Regulatory T-cell Transcriptomic Reprogramming Characterizes Adverse Events by Checkpoint Inhibitors in Solid Tumors

Immune checkpoint inhibitors (ICI), such as anti–PD-1, anti–PD-L1, and anti–CTLA-4, have shown remarkable efficacy in the treatment of solid tumors. Despite their success, more than half of treated patients do not derive meaningful benefit from such treatments, and responses are often accompanied by autoimmune toxicities, termed immune-related adverse events (irAE), which may interfere with therapy and could lead to life-threatening conditions (1–2). These toxicities usually develop within the first few weeks to months after treatment initiation and can manifest as generalized symptoms including fatigue or fever, or can produce organ-specific damage leading to rash, arthritis, colitis, pneumonitis, and adrenal or thyroid insufficiency (3–4). Management of irAEs in general entails dose interruption/discontinuation of immunotherapy and systemic high-dose corticosteroids (5). To date, it is not possible to predict which patients will develop irAEs, which organs will be involved, and how severe an irAE may be, making patient management challenging. Although genetic predisposition, gut microbiome, epitope spreading, and cross-presentation of neoantigens have been implicated in the development of irAEs (6–7), the precise molecular mechanisms that drive them remain largely unknown. As autoimmunity and cancer represent two sides of the same coin, in terms of the operation of immune tolerance mechanisms, perhaps it is not surprising that as we manipulate the immune system to treat cancer through the use of ICI, we inevitably disturb the mechanisms that regulate self-tolerance, inducing an array of irAEs.

Immune suppression by Foxp3⁺ regulatory T cells (Treg) is essential and indispensable for maintenance of tolerance and prevention of autoimmunity, as illustrated by the fact that spontaneous autoimmune disease develops when Tregs are rendered deficient. For example, mutations in the human gene encoding Foxp3 impair Treg development and cause a fatal multiorgan autoimmune disease called immune dysregulation, polyendocrinopathy, enteropathy, and X-linked syndrome (8). In rodents, depletion of Foxp3⁺ Tregs causes similar autoimmune diseases (9).

In addition to preventing autoimmunity, Tregs suppress the antitumor immune response and promote tumor outgrowth (10). The tumor microenvironment (TME) is unique in that it is nutrient poor, hypoxic, and acidic. This provides a unique proliferative and functional advantage to Tregs, which rely on oxidative phosphorylation, compared with effector T cells, which are primarily glycolytic (11). Consistent with this, increased frequencies of intratumoral Tregs have been reported in almost all solid tumors (12), and the level of Treg infiltration is associated with unfavorable prognosis in various types of cancer (12–13). In the TME, Tregs acquire enhanced suppressor functions and an activated phenotype with expression of chemokine receptors such as CCR4 or CCR8 and upregulation of immune checkpoint proteins such as CTLA-4, PD-1, PD-L1, PD-L2, and IL1R2 (14). Thus far, efforts to therapeutically target Tregs in cancer have had limited success and have been accompanied by the development of side effects.

Tregs express high levels of CTLA-4 and PD-1, and recent evidence suggests that PD-1 expression by Tregs in the TME can predict the efficacy of anti–PD-1 immunotherapy (15). Therefore, we hypothesized that ICI could imprint on Tregs and induce their reprogramming, contributing to the development of irAEs. Herein, we demonstrate an intense “inflammatory” phenotype in CD4⁺CD25⁺CD127⁻ Tregs isolated from the blood of patients with advanced metastatic melanoma who developed irAEs after ICI therapy. As the use of ICIs in anticancer therapy is rapidly expanding, the frequency of irAEs is expected to significantly increase. Thus, it is urgent that we increase our understanding of the mechanisms underlying the development of irAEs in order to not only facilitate better management of irAEs but also develop prognostic biomarkers.

