An intra-tumoral niche maintains and differentiates stem-like CD8 T cells

Abstract

Tumour-infiltrating lymphocytes are associated with a survival benefit in several tumour types and with the response to immunotherapy^1-8. However, the reason some tumours have high CD8 T cell infiltration while others do not remains unclear. Here we investigate the requirements for maintaining a CD8 T cell response against human cancer. We find that CD8 T cells within tumours consist of distinct populations of terminally differentiated and stem-like cells. On proliferation, stem-like CD8 T cells give rise to more terminally differentiated, effector-molecule-expressing daughter cells. For many T cells to infiltrate the tumour, it is critical that this effector differentiation process occur. In addition, we show that these stem-like T cells reside in dense antigen-presenting-cell niches within the tumour, and that tumours that fail to form these structures are not extensively infiltrated by T cells. Patients with progressive disease lack these immune niches, suggesting that niche breakdown may be a key mechanism of immune escape.

Extended Data Fig. 1. Description of statistics and sub-group analyses of progression-free survival.

a, Descriptive statistics. Table details the demographic, disease stage, disease characteristic and immune infiltrate breakdown of the cohort of patients with kidney cancer. b, Martingale residual plot illustrating discovery of 2.2% CD8 ‘optimal cut’. c, Comparison of optimal cut, sub-optimal cut and median cut. d, CD8 T cell infiltration predicts time to progression in stage III (T3N0M0) patients. Patients were stratified into high (>2.2% CD8) or low (<2.2% CD8) based on the optimal cut identified in a cohort of all-stage patients. CD8^hi, n = 13; CD8^lo, n = 7. P = 0.0059, HR = 6.480, as determined using log-rank test. e, CD8 T cell infiltration significantly improves prognostication in patients with kidney cancer with high SSIGN (size, stage, grade, necrosis) scores. P ≤ 0.0001, as determined using log-rank test. Patients were were stratified into low (scores 1–6) and high (scores >6) SSIGN score groups and into low (<2.2% CD8) and high (>2.2% CD8) T cell infiltration. SSIGN^loCD8^lo: n = 11, SSIGN^loCD8^hi: n = 16, SSIGN^hiCD8^lo, n = 28, SSIGN^hiCD8^hi, n = 13.

Extended Data Fig. 2. CD8 T cell infiltration is associated with improved survival and is independent of standard risk assessment tools, tumour features and patient demographics.

a, b, Proportion of CD8 T cells in the tumours of patients that progress or die after surgery as compared to those without disease progression (a) or death (b). c, Disease stage, P = 0.6. d, Fuhrman nuclear grade, P = 0.4. e, UISS groups, P = 0.3. f, SSIGN groups, P = 0.3. g, Maximum tumour size in one dimension, in centimetres, R = 0.01, P = 0.3. h, Histologic subtype, P = 0.7. i, Patient age at the time of surgery, in years, R = 0.001, P = 0.9. j, Patient sex, P = 0.8. k, Patient race/ethnicity, P = 0.7. Median value is shown for a–f, h and j–k.

Extended Data Fig. 3. Flow cytometric comparison and in vitro functional studies of stem-like and terminally differentiated CD8 T cells.

a, Flow cytometry gating scheme. FSC-A and FSC-H are used to select for singlets. Live (APC–Cy7 negative) CD3⁺ events are then selected from this population of singlets. Lymphocytes are selected from this live CD3⁺ population on the basis of FSC-A and SSC-A, and CD4⁺ and CD8⁺ T cell populations are selected from the lymphocyte population. b, Expression of various molecules by stem-like (green) and terminally differentiated (red) CD8 T cells in human tumours measured by flow cytometry. c–e, Expression of TCF1 (c), CD28 (d) and TIM3 (e) as measured by flow cytometry, by stem-like and terminally differentiated CD8 T cells isolated from patients with kidney cancer (n = 6) and cultured in vitro for 3 days with 10 U of IL-2 and with (stimulated) or without (unstimulated) anti-CD3/CD28/CD2 bead stimulation at a 1:1 ratio. Median value is shown. f, Number of live stem-like and terminally differentiated intra-tumoral CD8⁺ T cells after 3 days of in vitro culture in IL-2 supplemented media. Live/dead staining was used to determine the proportion and number of live CD8 T cells by flow cytometry. g, Composition of the CD8 T cell compartment. In 60 human kidney cancer patients, proportion of CD8 T cells that are stem-like cells (PD-1⁺CD28⁺TIM3⁻) correlates with total T cell infiltration (%CD8 T cells of total cells), while proportion of terminally differentiated cells (PD-1⁺TIM3⁺) does not. h, Percentage of total CD8 T cells correlates with the percentage of total cells that are stem-like CD8 T cells.

