Integrated transcriptomic and proteomic validation identifies SLC35D3 as a tumor-selective surface antigen for colorectal and neuroendocrine carcinomas.

1
Introduction
Although antibody–drug conjugates (ADCs), bispecific T-cell engagers (TCEs) and chimeric antigen receptor (CAR) T cells have transformed the oncology therapeutic landscape, their clinical success is still limited by on-target, off-tumor toxicities arising when the selected antigen is not exclusively tumor-restricted [[1], [2], [3]]. Epidermal growth factor receptor (EGFR), epithelial-cell adhesion molecule (EPCAM) and HER2, which is encoded by ERBB2 (erb-b2 receptor tyrosine kinase 2) gene, for instance, are extensively utilized targets whose baseline expression in skin, gastrointestinal mucosa or hepatocytes narrows the therapeutic window and causes dose-limiting adverse effects [[4], [5], [6]]. Consequently, the identification of "background-silent" surface proteins — those which are abundant on malignant cells but virtually absent from critical organs and tissues — has become a central goal for next-generation immunotherapies.
The availability of large public transcriptomic resources now enables systematic pursuit of this objective. The GTEx consortium profiles a broad spectrum of histologically normal human tissues, and therefore offers a robust baseline against which tumor expression can be compared [7]. Complementarily, the TCGA provides RNA-seq data for a broad range of tumor types together with normal controls [8]. Leveraging these repositories jointly allows investigators to examine thousands of genes in silico before prioritizing the most promising candidates for experimental testing.
Here, we applied this dual-dataset strategy to colorectal cancer (CRC), the world's second leading cause of cancer death [9]. A three-tier filter retained transcripts that (i) encode membrane proteins, (ii) show a median GTEx expression <1 RPKM in every organ, and (iii) are ≥10-fold up-regulated in TCGA-COAD tumors relative to normal colon. Only two genes met all three criteria; among them, SLC35D3 (solute carrier family 35, member D3)—a little-studied nucleotide-sugar transporter with no ongoing drug-development programs, unlike targets such as LY6G6D [10] which was another one met the criteria—offered the greatest opportunity for first-in-class therapeutic exploration. Subsequent immunohistochemistry (IHC) on extensive tissue microarrays and quantitative flow cytometry confirmed that SLC35D3 protein expression correlates with its transcript signal and is localized on the plasma membrane of several cancer types while remaining undetectable in critical normal tissues including brain, heart, liver, lung and kidney. These observations nominate SLC35D3 as a previously unexplored yet highly druggable antigen.

2
Materials and methods
2.1
Public RNA-seq analysis
RNA-seq data for normal tissues were taken from GTEx v4, and data for normal colon and CRC from TCGA-COAD (stddata_2015_06_01). Swiss-Prot annotations were screened to identify genes encoding a signal peptide, extracellular domain, GPI anchor or at least one transmembrane helix, thereby selecting putative surface proteins. Genes whose median expression across 30 GTEx tissues was below 1 RPKM were kept, and those showing a ≥10-fold higher median expression in CRC than in normal colon were flagged as tumor-enriched. This workflow yielded proteins that are scarcely expressed in normal tissues but strongly up-regulated in CRC. Because GTEx and TCGA differ in sample procurement and processing, we avoided direct cross-cohort comparisons: the low normal-tissue expression filter was applied within GTEx and tumor-versus-normal fold changes were evaluated within TCGA-COAD. Therefore, no cross-cohort batch correction was performed, and absolute expression levels should be interpreted cautiously across datasets. Their broader relevance was assessed with the TCGA Pan-Cancer dataset (downloaded from the UCSC Xena Browser), which was plotted in R 4.3.2 with ggplot2 4.0.0, while normal-tissue profiles from GTEx were illustrated in GraphPad Prism v9 (Dotmatics, Boston, MA, USA) alongside clinically investigated TCE and CAR-T targets. For Table 1, statistical significance of tumor-versus-normal differences in TCGA-COAD was assessed using Welch's t-test.

