Introduction

Peripheral T-cell lymphoma (PTCL) accounts for 25% of non-Hodgkin’s lymphoma (NHL) patients in China and about 10% in the USA and Europe.1 2 The most common subtypes of PTCL in China include extranodal NK/T-cell lymphoma (ENKTL), PTCL not otherwise specified (PTCL-NOS), anaplastic large cell lymphoma (ALCL), and angioimmunoblastic T-cell lymphoma (AITL).3 Currently, chemotherapy remains the primary and first-line treatment for most PTCL subtypes, including anthracycline-based chemotherapy such as CHOP (cyclophosphamide, doxorubicin, vincristine, and prednisone) and CHOP-like regimens. However, the response rate to this regimen is low and is accompanied by rapid relapse in some subtypes.4 5 According to the data disclosed by the International T-Cell Lymphoma Project involving 1314 patients, the 5-year survival rate for PTCL patients varies according to PTCL subtypes, with PTCL-NOS and AITL at 32%, ALK-negative ALCL at 49%, and adult T-cell leukemia/lymphoma (ATLL) at 14%.2 Overall, the 5-year overall survival (OS) rate for PTCL is significantly lower than that for B-cell NHL (41% vs 53%; p=0.0004).6 To improve the limited therapeutic effectiveness and poor prognosis of PTCL, phase I and II clinical trials have been initiated to explore new treatment options, and the initial results have shown promising outcomes, such as the use of anti-PD-1 immunotherapy.

Anti-PD-1 therapy has revolutionized the treatment of solid tumors and lymphomas such as Hodgkin’s lymphoma. Two previous phase II studies have consistently illustrated the modest single-agent efficacy of anti-PD-1 antibodies in the treatment of relapsed/refractory (r/r) PTCL. The overall response rate (ORR) was around 33%; the median duration of response (DoR) ranged from 2.9 to 3.6 months in two small cohorts of PTCL patients (n=18 and 12, respectively).7 8 However, earlier studies had limited sample sizes and a restricted subtype representation of PTCL. Thus, larger queues are necessary to cover the major subtypes of PTCL. Recently, an open-label phase II study was conducted to evaluate the efficacy of geptanolimab in 89 patients with r/r PTCL. Of the 89 patients, the most common histological subtypes were PTCL-NOS (n=28 (31.5%)) and ENKTL (n=19 (21.3%)), which represent the majority of PTCL patients in China, the USA, and Europe.1 9 The study reported an ORR of 40.4% (95% CI 30.2% to 51.4%) and a median DoR of up to 11.4 months (95% CI 4.8 to not reached),10 exhibiting promising therapeutic activity and an acceptable safety profile in r/r PTCL patients.

In addition to programmed death-ligand 1 (PD-L1) and tumor mutation burden (TMB), genetic mutations are potent predictive factors for the therapeutic effectiveness of anti-PD-1 treatment by affecting the mutational load, tumor microenvironment, PD-L1 expression, and antigen presentation. Alterations in oncogenes/tumor suppressors, such as epidermal growth factor receptor (EGFR) mutations, are correlated with poor outcomes11 due to reduced infiltration of CD8+ T cells in the tumor microenvironment,12 and TP53 mutations are predictors of a good response to immune checkpoint inhibitor (ICI) treatment due to elevated TMB caused by TP53 mutation.13 Genetic alterations in NF-kB signaling have been associated with an inflammatory environment in solid tumors and lymphomas.14–16 Structural variations (SVs) in the 9p24.1 region and mutations in genes located within the 9p24.1 amplicon, such as JAK2, are associated with increased PD-L1 and/or programmed death-ligand 2 (PD-L2) protein expression.17 18 Loss-of-function mutations in human leukocyte antigen (HLA) class I and II molecules, such as mutations in β2-microglobulin (B2M) and class II transactivator (CIITA), can affect antigen presentation and treatment efficacy.14 19 However, genomic studies on the effectiveness of immunotherapy in PTCL have not been systematically conducted, which hinders the application of immunotherapy in PTCL. Comprehensive next-generation sequencing (NGS) of tumor samples aids in exploring biological alterations and candidates for the development of subsequent combination treatment regimens. To investigate the genomic features influencing the efficacy of anti-PD-1 therapy, we depicted the mutational landscape linked to anti-PD-1 treatment response in 109 patients with PTCL and revealed potential resistance mechanisms through in vitro experiments of candidate genetic mutations.