Extended Data Fig. 4. TCR sequencing analysis for stem-like and terminally differentiated CD8 T cells.

a, Gating scheme for fluorescence activated cell sorting of cell populations for stem-like and terminally differentiated cell populations from human kidney tumours. Terminally differentiated cells are PD-1-high and CD39⁺. Stem-like cells are PD-1⁺CD39⁻CD28⁺. b, Estimation of population overlap. PD-1 and CD39 expression by flow cytometry was modelled using a two-population Gaussian mixing model. The amount of each population falling within each sorting gate based on the relative proportions of the populations was determined and used to calculate whether TCRs found in both populations could be accounted for by contamination. c, Pre-sort flow cytometry plots for patients sorted for TCR sequencing. d, Ranking of stem-like (green) and terminally differentiated (red) TCR clones from most to 10th most dominant clone by percent of total TCR repertoire (log₁₀). e, Number of unique TCR clones detected in stem-like (green) and terminally differentiated (red) cell populations as a function of number of cells collected. f, Percentage of overlap detected as a function of number of cells collected. g, Tumour samples were taken from two physically distant sites within the same tumour and stem-like and terminally differentiated cells were sorted from each and TCR sequenced. Venn diagrams illustrate unique TCRs found between stem-like populations in sites A and B, between terminally differentiated populations in sites A and B, and between location mismatched stem-like and terminally differentiated populations (for example, stem-like-A/terminally differentiated-B, stem-like-B/terminally differentiated-A), in addition to overlap between stem-like and terminally differentiated T cell populations within a single site. h, Table indicating the number of stem-like and terminally differentiated T cells collected, inferred purity of each population, percent overlap detected calculated by the number of TCRs detected in either sample divided by the total TCRs in both samples, and the power to detect >20% overlap (assuming 2,000 unique TCRs per sample) for each patient sample.

Extended Data Fig. 5. Transcriptional and epigenetic analysis of T cell subsets in tumours.

a, Comparison of differentially expressed genes between human cancer and viral specific CD8 T cell subsets. RNA-seq from cancer subsets compared to RNA-seq data collected from yellow fever (YF) antigen specific CD8 T cells (GSE100745) during effector (14 days post-vaccination) and memory (4+ years post-vaccination) time points. The number of differentially expressed genes (DEG) versus naive CD8 T cells was determined using DESeq2. Venn diagrams show number of DEG shared or unique between viral and cancer subsets. Although the cancer subsets of T cells share many genes with the YF specific cells, there are also many distinct genes only expressed in cancer T cell subsets. b, DEGs were clustered using cluster affinity search technique (CAST). Clusters with greater than 5% of total genes are shown. Heat map shows z-score of averages from each group. c, Principal component analysis of T cell subsets form cancer and viral-specific CD8 T cells, performed on genes that were differentially expressed in any group versus naive cells. d, Comparison of cancer subsets to transient effector programs found in YF specific T cells. Previously we have identified transient gene expression signatures that are expressed in YF-specific effector cells, but return to a naive state after antigen is cleared. These genes not expressed in memory or naive cells are highly expressed in both cancer subsets suggesting a similarity to an effector cell. e, Pairwise comparison of transient effector program genes between effector and cancer subsets shows the relationship of this subset of genes re-initiated program (blue) and the transient effector program (red) compared between YF and cancer subsets. Dotted 45-degree line represents equal fold change versus a naive CD8 T cell in cancer and yellow fever cells. f, GSEA and network analysis of pathways associated with differentiation. Gene set enrichment performed with GSEA and visualized with Cytoscape. The most significant networks are shown. Red indicates enrichment of nodes in terminally differentiated T cells, while blue shows enrichment in stem-like T cells. g, Histogram shows the distribution of the continuous region size of DMRs. h, Histograms show the relative frequency of DMRs within 10kb of transcription start sites. i, Global changes in methylation. Violin plots show the distribution of total methylation within identified DMRs in naïve, stem-like, and terminally differentiated cells. j, DMR patterns of differentiation. DMRs identified in Fig. 2d were clustered using CAST. Box plots show the interquartile range and mean of DMRs in each cluster by cell type k, Histograms show the total methylation from 0–100% in regions near important genes. Dot plots show the methylation of each CpG motif within highlighted regions of interest. l, Transcriptionally active transcription factors have over-represented binding in epigenetically modified regions of chromatin. Plots show the enrichment of transcription factor binding sites within differentially methylated regions in each cell type on the x-axis, and the y-axis shows the enrichment of transcription factor binding sites within the promoters of differentially expressed genes. Colour of dots represents the relative expression in stem-like (green) or terminally differentiated (red) cells, and the size of the dot is proportional to total expression of the transcription factor.

