Graphical abstract

   graphic file with name fx1.jpg
   [48]Open in a new tab

Highlights

     * •
       TIL infusion product is heterogeneous and comprises CD4, CD8, and
       (DN) T cells
     * •
       CD3^+CD8^−CD4^−(DN) TILs restrict CD8^+ TIL expansion, correlating
       with worse TIL-ACT
     * •
       CD8^+ TIL expansion is restricted by (DN) TILs via Fas-FasL, TGF-β,
       and IL-10 signaling
     * •
       Nine TME gene signatures and 14 intracellular T cell genes hold
       prognostic value of TIL-ACT
     __________________________________________________________________

   Liu et al. establish a transcriptomic and translational approach across
   outcomes and characteristics of tumor-infiltrating T cell (TIL)
   products and baseline tumor microenvironment from high-risk
   nasopharyngeal carcinoma patients receiving TIL-based immunotherapy.
   They identify the role of CD3^+CD4^−CD8^− double-negative TILs and
   other biomarkers in the prediction of TIL immunotherapy outcome.

Introduction

   Adoptive cell therapy (ACT) has become a well-established treatment
   option for cancer patients.[49]^1 ACT using ex vivo expanded
   tumor-infiltrating lymphocytes (TILs), which has advantages that derive
   from native genetically unmodified cells and comprises polyclonal cells
   capable of targeting multiple tumor antigens, is becoming more and more
   widespread in the treatment of solid tumors.[50]^2 Despite the exciting
   clinical achievement of autologous TIL-ACT in solid tumors, such as
   melanoma[51]^3 and cervical,[52]^4 colorectal,[53]^5 and breast
   cancer,[54]^6 in recent years, only a minority of patients attain
   long-term benefits with TIL-ACT. Identifying patients who are most
   likely to benefit from TIL-ACT remains a key challenge.

   The clinical success of immunotherapy including ACT and immune
   checkpoint blockade (ICB) is influenced by the features of the tumor
   immune context, such as the presence of activated or exhausted T cells;
   infiltration with immune regulatory myeloid cells or other categories
   of immune cells; and gene expression programs that may influence the
   recruitment, expansion, and activity of T cells.[55]^7 In a
   high-dimensional analysis of ACT products in patients with metastatic
   melanoma, TIL-ACT responders were reported to retain a pool of
   stem-like neoantigen-specific TILs that are absent in ACT
   non-responders.[56]^8 However, the clinically validated biomarkers
   associated with disease progression in TIL-ACT are still unclear due to
   the heterogeneity of tumor microenvironments (TMEs).

   Nasopharyngeal carcinoma (NPC) is strongly associated with Epstein-Barr
   virus (EBV) infection, and the highest rates occur in South China,
   Southeastern Asia, and North Africa.[57]^9 A large number of
   lymphocytes home into NPC tissues for the novel viral antigens
   presented by NPC cells, including EBNA1, LMP1, and LMP2. Thus, NPC
   represents a good model for cancer immunotherapy.[58]^10 We previously
   identified the safety and objective clinical response, along with
   increased EBV-specific immunity, in vivo after autologous TIL infusion
   in a phase 1 study in locoregionally advanced NPC patients.[59]^11
   However, the patients who received autologous TIL treatment following
   concurrent chemoradiotherapy (CCRT) did not obtain a significant
   survival benefit in a phase 2 clinical trial in advanced NPC patients
   with plasma EBV (pEBV) load ≥4,000 copies (n = 156,
   [60]NCT02421640).[61]^12 To address the issue of who will likely
   benefit from TIL-ACT among these EBV DNA-selected high-risk advanced
   NPC patients, we conducted a retrospective analysis of a cohort from
   this phase 2 clinical study ([62]NCT02421640) to screen the variants
   and biomarkers that predict clinical outcome after treatment with TIL
   infusion products. We also investigated the underlying molecular
   mechanisms that influence clinical outcomes.

Results

T cell diversity and outcome relevance of TIL infusion products

   To evaluate the associations of TME features and TIL infusion product
   compositions with the clinical outcomes of TIL-ACT, as well as the
   underlying molecular mechanisms influencing TIL-ACT outcomes, we
   utilized tumor tissue biospecimens and TIL infusion products procured
   from NPC patients who received adjuvant autologous TIL therapy after
   CCRT in a phase 2 clinical trial study ([63]NCT02421640; [64]Table S1).
   The methodology is outlined in [65]Figure 1A. Sixty-two tumor samples
   from patients treated with TIL plus CCRT (n = 22) or CCRT alone (n =
   40) were analyzed by bulk RNA sequencing (RNA-seq), and 47 TIL infusion
   products from patients treated with TIL plus CCRT were analyzed by flow
   cytometry. In addition, 26 of 47 TIL infusion products were evaluated
   by single-cell (sc)RNA-seq. After quality filtering, we obtained 65,978
   high-quality T cells and identified 14 T cell clusters with distinct G1
   phase, S phase, and G2M phase ([66]Figures 1B, [67]S1A, and S1B). As
   expected, CD3 expression was distributed across all the 14 clusters
   ([68]Figure S1C). Based on the expression of canonical marker genes, 11
   clusters were assigned to CD3^+CD8^+ T cells (89.8% of total cells),
   including C0_activated_CD8_T_cells, C1_cytotoxic_CD8_T_cells,
   C2_proliferating_CD8_T_cells-PCNA, C3_proliferating_CD8_T_cells-LTB,
   C4_CD8_MHC_II_T_cells, C5_proliferating_CD8_T_cells-TNFRSF8,
   C6_proliferating_CD8_T_cells-TOP2A, C9_effector_memory_CD8_T_cells,
   C10_proliferating_CD8_T_cells-LGALS1,
   C12_tissue_resident_memory_CD8_T_cells, and C13_stem_like_CD8_T_cells,
   and the CD3^+CD4^+ T cells (6.3% of total cells) were separated into
   two clusters, including C7_naive_CD4_T_cells and C11_Th2_CD4_T_cells
   ([69]Figures 1C and [70]S1D). In addition, the CD3^+CD4^−CD8^−
   double-negative (DN) T cells (3.9% of total cells) were a separate
   cluster: C8_DN_T_cells, referred to as (DN) TILs ([71]Figure 1D).
   Similar percentages of the CD3^+CD8^+, CD3^+CD4^+, and CD3^+CD4^−CD8^−
   T cell subsets were validated by flow cytometry in 47 TIL samples
   ([72]Figure 1E). The (DN) TIL frequency in these 14 T cell clusters was
   significantly lower in patients with non-progression than in those with
   progression ([73]Figures 1F and 1G). A lower frequency of (DN) TILs
   correlated with favorable progression-free survival (PFS) and overall
   survival (OS) in patients undergoing TIL-ACT ([74]Figures 1H, [75]S1E,
   and S1F).