Materials and methodsPatients and sample collection

A total of 109 patients diagnosed with PTCL were recruited from multiple medical institutions from July 12, 2018 to August 15, 2019, across China, including Jiangsu Cancer Hospital, Linyi Cancer Hospital and Cancer Center, Union Hospital, Tongji Medical College, and Huazhong University of Science and Technology. Patients with at least one bidimensionally measurable lesion, as defined by the Lugano classification,20 were included, and formalin-fixed paraffin-embedded (FFPE) tumor tissues were obtained and confirmed by pathologists for diagnosis and tumor purity. Among 109 patients, 89 underwent anti-PD-1 (geptanolimab) therapy after experiencing at least one prior systemic therapy failure in a phase II clinical trial (NCT03502629). These 89 patients were categorized into 3 subgroups, 36 of which showed durable clinical benefit (DCB; complete or partial regression), 45 exhibited no durable benefit (NDB; stable or progressive disease), and 8 were considered unassessable as evaluated by the independent radiological review committee according to the Lugano classification.20 Concurrently, targeted sequencing analysis was performed on tumor tissues from 73 patients collected prior to the initiation of anti-PD-1 therapy, as previously described.10 Patients’ characteristics are listed in table 1. Written consent was obtained from all patients according to the ethical regulations of each participating hospital. This study was conducted in accordance with the principles of the Declaration of Helsinki.

Table 1

The demographic and clinical characteristics of enrolled patients

Targeted NGS and data processing

The standard protocol for sequencing and data analysis has been previously described.10 Genomic DNA extracted from FFPE or peripheral blood mononuclear cells (PBMC) was amplified using four pools of primer pairs that target the coding regions of 440 genes relevant to cancer diagnosis, treatment, and prognosis . Amplicons were ligated with barcoded adaptors using an Ion AmpliSeq Library Kit 2.0 (Thermo Fisher Scientific). The quality and quantity of the amplified library were assessed using a 2100 Bioanalyzer (Agilent Technologies) and Qubit (Invitrogen). Emulsion PCR was employed for barcoded libraries conjugated with sequencing beads and enriched using the Ion Chef system (Thermo Fisher Scientific), followed by sequencing on an Ion GeneStudio S5 sequencer (Thermo Fisher Scientific).

The raw reads were mapped onto the hg19 reference genome using Ion Torrent Suite (V.5.10). The Torrent Coverage Analysis plugin calculates the coverage depth. The Torrent Variant Caller plugin (V.5.10) was employed for the identification of single-nucleotide variants and short insertions or deletions (INDELs). Variant eEffect predictor (VEP) annotated variants of the VEP (V.88) using the databases from COSMIC V.86 and the Genome Aggregation Database r2.0.2. Variants with coverage of 25 or above, allele frequency of 5% or above, and actionable variants with allele frequency of 2% or above were maintained.

TMB analysis

The TMB computation employed the sequenced regions of ACTOnco+ to approximate the quantity of somatic nonsynonymous mutations per megabase in all protein-coding genes. This calculation applied a machine learning model that considered all somatic variants with cancer hotspot correction, as previously described.10 TMB is commonly expressed as mutations per megabase (muts/Mb).

Cell culture and reagents

Jurkat cells were cultured in RPMI-1640 medium supplemented with 10% fetal bovine serum (FBS) and 1% penicillin/streptomycin. The cell line was authenticated by STR analysis. All the cells were incubated in a humidified incubator at 37°C with 5% CO2. Tofacitinib was purchased from MedChemExpress.

Construction of knockout cell lines and stable cell lines

YV-Cas-LV002 and YKO-LV001 single-gRNA plasmid vectors were transfected into 293 T cells, and the transfection medium was subsequently replaced with a fresh complete medium (Ubigene) 5 hours later. After 48–72 hours, the supernatant was centrifuged and filtered to collect lentivirus. Jurkat cells were then exposed to lentiviruses harboring Cas9 and JAK3-gRNA along with polybrene (Ubigene). 24 hours later, JAK3_01-G1-knockout (KO) cells were selected using a medium containing puromycin (2 µg/mL) and hygromycin (500 µg/mL), and the monoclonal strains were verified by RT-qPCR. The same procedure was replicated to construct carriers and package lentiviruses carrying wild-type JAK3, mutant JAK3 (p.E958K, p.A573V, p.M511I), and empty vectors. Transfections were performed in KO cells to establish stable JAK3 expression cell lines (Yuanjing Biotechnology).

Co-immunoprecipitation assay

Co-immunoprecipitation (Co-IP) was conducted as described previously21 (The deubiquitinase USP11 ameliorates intervertebral disc degeneration by regulating oxidative stress-induced ferroptosis via deubiquitinating and stabilizing Sirt3). Jurkat cells were centrifuged at 500×g for 10 min at 4°C and washed twice with 1×phosphate-buffered saline (PBS). The cells were lysed using lysis buffer and protease inhibitors. Centrifuge at 12,000–16,000×g for 10 min at 4°C to collect the supernatant and incubate with STAT3 antibody (7 µg; catalog no. 9139; Cell Signaling Technology) on a rocking mixer at 4°C for 4–6 hours to form antigen-antibody complexes. The antigen-antibody complexes were then added to the prepared Protein A/G magnetic beads, incubated on a rocking mixer at 4℃ for 4–6 hours, and washed three times. The washed antigen-antibody-bead complexes were denatured with 100 µL of 5×SDS PAGE loading buffer and subjected to SDS-PAGE analysis.

