrative framework for identifying candidate therapeutic targets using publicly available paired patient multi-omics datasets, with subsequent biological and translational validation representing future work. Pipelines capable of integrating transcriptomic and proteomic data provide a more accurate and functionally relevant overview of target abundance than transcript-level data alone. This is especially important in cancer, where post-transcriptional regulatory mechanisms often uncouple mRNA levels from protein expression16. A pipeline that accounts for both RNA and protein abundance thus offers a substantial improvement in the selection of potential actionable targets. Previous studies have taken different approaches to mapping the cell surface proteome and identifying cancer-specific proteins. For example, Bausch-Fluck et al.76 applied a machine learning-based predictor (SURFY) trained on high-confidence experimental data to define a surface proteome of 2886 proteins with 93.5% accuracy across cell types. Similarly, Cunha et al.77,78 curated 3700 transmembrane proteins and validated 573 genes across tumor and normal tissues using qPCR, emphasizing differential expression patterns and tumor specificity. Building on these efforts, and also on strategies from previous studies35,36, we developed a bioinformatic pipeline that, unlike earlier methods, leverages publicly available data from TCGA and performs donor-specific tumor vs normal tissue comparisons across two omics layers, ensuring that the identified candidate targets are transcriptionally and translationally relevant. However, it is important to acknowledge that there are limitations associated with the TCGA dataset. An additional limitation relates to the use of the RPPA proteomic data, which represents a targeted antibody-based platform and therefore does not provide full proteomic-wide coverage. Our use of RPPA reflects the availability of matched transcriptomic and proteomic datasets within TCGA, enabling integration of orthogonal evidence from paired tumor-normal samples. While this approach does not capture the entire proteome, it provides valuable protein-level validation within the constraints of currently available large-scale patient datasets. Future integration of unbiased mass-spectrometry-based proteomics and single-cell proteogenomic datasets would further enhance the comprehensiveness and depth of this pipeline. For instance, there is low comprehensive comparative data for certain cancer types, such as ovarian cancer, exemplified by the limited representation of tumor and normal donor data for RNA Seq and proteomics, as also noted in79. Thus, our pipeline may not be applicable across all TCGA cancer types. Addressing this limitation requires supplementing the data with additional datasets that provide a balanced number of tumor-normal samples across diverse cancer types, integrating various RNA and proteomic technologies to identify targets and validate their expression comprehensively. Indeed, our approach also leverages multiple databases such as NCBI (Gene/Protein), GeneCards, HGNC, GO, KEGG, GEPIA, HPA, UniProt, The therapeutic target database, and DRUGBANK, among others (see methods), and data from literature to enhance the identification and selection of potential targets (Supplementary File S6). In this study, we focused on breast cancer as it is one of the most prevalent cancers in the world1 and there is comprehensive transcriptomic and proteomic data in the TCGA. By applying stringent filtering criteria and mapping both transcriptomic and proteomic data, we narrowed down the list of targets from an initial 60660 mRNA and 487 proteins to 320 common differentially expressed mRNAs and proteins in breast cancer vs normal tissue (Figs. 2-4). Subsequently, we focused on the 55 cell surface proteins found within the 320 potential targets, all with links to cancer pathways and cell adhesion (Fig. 4). Among these 55 targets, 23 hits (Table 1) were already well-known cancer targets, which validated our selection approach. Importantly, our pipeline identified candidate six cell surface targets (DDR1, STING1, CDH1, HLA-DQA1, CLDN7, and MCT4) with no known or few mAbs in development (Supplementary Table S1T2), indicating that this pipeline can serve as a valuable tool for uncovering novel and/or underexplored therapeutic candidates. The ideal target for extracellular antibody therapy should be abundantly and uniformly expressed on the surface of cancer cells while being absent from normal tissue and should have a large extracellular domain with low levels of heterogenous post-translational modifications. However, it is difficult to find accurate readouts that ensure a target fulfils these criteria. Protein expression in cancer is complex and heterogeneous. While ideally the target should have higher abundance in both mRNA and protein, here we chose targets in which at least one dataset showed significantly