Figure 1.

   [76]Figure 1
   [77]Open in a new tab

   Study design and single-cell transcriptomic landscape of TIL infusion
   products

   (A) Graphical overview of the experimental design and bioinformatics
   workflow. In our clinical trial cohort, bulk RNA sequencing was
   performed on tumor tissue from NPC patients before CCRT+TIL (n = 22) or
   CCRT alone (n = 40), and TIL infusion products were isolated from
   pretreatment tumor tissue from these patients, expanded ex vivo, and
   subsequently prepared for single-cell (sc)RNA sequencing (n = 26) and
   flow cytometry (n = 47).

   (B) Uniform manifold approximation projection (UMAP) plot of 65,978
   quality-controlled T cells colored by 14 T cell subsets with the
   distinct G1 phase, S phase, and G2M phase. The cell-cycle phase of 14 T
   cell subsets was labeled in the UMAP plot and determined by the
   cell-cycle phase scores calculated by the Seurat package.

   (C) Heatmap showing the expression of the top 50 differentially
   expressed genes among cell subsets obtained by cell type annotation
   utilizing the expression of canonical marker genes in 14 cell
   subgroups. Information on the 14 cell subsets is displayed on the
   right.

   (D) UMAP plots showing the distribution of indicative T cell subsets
   between the non-progression group and progression group.

   (E) Dot plot showing the composition of CD3^+CD4^−CD8^−, CD3^+CD4^+,
   and CD3^+CD8^+ T cells in NPC TIL infusion products according to flow
   cytometry (n = 47). Scatter dot plots show individual values and mean ±
   SEM.

   (F) Bar plot comparing the frequency of 14 cell subgroups in the TIL
   infusion products from NPC patients in the non-progression (n = 15) and
   progression groups (n = 11). The data are presented as mean ± SEM. The
   Wilcoxon rank-sum test was used to determine the significance. ∗p <
   0.05; ns, not significant; adjusted for multiple comparisons using the
   Benjamini-Hochberg procedure.

   (G) Boxplot comparing the frequency of (DN) TILs in the non-progression
   (n = 28) and progression groups (n = 19) according to flow cytometry.
   Median and interquartile were shown in the boxplots. Wilcoxon rank-sum
   test was used to determine the significance.

   (H) Kaplan-Meier survival curves for progression-free survival (PFS)
   and overall survival (OS) in NPC patients stratified according to the
   abundance (high abundance vs. low abundance) of the C8_DN_ T_cells
   cluster inferred from (sc)RNA sequencing data (n = 26). p values were
   determined based on the two-sided log rank test. Optimal cutoff values
   were defined using the “survminer” R package.

   See also [78]Figure S1 and [79]Table S1.

Transcriptome characterization, origin, and composition of (DN) TILs in NPC

   The frequency of peripheral CD3^+CD8^−CD4^− (DN) T cells from NPC
   patients was notably increased relative to healthy donors ([80]Figures
   S2A and S2B). The transcriptome profile showed that (DN) TILs expressed
   high levels of genes encoding innate lymphocyte markers, such as
   FCGR3A, NCAM1, and NKG7; molecules characteristic of regulatory T cells
   (Tregs), including IKZF2, TNFRSF18, IL2RA, STAT5A, and ICOS, but not
   FOXP3 and CTLA4; suppressive and cytotoxic cytokines, such as
   interleukin (IL)-10, TGFB3, IFNG, and tumor necrosis factor (TNF); and
   activated (DN) T cell markers, such as CD69, XCL1, CD160, LTA, and
   CD44, compared with CD4^+ and CD8^+ TILs ([81]Figure 2A). In addition,
   Treg differentiation-related signaling pathways, such as transforming
   growth factor β (TGF-β) and IL-10, and the Treg gene signature were
   up-regulated in the (DN) TIL cluster ([82]Figures 2B and 2C). Flow
   cytometric analysis validated that the (DN) TILs from NPC patients
   exhibited high levels of IL-10, TGF-β, and IKZF2, while showing an
   absence of FOXP3 and CTLA4 expression compared with CD4^+ and CD8^+
   TILs ([83]Figures 2D and [84]S2C). Furthermore, we confirmed that the
   trajectory from peripheral to tumor-infiltrating (DN) T cells contained
   two distinct branches, including Cell Fate #1 (CF1) and Cell Fate #2
   (CF2), in treatment-naive NPC patients (GEO: [85]GSE162025
   ; n = 10, [86]Figure 2E). The CF1 (DN) T cells were mainly found in
   tumor tissues and exhibited increased expression of genes encoding
   characteristic molecules of activated Tregs, such as CD69, IKZF2,
   TNFRSF18, and IL-10, along with an increase in the Treg gene signature
   compared to CF2 (DN) T cells, which were mainly located in peripheral
   blood and peripheral (DN) T cells ([87]Figure S3). We observed a close
   resemblance between the characteristics of the CF1 cluster and the (DN)
   TIL gene signature in the baseline TMEs ([88]Figure 2F). Accordingly,
   we observed that the expression of IL-10, TGF-β, and IKZF2 was
   increased in (DN) TILs compared to peripheral (DN) T cells from NPC
   patients or healthy donors, but not the expression of FOXP3 and CTLA4
   ([89]Figures 2G and [90]S2D). These data suggest that (DN) TILs exhibit
   an activated IL-10^+TGF-β^+IKZF2^+FOXP3^−CTLA4^−Treg-like phenotype in
   the pretreatment TME.

Figure 2.

   [91]Figure 2
   [92]Open in a new tab

   Transcriptional characterization and developmental trajectory of
   CD3^+CD8^−CD4^− (DN) TILs

   (A) Heatmap showing the expression levels of genes encoding NK cell
   markers, Treg-related molecules, immune checkpoint proteins, cytokines,
   and (DN) T cell-related markers among CD8^+, CD4^+, and (DN) TIL
   clusters.

