Cancer immunotherapies transition endothelial cells into HEVs that generate TCF1+ T lymphocyte niches through a feed-forward loop

Abstract

The lack of T cell infiltrates is a major obstacle to effective immunotherapy in cancer. Conversely, the formation of tumor-associated tertiary-lymphoid-like structures (TA-TLLSs), which are the local site of humoral and cellular immune responses against cancers, is associated with good prognosis, and they have recently been detected in immune checkpoint blockade (ICB)-responding patients. However, how these lymphoid aggregates develop remains poorly understood. By employing single-cell transcriptomics, endothelial fate mapping, and functional multiplex immune profiling, we demonstrate that antiangiogenic immune-modulating therapies evoke transdifferentiation of postcapillary venules into inflamed high-endothelial venules (HEVs) via lymphotoxin/lymphotoxin beta receptor (LT/LTβR) signaling. In turn, tumor HEVs boost intratumoral lymphocyte influx and foster permissive lymphocyte niches for PD1^- and PD1⁺TCF1⁺ CD8 T cell progenitors that differentiate into GrzB⁺PD1⁺ CD8 T effector cells. Tumor-HEVs require continuous CD8 and NK cell-derived signals revealing that tumor HEV maintenance is actively sculpted by the adaptive immune system through a feed-forward loop.

Keywords: -tT(EX); antiangiogenic immunotherapy; endothelial fate mapping; high-endothelial venule; immune checkpoint blockade; lymphotoxin beta receptor; multiplex immunohistochemistry; pT(EX); single-cell RNA sequencing; t; tertiary lymphoid structure.

Figure 1.. SmartSeq2 sequencing of TU-HEV, LN-HEV and TU-EC

(A) Study design, comparing transcriptomics of TU-HEV, LN-HEV, and TU-EC. (B and C) tSNE plot, colored by cell origins (B) or expression of representative marker genes (C). (D) Expression of TU-EC, TU-HEV, and LN-HEV-specific and common genes. (E) Immunofluorescence of selected marker genes. Scale bars indicate 50 μm. (F) Violin plots of selected HEV and inflammation genes. (G) Gene regulatory network (GRN) predicted by SCENIC, colored by gene expression (round nodes) or regulon activity (square node) in TU-HEVs. Population distribution and median-quantile-min/max without outliers are shown in violin+boxplot (F). Data are from one scRNA-seq experiment. See also Figure S1.

Figure 2.. Characterization of the mouse and human tumor vasculature by droplet-based scRNA-seq

(A-C) UMAP plots, colored by unsupervised EC clustering in mouse PyMT and E0771 tumors. (A), the expression of core HEV markers (B), or in silico-selected HEV cells (C). (D) HEV fraction per EC subtype. (E) Expression of TU-HEV differentially expressed genes (DEGs) and core HEV markers in tumor EC subtypes and TU-HEVs split by different treatments. (F) Quantification of HEV density of UT, DPAg, anti-IFNγ, or DPAg plus anti-IFNγ in PyMT tumors. HEV number was determined by immunofluorescence staining with CD31 and MECA79 antibodies on frozen tissues. N tumors: 9–10 (G) Tumor growth curves of PyMT-bearing mice treated as in F. N tumors: 11–12 (H) Quantification of CD3⁺ T cells 50 μm around HEVs by immunofluorescence staining of DPAg or DPAg + anti-IFNγ-treated PyMT. N fields: 14–32. (I) Quantification of GrzB⁺ cells of PyMT tumors. N fields: 9–10 (J) qPCR gene expression of perforin from total tumor lysate. N samples: 4–6 (K) Expression of selected HEV and inflammation genes in HEVs in DPAg, DPAg + anti-IFNγ treated tumors, and other ECs. (L) UMAP human breast cancer EC dataset, colored by unsupervised EC clustering. (M-O) Validation of selected conserved TU-HEV markers shared between mouse and human datasets by RNAscope. In murine PyMT (N=1) and MC38 (N=1) tumors, ERG and Chst4/Glycam1 positivity identify ECs/HEVs, respectively (M, left panel), and the particle count of each RNA probe is quantified by QuPath (N). In the human breast tumor (N=1), PECAM1 and MECA79 positivity identify ECs/HEVs, respectively (M, right panel). Mean signal intensity of each probe in each PECAM1⁺ tile was measured (O) in QuPath. Scale bars indicate 20 μm. The mean ± SEM is shown (F-J). Median-quantile-min/max without outliers ± population distribution is shown in violin±boxplot (K,N,O). Statistics were assessed by Kruskal-Wallis test (F,I,J), Two-way ANOVA (G), Mann-Whitney test (H), and Wilcoxon test (N,O). Data are pooled from at least two independent experiments (F-J) or one experiment (A-E,L-O). See also Figure S2 and S3.