Cytomembrane and cytoplasmic protein extraction

Cytomembrane and cytoplasmic proteins were extracted using the Mem-PER Plus Membrane Protein Extraction Kit (catalog no. 89842, Thermo Scientific) according to the manufacturer’s instructions. Briefly, the cells were centrifuged at 300× g for 5 min to collect cell pellets. Add 0.75 mL permeabilization to the cell pellet and mix gently at 4°C for 10 min. Permeabilized cells were centrifuged at 16000× g for 15 min, and the supernatant containing cytoplasmic proteins was transferred to a new centrifuge tube. The pellet was incubated with lysis buffer, and the supernatant containing soluble membrane proteins and membrane-associated proteins was collected in a new centrifuge tube.

Western blot

Cell pellets were resuspended and lysed in RIPA lysis buffer containing 1% protein phosphatase inhibitor and 2% protease inhibitor. After centrifugation and protein quantification, 25–40 µg of total protein were loaded into the well of 10% BIS-Tris Gels, electrophoresed at 150V for 0.5–1 hour, then transferred onto PVDF membranes. The membranes were blocked with a solution of 5% BSA/nonfat-dried milk dissolved in TBST for 1 hour. The membranes were then incubated with primary antibodies specific for JAK3 (1:1000, 8827S, Cell Signaling Technology), p-JAK3 (1:1000, 5031S, Cell Signaling Technology), STAT3 (1:1000, 4904T, Cell Signaling Technology), p-STAT3 (1:1000, 9145, Cell Signaling Technology), PD-L1 (1:500, 13684S, Cell Signaling Technology), PI3K (1:1000, 4249, Cell Signaling Technology), MAPK (1:1000, 4685, Cell Signaling Technology), p-MAPK (1:1,000, 4370T, Cell Signaling Technology), GADPH (1:1000, 2118S, Cell Signaling Technology) overnight at 4 °C. After washing the membrane with TBST, secondary antibodies were added and incubated at room temperature for 1 hour. Following another round of washing, super ECL plus (Applygen) was applied to develop postexposure images.

Growth inhibition assay

Suspension cells (1×105 cells/well) were seeded into 96-well plates (Corning), and the gradient concentrations of the JAK3 inhibitor Tofacitinib were added to the wells, and each concentration was repeated three times. After 48 hours of treatment, cell viability was assessed using the Cell Counting Kit (CCK)-8 assay (Dojindo Laboratories), and the optical density (OD) at 450 nm was determined to calculate the cell growth inhibition rates or cell viability.

PD-L1 immunohistochemistry

The expression of PD-L1 was assessed in FFPE tumor biopsy samples using an anti-PD-L1 rabbit monoclonal antibody (SP263; Roche) and OptiView DAB immunohistochemistry (IHC) detection kit (Roche) as previously described.10 PD-L1 expression was quantified based on the estimated percentage of cells stained PD-L1 positively.

Flow cytometry for the detection of cell-surface PD-L1

For measurement of PD-L1 expression, Jurkat cells were stained with FITC-conjugated anti-human CD274 antibody (1:100, 393605, Biolegend) or isotype-control antibodies for 30 min at 4℃. The stained cells were analyzed by BD FACSDiva software on flow cytometer (BD, LSRFortessa). Data were analyzed using FlowJo Software.

BA/F3 syngeneic tumor models

4-week-old male C3H mice were acquired from Weitong-Lihua Company (Beijing, China) and maintained under specific pathogen-free conditions. To construct the BA/F3-derived syngeneic tumor model, a suspension containing 0.1 mL of 6×10⁶ BA/F3 cells mixed with 0.1 mL matrigel was subcutaneously inoculated into the flank region. When the syngeneic tumor volumes reached 80–150 mm³, mice were randomly allocated into nine experimental groups (n=7 per group) for intraperitoneal (i.p.) drug administration as follows: (1) Control group: 200 µL PBS; (2) anti-PD-1 antibody group: anti-PD-1 antibody (2 mg/kg) alone, delivered every 3 days; (3) Tofacitinib group: Tofacitinib (50 mg/kg) alone, administered every 3 days. Tumor volumes and body weights were measured every 3 days. Mice were euthanized by CO2 inhalation when tumors reached a diameter of 15 mm in the control group or presented with ulceration at the tumor site.

Single-cell sequencing analysis

To ensure data quality during the preprocessing phase, low-quality cells with fewer than 500 detected genes and those with over 25% mitochondrial counts were filtered out. Single-cell transcriptomic profiles were then processed through the standardized analytical workflow implemented in the Seurat package (V.4.0). Specifically, gene expression matrices were normalized through the ScaleData function. Principal component analysis was conducted on the normalized dataset, extracting the first 20 principal components. Uniform Manifold Approximation and Projection (UMAP) was generated using the RunUMAP algorithm (Seurat implementation).

Statistical analysis

Intergroup comparisons were conducted using Student’s t-test, Mann-Whitney-Wilcoxon test, or Kruskal-Wallis test. Correlations between categorical variables were calculated using Fisher’s exact test or χ2 test. Survival curves were drawn using the Kaplan-Meier method, and the log-rank test was used for survival analysis. The threshold was set at p<0.05. Proliferation curves were plotted using GraphPad Prism (V.9.0) software, and the inhibitory concentration (IC50) was determined by non-linear regression analysis. Western blot bands were quantified using ImageJ software. All statistical analyses used R (V.4.2.2), GraphPad Prism (V.9.0), and Image J software (V.1.8.0).