   (B) Gene set enrichment analysis (GSEA) showing significant
   upregulation of TGF-β (left) and IL-10 (right) signaling pathways in
   (DN) TILs compared with other TIL subsets.

   (C) Violin plot showing the expression of a regulatory T cell signature
   across (DN) TIL, CD4^+ TIL, and CD8^+ TIL clusters. A dashed line
   indicates the median of the signature score of the (DN) TIL cluster.
   Median and interquartile were shown in the boxplots. Non-parametric
   Kruskal-Wallis test was used to determine the significance.

   (D) Representative overlap histogram for the expression of IL-10,
   TGF-β, IKZF2, FOXP3, and CTLA4 among (DN) TILs, CD4^+ TILs, and CD8^+
   TILs determined by flow cytometry.

   (E) Developmental trajectory of 16,978 (DN) T cells from tumor tissues
   and peripheral blood of naive treated NPC patients (n = 10) inferred by
   Monocle 2 and CytoTRACE algorithm. Solid and dotted lines denote
   distinct cell fates based on expression profiles, with colors
   indicating the origin of (DN) T cells, pseudo-time, and CytoTRACE score
   (GEO: [93]GSE162025
   ).

   (F) Dot plot showing the correlation between (DN) TIL signature scores
   of TMEs in this study and each state of (DN) T cells (GEO:
   [94]GSE162025
   ; PB, Cell Fate_1, and Cell Fate_2). The sizes and colors of the
   circles represent the strength of the relationship, assessed using
   Spearman’s correlation test. PB, peripheral blood.

   (G) Representative overlap histogram for the expression of IL-10,
   TGF-β, IKZF2, FOXP3, and CTLA4 in the (DN) T cells from TIL infusion
   products, peripheral blood of NPC patients, or healthy donors
   determined by flow cytometry.

   See also [95]Figures S2 and [96]S3.

   Subsequently, the 2,609 (DN) TILs (C8 cluster) from 26 TIL infusion
   products could be clustered into six subsets ([97]Figure 3A). NCAM1
   encoding CD56 and FCGR3A encoding CD16 were expressed in two of the six
   subsets ([98]Figure 3B). Some previous studies had found that (DN) T
   cells could originate from the thymic T cell DN3 stage, double-positive
   (DP) stage, or single-positive (SP) stage.[99]^13^,[100]^14 However, we
   found that the expression of genes encoding T cell receptor (TCR)
   alpha-beta (αβ) and TCR gamma-delta (γδ) were distributed in all six
   (DN) TIL subsets, with the TCRαβ^+ T cells comprising 76% of the total
   (DN) TILs, suggesting that these (DN) TILs may mainly originate from T
   cells after the DP stage ([101]Figures S4A and S4B). To further
   investigate the origin of (DN) TILs, we analyzed the scRNA-seq and TCR
   sequencing (TCR-seq) datasets of three paired TIL infusion products and
   peripheral blood samples from NPC patients treated with CCRT plus
   TIL-ACT. We found that TCRαβ-positive cells were distributed within the
   CD3^+ T cell population, including CD4^+, CD8^+, and (DN) T cells, in
   both the TIL infusion products and peripheral blood ([102]Figures S4C
   and S4D). Further analysis revealed that 60.2% of TCRαβ-positive (DN)
   TILs shared common TCRs (10 shared clones) with 57.5% of CD8^+ TILs,
   and 63.4% of TCRαβ-positive (DN) TILs shared common TCRs (12 shared
   clones) with 49.5% of CD4^+ TILs. Among these shared TCR clones, 8 were
   shared by CD4^+, CD8^+, and (DN) TILs, whereas 33.4% of TCRαβ-positive
   (DN) TILs exhibited unique TCR clones ([103]Figure S4E). In contrast,
   only 31.3% of TCRαβ-positive (DN) T cells shared common TCRs (31 shared
   clones) with 16.6% of CD8^+ T cells, and 1.6% shared common TCRs (2
   shared clones) with 0.3% of CD4^+ T cells. Among these shared TCR
   clones, only 1 was shared by CD4^+, CD8^+, and (DN) T cells together,
   whereas 68.2% of TCRαβ-positive (DN) T cells exhibited unique TCR
   clones ([104]Figure S4F). We also observed that 20.4% of TCRαβ-positive
   (DN) TILs shared common TCRs (2 shared clones) with T cells from
   peripheral blood ([105]Figure S4G). These data suggest that most
   TCRαβ-positive (DN) TILs originate from CD4^+ and CD8^+ T cells (DP or
   SP stage T cells) in the TME, with only a small fraction recruiting
   peripheral (DN) T cells. Further comparing the transcriptional features
   of the CD56^− and CD56^+ (DN) TIL clusters, we found that innate
   lymphocyte markers, such as NCAM1, and several genes encoding Treg-like
   characteristics, such as IKZF2, TNFRSF18, TGFB3, and PRF1, were highly
   expressed in CD56^+ (DN) TILs. Most genes encoding activated
   suppressive Treg-like characteristics, such as IL2RA, ICOS, TIGITIL10,
   IFNG, CD69, XCL1, and CD44, were highly expressed in CD56^− (DN) TILs
   ([106]Figures 3C–3E). IKZF2 is a key factor in maintaining the
   suppressive function and stability of Tregs and is highly expressed in
   (DN) T cells,[107]^15^,[108]^16 both CD56^− (DN) TILs and CD56^+ (DN)
   TILs, and the activated DN Treg signature was the highest in CD56^−
   IKZF2^+ (DN) TILs compared with other (DN) TIL subsets ([109]Figures 3B
   and 3F). A higher frequency of CD56^− IKZF2^+ (DN) TILs was
   significantly associated with poor PFS and OS in NPC patients treated
   with CCRT plus TIL ([110]Figure 3G).

Figure 3.

   [111]Figure 3
   [112]Open in a new tab

   Transcriptomic characterization and association of CD56^− (DN) TILs
   with clinical outcome

   (A) UMAP plot depicting 2,609 quality-controlled (DN) TILs colored by
   six cell subsets.

   (B) UMAP plot of (DN) TILs, with cells colored based on the relative
   normalized expression of NCAM1 (encoding CD56), FCGR3A (encoding CD16),
   and IKZF2.