Peripheral blood samples were obtained from individuals diagnosed with cancer or autoimmune diseases and were freshly analyzed directly after collection. Patients with advanced metastatic melanoma, stage III(B–D)–IV, who were receiving treatment with anti–PD-1 (nivolumab 240 mg/2 weeks or pembrolizumab 200 mg/3 weeks, n = 26), were recruited through the First Department of Internal Medicine, Laikon General Hospital, National and Kapodistrian University of Athens (Greece). Samples were collected at the beginning of the third cycle of anti–PD-1 infusion. Patients with kidney (n = 6), non–small cell lung (n = 6), liver (n = 1), and bladder (n = 3) cancer who were receiving treatment with anti–PD-1 (nivolumab 240 mg/2 weeks or pembrolizumab 200 mg/3 weeks, between 8th and 12th cycles of infusion) were recruited through the oncology clinic of Alexandra University hospital, National and Kapodistrian University of Athens (Greece) and Department of Medical Oncology, University General Hospital of Heraklion (Greece). Individuals with cancer were staged according to the American Joint Committee on Cancer 8th edition (AJCC Manual).

Individuals with rheumatoid arthritis (RA; n = 9) were diagnosed according to the 1987 American College of Rheumatology criteria and had moderate disease activity [Disease Activity Score (DAS) > 4.2] according to the disease activity score based on the 28 joint counts. RA and healthy individuals (n = 14) were recruited through the Rheumatology and Clinical Immunology Department, 4th Clinical Pathology, Attikon University Hospital (Athens, Greece). Individuals with inflammatory bowel disease (n = 4) who were in remission or had moderate disease activity, based on standard clinical, endoscopical, and histologic criteria, were recruited through the GI-Unit, 3rd Academic Department of Internal Medicine, Sotiria Hospital, National and Kapodistrian University of Athens. Individuals with Hashimoto's thyroiditis (n = 7), who were diagnosed on the basis of increased thyroid-stimulating hormone (TSH), increased anti–thyroid peroxidase (TPO) and anti-Tg, and normal or low T₄/T₃ levels and receiving thyroxine replacement therapy, were recruited through the Endocrine Unit of 1st Department of Internal Medicine, Laiko Hospital (Athens, Greece). All individuals with autoimmune diseases did not receive steroids for at least 24 hours prior to blood drawing.

The Clinical Research Ethics Board of all hospitals approved this study. Informed written consent was obtained from all individuals prior to sample collection. All patient studies were conducted in accordance with ethical guidelines of the Declaration of Helsinki. Meta-analysis of healthy and RA samples was performed from Alissafi and colleagues (16). Clinical and demographic characteristics of all the patients included in this study are summarized in Supplementary Table S1. Objective response rate (ORR) is defined as the proportion of patients with a complete or a partial response to therapy.

Heparinized blood (20 mL) was collected from human subjects. Peripheral blood mononuclear cells (PBMC) were isolated on Histopaque-1077 (cat. #10771, Sigma) density gradient. Briefly, blood was diluted 1:1 with PBS and carefully layered over Histopaque medium. Ficoll gradients were centrifuged at 1,800 rpm for 30 minutes with no brake at room temperature. The PBMC layer was collected, and cells were washed with PBS.

Tregs were isolated from PBMCs using the CD4⁺CD25⁺CD127^dim/− Regulatory T Cell Isolation Kit II (cat. #130-094-775, Miltenyi Biotec) according to the manufacturer's instructions. Briefly, non-CD4⁺ and CD127^high cells were indirectly magnetically labeled with a cocktail of biotin-conjugated Abs and anti-biotin mAbs conjugated to MicroBeads. The labeled cells were subsequently depleted by separation through MACS column. In the second step, the CD4⁺CD25⁺CD127^dim/− Tregs were directly labeled with CD25 MicroBeads II and isolated by positive selection from the pre-enriched CD4⁺ T-cell fraction, by separation through MACS Column, and eluted as the positively selected cell fraction.

For analysis of Tregs, single-cell suspensions from human PBMCs were stained with conjugated antibodies against human CD4 (BioLegend; cat. #317428, RRID:AB_1186122), CD25 (BioLegend; cat. #302604, RRID:AB_314274), and CD127 (BioLegend; cat. #351316, RRID:AB_10900804). For Foxp3 intracellular staining, cells were fixed and stained using the Foxp3 Staining Set (Thermo Fisher Scientific; cat. #12-4774-42, RRID:AB_10670338) according to the manufacturer's instructions. Cells of interest were acquired on a FACS ARIA III (BD Biosciences) using BD FACSDIVA v8.0.1 software (BD FACSDiva Software, RRID:SCR_001456). Analysis was performed with FlowJo software (FlowJo, RRID:SCR_008520).