Extended Data Fig. 6. Quantitative immunofluorescence analysis of tumour immune infiltration.

a, Flow cytometry data illustrating the number of naive cells present intra-tumorally. Left, representative patient. Right, summary data. b, Comparative amounts of CD45RO expression on naive and stem-like intra-tumoral CD8 T cells. c, Workflow for immunofluorescence imaging analysis and immuno-map creation. Single channel immunofluorescence images are imported into CellProfiler. CD8⁺ and MHC-II⁺ objects are identified in the respective channel images. The XY location of each CD8⁺ and MHC-II⁺ object is exported. The TCF1 staining intensity is measured inside the CD8⁺ objects. These parameters are used to calculate MHC-II⁺ density, measure the distance from each CD8⁺ object to its nearest MHC-II⁺ neighbour, and to finally create immuno-maps for immunofluorescence images. d, Histo-cytometric analysis of tumour infiltrating immune populations. Location and fluorescence intensity of CD8⁺ and MHC-II⁺ cells were determined using CellProfiler. After image compensation, CD8⁺ and MHC-II⁺ cells were gated. TCF1 intensity of each cell is shown on histograms for each population below. Comparison of flow cytometry data from the same patient sample is also shown. e, Patients with kidney cancer with high CD8 infiltration determined by flow cytometry. Patients that were determined to have high CD8 infiltration by flow cytometry were selected for analysis by immunofluorescence. f, Haematoxylin and eosin stains of human kidney tumour. Selected slides from human kidney tumour shown in part e, to be highly infiltrated by T cells. Regions of tumour tissue are highlighted in yellow. g, Immunofluorescence imaging of kidney tumour. Selected tumours shown to be highly infiltrated by T cells. Tumour section was stained for MHC-II to identify antigen-presenting cells, and CD8 and TCF1 to identify stem-like and terminally differentiated CD8 T cell populations. Insets shows zoomed regions highlighted in the larger image. h, Dendritic cells populations, stem-like, and terminally differentiated CD8 T cells in three representative kidney cancer patients. i, Cellular spatial relationship map (middle) analysis and construction conducted as in Fig. 3e. j, CD8 expression of TCF1 preferentially occurs in dense APC zones. Amount of TCF1 expressed in each CD8 T cell graphed against the density of MHC-II around each T cell (MHC-II⁺ cells per 10,000 μm²). k, l, TCF1⁺ CD8 T cells are localized near dense MHC-II regions in other cancers. Prostate and bladder tumours were imaged for CD8, MHCII and TCF1. Regions of dense MHC-II aggregates are shown in grey and the location of TCF1⁺ CD8 T cells in green (l).

Extended Data Fig. 7. Comparison of tertiary lymphoid structures and antigen-presenting niches in kidney tumours.

a, Haematoxylin and eosin slides highlighting tertiary lymphoid structures (TLS) in kidney tumours with high (top) and low (bottom) CD8 T cell infiltration. Yellow boxes highlight areas shown in zoomed insets. b, Haematoxylin and eosin slide showing dense immune infiltration in a tumour with high CD8 T cell infiltration but lacking presence of TLS. Yellow boxes highlight areas shown in zoomed insets. c, Immunofluorescence staining illustrating organizational structure of human tonsil. CD8 staining is shown in red, MHC-II in green, TCF1 in yellow and DAPI (nuclei) in blue. White box highlights zoomed area shown in inset. Follicle and extrafollicular space shown as labelled. T cell zone shown in rightmost panel. d, Immunofluorescence staining illustrating tumour TLS. CD8 staining is shown in red, MHC-II staining in green and DAPI staining of nuclei in blue. White box highlights zoomed area shown in inset. Follicle and extrafollicular space shown as labelled. e, Immunofluorescence staining illustrating dense immune infiltration in TLS negative kidney tumour. CD8 staining is shown in red, MHC-II in green, TCF1 in yellow and DAPI in blue. White box highlights zoomed area shown in inset. Follicle and extrafollicular space shown as labelled. f, There is no significant difference in CD8 T cell infiltration between kidney tumours with and without TLS. CD8 T cell infiltration measured by flow cytometry and shown as percentage of CD8⁺ of total cells. Statistical analysis resultant from Mann–Whitney test is shown. g, Lack of correlation between proportion of CD8 T cells and CD19⁺ B cells in tumours. Linear regression results P = 0.6006 with R² = 0.02167. h, B cell infiltration between tumours with high or low CD8 T cell infiltration was not significantly different. B cell infiltration is shown as the percentage of CD19⁺ B cells of total cells. Statistical analysis resultant from Mann–Whitney test is shown. Median value shown in f and h.

