Deep spatial proteomics reveals a suppressive immune niche linked to immune evasion in renal cell carcinoma

Abstract

Aim: Clear cell renal cell carcinoma (ccRCC) presents an “immune paradox” where high CD8⁺ T cell infiltration correlates with poor survival and limited response to immune checkpoint blockade (ICB). Simple cell density metrics fail to capture the functional state of the immune microenvironment, suggesting an uncharacterized spatial mechanism underlying immune evasion and poor clinical outcomes. We aimed to develop an AI-driven spatial proteomics framework to decode this immunosuppressive phenotype.

Methods: We developed PhenoSSP, a hierarchical deep learning framework based on a Vision Transformer backbone, designed for single-cell phenotyping from 7-channel multiplex immunofluorescence (mIF) images. It was applied to 1,633 tissue microarray (TMA) cores from 834 ccRCC patients. We defined a density-normalized Spatial Interaction Score was defined to quantify FOXP3⁺ regulatory T cell (Treg) enrichment within a 30 μm radius of CD8⁺ T cells. Survival analyses were performed using Kaplan–Meier curves with log-rank tests and multivariable Cox regression models.

Results: PhenoSSP achieved a balanced accuracy of 71.0% and an F1-Macro of 70.7%, outperforming conventional methods. CD8⁺ T-cell density alone was not significantly associated with overall survival (OS, P = 0.057), whereas the Spatial Interaction Score was significantly associated with poor OS (log-rank test, P = 0.012) and was significantly higher in non-survivors (P = 0.032).

Conclusion: This study reveals a spatial basis for immune evasion and poor prognosis in ccRCC: the pre-existing Treg enrichment near CD8⁺ T cells, rather than the abundance of effector cells, was associated with impaired antitumor immunity. The Spatial Interaction Score serves as a candidate prognostic biomarker and provides a rationale for Treg-targeting combination strategies in patients harboring spatially defined suppressive niches.

Keywords

Kidney cancer / spatial biomarker / spatial proteomics / deep learning / tumor microenvironment / regulatory T cell / immune evasion

References

[1] Topalian SL,Pardoll DM.Immune checkpoint blockade: a common denominator approach to cancer therapy.Cancer Cell2015;27:450-61 PMCID:PMC4400238

[2] Fridman WH,Sautès-Fridman C.The immune contexture in human tumours: impact on clinical outcome.Nat Rev Cancer2012;12:298-306

[3] Nakano O,Naito Y.Proliferative activity of intratumoral CD8⁺ T-lymphocytes as a prognostic factor in human renal cell carcinoma: clinicopathologic demonstration of antitumor immunity.Cancer Res2001;61:5132-6

[4] Giraldo NA,Pagès F.Orchestration and prognostic significance of immune checkpoints in the microenvironment of primary and metastatic renal cell cancer.Clin Cancer Res2015;21:3031-40

[5] Braun DA,Bakouny Z.Interplay of somatic alterations and immune infiltration modulates response to PD-1 blockade in advanced clear cell renal cell carcinoma.Nat Med2020;26:909-18 PMCID:PMC7499153

[6] Motzer RJ, Escudier B, McDermott DF, et al.; CheckMate 025 Investigators. Nivolumab versus everolimus in advanced renal-cell carcinoma.N Engl J Med2015;373:1803-13 PMCID:PMC5719487

[7] Rini BI,Stus V.Pembrolizumab plus axitinib versus sunitinib for advanced clear cell renal cell carcinoma: 5-year survival and biomarker analyses of the phase 3 KEYNOTE-426 trial.Nat Med2025;31:3475-84 PMCID:PMC12532709

[8] Galon J,Sanchez-Cabo F.Type, density, and location of immune cells within human colorectal tumors predict clinical outcome.Science2006;313:1960-4