ResultsPatient characteristics

Among the 89 PTCL patients treated with geptanolimab, 36 (40.4%) were classified as responders, 45 (50.6%) as non-responders, and 8 (9.0%) as unassessable. The pathological subtypes included 30 (33.7%) cases of PTCL-NOS, 21.3% ENKTL (n=19), and 15.7% ALK-ALCL (n=14), which comprised the major PTCL subtypes in China and the USA.9 All the patients had previously undergone systematic treatment. The majority of patients with ENKTL had received prior asparaginase-based chemotherapy and other subtypes of PTCL, excluding ENKTL, who were mainly treated with anthracycline-containing chemotherapy. The median follow-up for progression-free survival (PFS) and OS was 2.63 and 12.12 months, respectively. 44 patients’ tumor tissues were available before treatment with geptanolimab.

The median PD-L1 expression was 60%, with an standard deviation (SD) of 30.03%. ENKTL (median, 85%; range, 0%–98%) had the highest median PD-L1 expression level, which was significantly higher than that of PTCL-NOS (median, 40%; range, 2%–95%), followed by ALK+ALCL (median, 72.5%; range, 30%–98%) and AITL (median, 47.5%; range, 42%–75%) (figure 1A). Patients with high PD-L1 expression levels (≥50%) were reported to obtain more benefit from geptanolimab treatment than those with low levels (median PFS 6.2 vs 1.5 months, p=0.002).10 Similarly, PD-L1 was significantly elevated (Mann-Whitney-Wilcoxon (MWW), p=0.0061) in patients with DCB compared with NDB patients in this study (figure 1B). Moreover, we leveraged an independent cohort of ENKTL treated with anti-PD-1 antibody to assess the relationship between PD-L1 expression and clinical outcome. The pretreatment levels of PD-L1 were significantly higher in DCB compared with NDB patients who have received anti-PD-1 therapy (p=0.004, online supplemental figure S1A). PD-L1 exhibited marginal significance among distinct subtypes (Kruskal-Wallis, p=0.055; figure 1A) and was still a predictor for PFS (HR: 0.99, 95% CI: 0.98 to 1.00, p=0.005) when adjusting for PTCL subtypes in the multivariate Cox regression algorithm (figure 1C and online supplemental figure S1B), indicating the strong and relatively independent predictive capability of PD-L1. Meanwhile, a negative correlation was observed between PD-L1 expression and the tumor/stromal cell ratio (Pearson’s correlation coefficient=−0.2, p=0.18; online supplemental figure S1C), indicating that PD-L1 expression may affect the tumor environment. Though ENKTL with the highest PD-L1 level had the longest PFS and the highest proportion of DCB (67%), AITL exhibited the longest OS and a relatively high proportion of DCB (50%) (figure 1C and online supplemental figure S1D), suggesting the existence of other underlying molecular mechanisms that may contribute to immunotherapy efficacy other than PD-L1.

Figure 1

The prognostic value of PD-L1 and tumor mutation burden (TMB) in peripheral T-cell lymphoma (PTCL) patients treated with geptanolimab. (A and D) Comparison of PD-L1 positivity percentage and TMB levels among various PTCL subtypes. Data are represented as the mean±SD. P value was calculated using the Mann-Whitney-Wilcoxon and Kruskal–Wallis tests. *p<0.05; **p<0.01. (B) Boxplot of the PD-L1 positive ratio in the no durable benefit (NDB) and durable clinical benefit (DCB) groups (NDB, n=28; DCB, n=53). The two-sided Mann-Whitney-Wilcoxon test was used. The centerline represents the median, and the box bounds represent the IQR. Whiskers span a 1.5-fold IQR. The box limits indicate the IQR (25th–75th percentile), with a centerline representing the median. *p< 0.05, **p< 0.01. (C) Kaplan-Meier models of progression-free survival (PFS) and overall survival (OS) for distinct PTCL subtypes. The p value was calculated using the log-rank test. (E) TMB is a significant predictor in the multivariate logistic regression adjusted for PTCL subtypes.

Although we did not find a direct association between TMB and treatment response (MWW, p=0.44; online supplemental figure S1E), the result showed that TMB was significantly different among various PTCL subtypes (Kruskal-Wallis, p=0.024; figure 1D). The TMB value in mycosis fungoides (MF) was higher than that in ALK-ALCL and PTCL-NOS (mean of 12.3 vs 1.2 and 2.6, non-synonymous mutations; MWW, p<0.05), whereas the ORR in MF was conversely lower (0% vs 62% and 29%; Fisher’s exact test, p<0.001; online supplemental figure S1A). Patients diagnosed with the MF subtype had poor PFS and OS (figure 1C). In multivariate Cox regression analysis adjusted for PTCL subtype, TMB was a significant biomarker (p=0.05; figure 1E).

