Portrayed the development of activated T cell to exhausted T cell based on the ratio of CD28 to CTLA4

Abstract

T cell exhaustion, to some extent, limits the clearance of viruses and tumor cells and accounts for the poor effectiveness of immune therapy. The generally accepted classification of exhausted T cells assigns them as progenitor and terminal statuses. However, the outcome of immune therapy benefits little from this classification. In this study, we constructed a new indicator with the ratio of CD28 to CTLA4 (ccRatio) to continuously characterize the development of T cell from activation to exhaustion. We showed that the ccRatio decreased as T cell exhaustion progressed. We also found that patients who benefitted from anti-CTLA4 treatment had a higher ccRatio value. In addition, by studying the genes that correlated with ccRatio, we found ccRatio is associated with T cell functions which is further confirmed in validation datasets. Furthermore, we found that transcription factor VDR can contribute to differentiation of T cell from activation to exhaustion. The establishment of ccRatio allows us to study T cell exhaustion in a continuous developmental perspective, which is conducive to understand the T cell differentiation and helped in improving existing immunotherapy methods.

Supplementary Information

The online version contains supplementary material available at 10.1007/s00262-025-04297-5.

Introduction

Exhausted T cells (Texs) derive from environments in which the host cannot thoroughly clear antigens, such as chronic infection and the tumor microenvironment [1, 2]. It is well accepted that Texs are weakened in function and proliferation [3]. This compromised function not only protects the host from tissue damage caused by an excessive immune response but also hinders the ability of T cells to clear foreign antigens [4]. From this viewpoint, T cell exhaustion seems to be a double-edged sword. Overactivity of T cells is always considered related to some autoimmune diseases [5, 6], such as systemic lupus erythematosus [7] and rheumatoid arthritis [8]. Researchers have found that inducing T cell exhaustion is a possible therapeutic approach for autoimmune disorders [9]. Moreover, Tex provide a route for tumor cell immune evasion and are also an obstacle for cancer immunotherapy. Therefore, understanding the mechanisms of T cell exhaustion development and finding a marker that helps to regulate T cell exhaustion are key points in the treatment of autoimmune diseases and cancers [10].

Immunotherapy is widely accepted in clinical cancer treatment, though a limited number of patients will benefit from this approach [11]. The major reason for the less-than-ideal results of immune therapy may be due to the heterogeneity of Tex groups. A broad classification has been applied to divide Texs into progenitor and terminal states [12, 13], with terminal Texs being considered the major cause of lack of response [14, 15]. PD-1 is the most commonly used marker in this field and inhibits T cell functions by binding to the PD-L1 ligand on the tumor cell surface. However, improvement in the 5-year survival rate for tumor patients is limited, increasing from 20% to only 40-50% [16]. There are exceptions, and PD-1 is highly expressed by some active T cells [17–19]. Thus, it is necessary to find a new marker to provide further support in treatment guidance for cancer patients.

As the most basic costimulator, CD28 is necessary for T cell activation [20, 21]. CTLA4 acts as a coinhibitor and competes with CD28 for binding to CD80/CD86. This kind of competitive binding of ligands renders CD28 and CTLA4 a good gene pair for studying T cell activity. CTLA4 is induced and externalized when T cells are activated by APCs [22]. High affinity for a ligand promotes binding to CD80/CD86 versus CD28. Binding with ligand fixes CTLA4 on the cell surface, and it accumulates as long as the antigen persists [22]. Accumulation of CTLA4 leads to transformation of cell activation, which is dominated by CD28, to an inhibition phase. During this process, how the ratio of CD28 to CTLA4 dynamically changes from T cell activation to exhaustion and whether the dynamic changes are related to T cell functions are not clear. Here, we perform comprehensive data analysis to explore whether the dynamic change in this ratio is related to progression of T cell exhaustion and whether the ratio serves as a good indicator in clinical immunotherapy.

