Skip to main content

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

  • Article
  • Published:

Spatial immune scoring system predicts hepatocellular carcinoma recurrence

Nature volume 640, pages 1031–1041 (2025)Cite this article

Subjects

An Author Correction to this article was published on 10 July 2025

This article has been updated

Abstract

Given the high recurrence rates of hepatocellular carcinoma (HCC) post-resection1,2,3, improved early identification of patients at high risk for post-resection recurrence would help to improve patient outcomes and prioritize healthcare resources4,5,6. Here we observed a spatial and HCC recurrence-associated distribution of natural killer (NK) cells in the invasive front and tumour centre from 61 patients. Using extreme gradient boosting and inverse-variance weighting, we developed the tumour immune microenvironment spatial (TIMES) score based on the spatial expression patterns of five biomarkers (SPON2, ZFP36L2, ZFP36, VIM and HLA-DRB1) to predict HCC recurrence risk. The TIMES score (hazard ratio = 88.2, P < 0.001) outperformed current standard tools for patient risk stratification including the TNM and BCLC systems. We validated the model in 231 patients from five multicentred cohorts, achieving a real-world accuracy of 82.2% and specificity of 85.7%. The predictive power of these biomarkers emerged through the integration of their spatial distributions, rather than individual marker expression levels alone. In vivo models, including NK cell-specific Spon2-knockout mice, revealed that SPON2 enhances IFNγ secretion and NK cell infiltration at the invasive front. Our study introduces TIMES, a publicly accessible tool for predicting HCC recurrence risk, offering insights into its potential to inform treatment decisions for early-stage HCC.

This is a preview of subscription content, access via your institution

Access options

Buy this article

  • Purchase on SpringerLink
  • Instant access to the full article PDF.

USD 39.95

Prices may be subject to local taxes which are calculated during checkout

Fig. 1: Spatial multi-omics analysis reveals associations between NK cell distribution and HCC recurrence.
Fig. 2: Identification of distinct spatial patterns of gene expressions for HCC recurrence prediction.
Fig. 3: Development of a TIMES scoring system for HCC recurrence risk stratification.
Fig. 4: SPON2+ NK cells are activated with migration potential to the HCC TC and in proximity to higher abundance of IFNγ+CD8+ T cells.
Fig. 5: Effect of Spon2 knockout on NK cell function in HCC mouse models.

Similar content being viewed by others

Data availability

The DSP sequencing, LC–MS/MS and 10x Visium data in this study have been deposited in the China National Center for Bioinformation (CNCB) database, under user accession projects OMIX005738, OMIX005736 and HRA006579. The human genome dataset GRCh38 used for 10x Visium data alignment was downloaded from https://www.10xgenomics.com/support/software/space-ranger/downloads. The accession number for the data of single-cell RNA-seq reported in this paper is CNP0000650 (https://db.cngb.org/search/project/CNP0000650). In addition, this study contains Extended Data Figs. 1–10, Supplementary Tables 1–38 and Supplementary Figs. 1–9. In Supplementary Tables 34–37, we reviewed existing clinical guideline on early-stage HCC treatment (Supplementary Table 34), previous comparative research on early-stage HCC treatment (Supplementary Table 35), previous studies on the prognostic value of NK cells in HCC (Supplementary Table 36) and SPON2-related research in fibroblasts (Supplementary Table 37). Source data are provided with this paper.

Code availability

A Code Ocean capsule (https://doi.org/10.24433/CO.7364332.v1) containing executable programming scripts and input and output data is accessible.

Change history

References

  1. Zhai, W. et al. The spatial organization of intra-tumour heterogeneity and evolutionary trajectories of metastases in hepatocellular carcinoma. Nat. Commun. 8, 4565 (2017).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

  2. Rumgay, H. et al. Global burden of primary liver cancer in 2020 and predictions to 2040. J. Hepatol. 77, 1598–1606 (2022).

    Article  PubMed  PubMed Central  Google Scholar 

  3. Wu, J. et al. A noninvasive approach to evaluate tumor immune microenvironment and predict outcomes in hepatocellular carcinoma. Phenomics 3, 549–564 (2023).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  4. Brown, Z. J. et al. Management of hepatocellular carcinoma: a review. JAMA Surgery 158, 410–420 (2023).

    Article  PubMed  Google Scholar 

  5. Singal, A. G., Kudo, M. & Bruix, J. Breakthroughs in hepatocellular carcinoma therapies. Clin. Gastroenterol. Hepatol. 21, 2135–2149 (2023).

    Article  PubMed  PubMed Central  Google Scholar 

  6. Elderkin, J. et al. Hepatocellular carcinoma: surveillance, diagnosis, evaluation and management. Cancers 15, 5118 (2023).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  7. Schürch, C. M. et al. Coordinated cellular neighborhoods orchestrate antitumoral immunity at the colorectal cancer invasive front. Cell 182, 1341–1359.e19 (2020).

    Article  PubMed  PubMed Central  Google Scholar 

  8. Casasent, A. K. et al. Multiclonal invasion in breast tumors identified by topographic single cell sequencing. Cell 172, 205–217.e12 (2018).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  9. Sun, C., Sun, H.-y, Xiao, W.-h, Zhang, C. & Tian, Z.-g Natural killer cell dysfunction in hepatocellular carcinoma and NK cell-based immunotherapy. Acta Pharmacol. Sin. 36, 1191–1199 (2015).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  10. Nersesian, S. et al. NK cell infiltration is associated with improved overall survival in solid cancers: a systematic review and meta-analysis. Transl. Oncol. 14, 100930 (2021).

    Article  CAS  PubMed  Google Scholar 

  11. Lee, H. A. et al. Natural killer cell activity is a risk factor for the recurrence risk after curative treatment of hepatocellular carcinoma. BMC Gastroenterol. 21, 258 (2021).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  12. Juengpanich, S. et al. The role of natural killer cells in hepatocellular carcinoma development and treatment: a narrative review. Transl. Oncol. 12, 1092–1107 (2019).

    Article  PubMed  PubMed Central  Google Scholar 

  13. Xue, J. S. et al. The prognostic value of natural killer cells and their receptors/ligands in hepatocellular carcinoma: a systematic review and meta-analysis. Front. Immunol. 13, 872353 (2022).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  14. Zhao, J.-J. et al. Interleukin-37 mediates the antitumor activity in hepatocellular carcinoma: role for CD57+ NK cells. Sci. Rep. 4, 5177 (2014).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  15. Chew, V. et al. Toll-like receptor 3 expressing tumor parenchyma and infiltrating natural killer cells in hepatocellular carcinoma patients. J. Natl Cancer Inst. 104, 1796–1807 (2012).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

  16. Wu, Y. et al. Monocyte/macrophage-elicited natural killer cell dysfunction in hepatocellular carcinoma is mediated by CD48/2B4 interactions. Hepatology 57, 1107–1116 (2013).

    Article  CAS  PubMed  Google Scholar 

  17. Sun, H. et al. Accumulation of tumor-infiltrating CD49a+ NK cells correlates with poor prognosis for human hepatocellular carcinoma. Cancer Immunol. Res. 7, 1535–1546 (2019).

    Article  CAS  PubMed  Google Scholar 

  18. Sun, H. et al. Human CD96 correlates to natural killer cell exhaustion and predicts the prognosis of human hepatocellular carcinoma. Hepatology 70, 168–183 (2019).

    Article  CAS  PubMed  Google Scholar 

  19. Wu, M., Mei, F., Liu, W. & Jiang, J. Comprehensive characterization of tumor infiltrating natural killer cells and clinical significance in hepatocellular carcinoma based on gene expression profiles. Biomed. Pharmacother. 121, 109637 (2020).

    Article  CAS  PubMed  Google Scholar 

  20. Pinyol, R. et al. Molecular predictors of prevention of recurrence in HCC with sorafenib as adjuvant treatment and prognostic factors in the phase 3 STORM trial. Gut 68, 1065–1075 (2019).

    Article  PubMed  Google Scholar 

  21. Li, J. H. & O’Sullivan, T. E. Back to the future: spatiotemporal determinants of NK cell antitumor function. Front. Immunol. 12, 816658 (2022).

    Article  PubMed  PubMed Central  Google Scholar 

  22. Fu, T. et al. Spatial architecture of the immune microenvironment orchestrates tumor immunity and therapeutic response. J. Hematol. Oncol. 14, 98 (2021).