higher abundance in cancer, as long as the other dataset did not show significantly lower levels in cancer compared to matched non-cancer tissue. This strategy accounts for the variability inherent to high-throughput omics data acquisition methods. Indeed, our chosen target DDR1 showed significantly higher mRNA abundance in cancer compared to normal tissue and non-significantly different abundance in the protein datasets (Figs. 2-4)20,21,80. While DDR1 was identified as a target for breast cancer, the antibodies developed here could be used to detect or monitor DDR1 expression in other cancer types (Supplementary Fig. S1F180,), contributing to a broader diagnostic toolkit. The role of DDR1 in the normal tissue is well understood, but the implications of DDR1 overexpression during cancer are still unclear. DDR1 is a collagen-activated receptor tyrosine kinase that undergoes autophosphorylation upon ligand binding, regulating cell adhesion, migration, and tissue organization under normal physiological conditions. The anti-DDR1 mAb 3E3 (WO 2013/034933 A1), isolated from immunized mice, inhibits DDR1 activation by binding to the extracellular region of DDR1. Importantly, 3E3 does not block collagen interaction, but still prevents cytoplasmic domain autophosphorylation, likely through an allosteric mechanism that interferes with receptor activation53. During cancer, DDR1 is not normally expressed in immune cells but it is aberrantly overexpressed in tumor cells, including breast cancer20,21, where it contributes to immune evasion by promoting collagen fiber alignment that forms a physical barrier against T cell infiltration. In line with this, the humanized anti-DDR1 antibody PRTH-101 isolated from immunized rabbits that entered phase I clinical trials (NCT05753722) in April 2023, disrupts collagen alignment and enhances T cell penetration into tumors54, highlighting DDR1 as an attractive immunotherapy target. Building on this, we developed novel anti-DDR1 human-derived mAbs with the potential to induce cancer cell death. Unlike existing mouse- and rabbit-derived mAbs in development, our human-derived antibodies should offer lower immunogenicity, making them better suited for therapeutic applications. Four unique mAbs, A12, B11, C12, and E8 were evaluated for their antigen-binding affinity, specificity, and early functional activity. The antibodies exhibited a range of antigen-binding affinities to purified recombinant DDR1, preliminary good specificity towards DDR1 on breast cancer cell lines, and cytotoxic and ADCC activity against DDR1 expressing cell lines, and a lack of binding or cytotoxicity to DDR1-negative cell lines. To our knowledge, these novel clones represent the first fully human anti-DDR1 antibodies derived from a naïve phage display library without prior immunization or affinity maturation, offering both binding affinity, high specificity and Fc-mediated cytotoxicity, which is not yet reported for existing mouse- or rabbit-derived candidates currently under development. ADCC is a key mechanism of action for many therapeutic antibodies and requires interaction between antibody Fc regions and Fcγ receptors on immune effector cells81. In this study, SPR analysis confirmed binding of the anti-DDR1 antibodies to FcγRIIIa, supporting their ability to engage immune effector cells (Supplementary Table S1T4 and Fig. S1F12). The observed target-specific cytotoxicity in DDR1-positive cancer cell lines and absence of cytotoxicity in DDR1-negative controls further supports functional ADCC activity (Figure 10). However, it is important to note that in vitro ADCC assays represent early functional characterization and do not establish clinical efficacy. Clinical therapeutic effectiveness requires extensive preclinical and clinical evaluation and is beyond the scope of this study. Future investigation of these novel mAbs is warranted in collagen-rich tumor microenvironments, along with in-vivo validation using xenograft mouse models. Additional antibody characterization including epitope mapping, immunogenicity profiling, invasion (transwell) assay, immunocytochemical, tissue cross-reactivity studies and stability assessments should be conducted to further evaluate therapeutic viability. It is important to emphasize that while DDR1 has been identified as a promising candidate therapeutic target based on multi-omics analysis and functional antibody validation, establishing clinical effectiveness requires extensive preclinical and clinical studies. Several DDR1-targeted therapies are currently under investigation, supporting the biological and therapeutic relevance of DDR1. The present study demonstrates the effectiveness of the pipeline for identifying candidate targets and generating functional antibodies, providing a foundation for further therapeutic development. In conclusion, our computational pipeline can identify targets for therapy or diagnostics by using paired tumor-normal