[9] Song X,Geng W.Spatial and single-cell transcriptomics reveal cellular heterogeneity and a novel cancer-promoting Treg cell subset in human clear-cell renal cell carcinoma.J Immunother Cancer2025;13:e010183 PMCID:PMC11748785

[10] Qureshi OS,Nakamura K.Trans-endocytosis of CD80 and CD86: a molecular basis for the cell-extrinsic function of CTLA-4.Science2011;332:600-3 PMCID:PMC3198051

[11] Wang G,Guo Z.IL-2-deprivation and TGF-beta are two non-redundant suppressor mechanisms of CD4⁺CD25⁺ regulatory T cell which jointly restrain CD4⁺CD25^- cell activation.Immunol Lett2010;132:61-8 PMCID:PMC2922908

[12] Greenwald NF,Moen E.Whole-cell segmentation of tissue images with human-level performance using large-scale data annotation and deep learning.Nat Biotechnol2022;40:555-65 PMCID:PMC9010346

[13] Chen RJ,Li Y. Scaling vision transformers to gigapixel images via hierarchical self-supervised learning. arXiv 2022; arXiv:2206.02647. Available from: https://doi.org/10.48550/arXiv.2206.02647. [accessed 18 May 2026]

[14] Dosovitskiy A,Kolesnikov A. An image is worth 16×16 words: transformers for image recognition at scale. arXiv 2020; arXiv:2010.11929. Available from: https://doi.org/10.48550/arXiv.2010.11929. [accessed 18 May 2026]

[15] Oquab M,Moutakanni T. DINOv2: learning robust visual features without supervision. arXiv 2023; arXiv:2304.07193. Available from: https://doi.org/10.48550/arXiv.2304.07193. [accessed 18 May 2026]

[16] He K,Xie S,Dollár P. Masked autoencoders are scalable vision learners. arXiv 2021; arXiv:2111.06377. Available from: https://doi.org/10.48550/arXiv.2111.06377. [accessed 18 May 2026]

[17] Loshchilov I. Decoupled weight decay regularization. arXiv 2017; arXiv:1711.05101. Available from: https://doi.org/10.48550/arXiv.1711.05101. [accessed 18 May 2026]

[18] He K,Ren S. Deep residual learning for image recognition. arXiv 2015; arXiv:1512.03385. Available from: https://doi.org/10.48550/arXiv.1512.03385. [accessed 18 May 2026]

[19] Cortes C.Support-vector networks.Mach Learn1995;20:273-97

[20] Breiman L.Random forests.Mach Learn2001;45:5-32

[21] Hosmer DW Jr., Lemeshow S, Sturdivant RX. Applied logistic regression. John Wiley & Sons; 2013.

[22] Kokhlikyan N,Martin M. Captum: a unified and generic model interpretability library for PyTorch. arXiv 2020; arXiv:2009.07896. Available from: https://doi.org/10.48550/arXiv.2009.07896. [accessed 18 May 2026]

[23] Bentley JL.Multidimensional binary search trees used for associative searching.Commun ACM1975;18:509-17

[24] Kaplan EL.Nonparametric estimation from incomplete observations.J Am Stat Assoc1958;53:457

[25] Newman AM,Green MR.Robust enumeration of cell subsets from tissue expression profiles.Nat Methods2015;12:453-7 PMCID:PMC4739640

[26] Cui H,Lu Y.TIMER3: an enhanced resource for tumor immune analysis.Nucleic Acids Res2025;53:W534-41 PMCID:PMC12230722

[27] Zheng L,Si W.Pan-cancer single-cell landscape of tumor-infiltrating T cells.Science2021;374:abe6474

[28] Schürch CM,Barlow GL.Coordinated cellular neighborhoods orchestrate antitumoral immunity at the colorectal cancer invasive front.Cell2020;182:1341-59.e19 PMCID:PMC7479520

[29] Miao D,Gao W.Genomic correlates of response to immune checkpoint therapies in clear cell renal cell carcinoma.Science2018;359:801-6 PMCID:PMC6035749