Results

The ratio of CD28 to CTLA4 decreases with progression of T cell exhaustion

Given that CD28 and CTLA4 alone can serve as indicators of T cell activity, we assessed their expression levels in 4 different T cell exhaustion statuses described by Jean-Christophe Beltra [23]. Both CD28 and CTLA4 showed decreasing expression trends during development from progenitor1 to intermediate status but rebounded in the terminal exhaustion stage (Fig. 1A-B). Since CD28 and CTLA4 competitively bind to the same ligand (CD80/CD86), their ratio can be taken as an advantage in charge of T cell activity. We found that the ratio of CD28 to CTLA4, denoted as ccRatio, decreased steadily from progenitor1 to terminal statuses (Fig. 1C), which suggests that ccRatio may be a better marker than CD28 or CTLA4 alone in determining T cell exhaustion progression. To confirm the decrease in ccRatio in Tex development, independent datasets that included both single-cell and bulk RNA sequencing data were collected. A decrease in ccRatio was observed in both datasets (Fig. 1D and supFig 1A for a single cell, supFig 1B-C for bulk). Although they used different markers to define the progenitor and terminal Tex status, ccRatio was consistently higher in progenitor cells.

Fig. 1.

ccRatio is a potential indicator of Tex progression. A-C) The expression of CD28 (A), CTLA4 (B) and ccRatio (C) in 4 different Tex states in Beltra Data. Progenitor1 (Prog1), Progenitor2 (Prog2), Intermediate (Int) and Terminal (Term) states were represented by 4 different colors. The data used for these figures were from GSE14987923 (n = 3 in each group). D) The value of log2 transformed ccRatio of each cell in Miller scRNA-seq data was shown in boxplot. The values for progenitor Tex cells (n = 2142) were shown in blue box, and the red one was Terminal states (n = 3388) that defined by the marker used in Miller paper. The P-value was calculated with two side t-test. The data used in this analysis were from GSE122713. E) ccRatio was calculated in chronic (light blue) and acute (red) virus infection with bulk RNA-seq data. Each dot represents one sample. The P-value was calculated with two side t-test. Data from GSE132028. Chronic n = 5, Acute n = 4. F) ccRatio was calculated in samples from paired TCGA normal (red) and tumor (light blue) samples (n = 681). The outlier samples were shown with dots. The P-value was calculated with two side t-test. G) The time of ‘disease no progression’ is collected from TCGA sample in cBioPortal database (9602 samples with both CD28 and CTLA4 expression and DFS information). The samples were separated into low and high group based on the mean value of ccRatio. The red line represents the high ccRatio group and blue one is the low ccRatio group. The p-value was calculated with Log-rank test. The lower panel listed the number of samples in each group in given time points listed in the x-axis of survival curve

It is generally acknowledged that the tumor environment and chronic infection can induce T cell exhaustion [10, 24, 25]. We found that compared with acute infection, ccRatio was significantly lower in the chronic infection group (Fig. 1E). Like that in the viral infection model, a decrease in ccRatio was also found in tumors compared with paired normal tissues (Fig. 1F). As shown above, a low ccRatio is related to late development of Texs, which is known to benefit tumorigenesis. Therefore, we further evaluated whether ccRatio is related to tumor survival. No progression survival time data were acquired from TCGA cohort and divided into two groups by the mean value of ccRatio. A significantly better no progression survival time was observed for the high ccRatio group than the low group by the Kaplan‒Meier survival test (Fig. 1G). Taken together, these concordant results indicate that ccRatio is a potential indicator related to Tex development.

Immune therapy-responsive patients have a lower ccRatio

T cell exhaustion is the main target of immune checkpoint blockade (ICB) therapy for tumor patients. However, the effectiveness of ICB is unsatisfactory in most cases. We observed a consistent directional trend across independent datasets: A high ccRatio—indicative of reduced Tex progression—was more prevalent in the ICB response group (Fig. 2A-C). Notably, results from Fig. 2A-B did not reach statistical significance, a limitation likely attributed to the small sample size of the cohorts analyzed. Despite this, the pattern of higher ccRatio in responders was reproducible across all datasets, supporting the potential relevance of ccRatio to ICB response. Furthermore, in a dataset distinguishing patients treated with anti-CTLA4 or anti-PD-1 monotherapy, anti-CTLA4-responsive patients exhibited a statistically significantly higher ccRatio compared to nonresponders (Fig. 2D). In contrast, no statistically significant difference in ccRatio was observed between responders and nonresponders in the anti-PD-1 group (Fig. 2E). Additionally, T cells with restored activity following ICB treatment showed a higher ccRatio than those in the control group (Fig. 2F). Collectively, these findings suggest that ccRatio is associated with anti-CTLA4 therapy outcomes in cancer treatment. This is supported by the statistically significant difference in anti-CTLA4-treated patients, as well as consistent directional trends across other ICB cohorts.