   (C) Volcano plot showcasing differentially expressed genes (DEGs)
   between CD56^− (DN) TILs and CD56^+ (DN) TILs. Red dots and blue dots
   indicate up-regulated genes (n = 770) and down-regulated genes (n =
   316) in CD56^− (DN) TILs, respectively. Selected genes were labeled.

   (D) Heatmap showing the expression of genes encoding TCR proteins, NK
   cell markers, immune checkpoint proteins, cytokines, and (DN) T
   cell-related markers among CD56^− and CD56^+ (DN) TIL clusters.

   (E) GSEA plots showing that Treg signaling pathways were significantly
   up-regulated in CD56^− (DN) TILs compared with CD56^+ (DN) TILs.

   (F) Violin plot showing the expression of an activated DN Treg
   signature across four (DN) TIL clusters (CD56^−IKZF2^+, CD56^−IKZF2^-,
   CD56^+IKZF2^-, and CD56^+IKZF2^+ (DN) TILs). A dashed line indicates
   the median of the signature score of the CD56^−IKZF2^+ (DN) TIL. Median
   and interquartile were shown in the boxplots. Kruskal-Wallis test was
   used to determine the significance.

   (G) Kaplan-Meier survival curves for PFS (top) and OS (bottom) in NPC
   patients stratified according to the abundance (high abundance vs. low
   abundance) of CD56^−IKZF2^+ (DN) TILs inferred from (sc)RNA sequencing
   data (n = 26). p values were determined based on the two-sided log rank
   test.

   See also [113]Figure S4.

Relationship of CD8^+ TIL subsets with (DN) TILs and TIL-ACT outcomes

   Cell-cell communication analysis revealed that (DN) TILs interact with
   CD4^+ and CD8^+ TIL subsets through the TGF-β, IL-10, and Fas/Fas
   ligand (FasL) signaling pathways ([114]Figure 4A). We further
   determined that the (DN) TILs from NPC patients could suppress the
   proliferation of naive CD4^+ and CD8^+ T cells, and the suppression of
   (DN) TILs was largely diminished by the administration of anti-IL-10
   and anti-TGF-β antibodies, as well as Fas-FasL inhibitor in vitro.
   Notably, the (DN) TILs exhibited suppressive ability toward naive CD4^+
   and CD8^+ T cells comparable to that of conventional CD4^+ induced
   (i)Tregs ([115]Figures 4B, 4C, [116]S5A, and S5B). However, a
   significant negative association was found between the frequency of
   (DN) TILs and C2_proliferating_CD8_T_cells-PCNA in the Spearman
   correlation analysis ([117]Figure 4D). Importantly, the frequency of
   (DN) TILs was significantly negatively associated with the total CD8^+
   TIL density in the TIL infusion products ([118]Figure 4E), suggesting
   that a high frequency of (DN) TILs will limit the expansion of CD8^+
   TILs in TIL infusion products. Moreover, we demonstrated that the (DN)
   TILs from NPC patients could suppress the proliferation of autologous
   ex vivo expanded CD8^+ TILs ([119]Figure 4F). In addition, the TIL
   infusion products exhibited cytotoxicity against human leukocyte
   antigen (HLA)-matched EBV^+ TW03 cells in a dose-dependent manner, but
   not the HLA-mismatched cytokine-induced killer (CIK) cells. However,
   the cytotoxicity of TIL infusion products containing a higher
   percentage of (DN) TILs was lower than that of TIL infusion products
   containing a lower percentage of (DN) TILs ([120]Figures S5C and S5D).
   Notably, the depletion of (DN) T cells from these TIL infusion products
   significantly enhanced their cytotoxicity against EBV^+ TW03 cells
   ([121]Figure S5E). These data suggest that (DN) TILs restrict CD8^+ TIL
   expansion and anti-tumor activity.

Figure 4.

   [122]Figure 4
   [123]Open in a new tab

   Associations of CD8^+ TIL subsets with (DN) TILs and clinical outcome

   (A) Dot color represents the communication probability of the specific
   ligand-receptor pairs, including TGF-β, IL-10, and Fas-FasL signaling
   between the indicated sender cluster and receiver clusters.

   (B) Representative histograms and statistical graph showing
   carboxyfluorescein succinimidyl ester (CFSE) dilution (left) and
   proliferation inhibition rates (right) of (DN) TILs with CD4^+ and
   CD8^+ naive T cells at a ratio of 4:1, 2:1, 1:1, and 1:2. The data are
   presented as mean ± SD, and three biological replicates were included.

   (C) Representative histograms and statistical graph showing CFSE
   dilution (left) and proliferation inhibition rates (right) of (DN) TILs
   with CD4^+ and CD8^+ naive T cells at a 1:1 ratio with neutralizing
   antibodies against TGF-β and IL-10, as well as the presence or absence
   of a Fas-FasL signaling antagonist. The data are presented as mean ±
   SD, and three biological replicates were included. p values were
   evaluated by t test (two-sided). ∗∗p < 0.01; ∗∗∗p < 0.001.

   (D) Bubble plot showing the correlation between the frequency of (DN)
   TILs and other TIL subsets in TIL infusion products (n = 26). p values
   were determined by Spearman correlation analysis. ∗p < 0.05.

   (E) Dot plot showing the correlation between the frequency of (DN) TILs
   and CD8^+ TILs in TIL infusion products (n = 26). p values were
   determined by Spearman correlation analysis.

   (F) Representative histograms and statistical graph showing CFSE
   dilution (left) and proliferation inhibition rates (right) of (DN) TILs
   with autologous expanded CD8^+ TILs at ratios of 4:1, 2:1, and 1:1.
   Conventional CD4^+ (i)Tregs were included as a control. The data are
   presented as mean ± SD, and three biological replicates were included.

   (G–K) Kaplan-Meier survival curves of PFS and OS in NPC patients
   stratified according to the abundance (high vs. low) of total CD8^+
   TILs (G), C2_proliferation_CD8_T_cells-PCNA (H), C4_MHC_II_CD8_T_cells
   (I), C6_proliferation_CD8_T_cells-TOP2A (J), and
   C12_tissue_resident_memory_CD8_T_cells (K) inferred from (sc)RNA
   sequencing data (n = 26). p values were determined based on the
   two-sided log rank test.

   See also [124]Figures S5 and [125]S6.