The mutational landscape of PTCL

We carried out deep targeted capture sequencing that covered 440 cancer-related genes in 73 patients with PTCL paired with normal blood samples. The median sequencing depth was 1161× (range: 90–4000×). Among the 73 PTCL patients, 27 (37.0%) had PTCL-NOS, 21.9% had ENKTL (n=16), and 13.7% had ALK-ALCL (n=10). The most frequent alterations were TET2 (12%, n=9), followed by TP53 (11%, n=8), FAS (10%, n=7), KMT2C (10%, n=7), KMT2D (8%, n=6), JAK3 (7%, n=5), and SETD2 (7%, n=5) (figure 2A). TET2 and FAS mutations were scattered across the genes to which they belonged, whereas TP53 mutations mainly occurred in the DNA binding and tetramerization domains (online supplemental figure S2A). TET2 mutations were frequently detected in PTCL-NOS (3/8), suggesting that PTCL-NOS is mainly driven by the epigenetic genes TET2 (11%, n=3) and KMT2C (11%, n=3; figure 2A). Similarly, TP53 mutations mainly occurred in ALK-ALCL (3/8), a subtype driven by TP53 mutations (30%, n=3; figure 2A). In contrast to the former two genes, FAS mutations mainly occurred in ENKTL (3/7), although ENKTL was not merely driven by FAS mutations (figure 2A). Interestingly, JAK/STAT pathway alterations were also vital in ENKTL, with 60% of JAK3 mutations and 100% of STAT3 mutations occurring in this subtype (figure 2A). Previous studies have also demonstrated that activating STAT3 mutations are common in ENKTL patients from the USA and Asia.22 23 Changes in DNA methylation have been implicated in the pathogenesis of T-cell lymphomas. Consistently, mutations in the DNA methylation regulator TET2 (12%, n=9) were the most frequent genetic alterations observed. In contrast, other gene mutations involved in epigenetic signaling, such as KMT2C (10%, n=7), KMT2D (8%, n=6), SETD2 (7%, n=5), and ARID1A (5%, n=4), were also common in our cohort. Except for the epigenetic/chromatin remodeling pathway (38%, n=28), the most common mutations were involved in the MAPK/Ras (14%, n=10), PI3K/AKT/mTOR (14%, n=10), and Wnt pathways (8%, n=6; figure 2A).

Figure 2

Mutational landscape of PTCL. (A) Mutation plot of genes of which the frequency is noted on the right in pretreatment patients. The type of genetic alteration (missense, splice, in-frame, nonsense, frameshift) and PTCL subtype are described in the legend, and the alterations present in ≥3% of cases were involved in the figure. (B) TP53 and FAS alteration on TMB level and PD-L1 expression. A two-sided student’s t-test was used. PTCL, peripheral T-cell lymphoma; TMB, tumor mutation burden.

Alterations in TP53 have been reported to be associated with decreased expression of gene sets related to innate or adaptive immune cells in prostate cancer.24 In PTCL patients, we observed TP53 mutations that could lead to higher TMB and lower PD-L1 expression levels (MWW, p=0.017 and 0.0026, respectively; figure 2B). TP53 mutations were not statistically associated with patients’ survival after treatment with the anti-PD-1 regimen (p=0.103 and 0.705, respectively; online supplemental figure S2B). Similarly, the mutant FAS gene, which mediates the deregulation of extrinsic apoptotic pathways, led to higher TMB in this study, but no significant prognostic value was observed (figure 2B, online supplemental figure S2B).

Impact of distinct mutational patterns on clinical outcome

Two mutational signatures were identified: signature 1 (SBS32) and signature 2 (SBS5). In our cohort, 55% of PTCL patients were determined to have signature 1, which was attributed to immunosuppression induced by prior treatment with azathioprine (figure 3A). Signature 1 had a high cosine similarity with SBS7a (0.502) and SBS2 (0.501) COSMIC signatures (online supplemental figure S3A). SBS2 is associated with APOBEC Cytidine Deaminase (C>T), and SBS7a is commonly found in tumors that are likely to be caused by UV radiation (C>T-TC/CC). Considering the correlation between signature 1 and APOBEC cytidine deaminase activity, we enriched APOBEC-related mutagenesis patterns, as previously described.5 In total, 13.5% (n=8) of the PTCL samples had an enriched APOBEC-related mutation signature (figure 3B), further supporting the role of APOBEC-mediated mutagenesis in PTCL carcinogenesis. Interestingly, patients with an APOBEC-enriched signature had significantly worse OS than those without APOBEC-lacked patients (p=0.031, figure 3C).