Figure 3.. Ontogeny of TU-HEVs

(A) Differentiation trajectory of the tumor vasculature predicted by Velocyto/ScVelo, based on tSNE plots by Palantir. Differentiation direction is indicated by arrows in the entire datasets (top) and split by treatment groups (bottom). (B) Study design of confetti tracing experiment. Recombination outcome after tamoxifen induction leads to the expression of either CFP, GFP, YFP or RFP. (C) Representative images of ECs after tamoxifen induction. Blood vessels with single- (left) or multiple-colored ECs (right). Scale bars indicate 50 μm. (D) Representative images of MECA79⁺ HEVs. Scale bars indicate 50 μm. (E) Pie chart showing the fraction of HEV vessels found with 1 to 4 colors. (F) Fraction of mitotic cells in TU-HEVs, LN-HEVs, and TU-ECs. Data are pooled from at least two independent experiments (C-E) or one scRNA-seq experiment (A,F). See also Figure S4.

Figure 4.. TU-HEVs dynamically arise upon immunotherapy and require continuous signals

(A and C) Growth curve of PyMT (A) and E0771 (C) tumors during and after DPAg treatment. N tumors: UT = 53; DPAg = 31; DPAg stop = 32 (A). N tumors: 11 for each cohort (C). 95% confidence interval (CI) of the curve is indicated by the grey line in A. (B and D) Quantification of TU-HEV density in PyMT (B) or E0771 (D) tumors by immunofluorescence tissue staining. N tumors: UT = 10; DPAg = 10; DPAg 8D off = 4; 18D off = 10 (B). N tumors: 11 for each cohort (D). (E-J) Immune cell characterization of PyMT (N = 10 for each cohort) (E, G, and H) or E0771 (N = 6 for each cohort) (F, I, and J) tumors upon DPAg treatment or after treatment cessation (8D off or 18D off) by flow cytometry. CD62L staining was used to discriminate naïve (CD62L⁺) CD8 T cells from the activated (CD62L⁻) CD8 T cells (G and I). (K and L) Immunofluorescence staining and quantification of MC38 tumor sections from Chst4-tdT reporter mice upon anti-CTLA4 + LTbR Ag (CAg) treatment (treatment ON) or after treatment cessation (treatment OFF). (K) Representative images of MECA79/vWF/CD31/Vimentin staining, and quantification of percentage of each subtype (HEV/PCV/EC/Mesenchymal/Other) among total tdTomato⁺ structures during treatment ON or OFF (N=2). Yellow and red arrowheads depict double- and single-labeled cells, respectively. Scale bars indicate 50 μm. (L) Quantification of CD3 density within 50 μm of MECA79⁺ or MECA79⁻tdTomato⁺ vessels. 95% confidence interval (CI) of the curve is indicated by the grey line in A. The mean ± SEM is shown (B,D,E-J,L). Statistics were assessed by Kruskal-Wallis test (B,D,E-J) or Wilcoxon test (L). Only statistical differences UT vs DPAg, DPAg vs 18D off, and DPAg vs 8D off are shown (E-J). Data were pooled from at least two independent experiments (A-J) or one experiment (K,L). See also Figure S4.

Figure 5.. CD8 T-cells and NK cells induce immunotherapy-dependent HEV formation via the LT/LTβR axis

(A-C) TU-HEV density (A, N = 7–15), tumor growth curve (B, N = 6–15), and apoptotic index (C, N=12–21) of PyMT-bearing mice treated with DPAg and depleted of CD8, CD4, NK cells or macrophages by administration of anti-CD8, anti-CD4, anti-NK1.1, anti-CSF1R, respectively. (D) Violin plot of Lta (LTα), Ltb (LTβ) and Tnfsf14 (LIGHT) expression in each immune cell cluster from the E0771 and PyMT dataset. (E) TF activities in each EC subtype predicted by SCENIC. (F) GSEA plots of Nfkb2 and Relb-regulated target genes in HEVs versus non-HEV ECs. (G) Waterfall plot, comparing regulon activities of TU-HEVs and non-HEV ECs in PCVs predicted by SCENIC. (H) Experimental design of HEV evaluation in PyMT LTβR^ECKO mice. (I-K) Quantification of LTβR⁺CD31⁺ (I), MECA79⁺CD31⁺ cells (J) by flow cytometry, and quantification of MECA79⁺CD31⁺ tumor vessels by immunofluorescence staining (K) of PyMT-OVA-bearing Cdh5-Cre^ER×LTβR^fl/fl (CRE⁺) or control (CRE⁻) mice treated with DC101 + aPD-L1 + aCTLA-4 (DPC). N tumors: CRE⁻ = 8; CRE⁺ = 16 (I and J). N tumors: CRE⁻ = 8; CRE⁺ = 15 (K). The mean ± SEM is shown (A-C,I-K). Population distribution and median-quantile-min/max without outliers are shown in violin+boxplot (D). Statistics were assessed by Krustal-Wallis test (A,C), Two-way ANOVA (B), and Mann-Whitney test (I-K). Only statistical differences UT vs DPAg IgG, and DPAg IgG vs DPAg immune cell-depleted tumors are shown in A-C. Data are pooled from two independent experiments (A-C,I-K) or one scRNA-seq experiment (D-G). See also Figure S5 and S6.