Total RNA was extracted using NucleoSpin RNA XS (Macherey-Nagel; cat. #740955.50) as per the manufacturer's protocol, and mRNA libraries were generated using the Illumina TruSeq Sample Preparation kit v2 (Illumina; cat. #RS-122-2001). Single-end 75-bp mRNA sequencing was performed on Illumina NextSeq 500 platform. Raw fastq sequencing reads were aligned against the human reference genome sequence (GRCh37/hg19) using the STAR aligned v2.5.0c, discarding all nonuniquely aligned reads (16). For read counting per annotated gene, we utilized the STAR (STAR, RRID:SCR_015899) function “–quantMode GeneCounts,” counting reads matching exons of the Ensembl V75 Genes annotation. Further processing was performed with the R Bioconductor (Bioconductor, RRID:SCR_006442) package edgeR v.3.14.0 (edgeR, RRID:SCR_012802) using unstranded reads. Reads were normalized for intrasample and intersample variances using the functions “calcNormFactors” and “estimateTagwiseDisp,” and further by the gene length as per Ensembl V75 Genes annotations, resulting in fragments per kilobase per million for each gene. Differential gene expression analysis was performed as previously described (16). Genes with P value < 0.05 and fold change |FC| > 1.5 were considered statistically significant. Heatmaps and boxplots were created in R with an in-house–developed script which is based on ggplot package. Venn diagrams were created using Venny 2.1.0 online tool.

Significant differentially expressed genes (DEG) were used for gene ontology analysis using g:Profiler Web server (G:Profiler, RRID:SCR_006809). Gene set enrichment analysis (GSEA; RRID:SCR_003199) was also performed to reveal enriched signatures in our gene sets based on the Molecular Signatures Database (MSigDB) v7.0. Gene sets were ranked by taking the −log₁₀ transform of the P value multiplied by the FC. Significantly upregulated genes were at the top and significantly downregulated genes were at the bottom of the ranked list. GSEA preranked analysis was then performed using the default settings. Enrichment was considered significant if FDR (q value) <5%. Pathway analysis was performed using Ingenuity Pathway Analysis (IPA, RRID:SCR_008653). We selected DEGs for each pairwise comparison using P value < 0.05 and |FC| ≥ 1.5 thresholds.

Statistical analysis was performed using unpaired Student t test in GraphPad Prism v5 software (GraphPad Prism, RRID:SCR_002798). Data are presented as mean ± SEM. P value < 0.05 was considered as indicative of statistical significance. All P values and n are reported in the figure legends. The investigators were not blinded to the identities of the samples. No randomization was performed. Compared samples were collected and analyzed under the same conditions. G*power analysis was performed (with 90% power and 0.001 type 1 error) to calculate the appropriate sample size. No data were excluded. All data showed normal distribution.

The RNA sequencing (RNA-seq) data have been deposited in the European Genome-phenome Archive (EGA accession number: EGAS00001004694)

The molecular mechanisms underlying the development of irAEs remain elusive, and their delineation will facilitate the design of translational approaches to induce robust antitumor immunity without disturbing peripheral tolerance. Herein, we explored whether ICI immunotherapy reprograms peripheral Tregs during cancer and could contribute to the development of irAEs. To do this, CD4⁺CD25⁺CD127⁻ Tregs were isolated from the blood of individuals with advanced melanoma who were treated with anti–PD-1 and subjected to transcriptomic analysis. Eleven of the patients experienced irAEs such as hypophysitis, thyroiditis, psoriasis, and colitis, and are denoted hereafter as Mel-irAEs, and 15 did not (denoted as Mel-ctrl; Supplementary Table S1). Although we did not observe significant differences in the frequency of CD4⁺CD25⁺Foxp3⁺CD127⁻ Tregs between Mel-irAEs and Mel-ctrl (Supplementary Fig. S1A), genome-wide RNA-seq analysis revealed 252 DEGs (|FC| ≥ 1.5 and P < 0.05) between the two groups (Supplementary Fig. S1B).

