De novo AITL cases were identified from January 2015 to September 2021 at the Department of Pathology of West China Hospital, Sichuan University. Three expert hematopathologists (S. Z., W. Z., and W. L.) independently reviewed all candidate cases using WHO-classified diagnostic criteria for AITL (4th edition, 2008/Revised 4th edition, 2017) and provided unanimous diagnoses. Patients with recurrent or secondary AITL were excluded. Relevant clinical information was obtained by reviewing electronic medical records. Follow-up information was collected in December 2022 through telephone interviews or electronic medical record reviews. Survival time was calculated from the date of pathological diagnosis to the date of death or last follow-up. The study was approved by the Ethics Committee on Biomedical Research, West China Hospital of Sichuan University. The committee waived the requirement for informed consent as the data for the patients included in the study were retrospectively analyzed.

Single-cell RNA sequencing (scRNA-seq) was performed at the Department of Respiratory and Critical Care Medicine, Frontiers Science Center for Disease-Related Molecular Networks, West China Hospital, Sichuan University, China. The scRNA-seq procedures included sample collection and processing, single-cell RNA library construction and sequencing, and scRNA-seq data processing and analysis. Details of scRNA-seq are provided in the Supplementary Materials and Table 1 .

The expression of IC (PD-1, CTLA-4, TIM3, LAG3, and TIGIT) on CD8⁺TILs was evaluated in fresh lymph node samples from 20 patients with de novo AITL using flow cytometry (FCM) with the following panels (FITC/PE/PerCP-Cy5.5/PE-Cy7/APC/APC- Cy7/AmCyan): (1) -/PD-1/CD8/CD4/CTLA-4/CD45/CD3/CD19 and (2) TIM3/LAG3/CD8/CD4/TIGIT/CD45/CD3/CD19. The detailed procedures have been described in our previous study (7).

The expression levels and types of IC (HAVCR2, CD244, PTGER4, LAG3, PDCD1, CD274, CD101, CTLA-4, CD160, LILRB4, and TIGIT) were analyzed in formalin-fixed and paraffin-embedded samples from 20 patients with de novo AITL using RNA sequencing. RNA sequencing, quality control, and data normalization were performed by Shanghai Rightongene Biotechnology Co., Ltd. (Shanghai, China). RNA sequencing data for AITL (n = 51) (GSE19069 and GSE58445) were downloaded from the GEO website (https://www.ncbi.nlm.nih.gov/gds/). The details of RNA sequencing are provided in the Supplementary Materials .

The average expression levels of CD8A, CD8B, GZMA, GZMB, and PRF1 detected by RNA sequencing were defined as the cytotoxic T lymphocyte (CTL) levels (8); the 20 AITL cases were divided into high- and low-CTL groups based on the mean CTL marker levels. The 189-exhaustion biology-related gene panel was established according to a previous study, which showed a positive association with CD8⁺TILs exhaustion and reflected exhaustion from 19 biological aspects (9, 10) (including IC, glycosylation, effector functions, IFN response, translation, cytokines, ubiquitination, mitochondria, chromatin/DNA repair, proteases, homing and migration, cytoskeleton/cell adhesion, metabolism, cell surface receptors and ligands, apoptosis/cell death/caspase and annexins, membrane biology and vesicle transport, cell cycle, signaling, and miscellaneous aspects) ( Supplementary Table 1 ). The 189-exhaustion biology-related gene panel was then used to analyze exhaustion-related biological changes in CD8⁺TILs from the high- and low-CTL groups.

Statistical analyses were performed using the Statistical Package for the Social Sciences (SPSS) software (version 26.0, SPSS Corp., Chicago, IL, USA). Spearman’s correlation coefficients were used to assess correlations among CD8⁺TILs, CD8A, CD8B, and IC. The chi-square test was used to compare the proportion of increased IC between the high- and low-CD8⁺TILs groups, as well as the CTL groups, and the differentially expressed genes (DEGs) between the high- and low-CTL groups. The cutoff values for CD8⁺TILs and CTL were defined as the points at which sensitivity and specificity were maximized in the receiver operating characteristic (ROC) curves for predicting overall survival (OS). Survival time was calculated using Kaplan-Meier analysis. A two-tailed p value less than 0.05 was considered statistically significant.