Figure 3

Distinct mutational patterns and their effect on clinical outcomes. The mutational spectrum of the two mutational signatures, denoted as signature 1 (SBS32) and signature 2 (SBS5), correspond to COSMIC signatures with similarities of 0.729 and 0.625, respectively. (A) The APOBEC enrichment score is shown on the y-axis; significantly enriched samples are labeled as red circles and non-enriched patients are presented as blue circles. 24 PTCL samples were enriched in signature 1 (yellow bars) and 20 showed enrichment in signature 2 (purple bars). (B) The Kaplan-Meier plot of OS shows that APOBEC-enriched (colored red) and non-APOBEC-enriched (colored blue) patients have significantly different prognoses. (C) Kaplan-Meier model of OS based on DNA VAF levels. Patients with high VAF values had significantly shorter OS than those with low VAF values. The p value was calculated using the log-rank test. OS, overall survival; PTCL, peripheral T-cell lymphoma; VAF, variant allele frequency.

To investigate whether the DNA variant allele fraction (VAF) could be used to predict the prognosis of immunotherapy, as previously reported,25 we divided the DNA-positive patients into two groups based on the mean DNA VAF level (high, n=15; and low, n=30). Patients with a higher baseline DNA VAF had shorter PFS (p=0.21, online supplemental figure S3B) and significantly poorer OS than those with a lower VAF (p=0.036, figure 3D).

JAK3 and EZH2 mutations lead to worse survival

Somatic mutations in specific genes, such as EGFR, may influence the efficacy and response to immune checkpoint therapy, such as EGFR mutations. In our study, a distinct baseline mutational landscape was observed between patients with DCB and NDB. Patients with DCB and NDB were both involved in the RTK-RAS and NOTCH pathways, while the specific genetic profiles were different (online supplemental figure S4A). In particular, SETD2, JAK3, and EZH2 mutations only occurred in NDB patients (11% vs 0%; figure 4A), whereas SYNE1 (14% vs 0%), EP300 (10% vs 0%), and SMARCA4 (10% vs 0%) were mainly enriched in DCB patients (figure 4A). Among these alterations, JAK3 and EZH2 mutations correlated with lower PD-L1 expression (Student’s t-test, p<0.001 and p<0.05, respectively; figure 4B). Consistently, the alternations of JAK3 (HR, 6.07; p=0.014) and EZH2 (HR, 4.76; p=0.027) led to significantly worse PFS than in wild-type patients (figure 4C). EZH2 mutations were also correlated with worse OS (HR, 7.46; p=0.004; figure 4C), whereas JAK3 mutations were not, probably because of the distinctly elevated TMB in JAK3-mutant patients (MWW, p<0.05; figure 4B), which was not observed in patients with EZH2 mutations (MWW, p=0.079; online supplemental figure S4C). Multivariable adjusted Cox proportional hazards analyses adjusted for sex, age, PTCL subtypes, and stage confirmed 15 and 17 genetic alterations as independent predictors of PFS and OS, respectively (online supplemental tables S1 and S2). Among these mutations, JAK3 (p=0.004) and ADAMTSL1 (p=0.018) mutations, which differed between DCB and NDB patients, were independent indicators of PFS (online supplemental tables S1 and S2). Additionally, mutations in ADAMTSL1 were associated with higher PD-L1 levels (MWW, p=0.088; online supplemental figure S4B). In contrast to the aforementioned genes, mutations in EP300 (Student’s t-test, p<0.001; figure 4B) and HNF1A (Student’s t-test, p=0.073; online supplemental figure S4B) were associated with higher levels of PD-L1 expression. Patients with EP300 mutations had better OS (HR<0.001; p=0.217) and PFS (HR<0.001; p=0.107) than those with wild-type genes (online supplemental figure S4D).

Figure 4

JAK3 activating mutations led to decreased PD-L1 expression. (A) Comparison of somatic mutation frequencies between patients with NDB (n=25) and DCB (n=24). Genes with alterations present in ≥5 cases of the cohort are shown. (B) Boxplots of JAK3, EZH2, and EP300 mutations on PD-L1 expression. Alterations in JAK3 also lead to higher TMB levels. The centerline represents the median. The upper and bottom of the box represent the IQR, and whiskers span 1.5-fold the IQR. *p< 0.05, ***p< 0.001. (C) The prognostic value of JAK3 and EZH2 alterations by Kaplan-Meier analysis. The p value was calculated using the log-rank test. (D) Following cell type annotation analysis, the single-cell RNA sequencing dataset was stratified into seven distinct cellular subpopulations: granulocytes, erythrocytes, monocytes, NK cells, T cells, fibroblasts and endothelial cells. (E) The percentage and number of T cells in each group. (F) and (G) Distribution of PD-L1 in single-cell dataset. DCB, durable clinical benefit; NDB, no durable benefit; TMB, tumor mutation burden.