   In CD8^+ TIL clusters, the frequencies of
   C2_proliferating_CD8_T_cells-PCNA, C4_MHC_II_CD8_T_cells,
   C6_proliferating_CD8-T_cells-TOP2A, and total CD8^+ TILs correlated
   with improved PFS and OS, and the frequencies of
   C3_proliferating_CD8_T_cells-LTB and
   C5_prolifearting_CD8_T_cells-TNFRSF8 were related to better PFS. In
   contrast, the frequencies of C1_cytotoxic_CD8_T_cells,
   C10_proliferating_CD8_T_cells-LGALS1, and
   C12_Tissue_resident_CD8_T_cells were linked to worse prognosis, and the
   frequencies of C0_activated_CD8_T_cells,
   C9_effector_memory_CD8_T_cells, and C13_stem_like_CD8_T_cells were not
   associated with prognosis in NPC patients treated with CCRT plus TILs
   ([126]Figures 4G–4K and [127]S6A). Notably, the CD8^+ TIL subsets (C1,
   C10, and C12) related to poor prognosis had lower expression of
   proliferation-related genes, such as TNFRSF8 (encoding CD30) and PCNA,
   as well as proliferation gene signatures, but had higher expression of
   exhaustion-related genes, such as TIGIT, as well as dysfunction gene
   signatures, compared to the CD8^+ TIL clusters (C2, C3, C4, C5, and C6)
   being related to favorable prognosis, indicating that these CD8^+ TIL
   clusters related to worse prognosis were the exhausted TILs
   ([128]Figures S6B and S6C). Furthermore, we demonstrated that the
   cytotoxicity of CD8^+ TILs against TW03 cells was significantly reduced
   following the administration of a neutralizing antibody targeting
   CD137, a marker of activated cytotoxic T cells[129]^17 and CD30 in
   vitro. CD137 was highly expressed in the C4_MHC_II_CD8_T_cells and
   C5_Proliferating_CD8_T_cells_TNFRSF8 clusters ([130]Figures S6D and
   S6E). These data suggest that distinct CD8^+ TIL subsets exhibit
   different impacts on TIL-ACT outcome.

Associations of baseline TME transcriptome features with TIL-ACT outcomes and
the composition of TIL infusion products

   Gene expression profiling analysis using gene signatures representing
   both immune cell and tumor characteristics[131]^18^,[132]^19^,[133]^20
   ([134]Table S2) was performed in 62 pretreatment NPC tissues from 22
   patients treated with CCRT plus TIL-ACT and 40 patients treated with
   CCRT alone ([135]Figure 5A). Cox regression analysis showed that the
   expression of natural killer (NK) cell, major histocompatibility
   complex (MHC) class II, and mismatch repair gene signatures was
   associated with PFS in the TIL+CCRT arm but not in the CCRT-alone arm
   ([136]Figure 5B). Kaplan-Meier survival analyses demonstrated that the
   patients with high levels of NK cell, MHC class II, CD8^+ cytotoxic T
   cell, CD39^−CD69^− stem-like T cell, NeoTCR8, B cell receptor (BCR),
   LAMP3^+ dendritic cell (DC), and interferon receptor signatures had
   longer PFS after CCRT plus TIL-ACT but not CCRT alone, whereas the
   patients with a high mismatch repair signature, which is indicative of
   malignant cells,[137]^21 had shorter PFS for TIL-ACT plus CCRT but not
   for CCRT alone ([138]Figures 5C and [139]S7).

Figure 5.

   [140]Figure 5
   [141]Open in a new tab

   Relationship between transcriptome features in the baseline tumor
   microenvironment and clinical outcome and TIL subset composition

   (A) Heatmap showing the expression of the indicative gene signatures of
   the baseline TME in 62 NPC patients.

   (B) Bubble plots showing the Cox regression analysis of gene expression
   signatures in patients treated with CCRT+TIL-ACT or CCRT alone. The p
   values and hazard ratios were estimated from a stratified Cox model
   with the low-gene-signature-score group as the reference group.

   (C) Kaplan-Meier survival curves for PFS in NPC patients treated with
   CCRT+TIL-ACT (n = 22) stratified according to the indicative gene
   signature expression (high score vs. low score) of NK cell, MHC class
   II, CD8^+ cytotoxic T cell, CD39^−CD69^− stem-like T cell, NeoTCR8,
   BCR, LAMP3^+ DC, interferon receptor, and mismatch repair signatures.

   (D) Heatmap showing the correlation between gene signature scores and
   the frequency of 14 TIL subsets in TIL infusion products, with
   significance assessed by the Spearman correlation. ∗p < 0.05, ∗∗p <
   0.01.

   See also [142]Figure S7 and [143]Table S2.

   Notably, the TCR signaling, T cell, macrophage, cytotoxic activity, and
   CD8^+ cytotoxic T cell signatures were positively associated with the
   frequency of the C6 subset, which related to better TIL-ACT outcomes,
   whereas the LAMP3^+ DC and co-activation molecules signatures
   negatively correlated with the C10 cluster, which related to poor
   TIL-ACT outcome. Notably, the frequency of (DN) TILs, which correlated
   with poor TIL-ACT outcome, was positively associated with the
   tumor-associated macrophage (TAM) signature but negatively correlated
   with the LAMP3^+ DC and interferon receptor signatures ([144]Figure
   5D). These data suggest that the transcriptome profile of the baseline
   TME influences the clinical outcomes of TIL-ACT and the composition of
   TIL subsets in TIL infusion products.

Identification of 14 intracellular T cell genes with prognostic value for
TIL-ACT

   To identify predictive molecular biomarkers of clinical efficacy, as
   well as the potential therapeutic optimization targets for TIL-ACT in
   NPC patients, we identified a total of 414 differentially expressed
   genes (DEGs), comprising 205 up-regulated genes and 209 down-regulated
   genes, in TIL infusion products from non-progression versus progression
   patients ([145]Figure S8A). Among the 205 up-regulated DEGs, 13 genes,
   namely HHLA2, SPOCK1, EPDR1, RHOU, HYAL3, CD86, HLA-DOA, TBXAS1, CLU,
   SCD, CCDC34, FAM111B, and CYB561, were significantly associated with
   favorable PFS and OS in the CCRT+TIL-ACT arm, as well as the 13-gene
   signature ([146]Figures 6A, [147]S8B, and S8C). We determined that
   these 13 genes are over-represented in the pathways related to T cell
   activation, T cell co-stimulation, and leukocyte cell-cell adhesion
   ([148]Figure 6B). Among the TIL subsets, these 13 prognostic genes were
   highly expressed in CD8^+ TIL clusters, including C2, C3, C5, C6, C10,
   and C4 ([149]Figure 6C). Interestingly, among 209 down-regulated DEGs
   in the non-progression group, only TNFRSF18, encoding checkpoint
   protein, which was highly expressed in the (DN) TIL cluster and
   enhanced proliferative suppression mediated by the (DN) TIL, was
   associated with poor OS in patients treated with CCRT plus TIL-ACT
   ([150]Figures 6D, 6E, [151]S8D, and S8E). These data indicate that
   these 14 prognostic-related intracellular T cell genes may serve as
   potential targets to remold the anti-tumor immunity of TIL infusion
   products and optimize TIL-ACT.