Extended Data Fig. 8. Highly infiltrated kidney tumours are well vascularized and contain lymphatic vessels.

a, Immunofluorescence staining of human tonsil and highly T cell infiltrated human kidney tumours showing tissue vascularization. Formalin-fixed paraffin embedded tissue was stained for CD8 (T cells), MHC-II (antigen-presenting cells), CD31 (endothelial cells) and DAPI (nuclei). b, c, Immunofluorescence staining of human tonsil and highly T cell infiltrated kidney tumours showing presence of lymphatics via Lyve1 (b) and Podoplanin/D2–40 (c). Formalin-fixed paraffin embedded tissue was stained for CD3 (T cells), MHC-II (antigen presenting cells), Lyve 1 or Podoplanin/D2–40 (lymphatics) and DAPI (nuclei). d, Flow cytometry analysis shows tumour vascularization in highly (red) and poorly (grey) infiltrated kidney tumour. Tumours were stained using antibodies listed in Supplementary Table 2, collected on a Becton Dickinson LSR-II, and analysed using FlowJo. e, Histogram of flow cytometry analysis showing increased CD31 staining in highly T cell-infiltrated kidney tumours (red) as compared to poorly infiltrated tumours (grey). Analysis completed as described in d. f, Summary data of flow cytometry analysis showing differences in vascularization between highly (red) and poorly (grey) T cell infiltrated kidney tumours and prostate tumours (black). Analysis completed as described in d. g, h, Tumour-infiltrating T cells are PD-1⁺. Flow cytometry analysis showing T cells infiltrating kidney tumours are PD-1⁺, suggesting the cells are not naive and present due to blood contamination and showing that the MFI of PD-1 on tumour-infiltrating T cells is not significantly different between highly (red) and poorly (grey) infiltrated tumours. i, Representative flow cytometry plots showing PD-1 and TIM3 expression on tumour infiltrated T cells in highly (red) and poorly (grey) infiltrated tumours. Populations shown are gated on live, CD3⁺CD8⁺ cells. Median value shown in f–h.

Extended Data Fig. 9. Descriptive statistics and quantitative immunofluorescence analyses of human kidney tumours.

a, Descriptive table enumerating patient characteristics of patients with kidney cancer, with and without progressive disease. b, Comparison of the number of CD8⁺ cells per 300 μm × 300 μm field in patients with and without progressive disease. The number of CD8⁺ cells per 300 μm × 300 μm field were enumerated using the methods outlined in Extended Data Fig. 6. c, The correlation between enumeration of CD8 T cells by flow cytometry and by immunofluorescence. On the x axis, CD8 T cells are measured as a proportion of total cells. On the y axis, CD8 T cells are measured as a proportion of total DAPI objects detected in the tumour section. d, Estimated number of 20× fields of view necessary to obtain an accurate assessment of level of CD8 T cell infiltration is 171 fields of view. Increasing number of random fields of view were sampled from images and the percent of cells that were CD8 positive by IF correlated to FACS from the corresponding sample. e, Histological comparison of patients with kidney cancer shown in Fig. 4 — a patient with kidney cancer with dense T cell infiltration and no disease progression (red, left) and a patient with kidney cancer with poor T cell infiltration and progressive disease (grey, right). f, Comparison of the number of MHC-II⁺ cells per 300 μm × 300 μm field in stage III (T3N0M0) patients with and without progressive disease. The number of MHC-II+ cells per 300 μm × 300 μm field were enumerated using the methods outlined in Extended Data Fig. 6. g, Comparison of the proportion of tumour area with greater than 5 MHC-II⁺ cells per 10,000 μm² between stage III (T3N0M0) patients with and without progressive disease. Statistical analysis resultant from Mann–Whitney test is shown. h, No significant difference in number of fields of view sampled between patients with and without progressive disease was detected. i, Density of MHC-II⁺ APCs and CD8 T cells in densely (left) or poorly (right) infiltrated kidney tumours. x-axis shows the number of CD8⁺ cells per 10,000 μm². y-axis shows the number of MHC-II⁺ cells per 10,000 μm². Regions of predominantly MHC-II+ cells are highlighted in yellow, regions of predominantly CD8⁺ cells in red, and regions of shared MHC-II⁺ cells and CD8⁺ cells in green.