    Article  PubMed  PubMed Central  Google Scholar 

  23. Wu, L. et al. Spatially-resolved transcriptomics analyses of invasive fronts in solid tumors. Preprint at bioRxiv https://doi.org/10.1101/2021.10.21.465135 (2021).

  24. Sorin, M. et al. Single-cell spatial landscapes of the lung tumour immune microenvironment. Nature 614, 548–554 (2023).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

  25. Elhanani, O., Ben-Uri, R. & Keren, L. Spatial profiling technologies illuminate the tumor microenvironment. Cancer Cell 41, 404–420 (2023).

    Article  CAS  PubMed  Google Scholar 

  26. Zhao, N. et al. Spatial maps of hepatocellular carcinoma transcriptomes highlight an unexplored landscape of heterogeneity and a novel gene signature for survival. Cancer Cell Int. 22, 57 (2022).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  27. Wu, R. et al. Comprehensive analysis of spatial architecture in primary liver cancer. Sci. Adv. 7, eabg3750 (2021).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

  28. Sun, Y. et al. Single-cell landscape of the ecosystem in early-relapse hepatocellular carcinoma. Cell 184, 404–421.e16 (2021).

    Article  CAS  PubMed  Google Scholar 

  29. Chang, J. et al. Genomic alterations driving precancerous to cancerous lesions in esophageal cancer development. Cancer Cell 41, 2038–2050.e5 (2023).

    Article  CAS  PubMed  Google Scholar 

  30. Li, L. et al. Comprehensive proteogenomic characterization of early duodenal cancer reveals the carcinogenesis tracks of different subtypes. Nat. Commun. 14, 1751 (2023).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

  31. Mund, A. et al. Deep Visual Proteomics defines single-cell identity and heterogeneity. Nat. Biotechnol. 40, 1231–1240 (2022).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  32. Xiang, H. et al. Region-resolved multi-omics of the mouse eye. Cell Rep. 42, 112121 (2023).

    Article  CAS  PubMed  Google Scholar 

  33. Qu, Y. et al. A proteogenomic analysis of clear cell renal cell carcinoma in a Chinese population. Nat. Commun. 13, 2052 (2022).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

  34. Cheung, C. C. L. et al. Residual SARS-CoV-2 viral antigens detected in GI and hepatic tissues from five recovered patients with COVID-19. Gut 71, 226–229 (2022).

    Article  CAS  PubMed  Google Scholar 

  35. Kaya, N. A. et al. Multimodal molecular landscape of response to Y90-resin microsphere radioembolization followed by nivolumab for advanced hepatocellular carcinoma. J. Immunother. Cancer 11, e007106 (2023).

    Article  PubMed  PubMed Central  Google Scholar 

  36. Xu, X. F. et al. Risk factors, patterns, and outcomes of late recurrence after liver resection for hepatocellular carcinoma: a multicenter study from China. JAMA Surg. 154, 209–217 (2019).

    Article  PubMed  Google Scholar 

  37. Lee, S. K. et al. Immunological markers, prognostic factors and challenges following curative treatments for hepatocellular carcinoma. Int. J. Mol. Sci. 22, 10271 (2021).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  38. Jin, S. et al. Inference and analysis of cell–cell communication using CellChat. Nat. Commun. 12, 1088 (2021).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

  39. Medaglia, C. et al. Spatial reconstruction of immune niches by combining photoactivatable reporters and scRNA-seq. Science 358, 1622–1626 (2017).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

  40. Predina, J. D. et al. Characterization of surgical models of postoperative tumor recurrence for preclinical adjuvant therapy assessment. Am. J. Transl. Res. 4, 206–218 (2012).

    PubMed  PubMed Central  Google Scholar 

  41. Feng, Y. X. et al. Sorafenib suppresses postsurgical recurrence and metastasis of hepatocellular carcinoma in an orthotopic mouse model. Hepatology 53, 483–492 (2011).

    Article  CAS  PubMed  Google Scholar 

  42. Lim, K. C. et al. Microvascular invasion is a better predictor of tumor recurrence and overall survival following surgical resection for hepatocellular carcinoma compared to the Milan criteria. Ann. Surg. 254, 108–113 (2011).

    Article  PubMed  Google Scholar 

  43. Roayaie, S. et al. A system of classifying microvascular invasion to predict outcome after resection in patients with hepatocellular carcinoma. Gastroenterology 137, 850–855 (2009).

    Article  PubMed  Google Scholar 

  44. Hack, S. P. et al. IMbrave 050: a phase III trial of atezolizumab plus bevacizumab in high-risk hepatocellular carcinoma after curative resection or ablation. Future Oncol. 16, 975–989 (2020).

    Article  CAS  PubMed  Google Scholar 

  45. Bajorin, D. F. et al. Adjuvant nivolumab versus placebo in muscle-invasive urothelial carcinoma. N. Engl. J. Med. 384, 2102–2114 (2021).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  46. Zhou, Y. et al. Metascape provides a biologist-oriented resource for the analysis of systems-level datasets. Nat. Commun. 10, 1523 (2019).

    Article  ADS  PubMed  PubMed Central  Google Scholar 

  47. Zhang, X. et al. CellMarker: a manually curated resource of cell markers in human and mouse. Nucleic Acids Res. 47, D721–D728 (2019).

    Article  ADS  CAS  PubMed  Google Scholar 

  48. Hu, C. et al. CellMarker 2.0: an updated database of manually curated cell markers in human/mouse and web tools based on scRNA-seq data. Nucleic Acids Res. 51, D870–D876 (2023).

    Article  CAS  PubMed  Google Scholar 

  49. Mootha, V. K. et al. PGC-1α-responsive genes involved in oxidative phosphorylation are coordinately downregulated in human diabetes. Nat. Genet. 34, 267–273 (2003).

    Article  CAS  PubMed  Google Scholar 

  50. Subramanian, A. et al. Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc. Natl Acad. Sci. USA 102, 15545–15550 (2005).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

  51. Lim, J. C. T. et al. An automated staining protocol for seven-colour immunofluorescence of human tissue sections for diagnostic and prognostic use. Pathology 50, 333–341 (2018).

    Article  PubMed  Google Scholar 

  52. Ng, H. H. M. et al. Immunohistochemical scoring of CD38 in the tumor microenvironment predicts responsiveness to anti-PD-1/PD-L1 immunotherapy in hepatocellular carcinoma. J. Immunother. Cancer 8, e000987 (2020).

    Article  PubMed  PubMed Central  Google Scholar 

  53. Yeong, J. et al. Prognostic value of CD8+PD-1+ immune infiltrates and PDCD1 gene expression in triple negative breast cancer. J. Immunother. Cancer 7, 34 (2019).

    Article  PubMed  PubMed Central  Google Scholar 

  54. Yeong, J. et al. High densities of tumor-associated plasma cells predict improved prognosis in triple negative breast cancer. Front. Immunol. 9, 1209 (2018).

    Article  PubMed  PubMed Central  Google Scholar 

  55. Bankhead, P. et al. QuPath: open source software for digital pathology image analysis. Sci. Rep. 7, 16878 (2017).

    Article  ADS  PubMed  PubMed Central  Google Scholar 

  56. Van Acker, H. H., Capsomidis, A., Smits, E. L. & Van Tendeloo, V. F. CD56 in the immune system: more than a marker for cytotoxicity? Front. Immunol. 8, 892 (2017).

    Article  PubMed  PubMed Central  Google Scholar 

  57. Sung, J. J. et al. Immunohistochemical expression and clinical significance of suggested stem cell markers in hepatocellular carcinoma. J. Pathol. Transl. Med. 50, 52–57 (2016).

    Article  PubMed  Google Scholar 

  58. Tsuchiya, A. et al. Hepatocellular carcinoma with progenitor cell features distinguishable by the hepatic stem/progenitor cell marker NCAM. Cancer Lett. 309, 95–103 (2011).