Fig. 2.

ccRatio was higher in ICB responsive samples. A-C) ccRatio (y-axis) was shown with boxplot for ICB responsive samples (yellow), and nonresponsive samples (blue) in three different independent sample sets. The p-value was calculated with two side t-test. Data used in panel A were from GSE115821 (response n = 3; nonresponse n = 34, anti-PD-1 and/or anti-CTLA4 treatment), panel B was from GSE118201 (response n = 3; nonresponse n = 3, PD-1 blockade treatment) and panel C was download from EMBL-EBI database with accession ID E-MTAB-12733⁴⁴. (response n = 36; nonresponse n = 84; anti-LAG3 + anti-PD-1 treatment) D-E) Similar with that in A-C, the ccRatio for anti-CTLA4 (D) treatment (response n = 8; nonresponse n = 45) and anti-PD-1 (E) treatment (response n = 17; nonresponse n = 29) were calculated in a public dataset and shown with boxplot. The yellow one indicates the samples benefit from treatment and the blue one is nonbenefit. The data used in D-E were from Chen P.L.⁴⁷ F) ccRatio (y-axis) for T cells that recovered after ICB treatment was calculated with a public dataset and shown in orange box, and the control was in blue. The from GEO GSE130406 (Recovered n = 3; Control n = 2; Ingenol treatment)

ccRatio is related to Tex progression features in the tumor microenvironment

To further characterize ccRatio in T cell exhaustion progression, we collected single-cell RNA-seq data from 15 different cancer types (supFig. 2A). Only datasets with sufficient CD3⁺CD28⁺CTLA4⁺ T cells (>100 cells) were retained. A summary of the datasets used in the following studies can be found in Supplementary Table 1. According to Beltra’s definition [23], each of the cells was assigned to one of the phenotypes—progenitor1 (TCF1>0 & CD69>0), progenitor2 (TCF1>0 & CD69=0), intermediate (TCF1=0 & CD69=0) and terminal (TCF1=0 & CD69>0). ccRatio for each cell was calculated, and a decrease was found as cell exhaustion progressed (Fig. 3A). It has been reported that T cells gradually lose TCF7 (TCF1) and CXCR5 expression and gain CX3CR1 and CD38 expression with the development of exhaustion [23]. Here, we found that ccRatio was lower in the TCF7^- (TCF1=0) and CXCR5^- (CXCR5=0) groups (Fig. 3B) and higher in the CX3CR1^- (CX3CR1=0) and CD38^- (CD38=0) groups (supFig. 2B). In addition to CTLA4, the cells showed a lower ccRatio in the other exhausted marker-positive group, such as PD-1, TIM3 (HAVCR2), CD39 (ENTPD1) and LAG3 (supFig. 2C). In addition, single-cell GSEA was performed on these tumor data according to constructed gene sets of ‘progenitor markers’ and ‘terminal markers’ [26]. The results revealed that the ‘high ccRatio’ group had a higher progenitor score and a lower terminal score (Fig. 3C-D), which is consistent with our results that ccRatio gradually decreased as T cell exhaustion progressed.

Fig. 3.

ccRatio was related with Tex progression in TME. A) log₂ transformed ccRatio for cells in each Tex group was shown with boxplot. P-value was calculated with two side t-test. (Cell number for each group: Prog1 n = 5786; Prog2 n = 569; Int n = 2482; Term n = 15,776) B) In the left panel, every single T cells were assigned to group with TCF7 expression larger than zero (TCF7⁺, purple n = 6355) or equal to zero (TCF7⁻, yellow n = 18,258) and the log2 transformed ccRatio was shown in boxplot. The figure in the right panel is similar with that in left, but with CXCR5⁺ (purple, n = 3797) and CXCR5⁻ (yellow, n 20,816) groups. The dots represent the outlier value. C) An example GSEA plot of cell in high ccRatio group was shown in left panel. In this example cell, the genes were arranged in descending order according to each gene expression value. Each black vertical line marks the genes in progenitor exhausted gene set. The purple line indicates the change of enrichment score. The red dashed line indicates the highest enrichment score of progenitor exhausted gene set in this example cell. In the right panel, the highest enrichment score of progenitor exhausted gene set for each cell was calculated with same method in left panel and shown with boxplot, and the cells were grouped into high ccRatio group (purple, n = 10,493) and low ccRatio group (yellow, n = 14,120) according to the mean value of ccRatio. D) Similar with that in figure C, but the example cell was from low ccRatio group, and the enrichment score was calculated based on terminal exhausted gene set. All the p-value was calculated with two side t-test. The data used for these figures are listed in supplementary Table 1