Focusing on Treg biology and immune-related genes, we identified a core signature of upregulated transcripts in Mel-irAEs compared with Mel-ctrl Tregs (Fig. 1A). These genes included those encoding chemokines and chemokine receptors involved in inflammation and leukocytic mobility, such as CXCL2, CXCL16, CXCL10, and CCR1, and enzymes such as amino acid catabolizing ARG1 and antioxidant SOD2. In contrast, genes implicated in the function of activated Tregs such as CCR3 and SERPINE2 or in metabolic processes such as ADAM12 and ATP6AP1L were significantly downregulated in Mel-irAE Tregs compared with Mel-ctrl Tregs (Fig. 1A and Supplementary Fig. S1B). Pathway analysis of DEGs showed that circulating Mel-irAE Tregs exhibited an inflammatory-like phenotype, consisting of leukocyte activation, inflammatory response, cytokine production, oxidative stress response, type I IFN signaling, and IFNγ signaling, compared with Mel-ctrl Tregs (Fig. 1B). In detail, GSEA demonstrated that Mel-irAE Tregs were enriched in proinflammatory processes such as “IFNγ response” (with transcription factors: IRF1, STAT1, STAT3, and NFKB1A being among the top enriched genes), “IFNα response” (with top enriched genes belonging to the IFN-induced genes, i.e., IFITR2-3, IFITM2, IFIH1), “TNFα signaling via NF-Kb” (with top enriched genes including RELB, IL18, and IL1B), and “IL6/JAK/STAT3 signaling” (with STAT1, STAT3, TNF, IL1B, IRF1, and IFNGR2 being among the most enriched genes; Fig. 1C). The enrichment in inflammatory pathways in Mel-irAE Tregs was retained (370 DEGs) when individuals were categorized according to their ORR (Supplementary Fig. S1C and Supplementary Table S1).

Treg metabolism drives their proliferation and immunosuppressive nature under steady state (17), whereas their destabilized metabolic fitness dictates development of autoimmunity (16). In this context, our GSEA analysis revealed a metabolic rewiring in Mel-irAEs compared with Mel-ctrl Tregs. This rewiring was characterized by “MTORC1 signaling” (with enrichment for TBK1, SERP1, and NAMPT genes), “Reactive oxygen species pathway” (with top enriched genes such as MPO, SOD2, and CAT), and “Fatty acid metabolism” (with top enriched genes such as HMGCS1, ACOX1, and ACSL1; Fig. 1C).

To provide mechanistic insights, IPA was performed, revealing IFNγ, IL6, IL1, IL33, TNF, NF-kb complex, STAT1, STAT4, STAT5, IRF1, IRF3, IRF5, and IRF7 as the top activated upstream regulators in Mel-irAE Tregs (Supplementary Fig. S1D). These data provide insight into the signals that drive intense reprogramming of circulating CD4⁺CD25⁺CD127⁻ Tregs in patients with advanced melanoma who develop irAEs following ICI immunotherapy toward a proinflammatory phenotype accompanied by enrichment in metabolic alterations.

Despite the extensive clinical characterization of irAEs, it remains unclear to what extent they share common pathogenetic features with spontaneously arising autoimmune syndromes. To this end, we performed transcriptomic analysis of CD4⁺CD25⁺CD127⁻ Tregs isolated from peripheral blood of healthy individuals and individuals with autoimmune manifestations analogous to the irAEs cohort (ulcerative colitis, Crohn's disease, Hashimoto's thyroiditis, and RA, see Supplementary Table S1, denoted as autoimmune). When compared with Tregs from healthy individuals, autoimmune Tregs demonstrated a deregulated transcriptomic program that was similar to the Mel-irAE Treg phenotype (Fig. 2A). GSEA analysis demonstrated an overrepresentation of Mel-irAE Treg DEGs in autoimmune versus healthy Tregs, including genes involved in IFNγ signaling (CXCL10, IFNGR2, BCL3, and CD63), leukocyte activation (IRF1, ARG1, NLRP3, IFNGR1, and STAT3), regulation of autophagy (MAP1LC3B, BNIP3L, FOXO3, LAMTOR5, and RAB5A), apoptotic cell death (FASLG, KREMEN1, IRF1, SOCS3, and BCL3), and regulation of metabolic processes (LDLR, ATF3, MAP3K8, ARE, RELB, ADIPOR1, BNIP3L, ARF1, and MAFF; Fig. 2A and B). Our findings demonstrate a shared transcriptomic profile among Tregs from patients with autoimmune diseases and those with irAEs. This profile is enriched for inflammatory, apoptotic, and metabolic pathways, and is characteristic of destabilized Tregs (18).