    Article  CAS  PubMed  Google Scholar 

  59. Phillips, D. et al. Highly multiplexed phenotyping of immunoregulatory proteins in the tumor microenvironment by CODEX tissue imaging. Front. Immunol. 12, 687673 (2021).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  60. Black, S. et al. CODEX multiplexed tissue imaging with DNA-conjugated antibodies. Nat. Protoc. 16, 3802–3835 (2021).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  61. Wu, H. et al. Postmortem high-dimensional immune profiling of severe COVID-19 patients reveals distinct patterns of immunosuppression and immunoactivation. Nat. Commun. 13, 269 (2022).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

  62. Li, R. et al. A body map of somatic mutagenesis in morphologically normal human tissues. Nature 597, 398–403 (2021).

    Article  ADS  CAS  PubMed  Google Scholar 

  63. Emmert-Buck, M. R. et al. Laser capture microdissection. Science 274, 998–1001 (1996).

    Article  ADS  CAS  PubMed  Google Scholar 

  64. Deng, M. et al. Proteogenomic characterization of cholangiocarcinoma. Hepatology 77, 411–429 (2023).

    Article  PubMed  Google Scholar 

  65. Schwanhäusser, B. et al. Global quantification of mammalian gene expression control. Nature 473, 337–342 (2011).

    Article  ADS  PubMed  Google Scholar 

  66. Li, T. et al. TIMER2.0 for analysis of tumor-infiltrating immune cells. Nucleic Acids Res. 48, W509–W514 (2020).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  67. Andersen, P. K. & Gill, R. D. Cox’s regression model for counting processes: a large sample study. Ann. Stat. 10, 1100–1120 (1982).

    Article  MathSciNet  Google Scholar 

  68. Therneau, T. M. & Grambsch, P. M. Modeling survival data: extending the Cox model. Stat. Biol. Health https://doi.org/10.1007/978-1-4757-3294-8 (2000).

    Article  Google Scholar 

  69. Li, J. et al. Biodegradable electrospun nanofibrous platform integrating antiplatelet therapy–chemotherapy for preventing postoperative tumor recurrence and metastasis. Theranostics 12, 3503–3517 (2022).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  70. Tang, M. et al. Three-dimensional bioprinted glioblastoma microenvironments model cellular dependencies and immune interactions. Cell Res. 30, 833–853 (2020).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  71. Wu, Y. et al. Blockade of T-cell receptor with Ig and ITIM domains elicits potent antitumor immunity in naturally occurring HBV-related HCC in mice. Hepatology 77, 965–981 (2023).

    Article  PubMed  Google Scholar 

  72. Xu, X. et al. Group-2 innate lymphoid cells promote HCC progression through CXCL2-neutrophil-induced immunosuppression. Hepatology 74, 2526–2543 (2021).

    Article  CAS  PubMed  Google Scholar 

Download references

Acknowledgements

This work was supported by the National Key R&D Program of China (2021YFC2300604 and 2022YFA1303200), the National Natural Science Foundation of China (82394452, 82022056, 92169118, 82222048, 32330062, 82373232 and 32470720), the CAS Project for Young Scientists in Basic Research (grant no. YSBR-068), the Natural Science Foundation of Anhui Provincial (2008085J35 and 2408085JX013), the A*STAR Biomedical Engineering Programme (C211318003), the Singapore National Medical Research Council (MOH-000323-00 and OFYIRG19may-0007), the Industry Alignment Fund-Industry Collaboration Fund (IAF-ICP I2201E0014), the Central Public-interest Scientific Institution Basal Research Fund (no. 11024316000202300001) and the Qingyun Project Research Fund (PB- KC012101). We thank L. Sun, H. Sun and Z. Tian for their invaluable assistance, support, guidance and contributions to this study; T. Liang for generously providing the plasmid used in this research; and Y. Wang, M. Xiao and X. Wang for their assistance with the mass cytometry assays.

Author information

Author notes

  1. These authors contributed equally: Gengjie Jia, Peiqi He, Tianli Dai, Denise Goh

Authors and Affiliations

  1. Department of Hepatobiliary Surgery, State Key Laboratory of Immune Response and Immunotherapy, Anhui Provincial Clinical Research Center for Hepatobiliary Diseases, Anhui Province Key Laboratory of Hepatopancreatobiliary Surgery, The First Affiliated Hospital of University of Science and Technology of China (USTC), University of Science and Technology of China, Hefei, China

    Gengjie Jia, Peiqi He, Tianli Dai, Jiabei Wang, Mengyuan Sun, Fuling Li, Yao Liu, Qingping Tao, Wei Wang, Ahitsham Umar, Björn Nashan, Lianxin Liu & Cheng Sun

  2. Chinese Academy of Sciences Key Laboratory of Innate Immunity and Chronic Disease, Key Laboratory of Immune Response and Immunotherapy, Institute of Immunology, School of Basic Medical Sciences, Division of Life Sciences and Medicine, University of Science and Technology of China, Hefei, China

    Gengjie Jia, Peiqi He, Tianli Dai, Mengyuan Sun, Fuling Li, Qingping Tao, Wei Wang, Ahitsham Umar & Cheng Sun

  3. Clinical Research Hospital of Chinese Academy of Sciences, University of Science and Technology of China, Hefei, China

    Gengjie Jia, Peiqi He, Tianli Dai, Jiabei Wang, Mengyuan Sun, Fuling Li, Yao Liu, Qingping Tao & Wei Wang

  4. Genome Analysis Laboratory of the Ministry of Agriculture and Rural Affairs, Agricultural Genomics Institute at Shenzhen, Chinese Academy of Agricultural Sciences, Shenzhen, China

    Gengjie Jia & Shuxia Hao

  5. Institute of Molecular and Cell Biology (IMCB), Agency for Science Technology and Research (A*STAR), Singapore, Singapore

    Denise Goh, Felicia Wee, Jeffrey Chun Tatt Lim & Joe Yeong

  6. Duke-NUS Medical School, Singapore, Singapore

    Tony Kiat Hon Lim & Nye-Thane Ngo

  7. Hunan Cancer Hospital, The Affiliated Cancer Hospital of Central South University, Changsha, China

    Yongchang Zhang

  8. State Key Laboratory of Genetic Engineering and Collaborative Innovation Center for Genetics and Development, School of Life Sciences, Institutes of Biomedical Sciences, Human Phenome Institute, Fudan University, Shanghai, China

    Chen Ding

  9. Department of Anatomical Pathology, Singapore General Hospital, Singapore, Singapore

    Joe Yeong

  10. Singapore Immunology Network (SIgN), Agency for Science Technology and Research (A*STAR), Singapore, Singapore

    Joe Yeong

  11. Cancer Science Institute, National University of Singapore, Singapore, Singapore

    Joe Yeong

Authors
  1. Gengjie Jia
    View author publications

    Search author on:PubMed Google Scholar

  2. Peiqi He
    View author publications

    Search author on:PubMed Google Scholar

  3. Tianli Dai
    View author publications

    Search author on:PubMed Google Scholar

  4. Denise Goh
    View author publications

    Search author on:PubMed Google Scholar

  5. Jiabei Wang
    View author publications

    Search author on:PubMed Google Scholar

  6. Mengyuan Sun
    View author publications

    Search author on:PubMed Google Scholar

  7. Felicia Wee
    View author publications

    Search author on:PubMed Google Scholar

  8. Fuling Li
    View author publications

    Search author on:PubMed Google Scholar

  9. Jeffrey Chun Tatt Lim
    View author publications

    Search author on:PubMed Google Scholar

  10. Shuxia Hao
    View author publications

    Search author on:PubMed Google Scholar

  11. Yao Liu
    View author publications

    Search author on:PubMed Google Scholar

  12. Tony Kiat Hon Lim
    View author publications

    Search author on:PubMed Google Scholar

  13. Nye-Thane Ngo
    View author publications

    Search author on:PubMed Google Scholar

  14. Qingping Tao
    View author publications

    Search author on:PubMed Google Scholar

  15. Wei Wang
    View author publications

    Search author on:PubMed Google Scholar

  16. Ahitsham Umar
    View author publications

    Search author on:PubMed Google Scholar

  17. Björn Nashan
    View author publications

    Search author on:PubMed Google Scholar

  18. Yongchang Zhang
    View author publications

    Search author on:PubMed Google Scholar

  19. Chen Ding
    View author publications

    Search author on:PubMed Google Scholar

  20. Joe Yeong
    View author publications

    Search author on:PubMed Google Scholar

  21. Lianxin Liu
    View author publications

    Search author on:PubMed Google Scholar

  22. Cheng Sun
    View author publications

    Search author on:PubMed Google Scholar

Contributions

C.S., L.L. and J.Y. conceived and directed the study. G.J. designed the overall data analysis strategies and developed the artificial intelligence approaches. L.L., J.W., J.C.T.L., Y.L., W.W., Y.Z., B.N. and Y.L. collected clinical samples for paraffin embedding. P.H. and F.L. performed the 10x Genomics Visium profiling and analysis. C.D. and F.L. conducted the proteomic analysis. P.H., F.L. and S.H. performed the Nanostring DSP profiling and analysis. G.J. and S.H. constructed the machine learning models using the transcriptomics data and developed the TIMES scoring system using the proteomics data. F.W., P.H., F.L., G.J. and S.H. performed the mIHC. G.J., T.L.K.H., N.-T.N., J.W., F.W., F.L., P.H. and S.H. validated the TIMES scoring system. P.H. analysed the single-cell RNA-seq data. P.H. and M.S. conducted the PhenoCycler mapping and analysis. M.S. and Q.T. established the 3D-bioprinting model. P.H. and F.W. performed the mIHC studies and survival analysis. T.D. conducted the mouse model studies. P.H. and A.U. built the TIMES website. C.S. and L.L. supervised the research. C.S., G.J. and D.G. drafted and revised the manuscript. All authors approved the final version of the manuscript.

