Fig 2: Identification of IFN-HEVs by pan-cancer vascular endothelial cell atlas(A) Uniform manifold approximation and projection (UMAP) plot of the Slingshot-derived VEC trajectory, depicting two major divergent trajectories. Trajectory 1 (T1) and trajectory 2 (T2) originated from VenEC-CLU (Clusterin) to ArtECs or TS-ECs, respectively. IFN-CapEC was the branchpoint of these two trajectories. VenEC-CLU, CLU+ venous endothelial cell; ArtEC, arterial endothelial cell; TS-EC, tip- and stalk-like endothelial cell; IFN-CapEC, IFN-activated capillary endothelial cell.(B) UMAP plot of ACKR1+RGCC+capillary venous (CV) endothelial cells from a pan-cancer scRNA atlas (NGDC: PRJCA018695), colored by cell types. CV1.Quiescent, quiescent CV; CV2.ESM1, ESM1+ CV; CV3.CDKN1A, CDKN1A+ CV; CV4.Intermediate.KLF4, KLF4+ intermediate CV; CV5.MFSD2A, MFSD2A+ CV; CV6.IGFBP5, IGFBP5+ CV; CV7.ISG15, ISG15+ CV; CV8.IFN-HEV.CXCL9, CXCL9+ IFN-HEV; CV9.HEV.IL33, IL33+ conventional HEV.(C) Distribution of CV subsets across CapEC-transition-1, IFN-CapEC, and CapEC-transition-2 clusters. CapEC-transition-1, transitional capillary-like EC-1; CapEC-transition-2, transitional capillary-like EC-2.(D) DEG analysis revealed distinct gene expression profiles among CV7.ISG15, CV8.IFN-HEV.CXCL9, and CV9.HEV.IL33.(E) Comparison of putative distribution of CV7.ISG15, CV8.IFN-HEV.CXCL9, and CV9.HEV.IL33 within TLS ROIs from NPC and CN tissues. Two-sided unpaired Student’s t test or Wilcoxon rank-sum test was conducted, depending on whether the data passed the normality test. ∗∗∗∗p < 0.0001; n.s., p > 0.05.(F) Diffusion component map indicating the differentiation order of tumor-enriched trajectory (T2). CapEC-transition-1, transitional capillary-like EC-1.(G) Radar plot showing the scaled activities of enriched REACTOME pathways in IFN-HEV and TS-EC.(H) Loess regression smoothened expression (y axis) of the representative genes of chemokine, IFN, collagen degradation, and VEGF signaling in pseudotime with the 95% confidence intervals (CIs).(I) GSEA plot for upregulated genes of tumor TLSs from the GeoMx dataset. NES, normalized enrichment score. p values were determined by a one-tailed permutation test by GSEA.

Fig 3: Establishment of a pretreatment CXCL9-TLS response-predictive scoring system to predict the effect of immunotherapy(A) Representative mIHC images of IFN-HEVs and CXCR3+CD4+ T cells among NPC patients exhibiting complete response (CR), stable disease (SD), and progressive disease (PD) in response to ICB therapy.Scale bar, 50 μm.(B) Clinical characteristics and immunotherapy efficacy of the Sun Yat-Sen University Cancer Center (SYSUCC) training cohort (n = 30).(C) Frequency of patients exhibiting response (R) and non-response (NR) to ICB therapy, stratified on the best cutoffs of CXCR3+ cells, CD4+ cells, CXCL9+ cells, MECA-79+ cells, TLS density, CXCL9+MECA-79+ cells, CXCR3+CD4+ cells, and average distance between these cells. Patient numbers of each group are shown in brackets. Chi-square test. ∗p < 0.05, ∗∗p < 0.01.(D) Formula of CTRscore. CTRscore, CXCL9-TLS response-predictive scoring system.(E) Smoothed area under the curve (AUC)-receiver operating characteristic (ROC) plots for CTRscore, TLS density, CXCL9+MECA-79+ cells, CXCR3+CD4+ cells, and average distance between these cells in the SYSUCC training cohort (non-responders vs. responders).(F) Distribution of patients with high or low CTRscore (CTRscore-H/-L) in non-responders and responders of the SYSUCC training cohort (n = 30).(G) Clinical characteristics, immunotherapy efficacy, and distribution of patients with high or low CTRscore in non-responders and responders of the First Affiliated Hospital of Guangxi Medical University (GXMU) cohort (n = 34).(H) Clinical characteristics, immunotherapy efficacy, and distribution of patients with high or low CTRscore in non-responders and responders of the Sun Yat-sen Memorial Hospital, Sun Yat-sen University (SYSUMH) cohort (n = 29).(I) Clinical characteristics, immunotherapy efficacy, and distribution of patients with high or low CTRscore in non-responders and responders of the ShenShan Medical Center (SSMC) cohort (n = 20).

Fig 4: CXCL9+ IFN-HEV correlates with the entry of CD4+ T cells into tumor TLSs(A) Representative images of tumor FFPE tissues showing TLS regions, IFN-HEV signature score, and the CD4, CD8A, and CD20 expression level of each spot from the spatial transcriptome dataset of clear cell renal cell carcinoma (ccRCC) (GEO: GSE175540). Only spots containing endothelial cells were included.(B) Correlation analyses between the IFN-HEV signature score and the expression level of CD4, CD8A, and CD20 (20019 spots, Pearson correlation test). Only spots containing endothelial cells were included.(C) IFN-HEV signature score between spots inside and outside the TLS region in the ccRCC spatial transcriptome dataset (GEO: GSE175540). Two-sided unpaired Wilcoxon rank-sum test.(D) Representative flow cytometry plots showing proportions of IFN-HEVs (CXCL9+MECA-79+), CD4+ and CD8+ T cells in NPC tissues (left). Correlations between IFN-HEV density and CD4+ T cell density or the CD4+/CD8+ T cell ratio in NPC tissues were concluded (right). n = 16, Pearson-correlation test.(E) Proportions of IL-4+, CXCR5+, T-bet+, IL-17+, and FOXP3+ CD4+ T cells in NPC tissues with low or high IFN-HEV infiltration (n = 5). Data are presented as mean ± SEM. Two-sided unpaired Student’s t test. ∗p < 0.05; ∗∗p < 0.01; n.s., p > 0.05.(F) Representative mIHC images of IFN-HEVs and CXCR3+CD4+ T cells in TLSs of the internal NPC FFPE cohort. Scale bar, 50 μm.(G) Average distance from CXCR3−CD4+ or CXCR3+CD4+ T cells to IFN-HEVs in the internal NPC FFPE cohort (n = 120). No detectable IFN-HEVs in 15 slides. Two-sided paired Student’s t test. ∗∗∗∗p < 0.0001.(H) Correlation analysis of density of IFN-HEVs and CXCR3+CD4+ T cells in our NPC FFPE cohort (n = 135). Pearson correlation test.

Fig 5: CXCL9 inhibits M2 macrophage-induced pro-tumor signaling in co-cultured A549 cells. (A) Representative Western blot bands showing the protein expression levels of VEGF-C, MMP2, MMP9, ERK, p-ERK, AKT, and p-AKT in A549 cells and M2 macrophages under different treatment conditions. β-actin was used as the loading control. (B-H) Quantitative analysis of relative protein expression levels normalized to β-actin. Data are presented as the mean ± standard deviation (n=3). (B) Relative expression of VEGF-C. (C) Relative expression of MMP2. (D) Relative expression of MMP9. (E) Relative expression of total ERK. (F) Relative expression of p-ERK. (G) Relative expression of total AKT. (H) Relative expression of p-AKT. *, indicates a significant difference (P<0.05).