2.2
Cell lines
The human pancreatic islet cell carcinoma line QGP-1 (Japanese Collection of Research Bioresources Cell Bank, Osaka, Japan; Cat. No. JCRB0183), human small-cell lung carcinoma line NCI–H1930 (American Type Culture Collection (ATCC), Manassas, VA, USA; Cat. No. CRL-5906), human colorectal carcinoma line LoVo (ATCC; Cat. No. CCL-229), human gastric carcinoma line SNU-16 (ATCC; Cat. No. CRL-5974), and human colorectal carcinoma line HCT 116 (ATCC; Cat. No. CCL-247) were used. QGP-1, NCI–H1930, HCT 116, and SNU-16 cells were maintained in RPMI-1640 medium supplemented with 10% heat-inactivated fetal bovine serum (FBS; 56 °C, 30 min) at 37 °C in a humidified atmosphere of 5% CO2. Adherent cells were detached with 0.25% trypsin-EDTA and they were passaged once or twice per week by reseeding into fresh culture flasks containing the same medium. LoVo cells were cultured in DMEM/F-12 medium (Thermo Fisher Scientific, Waltham, MA, USA) supplemented with N-2 Supplement (Thermo Fisher Scientific), 20 μg/mL insulin (Thermo Fisher Scientific), 20 ng/mL recombinant human EGF (Thermo Fisher Scientific), 10 ng/mL recombinant human basic FGF (Sigma-Aldrich, St. Louis, MO, USA), 2.9 mg/mL glucose (Sigma-Aldrich), and 4 mg/mL AlbuMAX™ I Lipid-Rich BSA (Thermo Fisher Scientific) at 37 °C in a humidified atmosphere of 5% CO2. LoVo cells were detached with Accutase (Innovative Cell Technologies, San Diego, CA, USA) approximately every two weeks and passaged by reseeding into new flasks with fresh complete medium.

2.3
Tissue microarray and immunohistochemistry
Tissue microarrays (TMAs) containing colorectal carcinoma (CO992a), small-cell lung carcinoma (LC818), pancreatic neuroendocrine tumor (PA806), a multi-tumor panel (MC2082b) and normal tissues (FDA999l) were purchased from US Biomax (Derwood, MD, USA). As procedural controls, formalin-fixed, paraffin-embedded (FFPE) cell blocks were prepared using iPGell (Genostaff, Tokyo, Japan) from cell pellets of (i) HCT 116 cells transiently transfected with a human SLC35D3 expression vector (NCBI RefSeq: NP_001008783.1; hereafter referred to as HCT 116-hSLC35D3), (ii) mock-transfected HCT 116 cells (HCT 116-mock), and (iii) the endogenously SLC35D3-positive line NCI–H1930. The SLC35D3 expression vector was based on pCMV-Entry (OriGene Technologies, Rockville, MD, USA).
Immunohistochemistry was performed on a BOND-III autostainer (Leica Biosystems, Nussloch, Baden-Württemberg, Germany) using a rat monoclonal anti-SLC35D3 antibody 5H11-D10 (10 μg/mL), which was generated under contract with Cell Engineering Co., Ltd (Osaka, Japan). 4 μm sections were deparaffinized, subjected to heat-induced epitope retrieval in ER1 buffer (Leica Biosystems) at 95 °C for 20 min, and endogenous peroxidase activity was quenched. After a 30-min protein block (Agilent Technologies Denmark ApS (Dako), Glostrup, Denmark), sections were incubated with the primary antibody for 30 min at room temperature. Following washes, rabbit anti-rat IgG (Jackson ImmunoResearch, West Grove, PA, USA) at 10 μg/mL for 30 min at room temperature and an HRP-labelled polymer reagent (Leica) for 30 min at room temperature were applied sequentially. The signal was developed using 3,3′-diaminobenzidine, and the nuclei were counterstained with hematoxylin. The slides were then dehydrated and mounted in malinol. After staining, the slides were digitally scanned with a NanoZoomer (Hamamatsu Photonics, Hamamatsu, Shizuoka, Japan), and histopathological evaluation was outsourced to MCO Co., Ltd. (Tokyo, Japan).