Corresponding authors

Correspondence to Joe Yeong, Lianxin Liu or Cheng Sun.

Ethics declarations

Competing interests

The authors declare no competing interests.

Peer review

Peer review information

Nature thanks George Miller, Christian Schürch and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Peer reviewer reports are available.

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Extended data figures and tables

Extended Data Fig. 1 Integrative Analysis of Spatial Transcriptomics in HCC Tissue Compartments and NK Cell Protein Expression Related to HCC Recurrence.

a, Hematoxylin and Eosin (H&E), Glypican 3 (GPC3), CD34, glutamine synthetase (GS), and Reticulin staining to delineate HCC tissue compartments. Tissue compartments including invasive front (IF) and tumor center (TC) were outlined using black dashed lines. These high-resolution images of other cases have been uploaded to the Open Science Framework public platform (https://osf.io/db3uf/?view_only=d8431249989045bab050581751199298). Experienced hepatopathologists Y.Z. and J.Y. thoroughly re-reviewed our selected sections to confirm delineations of adjacent stroma (AS), IF, and TC across all cases utilizing established HCC markers. b, Visium transcriptome clustering patterns differ in various pathological regions within tissues. K-Means clustering results of 10x Visium samples independently (top panel) and their Uniform Manifold Approximation and Projection (UMAP, bottom panel). c, Integrated UMAP embedding of all spatial spots from donors. Integrating the UMAP clustering results of 18 Visium samples from 17 patients. Coloration distinguishes spatial compartments. d, Comparison of CD3−CD56+ NK cell proportions at IF between non-recurrent (non-REC; n = 31 patients) and current (REC; n = 30 patients) HCC patients. Unpaired two-tailed Student’s t-test, P = 0.035. This enabled delineation of CD3−CD56+ ordinary NK cells (shown here), CD3−CD16+CD56+ cytotoxic NK cells (shown in d), and CD3−CD57+ mature NK cells (shown in Fig. 1f) subsets at the single cell level. For non-REC, max (3.13608), min (0.05872), median (0.97038), mean (1.32004), Q1 (0.61261), Q3 (2.05738); For REC, max (2.80434), min (0.06898), median (0.82503), mean (0.89512), Q1 (0.48859), Q3 (1.29746). IQR = Q3-Q1, Upper Whiskers = Q3 + 1.5*IQR, Lower Whisker = Q1-1.5*IQR; the whiskers extending to the most extreme data points. e, Comparison of CD3−CD16+CD56+ NK cell proportions at IF between non-REC (n = 31 patients) and REC (n = 30 patients) patients. Unpaired two-tailed Student’s t-test, P = 0.003. For non-REC, max (0.603270), min (0.005111), median (0.150094), mean (0.183848), Q1 (0.053496), Q3 (0.253224); For REC, max (0.250404), min (0.003508), median (0.058713), mean (0.081717), Q1 (0.025158), Q3 (0.136488). IQR = Q3-Q1, Upper Whiskers = Q3 + 1.5*IQR, Lower Whisker = Q1-1.5*IQR; the whiskers extending to the most extreme data points. f, Disease-free survival (DFS) analysis of stratified HCC patients. CD3−CD56+NKhigh (nnon-REC = 17 patients, nREC = 14 patients) and CD3−CD56+NKlow (nnon-REC = 14 patients, nREC = 16 patients) subgroups were separated using the median ratio of CD3−CD56+ NK abundance relative to DAPI+ cell abundance at IF (0.9245). Shaded areas correspond to 95% confidence intervals and central lines indicate medians. The survival curves of the two subgroups cannot be clearly distinguished, as the log-rank test determined this comparison to be not significant, with notation of ns. g, DFS analysis of stratified HCC patients. CD3−CD16+CD56+NKhigh (nnon-REC = 19 patients, nREC = 12 patients) and CD3−CD16+CD56+NKlow (nnon-REC = 12 patients, nREC = 18 patients) subgroups were separated using the median ratio of CD3−CD16+CD56+ NK abundance relative to DAPI+ cell abundance at IF (0.0827). Shaded areas correspond to 95% confidence intervals and central lines indicate medians. The survival curves of the two subgroups cannot be distinguished with significance, as suggested by the log-rank test and marked as ns. Statistical significance: *P < 0.05, **P < 0.01, and ns—not significant.

Source data

Extended Data Fig. 2 Liquid Chromatography-Tandem Mass Spectrometry analysis of HCC tissue and prognosis analysis of HCC patients in TCGA database.

a, Leukocyte-enriched areas from proteomics analysis of six HCC tissues (isolated by laser-capture microdissection) using Liquid Chromatography-Tandem Mass Spectrometry. b, Protein-based enrichment scores of leukocytes (T, B, and Myeloid cells) at AS, IF, and TC. Data are means ± 2 s.e.m. Sample sizes: AS (6 ROIs from 2 non-REC patients vs. 5 ROIs from 4 REC patients), IF (9 ROIs from 2 non-REC patients vs. 21 ROIs from 4 REC patients), and TC (6 ROIs from 2 non-REC patients vs. 14 ROIs from 4 REC patients). Unpaired two-tailed Student’s t-test, finding none to be significant. c, Protein expression profile across HCC tissue compartments from recurrent and non-recurrent patients. Bars depict the number of proteins detected by liquid chromatography-tandem mass spectrometry in 61 regions of interest (ROIs) from 6 HCC patients. ROIs are categorized by tissue compartments: invasive front (IF, n = 30 ROIs), adjacent stroma (AS, n = 11 ROIs), and tumor center (TC, n = 20 ROIs). Samples are derived from 2 non-recurrent (non-REC) and 4 recurrent (REC) HCC patients. The color-coded bar below the X-axis distinguishes non-REC (blue) from REC (red) patient samples within each compartment. d, The clustering patterns of compartments in different ROIs. Principal components analysis was conducted using 61 ROIs sampled at AS, IF, and TC from all the 6 HCC patients (21 ROIs from the 2 non-REC HCC patients and 40 ROIs from the 4 REC HCC patients). The X-axis and Y-axis denote the explained variance for the first and second principal components. Ellipses show the ROI clustering results at a 95% confidence level. e, Kaplan-Meier curves of disease-free survival (DFS) stratified by the expression levels of SPON2, ZFP36L2, HLA-DRB1, VIM, and ZFP36. From left to right, Kaplan-Meier curves are shown for DFS stratified according to the expressions of SPON2, ZFP36L2, HLA-DRB1, VIM, and ZFP36, respectively. The PanCancer Atlas cohort of 348 HCC patients from TCGA database were divided into high-expression (n = 174 patients) and low-expression (n = 174 patients) subgroups by the median expression of each gene. Shaded areas correspond to 95% confidence intervals and central lines indicate medians. The log-rank test determines the P value for comparing the DFS curves of the two subgroups, finding none to be significant. f, Comparisons of SPON2, ZFP36L2, HLA-DRB1, VIM, and ZFP36 expressions between patients with non-recurrent HCC (non-REC, n = 168 patients) and with recurrent HCC (REC, n = 131 patients). The PanCancer Atlas cohort of 299 HCC patients from TCGA database were used, and these comparisons were conducted using unpaired two-tailed Student’s t-tests.