Extended Data Fig. 10. Comparison of densely and poorly infiltrated kidney tumours by PDL-1 staining and by quantitative immunofluorescence.

a, Representative patients with densely infiltrated and poorly infiltrated kidney tumours whose disease has not progressed or has progressed, respectively. Whole-slide scans are shown for haematoxylin and eosin, anti-PD-L1, and immunofluorescence (CD8, MHC-II, DAPI) stains, with zoomed insets of immunofluorescence data. Yellow circles highlight the location of tumour tissue on the haematoxylin and eosin slide. Yellow boxes highlight the areas shown in the zoomed insets of immunofluorescence images. Immunofluorescence data are quantitatively analysed and mapped to show the density of MHC-II⁺ cells and the XY location of CD8⁺ T cells in the rightmost panel. Anti-PD-L1 scans are marked as ++ (positive-high), + (positive-low), or − (negative), as scored by board-certified pathologists. b, Patients in a are highlighted in red (highly infiltrated, non-progressors) and grey (poorly infiltrated, progressors) to show the percentage of CD8 T cell infiltration by flow cytometry. c, PD-L1 staining was scored by board-certified pathologists as positive-high, positive-low and negative. There is no significant difference between the percent CD8 T cell infiltration amongst these categories by ANOVA with Holm–Sidak correction. Median value shown. d, Progression free survival for patients with positive-high (PD-L1 high), positive-low (PD-L1 low), and negative (PD-L1 negative) kidney tumours. There was no significant difference in progression-free survival between the groups by Mantel–Cox log rank test (P = 0.6106) or by log rank test for trend (P = 0.3374).

Fig. 1. The anti-tumour T cell response is supported by a stem-like CD8 T cell, which gives rise to terminally differentiated CD8 T cells in the tumour.

a, Proportion of CD8 T cells in kidney tumours shown as percent of total cells (n = 68). b, Disease progression after surgery in patients with kidney cancer stratified into high or low CD8 T cell infiltration (±2.2%) based on optimal cut methods. Time to progression is the number of days from surgery until death or progression by RECIST criteria (n = 66). c, Gating strategy to identify intra-tumoral CD8 T cell populations. Populations shown are gated on live, CD3⁺ and CD8⁺. d, Expression (mean fluorescence intensity (MFI)) of activation markers, checkpoint molecules and transcription factors by TIM3⁺ and TIM3⁻ CD28⁺ subsets, gated as in c. e, f, Stem-like (TIM3⁻CD28⁺) and terminally differentiated (TIM3⁺) populations were sorted from kidney tumours, labelled with CellTrace violet, and cultured with anti-CD3/anti-CD28 beads and 10 U ml⁻¹ of IL-2 for 4–5 days. Proliferation index and percentage of cells divided is shown. g, h, Expression of TIM3, PD-1 and CD244 after cells undergo proliferation. Summary plots from in vitro activation experiments compared to fold change in MFI observed between the populations in vivo. i, TCR repertoires of stem-like and terminally differentiated T cells sorted as shown in Extended Data Fig. 4. TCR clones are represented by the number of reads detected in either T cell population. j, TCR repertoire overlap between stem-like and terminally differentiated T cells. The proportion of the detected TCR repertoire in each patient that is unique to each population or shared between the two is shown. k, l, Generation of checkpoint-high cells correlates with total T cell infiltration. Patients were classified as having a low (<20%) or high (>20%) fraction of TIM3⁺ terminally differentiated cells. Data show sample patients (k), and summary data in kidney (n = 49), prostate (n = 28) and bladder tumours (n = 8) (l).