2.4
Flow cytometry analysis
SLC35D3 expression on the cell surface was quantified by flow cytometry. The adherent cells were detached with Cell Dissociation Buffer (Thermo Fisher Scientific), resuspended in phosphate-buffered saline (PBS) containing 5% FBS, and transferred to 96-well plates. Samples were incubated for 30 min at 4 °C with the primary rabbit anti-SLC35D3 polyclonal antibody at 5 μg/mL (LSBio, USA); rabbit IgG isotype control (Cell Signaling Technology, Danvers, MA, USA, clone: DA1E) served as the negative control. After washing with 5% FBS/PBS, an APC-conjugated anti-human IgG secondary antibody (Jackson ImmunoResearch) containing Fixable Viability Stain 450 (BD Biosciences, Franklin Lakes, NJ, USA) was applied for 45 min at 4 °C. The Stained cells were washed, and fluorescence was acquired on a FACSCanto II flow cytometer (BD Biosciences). Data were analyzed in FlowJo v10 (BD Biosciences); dead cells positive for the viability dye were excluded by gating, and the distribution of APC fluorescence in live cells was displayed as histograms. Mean fluorescence intensity (MFI) values for the anti-SLC35D3 and the isotype-control samples were calculated, and the MFI ratio (anti-SLC35D3/isotype) was used as an index of SLC35D3 expression. Cell-line mRNA expression values (log2[TPM+1]) were obtained from the Cancer Cell Line Encyclopedia (CCLE) via DepMap Public 25Q2 (https://depmap.org/portal/).

3
Results
A stepwise, data-driven workflow was undertaken to identify membrane proteins that are silent in vital organs yet markedly over-expressed in colorectal cancer. Initially, the gene list was narrowed down to include only those predicted to localize to the cell surface on the basis of Swiss-Prot annotations. These candidates were then intersected with GTEx v4 RNA-seq data covering 30 histologically normal tissues. The final step involved retaining only those genes whose median expression did not exceed 1 RPKM in any organ. When the resulting list was compared with TCGA-COAD tumors, only two transcripts—LY6G6D and SLC35D3—displayed a ≥10-fold elevation over normal colon (Table 1). Given that LY6G6D is already under clinical evaluation, the poorly characterized nucleotide-sugar transporter SLC35D3 was prioritized for further study.
An analysis of TCGA Pan-Cancer expression data revealed that SLC35D3 mRNA levels are elevated in colon and rectal adenocarcinoma and pheochromocytoma/paraganglioma, while levels in normal samples remained at background levels (Fig. 1A). A complementary GTEx heat-map further highlighted the limited expression of SLC35D3 compared with well-established therapeutic antigens such as EGFR and EPCAM, which show widespread expression in several vital organs (Fig. 1B). These findings suggest that SLC35D3-directed interventions may have a wider therapeutic window.
To verify that the observed transcript enrichment is reflected at the protein level, we engaged Cell Engineering Co., Ltd. to produce a rat monoclonal antibody, 5H11-D10, for immunohistochemical analysis. The optimized protocol was then applied to commercial TMAs. In the TMA CO992a, 16 of the 30 primary colorectal adenocarcinomas (53%) were positive; the same proportion of paired lymph-node metastases were likewise positive, indicating that antigen expression is maintained during dissemination (Fig. 2A, Supplementary Table S1). An independent multi-tumor TMA MC2082b reproduced these findings, with 3/8 colorectal cores staining. Outside the gastrointestinal tract, SLC35D3 protein was detected in 32/80 small-cell lung cancers (40%) on TMA LC818 and in 3/8 pancreatic neuroendocrine tumors (38%), whereas no staining was observed in non-neuroendocrine pancreatic adenocarcinomas or in non–small-cell lung cancers in TMA MC2082b. Taken together, SLC35D3 protein was detected across multiple tumor types, with an overall positivity of 42% among malignant cores, suggesting its potential as a broadly applicable therapeutic target.
Off-tumor expression was evaluated using the FDA999l normal TMA, comprising 32 types of normal organ (Fig. 2B). Weak immunoreactivity was confined to specialized endocrine or neuroendocrine populations, including cells in the adrenal medulla, islets of Langerhans, and anterior pituitary; scattered enteroendocrine cells in the gastric and small intestinal mucosa; occasional non-neuronal cells within the small intestinal submucosal plexus; and rare neuroendocrine cells in the prostate. No SLC35D3 staining was detected in other tissues, including those most commonly associated with dose-limiting toxicity, such as heart, brain (including retina), liver, and lung. This restricted expression pattern is consistent with the GTEx mRNA profile (Fig. 1B) and the normal-tissue IHC survey (Fig. 2B), and suggests a potentially favorable on-target/off-tumor safety window in major organs, although dedicated cross-reactivity assessment and toxicology studies will be required before clinical translation.
The FFPE control cell blocks confirmed assay performance and antibody specificity: HCT 116-mock was negative, whereas HCT 116-hSLC35D3 and NCI–H1930 showed positive staining patterns (Fig. 2C). While both cell block controls and TMA specimens consistently exhibited granular and/or diffuse immunohistochemical staining patterns rather than distinct membranous staining, we sought to directly assess whether SLC35D3 is localized on the cell surface—a prerequisite for antibody binding and CAR engagement. Five human cancer cell lines with divergent CCLE transcript levels were analyzed by flow cytometry. HCT 116 cells, with a CCLE value of 0.0 log2 (TPM+1), showed minimal shift and largely overlapped with isotype control histograms. In contrast, LoVo cells, QGP-1 cells, NCI–H1930 cells, and SNU-16 cells, which expressed measurable transcript levels, displayed clear rightward shifts in fluorescence peaks (Fig. 3). These results confirm the presence of SLC35D3 on the cell surface of multiple tumor cell lines, thereby establishing SLC35D3 as a potentially tractable target for antibody- or CAR-based therapies.