Source data

Extended Data Fig. 3 TIMES demonstrates superior performance in predicting HCC recurrence compared to existing prognostic markers.

a, Comparison of prediction accuracies of ZFP36L2, ZFP36, VIM, and HLA-DRB1, as well as their combinations with SPON2. The prediction accuracy of an extreme gradient boosting model built on SPON2-labeled mIHC data was 89.77%, the highest among models constructed on mIHC data labeled by each of the five identified biomarkers (i.e., SPON2, ZFP36L2, ZFP36, VIM, and HLA-DRB1). We used SPON2’s accuracy as a baseline and subtracted it from the prediction accuracies of the other four biomarkers and their combinations with SPON2, in order to highlight the decreased or increased prediction accuracy compared to SPON2 alone. The X-axis lists the gene names (“+” indicates gene combinations), and the Y-axis shows the differences in prediction accuracies resulting from the subtraction. b, Comparisons of TIMES scores assigned to patients at different TNM stages. A bar chart compares TIMES scores that were assigned to the patients at early (I, II; n = 124 patients) and late (III, IV; n = 31 patients) stages, as defined by the Tumor Node Metastasis (TNM) system. Colored dots represent the scores of individual patients, and data are means ±2 s.e.m. Unpaired two-tailed Student’s t-test, P = 0.04. c, Comparisons of TIMES scores assigned to non-REC and REC patients. Left-to-right panels show ‘unstratified’ group (nnon-REC = 91 patients, nREC = 72 patients), TNM-stratified subgroups as ‘TNM < II’ (nnon-REC = 22 patients, nREC = 16 patients) and ‘TNM ≥ II’ (nnon-REC = 46 patients, nREC = 43 patients), and BCLC-stratified subgroups as ‘BCLC = A’ (nnon-REC = 17 patients, nREC = 9 patients) and ‘BCLC = B or C’ (nnon-REC = 53 patients, nREC = 51 patients). Colored dots represent the scores of individual patients, and data are means ±2 s.e.m. Unpaired two-tailed Student’s t-test, P = 7.3 × 10−16 for comparison within ‘unstratified’ group, P = 7.1 × 10−6 for comparison within ‘TNM < II’ subgroup, P = 7.4 × 10−14 for comparison within ‘TNM ≥ II’ subgroup, P = 0.036 for comparison within ‘BCLC = A’ subgroup, and P < 2.2 × 10−16 for comparison within ‘BCLC = B or C’ subgroup. d, Left: DFS curves stratified by TNM, with patients categorized into I (n = 53 patients), II (n = 71 patients), III (n = 16 patients), and IV (n = 15 patients) subgroups. The log-rank test determines P values for the comparisons between DFS curves, and only the significant comparisons are marked with notation as *P < 0.05 and **P < 0.01. Right: DFS curves stratified by BCLC, with patients categorized into A (n = 26 patients), B (n = 19 patients), and C (n = 85 patients) subgroups. Shaded areas correspond to 95% confidence intervals and central lines indicate medians. P = 0.002 for comparison between ‘TNMI’ and ‘TNMIII’ subgroups, P = 0.04 for comparison between ‘TNMI’ and ‘TNMIV’ subgroups, P = 0.05 for comparison between ‘TNMII’ and ‘TNMIII’ subgroups; P = 0.03 for comparison between ‘BCLCA’ and ‘BCLCB’ subgroups, P = 0.005 for comparison between ‘BCLCA’ and ‘BCLCC’ subgroups. e, Univariate Cox regression results for DFS. First column: TIMES and 17 clinical factors, of which natural logarithms of hazard ratio values, \({\rm{ln}}({\rm{HR}})\), are significantly different from 0 (P < 0.05 in the univariate regressions here). Second column: \({\rm{ln}}({\rm{HR}})\) values along with the 95% confidence intervals shown as horizontal bars (third column), showing means of \({\rm{ln}}({\rm{HR}})\) ± 2 s.e.m. Last column: P values from the Wald test, indicating the statistical significance of the corresponding \({\rm{ln}}({\rm{HR}})\). f, DFS multivariate Cox regression on TIMES and 17 clinical factors that were identified significant (P < 0.05 from the Wald test) in univariate Cox regressions. Tumor differentiation grade: no higher than Grade 2, between Grade 2 and Grade 3 (G2-G3), and no lower than Grade 3 (G3). Second column: \({\rm{ln}}({\rm{HR}})\) values along with the 95% confidence intervals shown as horizontal bars (third column), showing means of \({\rm{ln}}({\rm{HR}})\) ± 2 s.e.m. Last column: P values from the Wald test * P < 0.05, *** P < 0.001, and **** P < 0.0001. n = 254 patients. g, TIMES demonstrates superior performance in predicting HCC recurrence compared to NK cell subsets, including CD3−CD57+ mature NK, CD3−CD16+CD56+ cytotoxic NK, and CD3−CD56+ ordinary NK cells. Using a separate dataset of mIHC staining, receiver operating characteristic curves and area under the curve (AUC) of the TIMES model (AUC = 0.996) were compared against models built on the abundances of three NK cell subsets in the IF or TC compartments. The TIMES scoring system, which incorporates spatial information from multiple biomarkers, outperformed these prediction models based solely on NK cell subsets. h, Comparative predictive performance of TIMES versus SPON2 expression at TC for HCC recurrence. Receiver operating characteristic curves and area under the curve (AUC) of the TIMES model (AUC = 0.82) were compared against that built from SPON2 expression at TC (AUC = 0.59). i, Distributions of TIMES scores for anti-PD1 immunotherapy recipients. According to responsiveness to immunotherapy, patients can be categorized as subgroups of progressive disease (PD, in blue, n = 12 patients) and of partial response (PR, red, n = 13 patients) by RECIST (Response Evaluation Criteria in Solid Tumor) criteria. The X-axis represents TIMES scores, while the Y-axis displays the counts of the corresponding scores. The unpaired two-tailed Wilcoxon rank-sum test was employed to assess score distribution differences. The median difference (MD) between scores from PR patients and PD patients was 0.267, yielding a P value of 1.61 × 10−3. Statistical significance: *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001.

Source data

Extended Data Fig. 4 An open web-tool has been developed to facilitate clinical applications of our TIMES scoring system.

a, Home page of the web-tool at http://sun.times.ustc.edu.cn/. This web-tool allows users to upload standard pathological images for HCC recurrence risk prediction (see the webpage for a tutorial and Supplementary Video 1 for a brief introduction to the TIMES scoring system). It is designed to require minimal additional work from clinicians and is tolerant to incomplete measurement input, ensuring its applicability even in cases where certain data points may be missing (b, c). b, c, High concordances between TIMES scores computed using different numbers of tiles from each compartment (AS, IF, and TC). The scatterplots display the relationship between the original TIMES scores (X-axis), calculated using whole slide images (WSI), and the TIMES scores computed using a limited number of tiles (Y-axis) from each compartment. In b, TIMES scores were calculated using 2 tiles per compartment, while in c, 3 tiles per compartment were used. The concordance between the scores was assessed using Spearman’s rank correlation coefficient (ρ). The statistical significance of the correlation was determined using a paired two-tailed Student’s t-test, with the null hypothesis that the correlation coefficient is equal to zero. Both correlations were found to be highly significant, and P < 2.2 × 10−16 for Spearman correlation coefficients and for linear regression coefficients, indicating a strong agreement between the TIMES scores computed using WSI and those obtained using a limited number of tiles. These results demonstrate the robustness and stability of the TIMES scoring system, even when applied to a reduced number of tissue subregions; in other words, the TIMES score was not compromised by incomplete measurement input (see Methods). d, Tentative guide for escalation of treatment strategy (using immunotherapy as a paradigm) based on TIMES score. In a diverse cohort of HCC patients with varying risks of recurrence and potential differences in immunotherapy response, our TIMES scoring system may offer an initial guideline. Patients are stratified based on their TIMES scores: (1) Risk Stratification: Patients with TIMES scores greater than 0.5 are identified as having a high risk of recurrence. (2) Immunotherapy Consideration: Among the high-risk patients, those with TIMES scores exceeding a specific threshold are recommended for immunotherapy due to the anticipated high responsiveness. Establishing this threshold necessitates comprehensive data collection from a substantial number of immunotherapy recipients to refine prediction model training. This tentative guide aims to assist in identifying HCC patients who may benefit from immunotherapy, laying the foundation for further optimization as additional data becomes available for model enhancement.