Based on the inclusion criteria, 40 patients with de novo AITL were included in our cohort. The median patient age was 60 years (range, 38–83 years). Men accounted for 65% (n = 26) and women accounted for 35% (n = 14) of the sample. Eighteen (64.3%) patients were at an advanced stage (III/IV), and 28 (75.7%) were in a generally good condition (PS 0-1). The basic information is summarized in Table 2 . Most patients (n = 24, 64.9%) received chemotherapy with the CHOP, R-CHOP, or ICE regimens, whereas a few patients (n = 5, 13.5%) received other treatment options, including surgery and immune-related therapies. Overall, 19 (73.1%) patients responded positively to the treatment. Eight patients (21.6%) refused treatment after diagnosis because of advanced disease stage, compromised general condition, and chemotherapy intolerance. Treatment-related data for three patients were not available.

After quality control and filtering, we obtained single-cell transcriptomes for 14 406 cells from the lymph node tissues of two patients with AITL (AITL-1, CD8⁺TILs:60.9%; AITL-2, CD8⁺TILs:48.6%). Through principal component analysis (PCA), t-distributed stochastic neighborhood embedding analysis (t-SNE), and uniform manifold approximation and projection (UMAP), three main clusters were identified, including T-TILs (CD3D, CD3E, CD3G), B-TILs (CD19, MS4A1, CD79B), and macrophages (CD68, CD14), among which the T-TILs comprised naive T-TILs (IL7R, TCF7), CD4⁺TILs (CD4), and CD8⁺TILs (CD8A, CD8B) ( Figure 1 ; Supplementary Figures 1 , 2 ). AITL-1 showed higher expression of CTL and multiple IC genes in CD8⁺TILs than those in AITL-2 ( Figure 2A ); CD8⁺TILs from AITL-1 and AITL-2 were further divided into two clusters, including CD8-1 and CD8-2 cells ( Supplementary Figure 3 ). Compared to that in CD8-1 cells, the expression of CTL and multiple IC genes was higher in CD8-2 cells ( Figure 2B ).

The mean proportions of CD8⁺TILs, CD8⁺PD-1⁺, CD8⁺CTLA-4⁺, CD8⁺TIM3⁺, CD8⁺LAG3⁺, and CD8⁺TIGIT⁺ in AITL (n = 20) ( Figures 3A–E ) were 40.75% (range, 4.1–78.9%), 14.39% (range, 0.19–45.53%), 9.09% (range, 0.11–44.28%), 2.63% (range, 0–10.66%), 3.36% (range, 0.22–14.93%), and 13.57% (range, 0.24–47.88%), respectively ( Table 3 ). The proportion of CD8⁺TILs in AITL was positively correlated with that of CD8⁺TIM3⁺ (r = 0.72, p < 0.001), CD8⁺LAG3⁺ (r = 0.70, p < 0.001), CD8⁺CTLA-4⁺ (r = 0.57, p = 0.008), CD8⁺TIGIT⁺ (r = 0.55, p = 0.012), and CD8⁺PD-1⁺ (r = 0.54, p = 0.03) ( Figure 4 ). Twenty patients were divided into high- and low-expression groups according to the mean proportions of CD8⁺TILs and IC, respectively. Compared to the low CD8⁺TILs group, the number of cases with IC ≥ 3 increased in the high CD8⁺TILs group (20% vs. 90%, p = 0.003) ( Supplementary Table 2 ).

The mean expression levels of the CTL and IC genes are shown in Supplementary Table 3 . The CTL showed strong positive correlations with HAVCR2, LAG3, PTGER4, and CD244 ( Figure 5 ). Further, 20 patients were divided into high- and low-expression groups according to the mean expression of CTL and IC, respectively. Compared to those in the low CTL group, the number of cases with IC ≥ 6 tended to increase in the high CTL group (8.3% vs. 50%, p = 0.058) ( Supplementary Table 3 ).