4
Discussion
Leveraging complementary public RNA-seq resources and extensive protein-level validation, this study identifies SLC35D3 as a background-silent, tumor-associated cell surface antigen. This antigen meets three key criteria for next-generation therapeutics: selective expression in malignant tissue, negligible expression in vital organs, and confirmed plasma-membrane accessibility. This represents the first comprehensive evaluation of SLC35D3 across large patient cohorts and diverse tumor entities.
SLC35D3 is a multi-pass membrane protein with 10 predicted transmembrane domains [11]. While it has long been annotated as an “orphan” nucleotide-sugar transporter, recent work has suggested transporter activity consistent with UDP-glucose transport [12]. Notably, although SLC35D3 is generally regarded as being mainly localized to the endoplasmic reticulum (ER) and early endosomes [11], flow-cytometric profiling confirmed its cell-surface expression in several cell lines derived from colorectal, small-cell lung, pancreatic neuroendocrine, and gastric cancers. In addition, while clinical specimens often exhibited a granular and/or diffuse IHC pattern, rather than a sharp membranous staining, NCI–H1930 cells showed a similar pattern yet demonstrated cell surface expression by flow cytometry. These findings suggest that SLC35D3 is most likely accessible to therapeutic biologics. Several mutational and functional studies have shown that SLC35D3 is involved in dopamine signaling/neuronal function and adipocyte differentiation [11,13] and in the biogenesis of platelet dense granules [14,15]. Reports have also implicated members of the SLC35 family in tumor-associated glycosylation processes [[16], [17], [18], [19]]. Altered glycosylation patterns are cancer hallmarks contributing to progression, metastasis, and immune evasion [20,21]. More recently, the expression of SLC35D3 mRNA in lymph-node metastasis has been identified as a potential risk factor for a poor prognosis of colon cancer patients [22]. In addition, SLC35D3 has been suggested to promote colorectal tumor malignancy by relieving ERK suppression through inhibition of AMPK signaling [23]. Together, these reports suggest that SLC35D3 upregulation may reflect tumor-associated metabolic and/or glycosylation reprogramming and may be linked to malignant phenotypes. However, the mechanistic basis for its apparent plasma-membrane accessibility and its causal role in tumor progression remain to be established. This limited prior knowledge represents both challenge and opportunity; while mechanistic insights remain scarce, this target offers genuine novelty for first-in-class therapeutic development.
It is imperative to acknowledge the limitations of the study. Firstly, our stringent GTEx cutoff (median <1 RPKM in every tissue) was chosen to prioritize “background-silent” candidates, but different cutoffs (e.g., 0.5 or 2 RPKM) could change the candidate list. Secondly, bulk GTEx RNA-seq may under-represent rare cell populations with low-level SLC35D3 expression, and the TMA of normal tissue revealed reactivity in scattered neuroendocrine cells. Moreover, mouse studies have reported enriched Slc35d3 expression in striatal neurons involved in dopamine signaling [24]. Thus, on-target effects in neuroendocrine and/or neuronal cell types cannot be excluded based on bulk GTEx profiling alone, and tissue cross-reactivity testing and focused safety/toxicology evaluations