Changes in ccRatio are related to T cell function

To further study genes and functions related to ccRatio, correlation analysis was performed between gene expression and ccRatio for each cancer type (supFig. 3). In view of dropout events in scRNA-seq, which will result in unreasonably high correlation due to the excessive 0 value, the imputation algorithm ‘ALRA’ was used before correlation calculation [27]. The correlating genes were considered functionally related to ccRatio. We found that ccRatio-correlating genes were enriched in immune-related functions in most cancer types, such as ‘antigen processing and presentation,’ which was the most enriched (Fig. 4A). A similar conclusion was also made in our previous results, which revealed ccRatio-related genes to be enriched in ‘response to virus’ function in viral infection [28].

Fig. 4.

ccRatio is related with T cell function. A) The functional enrichment was done with genes that fulfill the cutoff of fdr < 0.05 and R² > 0.2 in each cancer types separately. In left panel, Fisher combined p-value was calculated, and top enriched function was shown with barplot. The length (x-axis) and the color of the bar represent the combined p-value of gene ontology enrichment. In right panel, the enriched p-value for each function (y-axis) in each cancer type (x-axis) was shown with different color as shown in legend. And the number of enriched genes for each function in each cancer type was marked with different size of dot. B) The functional enrichment results for goterm ‘T Cell Activation’ in each cancer type were shown in barplot. The length of the bar indicates the number of genes enriched in the goterm. The color of each bar indicates the p-value, and the nonsignificant results were shown in blue. C) The correlation relationship of genes in function ‘T Cell Activation’ was shown with circos plot. Only genes that correlated with ccRatio in at least 5 different cancer types were shown in figure. Different color in outside upper semicircle indicates different cancer types. And the black line in outside lower semicircle represents the genes. The correlation relationship between each gene and ccRatio in each cancer type was labeled with inner line, and the red line indicates positive correlation and the blue one is negative correlation

Moreover, the function of ‘T Cell Activation’ shows significant enrichment in 14 out of 15 cancer types (Fig. 4B). Most of the genes in this pathway correlated negatively with ccRatio, such as TIGIT which is another famous T cell coinhibitor (Fig. 4C). We found the gene RPS3, ZFP36L2 and RPL22 are positive correlated with ccRatio, and within which RPS3 shows significant correlation in 10 cancers. Although there is no direct evidence shows that RPS3 is regulated with T cell activity, other researchers found it is a subunit of NF-κB and participate in regulating cellular activation responses [29]. Here, we present a potential relationship between RPS3 and ccRatio and suggest that RPS3 can further take roles in the T cell differentiation from activation to exhaustion.

Validating the relationship between ccRatio and T cell function with viral infection data

To validate the relationship between ccRatio and T cell activity, a dataset with 43 PBMC samples which include 18 acute COVID-19 infection samples, 9 pre-infection samples, 9 after infection samples and 7 control normal samples was collected. In total, 4748 CD3⁺CD28⁺CTLA4⁺ cells were obtained for further validation. The data were treated and analyzed with the same pipeline as that in tumor samples. Most of the top enriched functions (88%, 15 /17) in tumor data (Fig. 4A) can be reproduced in validation datasets (Fig. 5A). And the function of ‘T Cell Activation’ was also significantly enriched in all four data conditions (Fig. 5B). Besides that, we also interrogate whether similar correlation could be found between ccRatio and the genes in goterm ‘T cell activation’ in the validation dataset with that in tumor data. By extracting the genes from ‘T Cell Activation’ goterm, we found most of the ccRatio-related genes in tumor keep the significant correlation in validation dataset (52%-76%, Fig. 5C). Taken these results together, our analysis suggests that ccRatio can work as a continuous indicator to depict the changes in T cell activity.