JAK3A573V and JAK3M511I activating mutations inhibit PD-L1 expression and suppress the infiltration of T cells

JAK3 mutations are relatively common in PTCL26 27 and contribute to PTCL pathogenesis by influencing the JAK/STAT pathway.28 This disorder has been recognized as a therapeutic target with potential implications for immunotherapy.29 30 Previous studies have demonstrated that somatic and germline amino acid variants in the JAK3 gene can promote PD-L1 induction, leading to long-term benefits.31 In our cohort, we observed mutations in JAK3 were negatively correlated with PD-L1 expression. Meanwhile, patients harboring mutant JAK3 did not have mutations in alternative components of the JAK/STAT pathway, such as JAK1, JAK2, or STAT3. The JAK3 p.M511I mutation occurs between the JAK homology domain (JH)2 and JH3 domains located on the protein surface, which may potentially impact the structure of JAK3 and its downstream proteins (online supplemental figure S5). The JAK3 p.A573V mutation occurs in the JH2 domain, which is a kinase-like or “pseudo-kinase” domain (online supplemental figure S5). Although this kinase domain lacks catalytic activity, it plays a regulatory role in the activity of JH1, the domain used to encode kinase proteins. Mutations in this domain often lead to altered JAK kinase activity. The p.E958K mutation occurred in the JH1 domain, which may have affected the activity of the kinase protein (online supplemental figure S5). Among the three types of mutations, we observed that the JAK3 p.A573V mutation could lead to reduced PD-L1 expression compared with JAK3 wild-type patients (44.91 vs 66.99, p=0.16) in lymphoma patients (n=256) collected from the TCGA database (online supplemental figure S6).

To explore the functional consequences of these mutations, we constructed expression vectors for the wild-type, p.A573V, p.M511I, and p.E958K variants of JAK3. The proliferation of the Jurkat cell line was reduced by the p.M511I mutant compared with the p.A573V and p.E958K mutants (p<0.001) (online supplemental figure S7A). These cellular effects were consistent with inactivation of the PI3K and MAPK signaling pathways, characterized by diminished levels of PI3K and p-MAPK (online supplemental figure S7B). However, in BA/F3 cells, p.M511I and p.A573V mutant cells grew faster than wild-type cells (p<0.001) (figure 5A), and the tumor formation rate of BA/F3 cells harboring p.M511I and p.A573V mutants was higher than those with the wild-type vector (figure 5B). Compared with JAK3 WT, p.A573V and p.M511I variants showed relatively higher p-JAK3/JAK3 levels than the wild-type (figure 5C).

Figure 5

JAK3 mutations decreased PD-L1 expression through STAT3 inactivation. (A) Viability assays of BA/F3 cells expressing empty vector, JAK3 WT, p.E958K, p.A573V, or p.M511I expression vectors for up to 72 hours. **(B) Tumor weights and growth curves of JAK3 wild-type, JAK3, A573V and JAK3 M511I mutant mice models. ***p<0.001, *(C) Jurkat cells with knocked-out JAK3 were transferred into empty vector, wild-type vector, p.E958K, p.A573V, or p.M511I expression vectors. Western blotting was performed to analyze the expression of JAK3, p-JAK3, STAT3, p-STAT3, and PD-L1. (D) Jurkat and BA/F3 cell lines with knocked-out JAK3 were transferred into empty, wild-type, p.E958K, p.A573V, or p.M511I expression vectors. PD-L1 protein levels in Jurkat and BA/F3 cell lines were detected using western blotting. (E) Surface PD-L1 expression was determined by flow cytometry. (F) Co-immunoprecipitation and immunoblotting were performed using the indicated antibodies. (G) Cell viability assays with Tofacitinib (50, 100, 150, and 200 µM) for 72 hours in empty vector, JAK3 WT, and novel JAK3 mutant Jurkat cells. Data are shown as the mean±SD of three independent experiments. (H) JAK3 mutant BA/F3 cells are sensitive to the JAK3 inhibitor Tofacitinib. **p<0.01; ***p<0.001; ****p<0.0001.

Since oncogene activation drives cancer development, we conducted BA/F3 syngeneic tumor models to elucidate the impact of JAK3 A573V and JAK3 M511I mutations on the tumor microenvironment. Single-cell RNA sequencing of mice tumor tissues revealed that seven immune cell types were annotated, including granulocytes, monocytes, and T cells (figure 4D). JAK3 p.M511I and p.A573V mutations lead to lower levels of T-cell infiltration (figure 4E) and PD-L1 expression (figure 4F,G, p<0.0001) than JAK3 wild-type tumor, significantly p.A573V mutated tumor with lowest T-cell infiltration (0.15% vs 1.38%) (figure 4E) and PD-L1 expression (0.07 vs 25.96, p<0.0001) (figure 4F,G).

JAK3A573V and JAK3M511I inhibit PD-L1 expression and mediate sensitivity to Tofacitinib

We then explored whether JAK3 mutations could affect PD-L1 expression in Jurkat and BA/F3 cell lines. Jurkat cell lines harboring JAK3 p.A573V and p.M511I mutations displayed a reduction in PD-L1 expression. In contrast, cells expressing p.E958K vectors showed a slight decrease in PD-L1 levels, but this was not significant (figure 5C and online supplemental figure S8), which is consistent with the findings in the clinical cohort (figure 4B). Likewise, in BA/F3 cells, JAK3 p.A573V and p.M511I mutations reduced the expression of PD-L1 compared with cells containing JAK3 wild-type and overexpression vectors (figure 5D). As PD-L1 is present not only on the tumor cell membrane but also in the cytoplasm, we further explored whether JAK3 mutations affect the spatial heterogeneity of PD-L1 distribution. Similarly, cytoplasmic PD-L1 levels in p.A573V and p.M511I mutant cells were slightly lower than those in wild-type cells (figure 5D). Flow cytometry was performed on live cells to detect cell surface expression of PD-L1 protein. Compared with Jurkat cells with wild-type JAK3, JAK3 mutated Jurkat cells exhibited a decrease in surface staining for PD-L1 (figure 5E). Among these, the JAK3A573V-mutated Jurkat cells exhibited the lowest surface expression of PD-L1 protein (figure 5E).