Fig. 2. Stem-cell differentiation to the terminally differentiated state is associated with transcriptional and epigenetic changes.

a, Heat map of transcription factors, proliferation-related genes, checkpoint molecules, cytotoxic molecules, co-stimulatory molecules, survival genes and migration and adhesion genes. Figure shows the z-scored data. TD, terminally differentiated. b, GSEA comparison to mouse CXCR5⁺ and TIM3⁺ subsets of CD8 T cells. Gene sets were created from CXCR5 stem-like and TIM3⁺ exhausted CD8 subsets from LCMV infection. Plots show enrichment score (ES) against genes upregulated (red) and downregulated (green) in mice. c, Summary of the number of epigenetic changes occurring as CD8 T cells undergo differentiation. Illustration shows the number of DNA methylation changes occurring as cells differentiate. d, Green regions show methylated and demethylated regions as cells transition from naive to stem-like cells, and red shows these events as cells transition from stem-like to terminally differentiated. e, Specific epigenetic changes near important differentially expressed genes. Histograms show the total methylation from 0–100% in regions near important genes. Highlights show significantly differentially methylated regions. Dot plots show the methylation of each CpG motif within this highlighted domain.

Fig. 3. APCs form a supportive, intra-tumoral niche for TCF1⁺ stem-like CD8 T cells.

a, Identification of APC subsets in kidney (red, n = 53), bladder (green, n = 7) and prostate tumours (blue, n = 33). b, Correlation between CD8 T cells and APC populations. Percentage of total cells in the tumour that were CD8⁺ T cells and dendritic cells (CD11c⁺MHC-II⁺) or macrophages (CD68⁺CD11b⁺) in patients from a. Spearman correlation coefficient is shown. c, Immunofluorescence for MHC-II staining identifies APCs, whereas CD8 and TCF1 identify stem-like and terminally differentiated CD8 T cell populations in a representative patient with kidney cancer. Insets show regions highlighted in the larger image. Blue arrows denote examples of TCF1⁺ CD8 T cells. d, Cellular spatial relationship map. After acquiring XY coordinates of MHC-II⁺ cells, MHC-II cellular density was calculated (number of MHC-II⁺ cells per 10,000 μm²). XY location of CD8 T cells are overlaid with MHC-II density contour. CD8 cells were designated TCF1 positive or negative using histo-cytometry (Extended Data Fig. 6). e, MHC-II cellular density surrounding TCF1⁺ or TCF1⁻ subsets. MHC-II density at the corresponding XY coordinates of each CD8 T cell is shown. f, Distance between CD8 T cells and the closest MHC-II⁺ cell. g, Numerous regions of high MHC-II density correlates within increased number of TCF1⁺ cells in multiple tumour types. y axis shows proportion of the tumour with MHC-II density >5 MHC-II⁺ cells per 10,000 μm², with average number of TCF1⁺ CD8 T cells in the tumour on the x axis.

Fig. 4. Loss of APC niche is associated with impaired CD8 T cell response and disease progression.

a–d, Patients with dense T cell infiltration and no disease progression (red, left) and one with poor T cell infiltration and progressive disease (grey, right). a, Haematoxylin and eosin whole-slide images. Tumour is outlined in yellow. b, Whole-slide immunofluorescence images. MHC-II (yellow), TCF1 (green), CD8 (red) and DAPI (blue). c, Immunomap of APC density in tumours from b constructed as in Fig. 3 and Extended Data Fig. 7. d, Insets show highlighted regions from b and c, illustrating regions of high MHC-II⁺ density and stem-like T cell infiltration in kidney tumours. e, Comparison of the number of MHC-II⁺ cells per 300 μm × 300 μm field in patients with (n = 13) and without (n = 13) progressive disease. f, Comparison of the proportion of tumour area with >5 MHC-II⁺ cells per 10,000 μm² between patients with and without progressive disease. Mann–Whitney test result is shown. g, Patients with high MHC-II⁺ cell density had improved progression free survival. log-rank statistical analysis yields P = 0.04 and HR = 3.226. h, Immunomaps illustrating regions of MHC-II⁺ cell density (yellow), CD8⁺ cell density (red) or shared density (green) in tumours from a–d. i, Patients without progressive disease have more areas where the density of MHC-II⁺ cells (top), CD8⁺ cells (middle) or MHC-II⁺ and CD8⁺ cells (bottom) exceeds 5 cells per 10,000 μm².