Fig. 5.

ccRatio is related with T cell function in validation dataset. A) The functional enrichment was done with genes that fulfill the cutoff of fdr < 0.05 and R2 > 0.2 in 4 conditions of validation dataset. In left panel, Fisher combined p-value was calculated and showed in barplot for the function goterm. The length (x-axis) and the color of the bar represent the combined p-value of gene ontology enrichment. In right panel, the enriched p-value for each function (y-axis) in each condition (x-axis) was shown with different color as indicated in legend. And the number of enriched genes for each function in each condition was marked with different size of dot. B) The functional enrichment results for goterm ‘T Cell Activation’ in validation dataset was shown in barplot. The length of the bar indicates the number of genes in the goterm that are correlated with ccRatio. The color of each bar indicates the p-value. C) The stacked bar chart displays the percentage of genes in T cell activation function that are related to ccRatio in the cancer dataset and validation dataset. Red represents the percentage of genes associated with ccRatio in both datasets, while blue represents the percentage of genes associated with ccRatio only in the validation dataset

The common modulator of ccRatio across different cancer types

Given that ccRatio is based on gene expression of CD28 and CTLA4, transcription factors (TFs) are pivotal regulators of ccRatio. In our correlation analysis, 269 TFs were identified to correlate with ccRatio in different cancers (Supplementary Table 2). Most of these TFs show cancer-specific regulation of ccRatio, among which BATF, PRDM1 and VDR were the top three TF with the highest degree (Fig. 6A-B).

Fig. 6.

VDR is a potential regulator of ccRatio. A) Barplot shows the number of TF (y-axis) that reproducibly correlated with ccRatio in different number of cancer types (x-axis). The top 3 TFs with highest repetitive was labeled. B) The correlation value and p-value were shown with dot plots for top 3 TFs in different cancer types. The color indicates the spearman correlation value, and the dot size indicates the log2 transformed p-value. The gray dots represent the spearman correlation value do not fulfill significant test. C) FACS plots for T cells that treated with CD3 and control in day 4, day 8 and day 12. The x-axis is the value for CTLA4, and y-axis is for CD28. (example plot from one donor sample) D) The pie chart shows the percentage of cells with CD28⁻CTLA4⁻ (gray), CD28⁺CTLA4⁻ (yellow), CD28⁺CTLA4⁺ (red) and CD28⁻CCTLA4⁺ (blue) T cells in each panel of figure C. (Data summary from 4 donor samples) E) Barplot shows the mean ccRatio of T cells with CD3 stimulation, CD3 & calcitriol treatment, CD3 & calcifediol treatment and no treatment control. The data were measured in day 0, day 4, day 8 and day 12 for each treatment. And each condition contains 4 donor samples

VDR has been documented as a modulator of the immune system [30, 31] and is widely used as a therapeutic target in many diseases [32–34]. To study the function of VDR in regulating ccRatio in the T cell exhaustion system, persistent CD3 stimulation was used to induce Texs in vitro (details in methods). Expression of the common marker TIM3 increased compared with the nontreatment control (supFig. 4). Meanwhile, an increase in CTLA4 expressed cells can be seen in CD3 treatment group (Fig. 6C-D). During the cell differentiation, the proportion of CD28^-CTLA4⁺ cell increased and CD28⁺CTLA4⁺ cells decreased, which indicates the T cells undergo a process that acquiring CTLA4 expression and losing CD28 expression. To investigate the roles of VDR in this differentiation process, two drugs that target VDR were used: calcitriol, which is used as a VDR agonist; and calcifediol, which is used as a VDR repressor. The ccRatio is calculated with the geometric mean of fluorescent intensity (MFI) of CD28 and CTLA4. A decline of ccRatio can be observed during T cell differentiation in CD3 stimulation group (Fig. 6E). And in calcitriol treatment group, ccRatio significantly decreased in day 12 comparing with day 8 (Mann–Whitney U test P-value=0.02857), while, in calcifediol group, the ccRatio hardly decreased in day 8 and has limited reduction in day 12 when comparing with day 8 (Mann–Whitney U test P-value= 0.3429) (Fig. 6E). These results indicate VDR takes potential roles in the process of T cell differentiation from activation to exhaustion, which may be an underlying mechanism of its modulator roles in the immune system.

Discussion

T cell exhaustion represents a critical barrier to the effective immune clearance of tumors and chronic pathogens; optimizing immunotherapy outcomes therefore necessitates the precise characterization of this dynamic process [2, 10]. Conventional exhausted T cell (Tex) markers—including TCF1, TIM3, LAG3 and PD-1—and the classification of Tex into discrete progenitor and terminal subsets have demonstrated limited clinical utility, primarily because they fail to capture the continuous nature of T cell differentiation [3, 14]. In this study, we established the CD28/CTLA4 ratio (ccRatio) as a novel continuous indicator that dynamically tracks the transition of T cells from activation to exhaustion, thereby addressing the unmet need for a more refined marker of Tex progression.