To examine whether inflammatory reprogramming of Tregs is specific to patients with melanoma experiencing irAEs following anti–PD-1 immunotherapy, we performed transcriptomic analysis of Tregs from individuals with kidney, liver, bladder, and non–small cell lung cancer who did or did not experience irAEs following anti–PD-1 immunotherapy (denoted as CA-irAEs or CA-ctrl respectively, Supplementary Table S1). CA-irAE Tregs had a distinct transcriptomic profile, with 225 DEGs compared with CA-ctrl Tregs (Fig. 3A). Pathway analysis revealed an inflammatory profile in CA-irAE Tregs, noticeably similar to that of Mel-irAE Tregs, with leukocyte activation, cytokine production, immune effector process, and response to stress being among the top enriched pathways (Fig. 3B). GSEA analysis further confirmed a destabilized inflammatory phenotype in CA-irAE Tregs that was similar to that seen in Mel-irAE Tregs. In detail, IFNγ response (through IRF1, IRF2, IRF7, IRF9, STAT1, STAT2, CCL5, CXCL9, CXCL10, and CD274 genes), IFNα response (with top enriched genes such as IRF1, IRF2, IRF7, IRF9, STAT2, CXCL10, IFITs, and IFIs), and IL2–STAT5 signaling (IL1R2, EOMES, BCL2L1, SOCS1, CXCL10, SOCS2, and TNFSF10) are among the most enriched pathways (Fig. 3C). Noticeably, among the top activated upstream regulators of circulating Tregs in CA-irAEs were IFNγ, IFNα, IL17A, TBX21, NF-kB, HIF1A, IRF1, RUNX1, IL1B, TNFSF10, STAT1, STAT3, and STAT6 (Fig. 3D and Supplementary Fig. S2A). Taken together, these findings identify a distinct proinflammatory transcriptomic profile in Tregs isolated from individuals developing irAEs that is common across different types of cancer.

Because peripheral blood remains the most accessible tissue and development of irAEs occurs throughout the body, we reasoned that identification of signatures in circulating Tregs could characterize the development of irAEs irrespective of cancer type. For this reason, we compared gene expression in Mel-irAEs, CA-irAEs, autoimmune, and healthy Tregs (Fig. 4A and Supplementary Fig. S2B). Comparison of gene sets among the three disease settings revealed 93 common DEGs (1.8% overlap) comprising a prominent core signature of 19 genes involved in Treg immune-related processes and regulation of inflammatory responses (including SOCS1, SOCS3, IFNG, RORC, IL17RE, ARG1, AREG, and SEMA6B; Fig. 4A). Pathway analysis illustrated that the 93 shared DEGs were mostly involved in inflammatory processes, with IFNγ-mediated responses, immune activation, and response to stress being highly enriched (Fig. 4B). Collectively, our findings reveal a universal Treg inflammatory signature across irAEs in diverse cancer types that is shared with autoimmune Tregs, suggesting that targeting of this signature specifically in Tregs might hold the promise for designing novel therapies to manage irAEs as well as for development of liquid “biopsies.”

Herein, we have described an intense inflammatory transcriptional reprogramming of Tregs in the periphery of individuals with cancer developing irAEs following anti–PD-1 immunotherapy, with transcripts such as IFNG, STAT1, RORC, and STAT3 being highly enriched. This is in agreement with the notion of an increased plasticity or fragility of Tregs in diverse pathogenic settings, including autoimmunity and cancer (13, 18). In addition, Tregs from individuals with irAEs demonstrated a metabolic rewiring, which is consistent with the Treg dysfunction observed in individuals with autoimmune diseases (16). The acquired phenotype of Tregs supports the notion that tolerance is disturbed upon ICI immunotherapy in individuals with solid tumors who develop irAEs. The question that remains to be answered is why some individuals develop irAEs while others do not. Development of irAEs may occur through similar mechanisms to those that lead to autoimmune diseases, where the balance between effector and regulatory immune responses is driven by diverse parameters such as genetic and environmental factors. Thus, expression of specific HLA alleles, gene polymorphisms, and environmental factors, such as viral infections, in combination with Treg-mediated breakdown of tolerance, could be decisive in determining whether ICI immunotherapy leads to the development of irAEs. The notion of the autoimmune nature of irAEs is supported by the observation that TNF blockade, an established therapy in multiple autoimmune diseases, leads to amelioration of irAE symptoms in experimental mice (19). To this end, a comprehensive analysis of HLA alleles in patients with irAEs may provide solid evidence for genetic susceptibility to developing irAEs upon ICI immunotherapy.

