Malignant LNs with FL and DLBCL had been collected in comparison to tonsils from patients undergoing tonsillectomy ( Supplementary Table S1 ) at the Medical Faculty of the Julius-Maximilians-University Würzburg. After isolation of mononuclear cells (MNC) using cell sieves, cells were frozen and stored in liquid nitrogen at the Institute of Pathology.

Human tissue samples of FL, DLBCL, and reactive LNs ( Supplementary Table S1 ), together with their clinical data, were available at the Institute of Pathology. All relevant histologic examinations were completed. Ethical approval was obtained for this study, i.e., the data were used according to the ethical guidelines of the Medical Faculty and approved by the Institutional Ethics Committee of the Julius-Maximilians-University (149/23 and 136/21). Informed consent was obtained after a written explanation of the nature and possible consequences of the studies (136/21), while in 149/23 it is residual diagnostic material that has been approved for research by the Ethics Committee. Therefore, our study adheres to the Declaration of Helsinki.

A total of 5 × 10⁶ cells were used for flow cytometry analysis. Viable cells were first identified using the Zombie NIR™ Fixable Viability Kit or Zombie Green™ Fixable Viability Kit (both from BioLegend). Following viability staining, cells were incubated with Human TrueStain FcX™ (BioLegend) for 10 minutes to block Fc receptors. Surface staining was then performed at room temperature (RT) for 15 minutes using the following fluorophore-conjugated antibodies ( Supplementary Table S2 ): BV510- or PerCP or FITC-conjugated CD4 (clone OKT4, BioLegend), APC-Cy7- or PE-Cy7-conjugated PD-1 (clone NAT105, BioLegend), PerCP-conjugated CD19 (clone, HIB19), Pacific Blue-conjugated CXCR5 (clone I252D4, BioLegend), PerCP-conjugated CD107a (clone H4A3, BioLegend), PE-Cy7-conjugated ICOS (clone C398.4A, BioLegend), and PE-conjugated TIA-1 (clone 2G9A10F5, Beckman Coulter). Following surface staining, cells were fixed and permeabilized using the FOXP3/Transcription Factor Staining Buffer Set (Invitrogen) according to the manufacturer’s instructions. Intracellular staining was subsequently performed using FITC-conjugated FOXP3 (clone 206D, BioLegend), APC-conjugated Granzyme K (clone GM26E7, BioLegend), and BV510-conjugated Granzyme B (clone GB11, BD Horizon™). Data were acquired on a FACSCanto II (BD Biosciences) flow cytometer and analyzed with FlowJo^® v10.8.1 (Treestar Inc., Ashland, OR, USA).

Consecutive formalin-fixed paraffin-embedded (FFPE) sections were used to localize T_{F K} cells. The procedure of staining and image analysis has been described in detail (15). Briefly, the tissue sections were processed through deparaffinization followed by heat-induced antigen retrieval in citrate buffer (pH 6.0). After blocking with Dako protein block (DAKO, S3022) for 1 h at room temperature, sections were incubated with primary antibodies ( Supplementary Table S3 ) for 1 h, washed three times with TBST, and then stained with appropriate secondary antibodies ( Supplementary Table S3 ) alongside DAPI for 1 h at room temperature. Following final washes, slides were mounted with Fluoromount-G (ThermoFisher).

A Zeiss LSM780 confocal microscope was used for image acquisition. The 20x/0.8 objective and the required channels were selected. The smart-set-up of the microscope was set to “best signal”, while the detection range for the four channels was set to minimize signal spillover from the different channels. Prior to image acquisition, the GC was located using DAPI staining of the nuclei. The gain of the laser had to be adjusted according to the tissue, as LNs and FLs stained differently under the microscope.

The raw confocal images were then processed using Fiji (Fiji is just ImageJ. The steps of deconvolution, segmentation and cell counting were achieved as described (15). In addition, background correction could now be included in the macro before deconvolution. Cell subtype counting was achieved by applying a second macro, which used the Colocalization Finder. The actual overlap of the two images was transferred as regions of interest (ROIs) to the ROI manager, where the pixel size of each single overlap ROI was measured. If the ROI was larger than 200 pixels or smaller than 20 pixels, the ROI was discarded and deleted. Subsequently, the results were transferred into an Excel spreadsheet containing the sample number, cell subtype as well as ROI number and size. This plugin was used once more to indicate possible locations of T_FK cells. T_FK cell candidates were confirmed by creating intensity profiles as described (15). The process of creating and saving intensity profiles could now be automated using a Fiji macro. The measurement of follicles and GC sizes was automated as well, after manually specifying and creating ROIs for all follicles and GCs according to CD19 and BCL6 staining. The results were saved automatically in an Excel spreadsheet. The allocation of T_FK cells to follicles and GCs was done according to the aforementioned ROIs of follicles and GCs. T_FK numbers and all other cell numbers were collected in an Excel spreadsheet where sample averages and percentages were calculated.