The progressive decrease in ccRatio with advancing T cell exhaustion was consistently validated across multiple single-cell and bulk RNA sequencing datasets, as well as in models of chronic infection and tumor microenvironment in our results. This stability of ccRatio originates from the competitive binding of CD28 and CTLA4 to the shared ligand CD80/CD86—a functional relationship that inherently renders their ratio reflective of T cell activation status [22, 35]. Unlike PD-1, which is expressed by both active and exhausted T cells [17, 19], ccRatio specifically distinguishes between distinct Tex developmental stages and correlates with patient prognosis, highlighting its superiority as a predictive biomarker. Notably, a high ccRatio was associated with favorable responses to anti-CTLA4 therapy, which aligns with the competitive ligand-binding mechanism of CTLA4 and provides a rational basis for stratifying patients undergoing this treatment [22, 36].

Functional enrichment and validation analyses confirmed that ccRatio correlates with core T cell biological processes—including antigen processing and presentation, and T cell activation—across 15 cancer types and COVID-19 infection models. The consistency of these results across tumor and acute viral infection models demonstrates that the association between ccRatio and T cell activity transition is not restricted to specific pathological conditions. Furthermore, leveraging ccRatio as a continuous indicator enabled the identification of Tex progression regulators via correlation-based mathematical models—such as the transcription factor VDR identified in this study. Our in vitro experiments further revealed that VDR modulates ccRatio during Tex differentiation: VDR agonists accelerated the decline in ccRatio, while VDR antagonists attenuated this decrease. This finding expands the known immunomodulatory functions of VDR [30, 31] and identifies a potential therapeutic target for manipulating T cell exhaustion.

A key strength of this study lies in the integration of multiple datasets and experimental models, which ensures the robustness of ccRatio as a universal indicator of Tex progression. However, several considerations should be made. First, First, CTLA4 is known to be induced upon T cell activation [37], meaning its expression alone does not indicate that T cells have entered an exhausted state. Exhausted T cells are defined by impaired effector function and proliferation, yet the boundary distinguishing ‘exhausted’ cells from activated or functionally compromised cells remains poorly defined. While numerous studies have supported the functional roles of exhausted cells [38, 39], our results further showed that functional genes are expressed in the CTLA4-single-positive T cell population—albeit with limited expression of IFNγ-responsive genes, suggesting a reduced responsiveness to external stimuli. Second, although we observed a decrease in ccRatio during T cell exhaustion, the competitive advantage of CTLA4 over CD28 in ligand binding represents only one of the mechanisms by which CTLA4 contributes to T cell exhaustion. CTLA4 can also enhance T cell motility, reversing the T cell receptor (TCR) ‘stop signal’ and disrupting the interaction between T cells and antigen-presenting cells (APCs) [35, 40–42]. Additionally, T cell exhaustion involves a complex network of signaling pathways, such as the phosphatidylinositol 3-kinase (PI3K) pathway [43]. Thus, the decrease in ccRatio is likely driven by multiple synergistic effects, and many of the molecular mechanisms underlying Tex differentiation remain to be fully elucidated.

In conclusion, ccRatio provides a continuous framework for understanding T cell exhaustion, bridging the gap between discrete Tex subsets and enabling precise tracking of differentiation dynamics. Its association with immunotherapy response and identification of VDR as a modulator offer practical implications for improving treatment efficacy. By capturing the gradual functional decline of T cells, ccRatio advances our understanding of immune regulation and paves the way for personalized immunotherapeutic strategies.

Materials and methods

Data used

The data used in Fig. 1A-C were obtained from GSE149879 [23]. This dataset includes bulk RNA-seq data of T cells derived from spleens of LCMV Clone13 infection mouse model. Four T cell subsets were defined in this mouse model, and a normalized read count matrix was downloaded for the analysis in this study. We used these data to investigate expression of CD28 and CTLA4, calculate the CD28/CTLA4 ratio (ccRatio) and compare the performance of other T cell exhaustion markers (e.g., PD-1, TIM3, LAG3) across different exhaustion stages.