Although Treg accumulation in the TME is a hallmark of tumors, and in several cases correlates with the aggressiveness of tumor cells and impaired responses to immunotherapy, development of irAEs could be mediated by deregulated immune responses in the peripheral lymphoid compartments. Thus, analysis of blood could shed light on the pathogenic mechanism of irAEs. Given that ICI immunotherapy is administered i.v., Tregs, which abundantly express checkpoint molecules (i.e., PD-1 and CTLA-4; ref. 13), will be targeted systematically, potentially leading to imbalance of peripheral tolerance. Our findings highlight reprogramming of peripheral Tregs upon ICI immunotherapy in individuals with irAEs toward an inflammatory signature. Together, our data suggest peripheral immune responses should be further explored in the effort to map the mechanisms contributing to irAEs development. Identification of mechanisms leading to breakdown of tolerance and irAEs development in ICI-treated patients with cancer will deepen our understanding of immune deregulation in cancer and contribute to efficient management of irAEs. It may also provide tools for stratification of patients with increased susceptibility to irAEs.

A. Bamias reports personal fees from Roche and MSD; grants and personal fees from AstraZeneca and Bristol Myers Squibb; and grants from Pfizer outside the submitted work. D.T. Boumpas reports grants, personal fees, and nonfinancial support from European Research Council - H2020 EU and Horizon 2020 European Union during the conduct of the study. H. Gogas reports grants from Bristol Myers Squibb, Pfizer, and Novartis and personal fees from Pierre Fabre and MSD outside the submitted work. T. Alissafi reports grants, personal fees, and nonfinancial support from European Research Council - H2020 EU and Hellenic Foundation for Research and Innovation, and personal fees and nonfinancial support from Horizon 2020 European Union. P. Verginis reports personal fees and nonfinancial support from European Research Council - H2020 EU and Horizon 2020 European Union during the conduct of the study. No disclosures were reported by the other authors.

M. Grigoriou: Data curation, formal analysis, investigation, visualization, methodology, writing–review and editing. A. Banos: Formal analysis. A. Hatzioannou: Formal analysis. A. Kloetgen: Data curation, software, visualization. P. Kouzis: Resources. D. Aggouraki: Resources. R. Zakopoulou: Resources. G. Bamias: Resources. E. Kassi: Resources. D. Mavroudis: Resources. A. Bamias: Resources. D.T. Boumpas: Resources, funding acquisition, writing–review and editing. A. Tsirigos: Software, writing–review and editing. H. Gogas: Resources, writing–review and editing. T. Alissafi: Conceptualization, formal analysis, supervision, funding acquisition, investigation, writing–original draft, project administration, writing–review and editing. P. Verginis: Conceptualization, formal analysis, supervision, funding acquisition, investigation, writing–original draft, project administration, writing–review and editing.

The authors thank Giannis Vatsellas for assisting with RNA-seq and for providing technical advice; Anastasia Filia for providing bioinformatics support; Nikos Malissovas for providing technical assistance; Antigoni Pieta for assisting with patient sample collection; Anastasia Apostolidou for assisting with flow cytometry; and Dora Togia for financial and administrative management. This research project has received funding from the European Research Council (ERC) under the European Union's Horizon 2020 Research and Innovation program (grant agreement number 947975) and from the Hellenic Foundation for Research and Innovation (H.F.R.I.) under the “2nd Call for H.F.R.I. Research Projects to Support Post-Doctoral Researchers” (Project Number: 166) to T. Alissafi. D.T. Boumpas, T. Alissafi, and P. Verginis were supported by the ERC under the European Union's Horizon 2020 Research and Innovation program (grant agreement number 742390). D.T. Boumpas, T. Alissafi, and P. Verginis were supported by the European Union's Horizon 2020 Research and Innovation program under grant agreement number 733100.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.