Figure 6.

   [152]Figure 6
   [153]Open in a new tab

   Prognostic value analysis of 14 differentially expressed genes in TIL
   infusion products from non-progression and progression groups

   (A) Euler diagrams represent the overlapping genes enriched in the
   non-progression group and associated with PFS and OS. Thirteen
   overlapped genes were associated with non-progression, PFS, and OS. The
   right diagram represents 13 overlapped genes significantly associated
   with favorable PFS and OS. The colors of dots represent the influence
   of the gene on prognosis: red indicates a protective factor and green
   indicates a risk factor.

   (B) Pathway enrichment analysis of the 13 overlapped genes. The top 8
   significant pathways are shown.

   (C) Dot plot showing the expression of 13 overlapped genes for each T
   cell subset in NPC TIL infusion products. For each gene, dot colors
   represent the average expression in each cell type (scaled and
   log-normalized), whereas size reflects the percentage of cells with
   detectable expression in each TIL subset.

   (D) Dot plot showing the expression of TNFRSF18 for each T cell subset
   in NPC TIL infusion products.

   (E) Kaplan-Meier survival curves for PFS (top) and OS (bottom) in NPC
   patients stratified according to the expression of TNFRSF18 based on a
   median split inferred from (sc)RNA sequencing data (n = 26). p values
   were determined based on the two-sided log rank test.

   See also [154]Figure S8.

Discussion

   In 2024, the US Food and Drug Administration approved the usage of the
   first TIL drug, lifileucel, in advanced melanoma, supporting the
   potential for more widespread application of TIL therapy in the
   future.[155]^22 However, only a minority of patients obtain a survival
   benefit from TIL-ACT in NPC. Thus, it is crucial to know how to choose
   advanced NPC patients who are likely to obtain a clinical benefit from
   TIL-ACT. In this study, using single-cell and bulk RNA sequencing and
   flow cytometric analysis, we identified that the frequencies of
   CD3^+CD8^−CD4^− (DN) TILs and CD8^+ T cell clusters in TIL infusion
   products are associated with TIL-ACT outcome. Mechanically, the (DN)
   TILs exhibited a DN Treg phenotype and restricted CD8^+ TIL expansion
   through Fas-FasL, TGF-β, and IL-10 signaling. In addition, the
   transcriptome of the pretreatment TME could serve as a predictor of
   TIL-ACT outcome and was associated with the composition of TIL infusion
   products in these NPC patients.

   We identified that the CD8^+ TILs comprised 11 of 14 T cell clusters
   (89.8% of total cells) in 65,978 T cells from 26 TIL infusion products,
   suggesting that the CD8^+ T cell subset is the predominant constituent
   of TIL infusion products. However, only the frequency of the (DN) TIL
   cluster was significantly different in non-progression versus
   progression patients in the CCRT+TIL-ACT arm. Patients receiving
   TIL-ACT with a low frequency of (DN) TILs or high CD8^+ TILs in TIL
   infusion products had favorable PFS and OS and a longer restricted mean
   survival time (RMST) for PFS and OS than patients receiving CCRT alone
   ([156]Figure S1F), suggesting that the (DN) TILs hinder the clinical
   efficiency of TIL therapy, but the CD8^+ TILs benefit the clinical
   efficiency of TIL therapy. Some researchers have reported the existence
   of a CD3^+CD8^−CD4^− (DN) T cell subset in ex vivo-expanded TIL
   infusion products in clinical trials for autologous TIL-ACT in other
   solid cancers, but the association between the (DN) TIL proportion and
   clinical response of TIL-ACT, as well as the function of these (DN)
   TILs, was not mentioned in these studies.[157]^23^,[158]^24
   CD3^+CD8^−CD4^− (DN) T cells comprise one of the least studied T
   lymphocyte subgroups and express TCRαβ or TCRγδ but lack CD4 and CD8
   coreceptors.[159]^13 In previous reports, (DN) T cells exhibited dual
   functions under disease conditions; some researchers think that the
   peripheral (DN) T cells function as cytotoxic cells and use autologous
   ex vivo-expanded peripheral (DN) T cells to treat cancer
   patients.[160]^25^,[161]^26 Conversely, some researchers have found
   that peripheral (DN) T cells with a Treg-like phenotype can suppress T
   cell proliferation and send the death signal to CD4^+ T cells and CD8^+
   T cells with the same TCR specificity, preventing autoimmune
   disease.[162]^27^,[163]^28 Here, we found an increased proportion of
   (DN) T cells in peripheral blood from NPC patients compared to healthy
   donors ([164]Figures S2A and S2B). However, peripheral (DN) T cells did
   not exhibit a Treg-like phenotype ([165]Figures S2C, S2D, and [166]S3)
   and were not associated with the outcomes of NPC patients treated with
   either CCRT plus TIL-ACT or CCRT alone ([167]Figure S9). The
   transcriptome profile of the C8_DN_TIL cluster displayed an activated
   Treg-like phenotype: up-regulated genes encoding immune activation and
   characteristic Treg-like molecules, such as CD69, IKZF2, TNFRSF18,
   XCL1, FASLG, and IL-10, as well as Treg differentiation-related
   signaling pathways, such as TGF-β and IL-10 signaling. The (DN) TILs
   had an IL-10^+TGF-β^+IKZF2^+FOXP3^−CTLA4^− DN Treg phenotype, which is
   consistent with other reports of the peripheral (DN) T cell phenotype
   in autoimmune disease.[168]^29^,[169]^30^,[170]^31 Importantly, we
   further identified that the peripheral (DN) T cells could differentiate
   into activated Treg-like cells in tumor tissues from pretreated NPC
   patients (GEO: [171]GSE162025
   , [172]Figure S3). Accordingly, the tumor-infiltrated (DN) T cells have
   been identified in the pretreatment TMEs of several solid tumors, but
   the function of (DN) T cells in tumor tissue, including pro-tumor or
   anti-tumor effects, is dependent on the specific type of
   tumor.[173]^32^,[174]^33 In the present study, we further demonstrated
   that the (DN) TIL cluster comprised six clusters and expressed both of
   TCRαβ (76% of total cells) and TCRγδ. In addition, most TCRαβ-positive
   (DN) TILs shared TCR clones with CD8^+ and CD4^+ TILs, but a small
   subset of TCRαβ-positive (DN) TILs had unique TCR clones or shared
   clones with peripheral T cells ([175]Figure S4). These results suggest
   that most of the (DN) TILs were induced from CD8^+ and CD4^+ T cells in
   the TMEs of NPC patients. CD56 was highly expressed in two of these
   (DN) TIL subsets, and most of the activated Treg-related genes were
   highly expressed in CD56^− (DN) TILs, but IKZF2 was highly expressed in
   both CD56^− and CD56^+ (DN) TILs. The CD56^−IKZF2^+ (DN) TILs had
   predominantly activated suppressive Treg characteristics in (DN) TILs
   and were significantly associated with poor TIL-ACT outcome.