samples from both transcriptomic and proteomic datasets. Here, we used this pipeline to dissect combined transcriptomic and proteomic data from cancer and non-cancer tissue from the same patient. This approach allowed for the identification of 23 cell surface proteins as potential therapeutic targets. We selected DDR1 as a target, isolated new human antibodies against DDR1 from a naïve phage display library and demonstrated their cytotoxic activity against DDR1 expressing cells. Our bioinformatic pipeline provides a robust and practical framework for identifying candidate therapeutic targets using publicly available patient cohort multi-omics data. While further validation in vivo models is required to establish clinical applicability, this approach represents an important discovery step toward identifying promising targets and developing novel therapeutic antibodies. Methods R packages The detailed list of R packages used in the study are outlined in Supplementary Table S1T5. Transcriptomic and proteomic public data download and preprocessing This study analyzed publicly available human genomic data from The Cancer Genome Atlas (TCGA), in accordance with relevant guidelines and regulations. The original studies from which TCGA data were obtained received ethics approval and informed consent from all participants, as documented in their respective publications. No new human participants were recruited by the authors for this study, and no identifying information, images, or videos of human participants were obtained. The TCGA-BRCA RNA-Seq and RPPA datasets for donor normal and tumor tissues were downloaded from Genomics Data Commons (GDC) Data Portal(https://portal.gdc.cancer.gov). Differentially expressed genes (DEG) were identified from paired donor IDs that represent normal and tumor donor IDs with same TCGA submitter cases i.e., a total of 113 samples from the transcriptomic dataset and 33 samples from the proteomic dataset. We employed “TCGAbiolinks-2.30.4” package functions such as GDCquery, GDCprepare, GDCdownload for data curation which was focused on mRNA data involving 19,962 protein coding mRNAs and 487 proteins from the RPPA dataset. The pre-processing of RNA-Seq and proteomic raw read counts involved data filtering and data normalization. In the case of RNA Seq data, counts per million (CPM) values were calculated using the function ‘cpm’ from the “edgeR−4.0.16” package. Genes were kept for analysis if they had CPM > 0.5 in at least 10 samples, resulting in 16,404 mRNAs. This filtering step reduces the influence of lowly expressed genes and improves statistical power. The density distribution plot of log₂-CPM values before and after filtering is shown in Fig. 2b. In the case of the proteomic dataset, 1,797 proteins were omitted, resulting in a curated set matrix containing 457 proteins. The normalization process for both data sets was performed using the “edgeR−4.0.16” package function ‘calcNormFactor’ for RNA-Seq dataset, where the trimmed mean of M-values (TMM) normalization method82 was applied to minimize the impact of systemic biases and technical variations, and to ensure more accurate comparisons of gene expression levels across paired donor IDs. The “preprocessCore-1.64.0” package function ‘normalize.quantiles’ was applied for the proteomic dataset by adjusting the raw data with missing values based on the assumption that the data is missing at random83, as illustrated in Fig. 3b. mRNA and protein expression analysis DEGs analysis was performed using statistical methods on data comparison in “R-4.3.3”. For RNA-Seq dataset, gene-wise dispersions were estimated using “edgeR−4.0.16” package37 empirical Bayes approach to identify the genes that are differentially expressed between the different conditions of groups, and a quasi-likelihood negative binomial generalized log-linear model was fitted (glmQLFit). Differential expression was assessed using the quasi-likelihood F-test (glmQLFTest), which provides robust control of type I error. Statistical significance was defined as genes with adjusted P-value < 0.05 using the Benjamini Hochberg method84 where with a log2 fold change cutoff of > |0.1|. By applying these parameters, we identified 8,040 higher expression, 4,791 lower expression, and 3,573 non-significantly different expression for the tumor vs normal mRNA sample comparison. For RPPA dataset, differential protein abundance was determined using the “limma−3.58.1” package38, which fits a linear model to each protein followed by an Empirical Bayes moderated t-test to identify proteins whose expression is different. 