The data used in Fig. 1D were from GSE122713 [26], used as single-cell sequencing validation data to confirm the decrease in ccRatio along with Tex progression. We directly downloaded the barcodes.tsv.gz, genes.tsv.gz and matrix.mtx.gz from GEO database under this accession id. These 3 files were used as input for following clustering and UMAP plotting with R Seurat packages. Default ScaleData and RunPCA pipeline were implement. UMAP and FindNeighbor function were carried out with dims=1:3. The resolution for clusters finding was set to 0.5. The marker used to annotate proliferating, effector-like and progenitor T cells was as same as that in paper [26]. In addition, bulk RNA-seq validation datasets were acquired from GSE182035 and GSE188526 [44] (supFig. 1B-C).

ccRatio performance in chronic and acute viral infection was assessed in a bulk gene array dataset GSE132028 [45]. Pairwise tumor and normal samples were obtained from TCGA dataset. Specifically, these samples share the same patient ID and are distinguished by the 4th field in their sample barcodes: A value of ‘01’ indicates a tumor sample, while ‘11’ denotes a normal sample. No restrictions were imposed on tumor types, and a total of 681 patients were ultimately included in Fig. 1F. We used samples with paired normal control data in all available cancer types; 9602 samples with both CD28 and CTLA4 expression and DFS information were kept finally.

The data used in ICB treatment analysis were from E-MTAB-12733 [46], GSE115821 [47] and GSE118201. Data from GSE130406 involve sequenced T cells with restored cytokine secretion and control data. The cells with restored cytokine secretion were considered T cells that recovered after ICB treatment [48]. The patients involved in E-MTAB-12733 were treated with anti-LAG3 and anti-PD-1 therapy, and we only considered samples with both CD28 and CTLA4 expression. Complete and partial response were used the ‘response’ group; the samples with stable and progressive disease were used as the ‘no response’ group. Data from Chen P.L. [36] were used to assess ccRatio performance in anti-CTLA4 treatment and anti-PD-1 treatment. The processed gene expression matrix was directly downloaded from the corresponding database using the accession IDs listed in above. Unless otherwise specified, only samples expressing either CD28 or CTLA4 were retained.

All tumor single-cell RNA-seq data are listed in Supplementary Table 1. The single-cell data were filtered with the criterion that only cell with CD3>0 & CD28>0 & CTLA4>0 was retained. For data with a raw read count matrix, the data matrix was used to construct a Seurat object. For data with only raw sequencing data, the same single-cell data processing pipeline was used (see single-cell data processing).

The COVID-19 infection validation dataset was collected from GEO with the accession ID GSE239799. Same with that in tumor single-cell data, only cell with CD3>0 & CD28>0 & CTLA4>0 was kept for further validation.

Single-cell data processing

CellRanger (v7.0.1)) was used to quantify expression of genes in each cell. GRCH38 was used as a mapping reference. Other parameters were kept at their default settings. The Seurat objects for each dataset were merged. Genes with zero reads across all cells were excluded, and only those with UMI counts > 1 in at least two cells were retained. Additionally, cells with mitochondrial RNA expression exceeding 15% were removed. We used the ‘Adaptively Thresholded Low-Rank Approximation (ALRA)’ method to impute the zero value in the scRNA-seq data matrix [27]. The matrix was log normalized before performing the ‘choose_k’ and ‘alra’ functions. The imputed data matrix was used in all analyses. The other normalization and clustering were not performed here, as we did not study the cells in specific clusters.

Correlation analysis

Spearman correlation was performed to evaluate the association between ccRatio and the expression level of each individual gene across all T cell in the dataset. The data used in correlation analysis were the gene expression matrix from the output of ALRA imputation in single-cell data processing step. The ‘cor.test’ function was used to do the calculation in R. Genes with R²>0.2 were considered as genes related to ccRatio and used for functional enrichment analysis with the ‘clusterProfiler’ package. The level 3 data in gene ontology were used, and the goterms with p-value <0.05 in at least 5 cancer types were kept. The function of ‘T cell activation,’ which is level 5 terms, was also calculated.

ssGSEA Analysis

Single-cell GSEA was performed with the R package ‘GSVA’ (version 1.52.3) with the ssgsea function and default parameters. The gene sets corresponding to ‘progenitor exhausted’ and ‘terminal exhausted’ were retrieved from Miller et al [26], and the data processing details are provided in the second paragraph of the 'Data Used' section. R package ‘fgsea’ was used for figure presentation. In the boxplot of ssGSEA, cells were divided into high ccRatio and low ccRatio groups according to the mean ccRatio value. The ‘Anderson–Darling’ normality test was performed with ‘ad.test’ function in ‘nortest’ R package, and the p-value is both less than 0.05 (p-value < 2.2e-16) for ‘progenitor exhausted score’ and ‘terminal exhausted score’. The ‘wilcox-test’ was performed to compare the ‘progenitor/terminal exhausted score’ between ccRatio high and low groups.