   The cell-cell interaction analysis and in vitro data demonstrated that
   the suppressive effect of (DN) TILs on allogeneic naive CD4^+ and CD8^+
   T cells, as well as autologous CD8^+ effector T cells, was mediated
   through the Fas-FasL, IL-10, and TGF-β pathways. In addition, the (DN)
   TILs not only restricted CD8^+ TIL number but also suppressed the
   cytotoxicity of CD8^+ TILs against NPC cells ([176]Figures S5C–S5E).
   Most researchers think that the high proportion of CD8^+ TILs in the
   infusion product correlates with improved clinical response of TIL-ACT,
   but the CD8^+ TILs represent a heterogeneous cell population comprising
   tumor-specific and bystander T cells in varying differentiation status,
   including stem-like, effector, exhaustion, and senescence status; among
   them, the CD8^+CD39^−CD69^− stem-like neoantigen-specific TILs have
   been shown to correlate with improved clinical
   response.[177]^8^,[178]^34^,[179]^35^,[180]^36 Here, we observed that
   the frequency of total CD8^+ TILs and two proliferating CD8 TIL
   clusters (C2 and C6) and C4_MHC_II_CD8_T_cells with high proliferation
   ability and low dysfunction scores were associated with better
   survival, whereas the C1_cytotoxic_CD8_T_cells,
   C10_proliferating_CD8_T_cells-LGALS1, and
   C12_tissue_resident_memory_CD8_T_cells clusters with low proliferation
   ability and high dysfunction scores were associated with worse
   prognosis in NPC patients treated with CCRT plus TIL-ACT. In addition,
   three CD8^+ TIL clusters (C0_activated_CD8_T_cells,
   C9_effector_memory_CD8_T_cells, and C13_stem_like_CD8_T_cells) were not
   associated with the clinical outcomes of NPC patients treated with CCRT
   plus TIL-ACT ([181]Figure S6A). Taken together, these findings suggest
   that TILs with a stem-like phenotype are effective mediators of tumor
   control in NPC.

   The TME is a complex, heterogeneous cellular environment composed of
   various cell types and plays a significant role in clinical outcomes
   and the response to therapy.[182]^37 Recently, several (sc)RNA
   sequencing studies of pretreatment NPC tissues revealed that multiplex
   factors, such as CD70^+ malignant cells, LAMP3^+ DCs, and Tregs, alter
   the T cell fate, resulting in NPC immune
   escape.[183]^38^,[184]^39^,[185]^40^,[186]^41 In this study, we found
   that nine gene signatures in the pretreatment TME are associated with
   outcome in patients treated with CCRT plus TIL-ACT but not CCRT alone
   ([187]Figure S7). Among these signatures, the high levels of NK cell,
   MHC class II, CD8^+ cytotoxic T cell, CD39^−CD69^− stem-like T cell,
   NeoTCR8, BCR, LAMP3^+ DC, and interferon receptor signatures showed a
   survival benefit for TIL-ACT but not CCRT alone, whereas the mismatch
   repair signature, a malignant cell signature,[188]^21 was associated
   with worse survival after TIL-ACT but not CCRT alone. Associations were
   found between gene signature expression in the baseline TME and the
   composition of TIL infusion products. For example, TCR signaling, T
   cell, macrophage, cytotoxic activity, and CD8^+ cytotoxic T cell
   signatures were positively associated with the frequency of the
   C6_proliferacting_CD8_T_cells-TOP2A subset, which was related to better
   survival. In contrast, the LAMP3^+ DC and co-activation molecules
   signatures negatively correlated with the frequency of the
   C10_proliferating_CD8_T_cells-LGALS1 subset, which was related to poor
   survival. Notably, the frequency of the C8_DN_T_cells subset was
   positively associated with the TAM signature but negatively correlated
   with the LAMP3^+ DC and interferon receptor signatures. We further
   demonstrated that the LAMP3^+ DC cluster could interact with the (DN)
   TIL cluster through the binding of LGALS9 to its ligands CD44, CD45, or
   HAVCR2, a pathway associated with T cell exhaustion and death,[189]^42
   as observed in the pretreatment TME of NPC patients (GEO:
   [190]GSE162025
   ) ([191]Figure S10A), indicating that the LAMP3^+ DC cluster may
   promote exhaustion or apoptosis of (DN) TILs. In addition, the
   (sc)TCR-seq showed a substantial clonal overlap between CD4^+ and CD8^+
   T cells and (DN) T cells ([192]Figure S4), and we found that EBV^+ NPC
   cells could induce the expansion of (DN) T cells in vitro ([193]Figures
   S10B and S10C). These findings suggest that multiple factors in the
   TME, including suppressive conditions, tumor cells, and EBV antigens,
   may promote the expansion of (DN) TILs in NPC. The tumor
   antigen-specific conventional CD4^+ Tregs induced in TME and the
   antigen-specific suppression have been identified in
   cancers,[194]^43^,[195]^44 and Oliveira et al. has identified that HLA
   class II tumor-associated antigens could indirectly and directly induce
   the engagement of immunosuppressive CD4^+ Tregs in melanoma.[196]^45
   However, the underlying mechanisms driving (DN) TIL expansion in NPC
   require further investigation. Recently, neoantigen load and the
   transcriptional and epigenetic states in the TME were reported to
   correlate with treatment outcomes in a cohort of patients with
   metastatic melanoma receiving TIL therapy.[197]^46^,[198]^47