111 higher abundance, 156 lower abundance, and 190 non-significantly differently abundant proteins were identified, as illustrated in Fig. 2,3. Mapping targets for both omics’ dataset Protein targets in the RPPA dataset were first curated to ensure consistency with gene-level expression data. Targets that did not correspond to a valid gene symbol (e.g., acetylation events, phosphorylated proteins, or protein subunits) or contained non-standard characters such as punctuation were excluded and marked as missing values (Supplementary Files S2–S4). In addition, 48 peptide targets in our RPPA dataset were not listed in the TCGA Expanded_Ab_List_Updated.txt reference file (Supplementary File S4). Therefore, we performed manual mapping using multiple databases, including HGNC (HUGO Gene Nomenclature Committee, Home | HUGO Gene Nomenclature Committee), GeneCards (The Human gene database, GeneCards - Human Genes | Gene Database | Gene Search85), and NCBI Gene/Protein, by cross-referencing reference gene IDs and sequence information (Supplementary Files S2, S3). Where valid HUGO gene symbols were identified, these were incorporated into our R pipeline. Targets that could not be confidently mapped, or corresponded to post-translational modifications (PTM), protein subunits, or ambiguous symbols, were assigned as missing values and excluded from downstream analyses. Targets in the RPPA dataset that corresponded to phosphorylation-specific antibodies (e.g., acetylation events, phospho-proteins) were excluded to ensure unambiguous gene-level mapping between proteomic and transcriptomic datasets. Phosphorylation-specific RPPA analytes represent modification states of proteins rather than total protein abundance and therefore cannot be directly integrated with gene-level transcript expression data. For this reason, only total protein analytes with clear one-to-one gene mapping were retained for downstream integration analysis. Functional enrichment analysis We used the “clusterProfiler-4.10.1” package in the Bioconductor toolkit of “R-4.3.3” and the ‘compareCluster’ function with ‘enrichGO’ and ‘enrichKEGG’ to perform over representation (or enrichment) analysis. The enrichment analysis was performed using the GO (Gene Ontology Resource)41 and KEGG (KEGG: Kyoto Encyclopedia of Genes and Genomes) databases40. The final gene list included targets with a log2 fold change > |0.1| and a P-value of ≤ 0.05, and the gene names were converted into ENTREZID to represent genes in the NCBI database. The GO aspects of Cellular Component (CC), Biological Process (BP), and the Molecular Function (MF) were used in the overrepresentation analysis using the function enrichGO, and KEGG analysis was performed by the function enrichKEGG, with the partition P-valueCutoff = 0.01 and the estimated false discovery rate qvalueCutoff = 0.01. Finally, the Barplot and dotplot functions were used to plot the graphs. The DEGs that were identified from the transcriptomic and proteomic data were made into a subset based on their known roles as cell surface protein. We extracted ontology categories such as biological process, molecular functions, and signalling pathways from “UniprotR−2.4.0” package (UniProt database) in “R-4.3.3”. The type of transmembrane protein (single type I, type II, and type III membrane protein, as well as multi-pass membrane) were reported in this study for the selected subsets. GEPIA and HPA analysis of selected targets, database searching, and hit validation The GEPIA62 database, based on TCGA and GTEx data, utilized transcripts per million (TPM) metrics to ensure accurate quantile normalization for both total number and length of each transcript, allowing comparability of gene expression across TCGA pan-cancer datasets or matched normal samples from both TCGA and GTEx. Gene expression was determined based on the similar expression pattern that cluster along the chromosome, which is different than our pipeline which focuses exclusively on TCGA-matched tumor-normal samples to identify potential targets. GEPIA database was used to identify the expression levels of the upregulated cell surface targets in tumor and normal tissue types. We focused on genes that were greater than the top 10% of all the gene expression levels in the GEPIA database (http://gepia.cancer-pku.cn/), based on their tumor TPM normalized values. To ensure consistency, we customized the log2FC and P-value statistical parameters in GEPIA database to align with the statistical settings of our pipeline when curating the information. DDR1 protein levels were assessed in both healthy and tumor samples using immunohistochemistry (IHC) and subcellular localization datasets from HPA. “HPAanalyze-1.27.2” and “hpar-1.50.0” package (The Human Protein Atlas database) in “R-4.3.3” were used to retrieve the IHC dataset. The HPA IHC data are derived from antibody-based staining of human tissue sections, enabling visualization of protein distribution and intensity in both normal and cancer samples. Specifically, the datasets for normal tissue, pathology (cancer), and subcellular localization were downloaded using hpaDownload function, and subset with hpaSubset function. The data was visualized via the hpaVis functions (hpaVisTissue, hpaVisPatho, hpaVisSubcell). Various other databases