Statistical analysis

All tests (not specified) used in comparison between two variables were performed with a two-sided T test if the data were normally distributed; otherwise, the Mann‒Whitney test was applied with ‘wilcox.test’ function in R. Normal distribution was checked by the Shapiro–Wilk test (n<5000) or Anderson–Darling (n>5000). The combined p-value of different cancer types was calculated with the Fisher combined p-value method, using the ‘sumlog’ function in ‘metap’ R package. For the trend test in Fig. 1A-C, we performed the Shapiro–Wilk test to confirm the normal distribution of CD28, CTLA4 and ccRatio values. All three tests yielded p-values < 0.05, so we selected the Jonckheere–Terpstra test—a nonparametric statistical method—to verify whether the expression levels of CD28, CTLA4 and ccRatio exhibit a monotonic change during T cell exhaustion progression. The Jonckheere–Terpstra test was performed with ‘jonckheere.test’ function in ‘clinfun’ R package.

Details of experiments

Peripheral blood mononuclear cells (PBMCs) were isolated from 4 healthy donor blood by density gradient centrifugation. T cells were enriched from PBMCs using EasySepTM Human T Cell Isolation Kit (Steam Cell, #17951) according to manufacturer’s instructions. A total of 8*10⁴ T cells were plated in RPMI 1640 medium supplemented with 10% heat-inactivated FBS, antibiotic antifungal agents, 1 mM pyruvate (Gibco, 11360070), NEAA (Gibco, 11140050) and 50 µM β-mercaptoethanol (Sigma, m3148) in a 96-well plate. Cells were stimulated with T-Activator CD3/CD28 Dynabeads® (Life Technologies, 11131D) following manufacturer’s recommendations, and drugs (calcifediol-VDR inhibitors and calcitriol-VDR agonists, concentrations: 100 nM) were added. Every 48 h, cells were counted, washed and restimulated with a fresh batch of Dynabeads® and drugs. Dynabeads® and drugs were added on days 0, 2, 4, 6, 8 and 10. On days 0, 4, 8 and 12, surface antibody staining and flow cytometry detection were performed, cells were collected and washed with PBS once. Then, the samples were incubated at room temperature for 20 minutes in the best dilution of each of the following monoclonal antibodies: CD3 BUV395 (1:100), CD4-APC-H7 (1:100), CD8 Percp cy5.5 (1:200), CD38-APC (1:100), HLA-DR-BV786 (1:200), CD28-BB515 (1:200), CD152-PE (1:25), TIM3-BV421 (1:200) and PD-1-PE cy7 (1:200). After surface staining, the cells were washed twice with PBS; dead cells were removed with Fixable Viability Stain 440UV. Cytofix (BD Bioscience) was used to fix the cells on ice for 20 minutes. A BD LSRFortessaTM X-20 flow cytometer (BD Biosciences) was used for complete collection. Data analysis was performed with FlowJo (version 10) software. The gating strategy is in supplementary Fig. 4A.

Conflicts of interest

The authors declare no competing interests.

Informed consent

Informed consent was obtained from all subjects involved in the study.

Institutional review board

The blood samples used in this study are obtained from healthy adult donor from Guangzhou Blood Center, China. The use of the blood samples was approved by the Health Commission of Guangdong Province and Guangzhou Institute of Respiratory Disease.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Yuan Wang: Formal analysis, Methodology, Writing—original draft. Xinyue Mei: Validation experiment. Jiaying Zhong and Defeng Qi: sample collection. Jinpeng Cao and Junxiang Wang: data preparing. Shaojiang Zheng: manuscript revision. Zhongfang Wang: Conceptualization, Writing—review & editing and funding support.

Funding

This study was supported by the National Key Technologies Research and Development Program (2022YFC2604104, 2023YFC2306400), National Natural Science Foundation of China (82341244, 82271801, 82402059), R&D Program of Guangzhou National Laboratory (SRPG23‐ 005), S&T Program of Guangzhou Laboratory (SRPG22‐006) and Major Project of Guangzhou National Laboratory (GZNL2023A01009). Project supported by the Education Department of Hainan Province (Hnky2025-23). Academic Enhancement Support Program of Hainan Medical University (XSTS2025099, XSTS2025139). Hainan Province Science and Technology Special Fund (ZDKJ2021038).

Data availability

The data used in this study was included in supplementary materials and the scRNA-seq for bladder cancer generated by this study are available upon reasonable request to corresponding authors.