Figure 6.. TU-HEVs generate lymphocyte niches permissive for PD1^neg and PD1⁺ CD8 progenitor cells

(A and B) Flow cytometry quantification of PD1⁺ CD8 T cells (A) and CD8 T cells co-expressing IFNγ-GrzB, IFNγ-TNFα or IFNγ-IL2 (B) in UT (N = 10) and DPAg (N = 10) treated PyMT tumors. (C) Gating strategy for intratumoral PD1⁺TCF1⁻ TIM3⁻ cells, PD1⁺TIM3⁻TCF1⁺ pT_EX cells, and PD1⁺TCF1⁻TIM3⁺ tT_EX cells. (D and H) Flow cytometry quantification of PD1⁺TCF1⁻TIM3⁻ cells, pT_EX cells, tT_EX cells (D), and TCF1⁺PD1⁻ cells (H) in IgG (N = 9) and DPAg (N = 5) PyMT tumors. (E,F,I,J) Flow cytometry quantification of GrzB⁺ (E,I) and Ki67⁺ (F,J) PD1⁺TCF1⁻TIM3⁻, pT_EX and tT_EX cells (E,F) or TCF1⁺PD1⁻CD8 T cells (I,J) from IgG (N = 9) and DPAg (N = 5) treated PyMT tumors. (G) UMAP of CD8 T cell subsets in the PyMT dataset (left) and selected genes (right). (K) Differentiation trajectory inference by Velocyto/ScVelo. (L) Representative images of PyMT tumors from DPAg-treated mice. The scale bar indicates 50 μm. White head-arrows depict pT_EX cells; yellow head-arrows depict TCF1⁺PD1⁻ cells. Scale bar in the selections indicates 10 μm. (M-O) Percentage of pT_EX (TV N=17; HEV N=13) (M), TCF1⁺PD1⁻CD8 T cells (TV N=17; HEV N=13) (N), and tT_EX (TV N=27; HEV N=21) (O) among all CD8 T cells within 50 μm of HEVs or TVs in tumor sections of DPAg-treated PyMT-bearing mice. The mean ± SEM is shown (A,B,D-F,H-J,M-O). Mann-Whitney test (A,B,D-F,H-J,M-O). Data are pooled from at least two independent experiments (A,B,D-F,H-J,M-O) or one experiment (G,K). G and K derived from the UT and DPAg-treated PyMT tumors. See also Figure S7.

Figure 7.. Immune and vascular landscape by MILAN

(A) The relative proportion of different CD8 T cells within mouse MC38 tumors (left) and human breast cancer (right). (B) GrzB expression and Ki67⁺ percentage of the CD8 T cell subsets within mouse MC38 tumors (left) and human breast cancer (right). (C) Digital reconstruction of a representative human breast cancer section by MILAN. The selected area depicts HEV high area (with >25% HEV-ECs) or HEV low area (with <25% HEV-ECs) at tumor edge / bulk. Scale bar indicates 2 mm. Scale bar in the selections indicates 50 μm. (D) CD8 T-cells numbers in distinct human breast cancer locations. (E) Density plot depicting the distance between each indicated CD8 T cell subset to the closest HEV or blood vessels (BV). (F) Illustration of CD8 T cell subsets in the TU-HEV niche. (G and H) GSVA score of the human HEV signature in melanoma patients (CA209-038) from samples collected before (Pre) or during (On) immunotherapy (G), and NSCLC patients before aPDL-1 therapy (Atezolizumab) from the Poplar study (H). PD=progression disease; SD=stable disease; PR=partial responder; CR=complete responder. Median-quantile-min/max without outliers ± population distribution is shown in violin±boxplot (A,B,D,G,H). Mouse data are derived from four MC38 tumors treated with anti-CTLA-4 and LTβR Agonist (both from Oncurious). Human data are derived from five-six human untreated breast cancers. Wilcoxon test with the adjusted p-value using Holm correction. See also Figure S7.