   Our study showed that the expression of 14 intracellular T cell genes
   in TIL infusion products was linked to TIL-ACT outcome through overlap
   analysis. Among them, 13 up-regulated genes in TIL infusion products
   from patients with non-progression, including RHOU, HYAL3, HHLA2, CD86,
   SPOCK1, HLA-DOA, EPDR1, TBXAS1, CLU, SCD, CCDC34, FAM111B, and CYB561,
   predicted improved PFS and OS after TIL-ACT, and combining these 13
   genes into a signature (13-gene signature), the RMST of PFS and OS in
   the high-gene-signature group receiving CCRT plus TIL-ACT was 67.5 and
   75.10 months, respectively, exceeding the durations of 61.17 and 70.66
   months (RMST of PFS and OS) in patients receiving CCRT alone
   ([199]Figure S8B-S8C). We further found that the 13 genes were
   significantly enriched in high activation and proliferation TIL
   subsets, such as C4_MHC_II_CD8_T_cells,
   C5_proliferating_CD8_T_cells-TNFRSF8, and
   C2_proliferating_CD8_T_cells-PCNA, which were linked to better survival
   after TIL-ACT. In contrast, TNFRSF18 encoding checkpoint
   protein[200]^48 was down-regulated in the non-progression group and
   highly expressed in the (DN) TIL cluster and was related to worse OS
   after TIL-ACT ([201]Figure S8D). Notably, these genes may function as
   potential targets to alter T cell bioactivity and enhance the clinical
   efficiency of TIL-ACT.

   In recent years, PD-1 antibody immunotherapy plus chemoradiotherapy has
   shown outstanding clinical achievement in advanced NPC patients in
   clinical trials,[202]^49 and some biomarkers, such as high stromal MHC
   class II positive cell density and tumor PD-L1 expression, are linked
   to clinical benefits of PD-1 antibody treatment.[203]^50 However,
   pEBV-DNA >10,000 copies/mL is linked to poor outcome of anti-PD-1
   treatment,[204]^51 and patients with pEBV-DNA ≥4,000 copies/mL fail to
   achieve a survival benefit from sintilimab plus CCRT compared with CCRT
   alone.[205]^52 The pEBV-DNA load is a biomarker of recurrence and
   poorer prognosis in NPC patients.[206]^53 In our previous phase 2
   clinical trial, the median baseline pEBV-DNA copy number was 28,300
   copies/mL, indicating that most patients may not obtain benefit from
   immunotherapy, such as anti-PD-1 ICB treatment and
   TIL-ACT.[207]^12^,[208]^52 Viral infection has been reported to induce
   the expansion of (DN) T cells in the host,[209]^54 but the association
   and mechanism between heavy EBV DNA load and increased (DN) TILs in NPC
   patients need to be studied in the near future. If we selected the
   patients according to the frequency of CD8^+ TILs or (DN) TILs, the
   transcription level of prognostic genes (e.g., HHLA2) in the TIL
   infusion products, or gene signatures (e.g., MHC class II) in the
   baseline TME, we could observe longer survival in the CCRT+TIL-ACT arm
   compared with the CCRT-alone arm ([210]Figures S1F and [211]S8).

   In summary, we found that advanced NPC patients with a low (DN) TIL
   subset and high activating and proliferating CD8^+ TIL frequencies will
   obtain a survival benefit from TIL-ACT. In the functional analysis, we
   first determined that the (DN) TIL cluster functions as an activated
   Treg-like subset with a
   CD3^+CD4^−CD8^−IL-10^+TGF-β^+IKZF2^+FOXP3^−CTLA4^- T cell phenotype.
   The (DN) TILs hinder the clinical efficiency of TIL-ACT by suppressing
   CD8^+ TIL expansion and cytotoxic activity. In addition, the TME
   transcriptome status decides the composition of TIL infusion products
   and outcome of TIL-ACT. Taken together, our findings may aid in
   identifying suitable patients with advanced NPC for TIL-ACT and
   contribute to the development of strategies to enhance the clinical
   efficacy of TIL-ACT in the near future.

Limitations of the study

   This is a retrospective study based on a clinical trial in high-risk
   NPC patients at a single center. The association between the frequency
   of (DN) TILs in TIL infusion products and clinical benefit of TIL-ACT
   needs to be confirmed in a prospective multicenter clinical trial
   including several types of solid tumors. The functional validation of
   these (DN) TILs is limited due to the difficulty of in vitro culture
   and scarcity of patient samples in this clinical trial. The regulatory
   mechanism of (DN) TIL expansion in the TME and its relationship with
   EBV antigen gene expression in NPC cells was not explored in this
   study. Nevertheless, this identification of (DN) TIL number in TIL
   infusion products as a negative biomarker of the clinical efficacy of
   TIL-ACT in high-risk NPC patients informs the future developmental
   directions of TIL-based immunotherapy strategies that can be
   personalized for maximal clinical benefit.

Resource availability

Lead contact

   Requests for further information and reagents may be directed to and
   will be fulfilled by the lead contact, Jiang Li
   (lijiang2@mail.sysu.edu.cn).

Materials availability

   This study did not generate new unique reagents.

Data and code availability

     * •
       The generated bulk RNA-seq, scRNA-seq, and scTCR-seq data in this
       study have been deposited to Genome Sequence Archive (GSA) in BIG
       Data Center, Beijing Institute of Genomics (BIG), with the
       following accession number GSA-human: [212]HRA008590. Other
       publicly available data can be downloaded from the Gene Expression
       Omnibus (GEO) database at accession number GEO: [213]GSE162025. All
       raw data have been deposited in the Research Data Deposit (RDD)
       public platform at accession number RDD: RDDB2025729586
       ([214]www.researchdata.org.cn).
     * •
       This paper does not report original code. All custom code used for
       the analyses was written with existing software as detailed in the
       [215]STAR Methods section and is available upon request.
     * •
       Any additional information required to reanalyze the data reported
       in this work paper is available from the [216]lead contact upon
       request.

Acknowledgments