STAT3 plays a vital role in regulating the transcription of PD-L1 and is typically activated through the JAK-STAT3 pathway in PTCL.27 In our study, we detected a decrease in phosphorylated STAT3/total STAT3 protein caused by the p.A573V mutation at Tyr705 (figure 5C), with a concomitant reduction in PD-L1 expression (figure 5C). Thus, we performed a co-IP experiment to determine the relationship between STAT3 and PD-L1 expression. The results indicated that the interaction between STAT3 and PD-L1 in cells with JAK3 A573V, JAK3 E958K, and JAK3 M511I mutations led to a lack of interaction between STAT3 and PD-L1 compared with cells expressing wild-type and high-expression JAK3 protein (figure 5F). After inhibiting STAT3 phosphorylation with Stattic, the PD-L1 RNA levels in JAK3 mutant cell lines were essentially comparable, while in wild-type cell lines, the RNA levels of PD-L1 significantly decreased, indicating that the binding of STAT3 to the PD-L1 promoter region may be downregulated in JAK3 mutant cell lines (online supplemental figure S9). Finally, when Jurkat cells were treated with a gradient concentration of the JAK3 inhibitor Tofacitinib, cells with JAK3 mutations exhibited greater sensitivity to Tofacitinib compared with cells with JAK3 WT and empty vectors, especially JAK3 A573V and JAK3 M511I mutated Jurkat cells (figure 5G). Similarly, we found that JAK3 A573V (50% inhibitory concentration (IC50)=1.945 µM, 95% CI=1.447 to 2.650 µM) and JAK3 M511I mutant (IC50=0.028 µM, 95% CI=0.002 to 0.091 µM) BA/F3 cells were more sensitive to Tofacitinib while wild-type BA/F3 cells presented a higher IC50 value (IC50=3.036 µM, 95% CI=2.203 to 4.290 µM) (figure 5H), indicating that Tofacitinib may be a promising therapeutic strategy for JAK3 mutant PTCL patients.

Therapeutic potential of the JAK3 inhibitor Tofacitinib in the JAK3A573V and JAK3M511I mutated BA/F3 syngeneic mouse model

To assess whether the JAK3 inhibitor Tofacitinib could serve as a therapeutic strategy for patients with JAK3 A573V and JAK3 M511I mutated PTCL, wild-type JAK3 A573V and JAK3 M511I mutant BA/F3 cells were injected into C3H mice to construct syngeneic tumor models. These syngeneic tumor models were treated with PBS, Tofacitinib, or anti-PD-1 antibody. For anti-PD-1 antibody, tumor weights of JAK3 wild-type mice were significantly smaller compared with JAK3 A573V (0.005 g vs 0.211 g, p<0.0001, figure 6A,B) and JAK3 M511I mutant mice (0.005 g vs 0.223 g, p<0.0001, figure 6A,B). Tumor growth inhibition (TGI) rates with anti-PD-1 treatment were 91.5% for JAK3 wild-type mice, 49.7% for JAK3 A573V mutant, and 35.1% for JAK3 M511I mutant with anti-PD-1 treatment (figure 6C), indicating that JAK3 wild-type mice were more sensitive to anti-PD-1 therapy than JAK3 A573V and JAK3 M511I mutant mice. However, Tofacitinib showed significantly greater suppression of tumor growth than anti-PD-1 therapy in JAK3 A573V and JAK3 M511I mutant models. Tumor weights of JAK3 A573V (0.113 g vs 0.211 g, p<0.001, figure 6A,D) and JAK3 M511I (0.114 g vs 0.223 g, p<0.05, figure 6A,D) mutant mice treated with Tofacitinib were significantly smaller compared with anti-PD-1 treatment. TGI rates were 73.2% for JAK3 A573V and 76.8% for JAK3 M511I mutant mice with Tofacitinib treatment, significantly superior to anti-PD-1 therapy (73.2% vs 49.7%, 76.8% vs 35.0%, p<0.001) (figure 6E). Tofacitinib showed inferior efficacy compared with those treated with anti-PD-1 therapy in JAK3 wild-type mice (TGI rate 76.1% vs 91.5%, p<0.05) (figure 6A,E). The above results suggested that tumors with JAK3 A573V and JAK3 M511I mutations exhibited increased sensitivity to Tofacitinib but not to anti-PD-1 antibody. During the course of anti-PD-1 and Tofacitinib administration, there were no obvious changes in body weight (figure 6F).