For the cell-cell contact analysis of T_{F K} cells with FOXP3⁺ Treg cells, CD4⁺ T cells were first identified by a Fiji plugin and afterwards checked by CellProfiler for at least 25% overlay with signal in the FOXP3 channel. The outlines of the Treg cells were superimposed on the merged image and manually checked for cell-cell contact with T_{F K} cells.

We used publicly available scRNAseq and TCRseq datasets to query gene expression as well as TCR usage in T_{F K} cells of FL and DLBCL compared to tonsils. Samples and sequencing procedure can be found in (32).

All scRNA-seq data (32) were processed and analyzed using Seurat (v5.2.1). The FindAllMarkers function was applied to identify differentially expressed genes and to annotate clusters; however, gene set enrichment analysis (GSEA), KEGG pathway analysis, RITAN, and enrichR were not utilized in this study.

For TCR reconstruction and paired TCR clonotype calling, Cell Ranger (version 3.0.2) was used for variable diversity joining sequence assembly. TCR analysis was performed using scRepertoire (v2.2.1), where instead of random 1000, all sequenced cells were selected and single TCRα/TCRβ genes were retained for analysis.

In vitro stimulation with staphylococcal enterotoxin B (SEB; Sigma S4881-1MG) was performed as previously described (15). Cells from tonsils, FL and DLBCL were plated in 96 well U-plate (Greiner bio-one), 2*10^5 cells/well were stimulated with 1 μg/ml of SEB for 4 days. For assessment of degranulation capacity, PerCP-conjugated CD107a (H4A3, Biolegend) antibody, with or without monensin (Biolegend) and brefeldin A (Biolegend), was added during the last 3 hours of stimulation. Samples were acquired on a FACS Canto II (BD Biosciences) flow cytometer.

The killing assay was optimized based on the methodology described (33). Fluorescence-activated cell sorting (FACS) was used to isolate T_FK cells (CD4⁺CD25^–CXCR5⁺PD1⁺TIA-1⁺), T_FH cells (CD4⁺CD25^–CXCR5⁺PD1⁺TIA-1^–), and B cells (CD19⁺) from FL and DLBCL samples. Afterwards, B cells were incubated with 10 µg/mL anti-CD3 (clone: OKT3, BioLegend) for 1 hour at 37 °C to serve as target cells. Subsequently, anti-CD3 mAb-coated B cells were co-cultured with either T_FH or T_FK cells at a 1:1 ratio in a 96 well U-plate (Greiner bio-one) for 14 hours. After incubation, apoptotic target cells were quantified by flow cytometry following staining with the Zombie NIR™ Fixable Viability Kit (Biolegend) and APC-conjugated Annexin V (BioLegend) according to the manufacturer’s instructions.

All flow cytometric data are shown as mean ± SD and represent combined data from at least three independent experiments. The results were analyzed with Prism software (GraphPad) using Wilcoxon signed-rank test, Kruskal-Wallis One-way ANOVA. * p < 0.05, **p < 0.005, *** p < 0.001, ****p < 0.0001.

All statistical analyses and their respective graphs presented for IF staining were conducted created with GraphPad Prism by using T-test or Mann-Whitney test. * p < 0.05, **p < 0.005, *** p < 0.001, ****p < 0.0001.

Recently, we detected a subcluster of CD3⁺CD4⁺CD45RA^–CXCR5⁺ T_FH cells with a cytotoxic phenotype in peripheral blood and tonsils of rather healthy individuals (15). This observation prompted us to investigate whether GC-derived lymphoma harbor such cytotoxic T_FH – hereafter referred to as killer T_FH or T_FK cells. To address this, we performed flow cytometry analysis on samples from 13 FL and 9 DLBCL patients, comparing them with tonsil specimens from non-cancer patients ( Supplementary Figures S1A, B ).

While CD4⁺ T cells were equally abundant in tonsils, FL and DLBCL, only FL samples exhibited a comparable proportion of CXCR5⁺PD-1⁺ T_FH within the CD4⁺FOXP3^– conventional T cells (Tconv) relative to tonsils ( Figure 1A ). Both lymphomas contained significantly more FOXP3⁺ Tregs, which was reflected in more CD4⁺FOXP3⁺CXCR5⁺PD-1⁺ T_FR cells among CD4⁺ T cells, but not among Tregs for FL ( Figure 1B ). In contrast to FL, DLBCL samples exhibited a significant reduction in both T_FH and T_FR cells. Furthermore, DLBCL had significantly fewer T_FR among CD4⁺FOXP3⁺ Tregs compared to tonsil.