Source data

Extended Data Fig. 5 Spatial transcriptomic analyses revealed enriched activation and migration of SPON2+ NK cells.

a, UMAP embedding of all cells from donors with T cells, B cells, NK cells, and Myeloid projected. The UMAP plot shows annotations and colors for major immune cell types in the HCC ecosystem based on single-cell RNA sequencing data. b, UMAP embedding of all spatial spots from donors and relative gene marker expressions. The UMAP plot showcases the expression levels of SPON2, projected onto the same immune cell type map as shown in a, using single-cell RNA sequencing data. c, Differential expression of effector molecules in tumor-infiltrating SPON2+ and SPON2− NK cells from HCC patients. Analysis of single-cell RNA sequencing data from HCC tumor samples (n = 15 patients) comparing the expression levels of IFNG (encoding IFN-γ, P = 8.0 × 10−4) and PRF1 (encoding Perforin, P = 8.0 × 10−3) between SPON2+ NK cells (n = 15 patients) and SPON2− NK cells (n = 15 patients). Boxplots show the distribution of expression levels, with individual data points overlaid. Paired two-tailed Student’s t-test. d, Percentages of SPON2+ NK and SPON2− NK cells with NK-immunity-related pathways enriched (Pct. exp); average expression (Ave. exp). e, IFN-γ receptor gene expressions at TC. We separated SPON2high (n = 3 ROIs) and SPON2low (n = 4 ROIs) subgroups by SPON2’s median expression of CD57high ROIs (23.84). IFNGR1 (P = 0.016), IFNGR2 (P < 0.0001). Unpaired two-tailed Student’s t-test. f, Projections of enrichment scores based on SPON2, ZFP36, ZFP36L2, VIM, or HLA-DRB1, in combination with NK marker genes (see Methods). These enrichment scores (coded by color gradients) are projected onto the AS, IF, and TC compartments of tissues from REC or non-REC patients. From left to right panels, the enrichment scores denote the abundances of SPON2+ NK, ZFP36+ NK, ZFP36L2+ NK, VIM+ NK, and HLA-DRB1+ NK cells, respectively. g, Gene ontology (GO) enrichment analysis identifies biological pathway differences between SPON2+NKhigh and SPON2+NKlow spots at TC. The pathways showing most significant upregulations in SPON2+NKhigh spots compared to SPON2+NKlow spots at TC are listed. h, Ridge plots comparing IFNGR2 expression in SPON2+NKhigh and SPON2+NKlow at TC. The upper and lower panels show the distributions of IFNGR2 expression levels in SPON2+NKhigh and SPON2+NKlow spots at TC, respectively. Color gradient depicts the probability density. i, Gene set enrichment analysis for positive regulation of leukocyte migration and leukocyte mediated immunity at TC, based on differentially expressed genes between SPON2+NKhigh and SPON2+NKlow spots. NES: normalized enrichment score; statistical significance through a one-tailed permutation test (see Methods). j, Gene set variation analysis compares immune processes between SPON2highCD57high and SPON2lowCD57high ROIs at TC. The listed immune processes highlight distinctions between SPON2highCD57high (n = 3 ROIs) and SPON2lowCD57high (n = 4 ROIs) subgroups, categorized by the median expression level of SPON2 at CD57high TC (23.84). The columns denote the ROI samples. Color gradient depicts the enrichment level. Statistical significance: *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001.

Source data

Extended Data Fig. 6 3D-printed HCC model demonstrated the role of SPON2 in promoting the activation and migration of NK92 cells.

a, b, Representative images depicting NK92 cell migration toward the tumor region. NK92 cells were fluorescently labelled with PKH26 (in red), and HepG2 were labeled with PKH67 (green). They were tracked under a fluorescence microscope every 30 min for a period of 5 days while we selected the images at 12th, 24th, 48th, and 60th h in a, or each day for display in b. White circles highlight NK92 cells that infiltrated to tumor region. c, Flow cytometry gating strategy of SPON2+ NK92 cells in outer and inner circles of 3D bio-printed models. This gating strategy was used to identify SPON2+ NK92 cells that infiltrated to the inner circle or stayed at the outer circle of 3D-printed HCC models. d, Flow cytometry gating strategy to compare the expressions of IFN-γ and Perforin in SPON2+ NK92 cells and SPON2− NK92 cells. This gating strategy was used to evaluate IFN-γ and Perforin expressions in SPON2+ and SPON2− NK92 cells after 3 h of PMA and ionomycin stimulation. Numbers in the drawn gates indicate cell percentages. e, Proportions of SPON2+ NK92 to all the NK92 cells in outer and inner circles of 3D bio-printed models. Left: exemplar gating strategies (squares) for SPON2+ NK92 cell selection. Right: we compared proportion values between outer (n = 6 independent experiments) and inner circles (n = 6 independent experiments) in six bio-printed models (P = 0.0017). Paired two-tailed Student’s t-test. f, We measured IFN-γ and Perforin expressions in SPON2− and SPON2+ NK92 cells after 3 h of PMA and ionomycin stimulation. The proportions of IFN-γ+ NK92 cells (P = 0.018) and Perforin+ NK92 cells (P = 0.0002) were compared between SPON2− NK92 (n = 4 cells) and SPON2+ NK92 (n = 4 cells) cells. Paired two-tailed Student’s t-test. g, Flow cytometry verification of SPON2 expression in three cell lines. SPON2 expression in SPON2-overexpression NK92 (NK92OE-SPON2) cells and SPON2-knockdown NK92 (NK92KD-SPON2) cells relative to empty vector NK92 (NK92EV) cells. NK92KD-SPON2 cells were obtained via lentiviral shRNA transfection which achieved a 99% reduction in SPON2 expression. Numbers in the drawn gates indicate cell percentages. h, Representative images depicting the migration of NK92EV, NK92OE-SPON2, and NK92KD-SPON2 toward the tumor region. We documented videos of NK92EV (left), NK92OE-SPON2 (middle), and NK92KD-SPON2 (right; NK92 cells were colored in red) migration towards HepG2 cells (green) during a 12-h period. At the 12th h, images were captured by fluorescence microscopy (see Supplementary Videos 2–4). In these migration assays comparing NK92OE-SPON2 and NK92KD-SPON2 cells to NK92EV cells, NK92OE-SPON2 cells migrated to HepG2-cell-rich area within 12 h, while NK92EV and NK92KD-SPON2 cells did not. i, IFN-γ and Perforin expressions in NK92EV cells and NK92OE-SPON2 cells. The left two subfigures show the distributions of cells with different fluorescence intensities: IFN-γ (left) and Perforin (right) measurement. We use the maximal intensity value of ISO (vertical black dashed lines) as the threshold to call IFN-γ + (or Perforin+) and IFN-γ− (or Perforin−). We quantified the proportions of IFN-γ+ (or Perforin+) NK92 cells in NK92EV and NK92OE-SPON2 cells. In the right two subfigures, we compared these proportion values between NK92EV cells (n = 5) and NK92OE-SPON2 cells (n = 5) in three flow cytometry experiments (P = 8.0 × 10−4, P = 0.12), using paired two-tailed Student’s t-tests. Statistical significance: *P < 0.05, **P < 0.01, *** P < 0.001, ****P < 0.0001, and ns—not significant.

Source data

Extended Data Fig. 7 mIHC images and single-cell RNA sequencing analysis that revealed differences between SPON2+ NK and SPON2− NK cells in cell signaling interactions.

a, mIHC images depicting SPON2+ NK accumulation at TC. White arrows depict CD16+CD56+CD57+SPON2+CD3−CD163−CD68− NK cells. b, Crosstalk between different cell types identified using single-cell RNA sequencing data. The left panel illustrates the number of interactions between cell signaling ligands in various cell types, while the right panel depicts the strength of these interactions, as determined by CellChat analysis. Each dot represents a specific cell type, with different colors corresponding to distinct cell types. The size of the dot is proportional to the abundance of the cell type within the sample. The thickness of the lines connecting the dots indicates either the number (left) or strength (right) of the interactions between the connected cell types. Abbreviations: Epi., Epithelial cell; Endo., Endothelial cell; Mey., Myeloid cell. c, Top 50 signaling pathways that had largest fold changes between SPON2+ NK and SPON2− NK cells. Top 50 signaling pathways out of the 133 available signaling pathways are displayed and ranked by their respective fold changes between SPON2+ NK and SPON2− NK cells in the signaling axes received from other cell signal sources. Aside from the first five pathways that were not directly related to immune responses, IFN-II signaling pathway ranked the highest among those related to immune response. d, Heatmap showing the predicted ligand activity by NicheNet on genes highly expressed in SPON2+ NK. Darker purple indicates higher potential for regulation. Ligand-receptor analysis revealed that the IFN-II (IFN-γ) genes of SPON2+ NK cells were more sensitive to regulation from other cells in microenvironment. e, Hierarchical plot shows the communication network of SPON2+ NK, SPON2− NK, and other cells for IFNG- (IFNGR1+IFNGR2) signaling. In CellChat analysis, as shown in the left panel, SPON2+ NK and SPON2− NK cells served as signal-receiving target cells (represented by unfilled circles in middle) while these NK cells as well as other cell types acted as signal sources (filled color circles); Edge width indicates the communication probability and edge colors are the same as the signaling source. The right panel shows the communication network with other cell types used as signal-receiving target cells (unfilled circles in middle). Such analysis predicted an association of the IFNG-(IFNGR1-IFNGR2) signaling axis between CD8+ T and SPON2+ NK cells, but not between SPON2− NK and CD8+ T cells.