were used to search for known functions of potential targets, including The therapeutic target database50, DRUGBANK51, mAb sequence patent search (Natural Antibody, BLAST: Basic Local Alignment Search Tool (nih.gov), mAb sequence search (abYsis), commercial antibody search (Hand-picked Antibodies, ELISA Kits & Proteins for Research (antibodies-online.com), clinical therapeutic databases (Home | ClinicalTrials.gov), and patents (The Lens - Free & Open Patent and Scholarly Search) were used to search for mAb-associated patents and commercial mAbs. Where possible, to validate the potential hit and minimize false positives, analysis of published experimental literature was performed, focusing on ELISA, western blotting, and immunohistochemistry assays (Supplementary File S6). Searches were conducted between March 15-May 20, 2023. Cells, reagents, and antibodies All bacterial strains used in this study are described in Supplementary Table S1T6. ExpiCHO-S (Chinese hamster ovary) cells (Thermo Fisher Scientific cat#A29127), CHO-S cells (Thermo Fisher Scientific, cat# A11557-01), and FreeStyle 293-F human embryonic kidney (HEK) cells (Thermo Fisher Scientific, cat# R79007) were maintained as suspension cell cultures with shaking at 120 rpm at 37 ºC with 8% CO2, in either ExpiCHOTM expression media (Gibco, cat# A2910001) for ExpiCHO cells, Freestyle 293 expression media (Gibco, cat# 12338018) for HEK cells, and CD CHO AGTTM medium (Gibco, cat# 12490025) for CHO-S cells, supplemented with 100 U/mL Penicillin/Streptomycin (Pen-Strep, Gibco, cat# 15140122), and 0.4% anti-clumping agent (ACA) (Gibco, cat# 0010057DG). Cells were passaged and transfected as per the manufacturer’s instructions. MCF-7 and MDA-MB-231 cancer cells were obtained from Professor Kristofer Thurecht (UQ) or from National Centre for Cell Science (NCCS), Pune, India. MCF-7, MDA-MB-237, and HEK293-T cells (NCCS Pune, India) were grown in DMEM media (Gibco, cat# 11–965-092) supplemented with 10% fetal bovine serum (FBS, Sigma-Aldrich, cat# A3160402), 100 U/mL Penicillin/Streptomycin, 20 mM GlutaMAXTM (Gibco, cat# 35050061), and 1% (V/V) sodium pyruvate (Gibco, cat# 11360070). Raji cells (NCCS Pune, India) were grown in RPMI 1640 media (Gibco, cat# 11875119), supplemented with 10% FBS, 100 U/mL Penicillin/Streptomycin. Cells were passaged every 2 to 3 days using a 1:5 dilution ratio, and were maintained at 37 °C, 5% CO2 in a humidified incubator. Cells tested mycoplasma negative at both UQ and NCCS. The following commercial antibodies were used in this study: PE-conjugated anti-human DDR1 (CD167a) (Biolegend, cat# 334005/25, RRID: AB_2261656); Mouse anti-M13 antibody (Sino biological, cat# 11973-MM05T, RRID: AB_2857926); HRP conjugated goat anti-mouse IgG antibody (Cell Signalling, cat# 7076, RRID: AB_11125547); DyLight650 conjugated goat anti-mouse IgG antibody (Abcam; cat# 96874, RRID: AB_10679531); Alexa Fluor 647- AffiniPure Goat Anti-Human IgG (H+L) (Jackson ImmunoResearch, cat# 109–605-003, RRID: AB_2337880); and Rituximab (Ristova® from Roche, USA; 10 mg/ml, #batch H0185B01). For an isotype control, an in house manufactured human anti-HIV gp41 IgG1 antibody was obtained from the National Biologics Facility (NBF, UQ, Australia). Construction of hDDR1-GFP plasmid, transfection, and flow cytometry The CHO codon-optimized full length hDDR1 gene cloned into pTwistCMV was obtained from Twist Biosciences, USA (LZCBI174). To generate the hDDR1 extracellular and transmembrane (ECD-TM) domain construct (residues 1–438, NCBI Reference NP_054699.1), we subcloned the ECD-TM region into the pEGFP-N1 vector (Clontech, Takara Bio, San Jose, CA, USA). The ECD-TM fragment was PCR-amplified from LZCBI174, with primers engineered to introduce BamHI and NheI restriction sites. Both the PCR-amplified insert and pEGFP-N1 were digested with BamHI and NheI, gel purified and ligated in-frame with the C-terminal eGFP into pEGFP_N1, producing hDDR1(ECD-TM)-pEGFP-N1 (LZCBI178). Plasmid DNA was purified using PureLink HiPure Plasmid Maxiprep Kit (Invitrogen) and the construct was verified by Sanger Sequencing. Surface expression of hDDR1 was first confirmed by flow cytometry using commercial PE-Conjugated mouse anti-human DDR1 antibody. ExpiCHO cells were seeded to a density of 3–4 million cells/mL, greater than 98% cell viability and transfected with the pEGFPN1-hDDR1-GFP plasmid (1 µg DNA/mL cell suspension) using polyethyleneimine (PEI-Max, Polysciences, Taipei, Taiwan) at a ratio of 1:4 (w/v) prepared in OptiPROTM serum-free medium (Gibco, cat# 12309050) followed by incubation for 4–6 h, and shaking at 37 ºC. The ExpiCHO cells were diluted to 1 million cells/mL and incubated with shaking at 37 ºC for two days prior to flow cytometry analysis as follows. Cells were washed and resuspended at 0.8 × 106 cells/mL in PBS-1% FBS. Aliquots of 0.8 × 106 cells were incubated either alone or with 1/20 dilution of PE-conjugated anti-human DDR1 antibody