[30] Turajlic S,Xu H.Insertion-and-deletion-derived tumour-specific neoantigens and the immunogenic phenotype: a pan-cancer analysis.Lancet Oncol2017;18:1009-21

[31] Javed A.Role of natural killer cells in uveal melanoma.Cancers2020;12:3694 PMCID:PMC7764114

[32] Simoni Y,Fehlings M.Bystander CD8⁺ T cells are abundant and phenotypically distinct in human tumour infiltrates.Nature2018;557:575-9

[33] Dai S,Liu Z.Intratumoral CXCL13⁺CD8⁺T cell infiltration determines poor clinical outcomes and immunoevasive contexture in patients with clear cell renal cell carcinoma.J Immunother Cancer2021;9:e001823 PMCID:PMC7887366

[34] Qi Y,Lin Z.Tumor-infiltrating CD39⁺CD8⁺ T cells determine poor prognosis and immune evasion in clear cell renal cell carcinoma patients.Cancer Immunol Immunother2020;69:1565-76 PMCID:PMC11027701

[35] Granier C,Combe P.Tim-3 expression on tumor-infiltrating PD-1⁺CD8⁺ T cells correlates with poor clinical outcome in renal cell carcinoma.Cancer Res2017;77:1075-82

[36] Yakirevich E.Tumor mutational burden and immune signatures interplay in renal cell carcinoma.Ann Transl Med2020;8:269 PMCID:PMC7186632

[37] Murakami T,Takamatsu K.Multiplexed single-cell pathology reveals the association of CD8 T-cell heterogeneity with prognostic outcomes in renal cell carcinoma.Cancer Immunol Immunother2021;70:3001-13 PMCID:PMC10992910

[38] Campbell JR,Mesko PB.Fc-optimized anti-CCR8 antibody depletes regulatory T cells in human tumor models.Cancer Res2021;81:2983-94

[39] Tannir NM,McDermott DF.Nivolumab plus ipilimumab versus sunitinib for first-line treatment of advanced renal cell carcinoma: extended 8-year follow-up results of efficacy and safety from the phase III CheckMate 214 trial.Ann Oncol2024;35:1026-38 PMCID:PMC11907766

[40] Sharma P,Anandhan S.The next decade of immune checkpoint therapy.Cancer Discov2021;11:838-57

[41] Diab A,Bentebibel SE.Bempegaldesleukin (NKTR-214) plus nivolumab in patients with advanced solid tumors: phase I dose-escalation study of safety, efficacy, and immune activation (PIVOT-02).Cancer Discov2020;10:1158-73

[42] Dai J,Liang X.PBRM1 mutation as a predictive biomarker for immunotherapy in multiple cancers.Front Genet2022;13:1066347 PMCID:PMC9868445

[43] Friedhoff J,Jurcic C.BAP1 and PTEN mutations shape the immunological landscape of clear cell renal cell carcinoma and reveal the intertumoral heterogeneity of T cell suppression: a proof-of-concept study.Cancer Immunol Immunother2023;72:1603-18 PMCID:PMC10198938

[44] Batlle E.Transforming growth factor-β signaling in immunity and cancer.Immunity2019;50:924-40 PMCID:PMC7507121

[45] Wang J,van der Voort EI,Toes RE.Transient expression of FOXP3 in human activated nonregulatory CD4⁺ T cells.Eur J Immunol2007;37:129-38

[46] Patsoukis N,Strauss L.Revisiting the PD-1 pathway.Sci Adv2020;6:eabd2712 PMCID:PMC7500922

[47] Zhang Y,Kong Y,Nakai K.Toward graph-based decoding of tumor evolution: spatial inference of copy number variations.Diagnostics2025;15:3169 PMCID:PMC12731904

[48] Kong Y.Biomechanics-driven 3D architecture inference from histology using CellSqueeze3D.Adv Sci2026;13:e18706 PMCID:PMC12915120