Source data

Extended Data Fig. 8 In vitro models revealed the requirement for direct cell-cell contact between SPON2+ NK cells and CD8+ T cells.

a, Representative images of HCC stained with mIHC. Red arrows depict CD57+SPON2+CD3− NK cells, and yellow arrows depict the neighboring CD3+CD8+ T cells. They show the spatial proximity between the two cell types (identified based on the expression of multiple markers including CD3, CD8, CD16, CD57, CD56, CD68, CD163, and SPON2) across different patients and tissue regions. b, Representative images depicting CD8+ T cells in proximity to SPON2+CD57+ NK cells. Left: low-magnification overview showing IF and TC. Dashed box indicates high-magnification area. Right: high-magnification images of mIHC (example from REC patient L-8022). c, Numbers of CD8+ T cells proximal (within 15 μm) to CD57+SPON2+ NK (n = 5,142 cells) and CD57+SPON2− NK cells (n = 17,515 cells) from 28 patients. Two-tailed Mann-Whitney U test, P = 0.0079. Data are means ± s.e.m. d, Procedure for NKOE-SPON2 and NKKO-SPON2 generation. Human peripheral blood mononuclear cells (PBMCs) obtained from healthy donors, along with membrane-bound IL-21 (mbIL21) feeder cells, were cultured in 200 U/mL IL-2 for over a week to facilitate the expansion of NK cells. Subsequently, these NK cells underwent electroporation with Cas9 protein and SPON2-specific single guide RNA (sgRNA) to generate NKKO-SPON2 cells. Lentiviral transfection was used to induce increased SPON2 abundance and generate primary NK cell populations with SPON2 overexpression. The resulting NKOE-SPON2 and NKKO-SPON2 cells were stimulated with 200 U/mL IL-2, 20 ng/mL of IL-12, 50 ng/mL of IL-15, and 10 ng/mL of IL-18 for one day, and then co-cultured with CD8+ T cells and HepG2 cells. e, Flow cytometry analysis of SPON2 protein expression in NKOE-SPON2 and NKKO-SPON2 cells. Data was collected 3 days after transgene transfer into primary NK cells. f, g, Contact and transwell experiments of CD8+ T and SPON2-overexpression (or SPON2-knockout) NK cells. Using NKOE-SPON2 and NKKO-SPON2cells of a patient, we conducted five experiments in the first stage: CD8+ T cells alone (control), co-culture of NKOE-SPON2 and CD8+ T cells (contact), co-culture of NKKO-SPON2 and CD8+ T cells (contact), co-culture of NKOE-SPON2 and CD8+ T cells (transwell), and co-culture of NKKO-SPON2 and CD8+ T cells (transwell). In the second stage, two additional experiments involved co-culture of NKOE-SPON2 and CD8+ T cells (contact but with the presence of anti-IFN-γ for 24 h, an antibody that can block IFN-γ) and co-culture of NKOE-SPON2 and CD8+ T cells (contact but with the presence of IgG for 24 h, serving as a control). The expression changes of CD107a and GzmB in CD8+ T cells were subsequentially measured. (f) Representative flow cytometry profiles for a given patient. From left to right, there show the five experiments in the first stage and the two additional experiments in the second stage. (g) Comparisons of CD107a+CD8+ T cell proportions (top) and GzmB+CD8+ T cell proportions (bottom) between the experiments. The X-axis denotes these experiments with corresponding conditions marked by “+,” and the Y-axis depicts cell proportion values. Colored circles represent cell proportion values of individual patients (n = 9 patients). Paired two-tailed Student’s t-test. Data are means ± s.e.m. For CD107a: CD8+ T (Control) vs. NKOE-SPON2 + CD8+ T (Contact): P = 0.0002; CD8+ T (Control) vs. NKKO-SPON2 + CD8+ T (Contact): P = 0.035; NKOE-SPON2 + CD8+ T (Contact) vs. NKKO-SPON2 + CD8+ T (Contact): P = 0.0076; NKOE-SPON2 + CD8+ T (Contact) vs. NKOE-SPON2 + CD8+ T (Transwell): P = 0.012; NKKO-SPON2 + CD8+ T (Contact) vs. NKKO-SPON2 + CD8+ T (Transwell): P = 0.13; NKOE-SPON2 + CD8+ T (Contact Anti-IFN-γ) vs. NKOE-SPON2 + CD8+ T (Contact IgG): P = 0.010. For GzmB: CD8+ T (Control) vs. NKOE-SPON2 + CD8+ T (Contact): P = 0.0007; CD8+ T (Control) vs. NKKO-SPON2 + CD8+ T (Contact): P = 0.0029; NKOE-SPON2 + CD8+ T (Contact) vs. NKKO-SPON2 + CD8+ T (Contact): P = 0.0001; NKOE-SPON2 + CD8+ T (Contact) vs. NKOE-SPON2 + CD8+ T (Transwell): P = 0.22; NKKO-SPON2 + CD8+ T (Contact) vs. NKKO-SPON2 + CD8+ T (Transwell): P = 0.99; NKOE-SPON2 + CD8+ T (Contact Anti-IFN-γ) vs. NKOE-SPON2 + CD8+ T (Contact IgG): P = 0.25. h, i, Annexin V expression in HepG2 co-cultured with of NKOE-SPON2 or NKKO-SPON2 cells. Control: HepG2 cells alone. (h) Flow cytometry profiles. (i) Proportions of HepG2 with Annexin V expressed in total HepG2 cells. Circles represent samples (n = 4 independent experiments). Unpaired two-tailed Student’s t-test. Data are means ± s.e.m. Control vs. NKOE-SPON2: P = 0.0002; Control vs. NKKO-SPON2: P = 0.005; NKOE-SPON2 vs. NKKO-SPON2: P = 0.0039. j, k, Annexin V expression in HepG2 co-cultured with CD8+ T and either NKOE-SPON2 or NKKO-SPON2 cells. (j) Flow cytometry profiles. (k) Proportions of HepG2 with Annexin V expressed in total HepG2 cells (n = 4 patients). Unpaired two-tailed Student’s t-test. Data are means ± s.e.m. CD8+ T vs. CD8+ T + NKOE-SPON2: P = 0.0002; CD8+ T vs. CD8+ T + NKKO-SPON2: P = 0.033; CD8+ T + NKOE-SPON2 vs. CD8+ T + NKKO-SPON2: P = 0.0059; Statistical significance: *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, and ns—not significant.