for 1 hr at room temperature, followed by three washes with 200 µL PBS-1% FBS, centrifuged at 300 × g for 10 min, resuspended in PBS and then analyzed on a CytExpert software in a CytoFLEX S flow cytometer (Beckman Coulter) as described in86. Soluble biopanning Soluble recombinant protein biopanning was performed as previously described by Panagides et al.86, using 1013 cfu of a naïve human scFv phage display library (Jones-Mahler library) provided by NBF55,65, with the following modifications. The recombinant hDDR1 extracellular domain (amino acids 21–416 including a C-terminal hexahistidine tag, Sino Biological, cat# 10730-H08H) was used as the positive antigen, and a mix of hCD64 (Sino biological Pvt Ltd., cat# 10256-H08H), hCD32a (Sino biological Pvt Ltd., cat# 10374-H08H), and hCD16a (Sino biological Pvt Ltd., cat# 10389-H08H) were used as negative antigens for depletion of His tag specific binders. This commercial source of hDDR1 has been used in previous studies to isolate functional anti-DDR1 antibodies, giving confidence that it presents that correct DDR1 conformation to the phage library54. The wash stringency was increased by performing 3 washes in round 1 and 5 washes in rounds 2 and 3. Polyclonal and monoclonal ELISA Binding reactivity of phage pools from each biopanning round was assessed via polyclonal ELISA, as described previously86, but using recombinant hDDR1 as the antigen, and serial 1 in 10 dilutions of phage samples. To identify individual phage binders specific to hDDR1 positive antigen, monoclonal ELISA was performed on randomly picked E. coli clones from rounds 2 and 3, as described previously86. Clones were assessed based on the absorbance values, with control samples measured in triplicate. Cut-off values to determine positive clones were calculated as the mean plus three times the standard deviation (mean + 3xSD) of the triplicate control samples. Clones with absorbance at 450nm above the cut-off value were considered as positive binders. Sequencing and cross-reactivity testing of phage positive clones PCR products amplified from ELISA-positive phagemid clones were sent for Sanger Sequencing at the Australian Genome Research Facility (AGRF, Australia), using phagemid-derived forward and reverse primers given in Supplementary Table S1T7. Following sequencing, selected phage scFv clones were tested for cross-reactivity using soluble antigen by ELISA as above. These phage binders were rescreened against a 1 µg/mL mixture of His-tagged ECDs from hCD64, hCD32a, hCD16a, 1 µg/mL ECD of His-tagged hDDR2 (amino acids 22–399 including a C-terminal hexahistidine tag, Sino Biological, cat# 10209-H08H), or 1 µg/mL ECD of His-tagged hDDR1 antigen coated ELISA plate wells. Monoclonal-phage binding analysis To test binding of phage clones to cell-surface hDDR1-GFP fusion protein as described previously56, with the following modification: cells were blocked with 10 mL milk-PBS (MPBS), and phages were diluted 1:1 in 4% MPBS prior to incubation. Blocked cells (50 µL) and phage particles (50 µL) were combined. Centrifugation was performed at 300 × g for 10 min. Both primary (anti-M13 phage mouse) and secondary (goat anti-mouse DyLight650) antibodies were diluted in 2% MPBS. Fixation was carried out with 100 µL of 4% paraformaldehyde (PFA) for 15–20 min on ice. Finally, cells were resuspended in 200 µL PBS and analysed on the CytoFLEX S flow cytometer. Mean fluorescence intensity (MFI) was used as a measure of binding of the different mAbs to hDDR1-GFP fusion protein. Full-length IgG1 mAbs reformatting, expression, and purification Candidate phage clones were reformatted into human IgG1 (hIgG1) using the InFusion cloning system (Clontech, Takara Bio) and a set of mAbXpress vectors, as previously described67. These vectors contain human IgG1 constant region sequences. The variable regions were amplified from the purified phagemid vectors using primers outlined in Supplementary Table S1T7. Full length human mAbs were expressed and purified as described previously86, with following modification. Culture supernatants were harvested when cell viability dropped below 65% (8–10 days post-transfection) and loaded onto a prepacked 1 mL HiTrap MabSelect SuRe column (Cytiva, Macquarie Park, NSW, Australia) using an AKTA pure (Cytiva). Eluted proteins were neutralized to pH 7 with 1 M Tris-HCl pH 8, and buffer exchange was performed using Amicon Ultra Centrifugal Filters (cat# UFC903024) with a molecular weight cut-off (MWCO) of 30 kDa. Purified mAbs were assessed by SDS-PAGE under reducing and non-reducing conditions. To reduce protein samples, 40 mM DTT (final concentration) was added. Samples were heated to 95 ºC for 5 min, loaded onto 4–15% stain-free gradient gels (BioRad, cat#4568084), and the gel was visualized on a BioRad Chemidoc imaging system. Analytical size exclusion high-performance liquid chromatography (SE-HPLC) SE-HPLC was conducted to determine the mAb stability