The mean expression levels of CTL-related genes (CD8A, CD8B, GZMA, GZMB, and PRF1) (8) was 27.84 (range, 10.59–65.13). The 20 cases were divided into a high-CTL group (n = 8) and low CTL group (n = 12) based on the mean CTL level. Unsupervised hierarchical clustering revealed two distinct clusters of exhaustive biology-related genes, consistent with the two different CTL-level groups ( Figure 6A ). Twenty-four DEGs were distributed across 11 biological aspects between the high- and low-CTL groups. Further analysis of the DEG characteristics showed that among the total DEGs, 23 (95.8%) DEGs distributed in 11 aspects were involved in the high CTL group, including IC (HAVCR2, CD244), signaling (S100A11, GPR65), mitochondria (HIF1A), cytokines (IL-6 and IL-10), cell surface receptors and ligands (FASLG, SLAMF7, IFNGR1, NRP2), homing and migration (CXCL10, CXCL9, CCL5, CCR5, CCL4, CCRL2), membrane biology and vesicle transport (PLSCR1), apoptosis/cell death/caspase and annexin (CASP4), metabolism (GPD2), IFN response (IFIT3), and miscellaneous aspects (C1QC, CTSS). One (4.2%) DEG (NPRL3), distributed in miscellaneous aspects, was detected in the low-CTL group. Chi-square analysis revealed that more DEGs were involved in the high CTL group (p = 0) and covered more biological aspects than the low CTL group (p = 0.004) ( Figures 6B, E and Table 4 ). Besides, the result of gene-set enrichment analysis (GSEA) demonstrated that exhaustion biology-related genes were enriched in the high CTL group from our dataset (FDR = 0.005) ( Figure 7A ).

Following the protocol described above, AITL cases (n = 51) from the data by Iqbal et al. (GSE19069 and GSE58445) were divided into a high-CTL group (32 cases) and low-CTL group (19 cases) based on the mean CTL level (10.61). The pivotal 24 DEGs obtained from our results were compared between the two groups. Clustering results revealed two distinct clusters ( Figure 6C ). All 20 DEGs (distributed across 11 biological aspects) were involved in the high CTL group, including IC (HAVCR2), signaling (S100A11, GPR65), mitochondria (HIF1A), cytokines (IL-10), cell surface receptors and ligands (FASLG, SLAMF7, IFNGR1), homing and migration (CXCL10, CXCL9, CCL5, CCR5, CCL4, CCRL2), membrane biology and vesicle transport (PLSCR1), apoptosis/cell death/caspase and annexin (CASP4), metabolism (GPD2), IFN response (IFIT3), and miscellaneous aspects (C1QC, CTSS). The chi-square analysis revealed that more DEGs were involved in the high-CTL group (p = 0) and covered more biological aspects than those in the low-CTL group (p=0) ( Figures 6D, F and Table 4 ). One case in the low-CTL group clustered with the high-CTL group in the study by Iqbal et al. ( Figure 6C ). Further, the results of GSEA demonstrated that exhaustion biology-related genes were enriched in the high-CTL group from the dataset of Iqbal et al. (FDR = 0.006) ( Figure 7B ).

The correlations between the proportion of CD8⁺TILs, CTL levels, and clinical features are shown in Table 5 . Patients in the high CD8⁺TILs and high CTL groups were prone to demonstrate a lower response rate (CR + PR) to treatment and presented with more stage IV disease compared with those in the low CD8⁺TILs and low CTL groups.

Overall, follow-up data was available for 39 patients, with a follow-up rate of 97.5% (39/40). The follow-up time ranged from 0.3 to 38.5 months, and the median OS time was 24.7 months (95% confidence interval: 19.39–30.01 months; Figure 8A ). High CD8⁺TILs (≥ 45.9%) detected by FCM (p = 0.024, Figure 8B ) was a significant risk factor for patients with AITL; patients with high CTL levels (≥ 27.84) detected by RNA-seq (p = 0.181, Figure 8C ) also tended to have a poor prognosis.