Source data

Extended Data Fig. 9 Spon2-knockout (Spon2−/−) mice have increased tumor count and decreased IFN-γ+CD8+ T cells (subcutaneous and orthotopic models).

a, Construction procedure for in situ HCC models (see Methods). b, Tumor size comparison between Spon2−/− and WT mice. On Day 35, we harvested the HCC tumors from Spon2−/− and WT mice, measured their sizes, and counted tumor nodules. Representative images of the tumors are shown here with tumor nodules pointed by red arrows. c, Hepatosomatic index and tumor nodule comparisons between Spon2−/− and WT mice. We compared the hepatosomatic index values (the percentage of tumor weight on Day 35 relative to total body weight; P = 0.0015) and tumor nodule counts (P = 0.0167) between WT (n = 7 mice) and Spon2−/− (n = 6 mice) mice using an unpaired two-tailed Student’s t- test. Data are means ± s.e.m. d, Flow cytometry gating strategy of NK cells and CD8+ T cells. Numbers in the drawn gates indicate cell percentages. This involved a successive cell selection process with the yellow arrows pointing from one to another cell subgroups that were subsequently selected. e, Comparison of tumor-infiltrating CD45+CD8+ T cell proportions between WT and Spon2-knockout (Spon2−/−) mice. The Y-axis denotes the proportions of subcutaneous tumor CD45+CD8+ T cells relative to CD45+ cells in WT (n = 4 mice) and CD45+ cells in Spon2−/− (n = 4 mice) mice. The difference of the means was not significant (P = 0.75) based on an unpaired two-tailed Student’s t-test. Data are means ± s.e.m. f, Flow cytometry profiles for selecting specific T and NK cells from Spon2−/− and WT mice. From left to right, there show representative contour plots of IFN-γ+CD8+ T cells from Spon2−/− and WT mice, and IFN-γ+NK1.1+ cells from Spon2−/− and WT mice. These mice are orthotopic HCC models. Numbers in the drawn gates indicate the cell percentages. g, Subcutaneous tumors from Spon2−/− and WT mice on Day 18 (related to Fig. 5a–c). h, Orthotopic HCC tumors from Spon2−/− and WT mice on Day 22 (related to Fig. 5f,g). i, Construction procedure for subcutaneous HCC models that depleted NK, CD8+ T, or both cell types (see Methods). j, Tumor volume growth curves in WT and Spon2−/− mice treated with different antibodies. NK cell depletion resulted in significantly larger tumors compared to CD8+ T cell depletion in both Spon2−/− and WT mice (P < 0.0001). CD8+ T cell depletion led to significantly larger tumors in Spon2−/− mice compared to WT mice (P < 0.0001). In contrast, depletion of NK cells or both NK and CD8+ T cells resulted in no difference in tumor volumes between Spon2−/− and WT mice. Two-way ANOVA (n = 3 mice). Data are means ± s.e.m. k, Genotyping confirmation for NK cell-specific Spon2-knockout model construction. Using PCR, we confirmed that Spon2f/f was homozygous (left two columns) and Ncr1Cre+ was present (right two columns). The wild type allele of Spon2 is 277 bp, and the allele inserted by flox is 336 bp (left); The allele of Ncr1 with Cre activity is 364 bp (right). M: DNA markers (100 bp, 250 bp, 500 bp from bottom to top). l, Flow cytometry analysis of SPON2 protein expression in peripheral blood NK cells of WT and Spon2f/f-Ncr1Cre+ mice. SPON2 expression in peripheral blood NK cells of Spon2f/f-Ncr1Cre+ mice relative to peripheral blood NK cells of WT mice. Numbers in the drawn gates indicate the cell percentages. m, Subcutaneous tumors from Spon2f/f-Ncr1Cre− control mice and Spon2f/f-Ncr1Cre+ NK cell-specific SPON2 knockout mice on Day 23 (related to Fig. 5i,j). n, Tumor-infiltrating immune cell populations in SPON2 conditional knockout mice. Flow cytometric quantification of CD45+ NK cells (P = 0.0008), CD45+CD8+ T cells (P = 0.21), IFN-γ+ NK cells (P = 0.034), and IFN-γ+CD8+ T cells (P = 0.0001) in subcutaneous tumors from Spon2f/f-Ncr1Cre- (n = 5) and Spon2f/f-Ncr1Cre+ (n = 5) mice. Each data point represents an individual mouse. Unpaired two-tailed Student’s t-test. Data are means ± s.e.m. o, Subcutaneous tumors on Day 14 from HepG2 mouse models of human HCC injected with PBS, NKKO-SPON2, or NKOE-SPON2 cells (related to Fig. 5k,l). p, Construction procedure for HepG2 xenograft models of human HCC with human NKstimulated cells adoptive transfer (see Methods). q, Flow cytometry analysis of SPON2 protein expression in human NKstimulated and NK cells. Data were collected after 5 days of NK cell culture. r, Comparisons of subcutaneous tumor sizes on Day 14 among the HepG2 xenograft models of human HCC injected with PBS, NK, or NKstimulated cells. s, Volume growth curves of subcutaneous tumors extracted from the HepG2 xenograft models of human HCC injected with PBS, NK, or NKstimulated cells. Two-way ANOVA (n = 3 mice). Data are means ± s.e.m. PBS vs. NK: P < 0.0001; PBS vs. NKstimulated cells: P < 0.0001; NK vs. NKstimulated cells: P = 0.0014. Statistical significance: *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, and ns—not significant.

Source data

Extended Data Fig. 10 Detection of tumor-infiltrating leukocytes in HCC patients using mass cytometry and graphical summary of the anti-tumoral roles of SPON2+ NK cells.

a, Flow cytometry gating strategy of NK cells. Numbers in the drawn gates indicate the cell percentages. This gating strategy was used to select CD3−CD56+ NK cells. b, Classification of SPON2+ NK and other NK cells according to the expression level of SPON2. We show t-SNE projection of SPON2+ NK and other NK cells detected by mass cytometry. Upon categorizing the immune cell types based on SPON2 expression status at the protein level, 48.9% of SPON2+ cells corresponded to NK cells (also see Fig. 5m,n). This finding suggests that NK cells are the primary immune cell type expressing SPON2 in the HCC tumor microenvironment, further supporting the potential role of SPON2+ NK cells in HCC progression and their possible therapeutic relevance. c, t-SNE projection depicting IFN-γ expressions in HCC tissues. Detected by mass cytometry, IFN-γ expression is projected onto a cell type map that is the same as shown in the left panel of Fig. 5m. d, SPON2 expression in TC samples is positively correlated with patients’ overall survivals. The X-axis shows SPON2+ NK enrichment levels at TC, and the Y-axis denotes patients’ overall survivals in months. Spearman’s correlation coefficient ρ and P values are annotated (ρ = 0.59, P = 0.034), and a linear regression model is obtained and illustrated as a dashed red line (n = 13 samples from 13 patients, shown as blue points). Shaded areas correspond to 95% confidence intervals. e, The expression of SPON2 in NK cells is positively correlated with the expression of IFN-γ. The X-axis shows SPON2+ NK enrichment levels at IF and TC, and the Y-axis denotes IFN-γ+ NK enrichment levels at IF and TC. Two-tailed Spearman’s correlation coefficient ρ and P values are annotated, and linear regression models are obtained and illustrated as dashed red lines (n = 43 samples from 27 patients, shown as blue points). The correlation analysis suggested consistent trends between SPON2+ NK enrichment in HCC and IFN-γ production (ρ = 0.55, P = 1.6 × 10−4). Shaded areas correspond to 95% confidence intervals. f, Graphical summary of the anti-tumoral roles of SPON2+ NK cells. SPON2+ NK cell enrichment in the invasive front (IF) compartment of HCC tissue is associated with a lower risk of recurrence. These SPON2+ NK cells demonstrate functional activation with increased migration potential and cytotoxicity through elevated expression of IFN-γ and Perforin. Besides their direct cytotoxic activity, SPON2+ NK cells also interact with CD8+ T cells in anti-tumoral process. Therefore, HCC patients with a higher proportion of SPON2+ NK cells are expected to exhibit a lower risk of HCC recurrence.

Source data

Supplementary information

Supplementary Information (download PDF )

This file contains Supplementary Discussion, Supplementary Figs. 1–9, Supplementary Tables 17–38 and Supplementary References

Reporting Summary (download PDF )

Peer Review file (download PDF )

Supplementary Tables (download ZIP )

Supplementary Tables 1–17

Supplementary Video 1 (download MP4 )

Introduction to the Tumor Immune MicroEnvironment Spatial (TIMES) scoring system

Supplementary Video 2 (download MP4 )

The migration process of NK92 cells overexpressing SPON2 (NK92OE-SPON2) towards HepG2 tumor cells

Supplementary Video 3 (download MP4 )

The migration process of NK92 cells with SPON2 knockdown (NK92KD-SPON2) towards HepG2 tumor cells

Supplementary Video 4 (download MP4 )

The migration process of empty vector NK92 (NK92EV) cells towards HepG2 tumor cells.

Source data

Rights and permissions

Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Jia, G., He, P., Dai, T. et al. Spatial immune scoring system predicts hepatocellular carcinoma recurrence. Nature 640, 1031–1041 (2025). https://doi.org/10.1038/s41586-025-08668-x

Download citation

  • Received:

  • Accepted:

  • Published:

  • Version of record:

  • Issue date:

  • DOI: https://doi.org/10.1038/s41586-025-08668-x

This article is cited by

Search

Advanced search

Quick links

Nature Briefing

Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

Get the most important science stories of the day, free in your inbox. Sign up for Nature Briefing