and impurities for downstream functional studies on a Superdex 200 Increase10/300 GL (10 x 300 mm, 13µm, GE Healthcare, Chicago, IL, USA) column operated at 25 ºC using the Dionex Ultimate 3000 UHPLC system (Thermo Fisher Scientific, Waltham, MA, USA). Twenty-five µg protein was injected, and an isocratic elution was performed for 45 min at a flow rate of 0.5 mL/min, in a column pre-equilibrated with 50 mM sodium phosphate buffer, 300 mM NaCl, and 0.05% NaN3 at pH 6.8. All buffers were filtered with a 0.2 µm cutoff polyether sulfone membrane filtered (Pall Corporation) and degassed before use. Detection was performed by monitoring UV absorbance at 280 nm, and data were analyzed with Chromeleon software (Thermo Fisher Scientific, Waltham, MA, USA). The anti-hDDR1 mAb samples were overlayed on top of the IgG1 human isotype control antibody (NBF, UQ, Australia) standard to determine the purity, peak identification, and any modification with baseline and background correction. Reformatted full length mAb binding specificity analysis Binding of each monoclonal antibody (mAb) to various cell lines (MCF-7 and MDA-MB-231 constitutively expressing DDR169, and ExpiCHO87) was assessed to determine specificity for the target receptor. Flow cytometry assays were performed as previously described in Perera et al.56, with minor modifications. Briefly, ~0.8 × 10⁶ cells per sample were plated in 96-well round-bottom plates, washed, and incubated with each mAb (final concentration 5 µg/mL) for 1 h at 4 °C. After washing, cells were incubated with Alexa Fluor 647-conjugated goat anti-human IgG secondary antibody (1/400) for 1 h at 4 °C in the dark, washed, fixed in 0.5% PFA, and analyzed on a CytoFLEX flow cytometer, and data were processed using FlowJo v10.6.2 (FlowJo LLC). Expression of hDDR1 was confirmed using the positive control PE-conjugated anti-human DDR1 antibody at a final dilution of 1/20. MFI was used as a measure of binding of the different mAbs to DDR1. All experiments were performed in triplicates. Surface plasmon resonance (SPR) analysis Multicycle kinetics/affinity using capture was performed on a Biacore 8K+ instrument (Biacore Life Sciences, Cytiva) and analyzed with Biacore Insight Evaluation (version- 3.0.11. 15423) software to determine the binding affinities of the anti-hDDR1 mAbs. HBS-P+ buffer (10 mM HEPES, 150 mM NaCl, 3 mM EDTA, 0.05% v/v surfactant P20; pH 7.4) was used as running buffer and to dilute the analytes. A CM5 Series S sensor chip (Cytiva, cat# 29149603) with immobilized anti-His antibody was used to capture His-tagged hDDR1, hDDR2 and CD16a ligand (FcGR3a 176-Phe or 176-Val Variant) variants at 0.2 µg/mL onto flow cell 2 for 180 s at 10 µL/min. Afterwards, 1:2 serial dilutions of 100 nM to 0.78 nM and 250 nM to 3.9 nM of anti-hDDR1 mAbs (A12, B11, C12, E8) including Isotype control antibody were injected for hDDR1, hDDR2 and Fc gamma receptor binding over both flow cells at 30 µL/min. To determine antibody binding to hDDR1 and hDDR2 receptor, the association and dissociation phases were 300 s and 1200 s respectively. To determine antibody binding to Fc Gamma Receptor, the association and dissociation phases were 180 s and 600 s, respectively. After each cycle, the sensor surface was regenerated with 10 mM glycine pH 1.5 for 30 s at a flow rate of 30 µL/min for both flow cells. Binding kinetic assays were conducted at 25 ºC. Data were double reference subtracted by subtraction of flow cell 1 from flow cell 2, and subtraction of the 0 nM analyte cycles. Sensorgram curves were fitted using a 1:1 binding model. Kinetic parameters (association rate constant (ka), dissociation rate constant (kd) and the equilibrium dissociation constant (KD)) were derived based on the global fit for each analyte. Isolation of human peripheral blood mononuclear cells (PBMCs) and serum All methods were carried out in accordance with relevant guidelines and regulations. The use of human blood (isolated PBMCs) samples in this study was approved by the IIT Delhi Institute Ethics Committee under approval number VS-0825-REA78. Informed consent was obtained from all participants and/or their legal guardians prior to sample collection, and no identifying information, images, or videos of human participants are included in this manuscript. Peripheral blood samples (10 mL) were collected from five (2 times) healthy adult individuals at IITD Hospital. The human blood was diluted with an equal volume of sterile PBS at room temperature and gently mixed in a 50 mL falcon tube by manual rotation. Following this, the diluted blood was layered on top of Ficoll-HisepTm LSM (Himedia, cat# 1077) and centrifuged in a swing bucket rotor at 1500 rpm for 30 min at 18 °C. The cloudy layer containing PBMCs was carefully aspirated without disturbing the interphase and centrifuged again in equal volume of PBS to remove the excess Ficoll-HisepTm LSM. The supernatant was decanted, and PBMCs from all