T_FK cells are characterized by high expression of NKG7 and granzymes, particularly the tryptases GZMK and GZMA, with lower level of GZMB (15). Given that human NKG7 can be recognized by the anti-TIA-1 antibody (clone 2G9A10F5) - which exhibits well-documented cross-reactivity with both TIA-1 and NKG7 through recognition of a conserved pentameric epitope (GYETQ in TIA-1 and GYETL in NKG7 (34)). We therefore quantified the number of GZMK⁺TIA-1⁺ or GZMB⁺TIA-1⁺ T_FK cells within the CD4⁺FOXP3^–CXCR5⁺PD-1⁺ T_FH population ( Figure 1C ). Compared to tonsils, both FL and DLBCL showed a higher frequency of GZMK⁺TIA-1⁺ T_FK cells within T_FH or total CD4⁺ Tconv cells ( Figure 1D ). In contrast, the frequency of GZMB⁺TIA-1⁺ T_FK cells was similarly low in tonsils and FL. Notably, DLBCL samples exhibited a relative enrichment of GZMB⁺TIA-1⁺ T_FK cells within both the CXCR5⁺PD-1⁺ T_FH and the total CD4⁺ Tconv population. Furthermore, the positive correlation between T_FK cell abundance (both GZMK⁺TIA-1⁺ and GZMB⁺TIA-1⁺) and malignancy grade was confirmed by analyzing FL samples stratified by transformation stage ( Supplementary Figures S2A, B ). Given that T_FR cells have been implicated in suppressing GZMK⁺ cytotoxic T_FH cells in ectopic lymphoid follicles of patients with IgG4-related disease (35), we assessed the ratio of GZMK⁺TIA-1⁺ or GZMB⁺TIA-1⁺ T_FK cells to T_FR cells. Notably both T_FK cell subsets outnumbered T_FR cells in DLBCL samples, while this imbalance was not observed in FL ( Figure 1E ).

In summary, compared to tonsillar samples, FL exhibited an increased frequency of T_FK cells, predominately of the GZMK⁺TIA-1⁺ phenotype, accompanied by significantly elevated Tregs and T_FR cells. In contrast, DLBCL showed reduced numbers of T_FH and T_FR cells, but displayed enrichment of both GZMK⁺TIA-1⁺ and GZMB⁺TIA-1⁺ T_FK cells as well as Tregs.

Next, we aimed to determine whether tonsils, FL and DLBCL also contain cytotoxic non-T_FH, defined as CD4⁺PD-1⁺CXCR5^–TIA-1⁺ cells. Simultaneously we assessed CXCR5 expression level on the GZMK⁺TIA-1⁺ and GZMB⁺TIA-1⁺ T_FK cells. For classification, we distinguished between PD-1⁺CXCR5^med T_FH and PD-1⁺CXCR5^hi GC-T_FH cells (16). In both tonsils and FL, most CD4⁺FOXP3^–PD-1⁺ T cells express CXCR5 ( Figure 2A ). However, FL-derived CD4⁺FOXP3^–PD-1⁺CXCR5⁺ cells exhibited intermediate CXCR5 expression level. In DLBCL, where T_FH cells were less abundant ( Figure 1A ), the proportions of CD4⁺PD-1⁺CXCR5^– and CD4⁺PD-1⁺CXCR5⁺ cells were roughly equal, while GC-T_FH cells were underrepresented ( Figure 2A ). Interestingly, in tonsils, TIA-1⁺CD4⁺FOXP3^–PD-1⁺ T cells were similarly frequent among CXCR5^– and CXCR5^hi subsets, though granzyme expression tended to decrease with increasing CXCR5 levels ( Figures 2B, C ; Supplementary Figures S3A-C ). In FL, NKG7/TIA-1⁺ cells were predominantly non-T_FH cells, less frequent in CXCR5^med T_FH cells, and rare in CXCR5^hi T_FH cells. Most TIA-1⁺ cells co-expressed granzymes, particularly GZMK in FL, while GZMK levels remained comparable across CXCR5^–, CXCR5^med, and CXCR5^hiCD4⁺PD-1⁺ T cells. In DLBCL, no significant differences in cytotoxic phenotypes were observed among the analyzed T-cell subsets. However, CD4⁺FOXP3^–PD-1⁺ T cells displayed generally high expression of TIA-1, GZMK, and GZMB, independent of CXCR5 expression level.