will be required before clinical development of SLC35D3-targeted modalities. Integration with single-cell normal-tissue atlases (e.g., Human Cell Atlas) may further refine this assessment by revealing low-abundance SLC35D3-expressing cell types in critical organs. Thirdly, the use of TMA did not allow a comprehensive assessment of intratumoral heterogeneity, and the semiquantitative nature of IHC limits the precision with which expression levels can be evaluated. Accordingly, follow-up analyses using whole-tissue sections and quantitative approaches will be important to more precisely define expression heterogeneity and antigen density. Finally, we did not investigate whether SLC35D3 undergoes antibody-mediated internalization, which is particularly relevant for ADC development, nor did we evaluate functional activity as a TCEs or CAR-T target.
Notwithstanding the aforementioned limitations, the favorable expression landscape positions SLC35D3 as a compelling target for therapeutic exploitation. The positivity rates were substantial in both small-cell lung cancers (40%) and colorectal cancers (53%), underscoring their potential relevance in aggressive malignancies with pressing therapeutic needs [9,25]. From a translational pharmacology perspective, key next steps include (i) quantitative assessment of antigen density and heterogeneity in whole-tissue sections, (ii) evaluation of antibody internalization for ADC formats and antigen-dependent cytotoxicity for TCE and CAR-T formats, (iii) modality selection considering potential endocrine/neuroendocrine on-target liabilities, and (iv) in vivo safety studies.
In conclusion, through unbiased discovery followed by rigorous protein-level validation, we have identified SLC35D3 as a tumor-restricted, cell-surface antigen with broad representation across aggressive malignancies and minimal expression in essential organs. These characteristics warrant preclinical development of SLC35D3-targeted therapies such as ADCs, bispecific T-cell engagers and CAR-T cells, potentially expanding the arsenal of safe and effective therapies for colorectal and neuroendocrine cancers.

Declaration of generative AI and AI-assisted technologies in the manuscript preparation process
During the preparation of this work, the authors used ChatGPT (OpenAI, San Francisco, CA, USA) and DeepL (DeepL SE, Cologne, North Rhine-Westphalia, Germany) to improve the readability and language quality of the manuscript. They were used for English language editing, grammar checking, and literature search assistance. After using these tools, the authors reviewed and edited the content as needed and take full responsibility for the content of the published article.

Funding
This study was funded by 10.13039/501100002973Daiichi Sankyo Co., Ltd.

CRediT authorship contribution statement
Shunsuke Someya: Data curation, Investigation, Methodology, Validation, Visualization, Writing – original draft, Writing – review & editing. Tatsuya Inoue: Data curation, Formal analysis, Software, Visualization. Reimi Kawaida: Methodology, Project administration. Toshiaki Ohtsuka: Conceptualization, Supervision. Keisuke Fukuchi: Methodology, Project administration, Supervision, Writing – review & editing.

Declaration of competing interest
The authors declare the following financial interests/personal relationships which may be considered as potential competing interests: All authors are employees of Daiichi Sankyo Co., Ltd.