individuals were mixed and pooled.. AlamarBlue assay for cell viability Prior to the screen, to determine the cellular response within the assay, optimization of incubation time and plating density for each cell line was performed in triplicate (Supplementary Fig. S1F13a-e) according to the manufacturer’s instructions. MDA-MB-231, MCF-7, HEK293T, and CHO-S cells were incubated in 96-wells flat bottom plates at a concentration of 5000 cells/well in their respective complete medium (Cell Viability activity) in 5% CO2 at 37 ºC. After 24 h incubation, the cells were treated with 5 different concentrations of anti-hDDR1 antibody (1.25–20 µg/ml) and incubated for 24 h. Then, the AlamarBlue reagent (Thermo Scientific, cat# DAL1025) was added directly into the culture media at a final concentration of 10% v/v and the plate was wrapped in aluminium foil and returned to the incubator for 4 h for all cell lines except for MDA-MB-231 cells which required 10 h incubation. Absorbance of the plate was measured at 570 nm with reference at 600 nm using Gen5 software in an Epoch microplate reader (BioTek Instruments). As a positive control, Raji cells cultured in RPMI media and treated with Rituximab were used to validate the assay. IgG1 isotype control human antibody (NBF, UQ, Australia) was used as a negative control to compare with anti-hDDR1 mAbs. Also, the non-DDR1 expressing cell lines (HEK293T and CHO-S) were used as a negative control to compare with DDR1 expression cell lines (MCF-7 and MDA-MB-231) with respect to each anti-DDR1 mAb. As a positive control for cell lysis, ethanol was used at 1.25–20 % v/v concentrations. Furthermore, complete media only and cells only controls were used to calculate relative cell viability. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathrm{Re} lative\,Cell\,Viability\,(\% ) = \frac{{[(Eoxi600xA570) - (Eoxi570xA600)/(Ered570xC600) - (Ered600xC570)]treatment}}{{[(Eoxi600xA570) - (Eoxi570xA600)/(Ered570xC600) - (Ered600xC570)]cellsonly}} \times 100$$\end{document} Where the molar extinction coefficient (E) of oxidized AlamarBlue reagent at 570 nm = 80586 and at 600 nm = 117216, the absorbance of test wells is represented as A570 and A600, the molar extinction coefficient (E) of reduced AlamarBlue reagent at 570 =155677 and at 600 nm = 14652, and the absorbance of negative control wells (complete media only wells) are C570 and C600. Antibody-dependent cellular cytotoxicity (ADCC) assay Freshly isolated human PBMCs (see above) were used to perform this assay. PBMCs (50000 cells/well) and target cells (MCF-7, MDA-MB-231) were used as DDR1 positive breast cancer cells, (HEK293T, CHO-S) were used as DDR1 negative control cells; 5000 cells/well) were mixed at 10:1 [Effector (PBMCs): target cells] ratio were suspended in RPMI media and seeded in a 96-wells flat bottom plate for overnight incubation at 37 °C with 5% CO2. The next day, cells were treated with anti-hDDR1 monoclonal antibodies (at 3 different concentrations - 5 µg/ml (marked as “High”), 2.5 µg/ml (“Medium”), and 0.1 µg/ml (“Low”)) to a final volume of 100 µl per well. The positive control was Rituximab with Raji cell line (10:1 ratio) to validate the assay. As a positive control for cell lysis, ethanol was used at 5, 2.5 and 0.1% v/v concentrations. Additional wells containing cells in either RPMI (spontaneous release) or 1X lysis buffer (maximum release) were prepared. After overnight incubation at 37 °C with 5% CO2, the LDH activity in the cell culture supernatants was measured using the Cytoquant LDH Cytotoxicity Assay Kit (Invitrogen, cat# C20300/1) according to the manufacturer’s instructions. Percent lysis was calculated as follows: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\% {\text{ }}lysis{\text{ }} = \frac{{\left( {\exp erimentalrelease - spon\tan eousrelease} \right)}}{{(\max imumrelease - spon\tan eousrelease)}} \times 100$$\end{document} Absorbance at 490nm with reference at 680nm was measured using Gen5 software in an Epoch microplate reader (BioTek Instrument). Statistics All statistical analyses and graphs were generated using GraphPad Prism Software v.8 (GraphPad Software). For the antibody binding flow cytometry assay, MFI from three replicates were used to perform a One-Way ANOVA with a Tukey’s multiple comparisons test to assess the significance in antibody binding among different cell lines. A Two-Way ANOVA with a Tukey’s multiple comparisons test was used to evaluate the significant differences between factors such as varying antibody doses and different cell lines for ADCC. n.s. non-significant, *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001. Supplementary Information Supplementary Information 1. Supplementary Information 2. Supplementary Information 3. Supplementary Information 4. Supplementary Information 5. Supplementary Information 1. Supplementary Information 2. Supplementary Information 3. Supplementary Information 4. Supplementary Information 5.