Being CXCR5⁺, T_FK cells could home to GCs. However, T_FK cells in non-malignant tonsils are mainly localized extrafollicularly (15). To investigate their distribution in malignant tissues, we performed IF staining on FL samples, which maintain defined (though atypical) follicular structures with atypical GCs and partially preserved mantle zones, comparing them to reactive LNs as non-malignant controls. Using antibodies against CD19, CD4, TIA-1 and BCL6, we identified CD19⁺BCL6⁺ GC-B cells, CD4⁺BCL6⁺TIA-1^– T_FH and CD4⁺BCL6⁺TIA-1⁺ T_FK cells. In both tissues we found single T_FK cells, e.g. localized outside the follicles ( Supplementary Figures S4 , S5 ). Computer-assisted quantification uncovered comparable numbers of CD19⁺ B cells and total BCL6⁺ cells between FL and reactive LNs. However, FL samples contained fewer CD4⁺ T cells, fewer TIA-1⁺ cells, and a trend towards reduced T_FH cell numbers ( Figure 3A ). Strikingly, the percentage of CD4⁺TIA-1⁺BCL6⁺ T_FK cells within the CD4⁺BCL6⁺ T_FH population was significantly higher in FL compared to reactive LNs ( Figure 3B ), although their cell numbers—whether total per sample, follicular, or GC-localized—did not differ significantly ( Figure 3C ). Morphometric analysis showed that atypical follicles and GCs in FL tended to be larger than their benign ones in reactive LNs. A similar trend was observed when comparing low-grade (grade 1–2) and intermediate-grade (grade 2–3A) FLs ( Figure 3D ). This size difference became statistically significant when quantified per image per sample ( Figure 3E ). In contrast, neither the number of follicles/GCs nor their frequency correlated with disease severity ( Figure 3F ).

Thus, immunofluorescence analysis unequivocally identified CD4⁺TIA-1⁺BCL6⁺ T_FK within the atypical follicles and GCs of FL tissues ( Figure 4A ). Interestingly, their relative abundance in GCs/GC-like structures showed a significant stepwise increase with disease progression—first from reactive LN to FL overall, and then with disease severity ( Figure 4B ). This led to higher T_FK cell numbers in the atypical follicles and especially in “GCs” of grade 2-3A FLs compared to grade 1–2 FLs ( Figure 4C ). In summary, while CD4⁺TIA-1⁺BCL6⁺ T_FK cells are present in both reactive and malignant LNs, their follicular and GC localization is markedly more pronounced in FL than in mildly inflamed tonsils (15). Importantly, we observed that T_FK cell frequencies among T_FH cells and their GC homing propensity were highest in more advanced FL grades, suggesting a potential association between T_FK cell accumulation and disease malignancy.

To further characterize the NKG7⁺/TIA-1⁺ CD4⁺FOXP3^–PD-1⁺CXCR5⁺ in FL and DLBCL, we could rely on publicly available scRNAseq data (32). Spasevska et al. had been interested to define the phenotype of intratumoral Tregs in FL and DLBCL and sequenced CD4⁺ T cells from three FL, DLBCL and non-malignant tonsils each. Besides three Treg clusters, they defined 10 clusters of CD4⁺ Tconv cells of which two resembled T_FH cells (PDCD1, CXCR5, IL21, and TOX2) and one displayed upregulated RNA for GZMK, NKG7, CST7, GZMA, GZMB, and PRF1 (32).

We reanalyzed the data using Seurat version (v5.2.1) and projected them in a weighted-nearest neighbor uniform manifold approximation and projection (wnnUMAP) defining the same clusters (c0 – c12; Figure 5A ). Simplified UMAPs showed higher CXCR5, PDCD1, IL21 and TOX2 RNA expression in T_FH c5 and c8 as well as FOXP3 in c3 and c10 or LAG3 in c9/FOXP3^–LAG3⁺ Tregs (32) ( Figure 5B ). c1 [designated as GZM⁺ (32)] exhibited almost exclusive RNA expression of NKG7, CCL5, GZMK, GZMA, GZMB, and CST7 (encoding cystatin-F). However, cells in c1 also shared the expression of CXCR5, PDCD1, TOX2, or LAG3 with T_FH cells, activated FOXP3⁺ and FOXP3^–LAG3⁺ Tregs, albeit CXCR5 to a lesser extent than in T_FH cells. Still, this classified c1 as a subtype of follicular cells, before named T_FK cells by us (15). As a minimum it can be stated that c1 contained T_FK and that c3 contained T_FR cells as well as that c9/LAG3⁺ Tregs, expressing CXCR5, PD-1, TOX2, and IL-21, resembled T_FH-like cells.

In agreement with our flow cytometric data, the three sequenced DLBCL samples enclosed almost no non-cytotoxic T_FH cells (c5 and c8 or c9; Figure 5C ). However, while only a limited number of cells from GZM⁺ c1 could be identified in tonsils, FL exhibited approximately 10% and DLBCL 20% cytotoxic CD4⁺ T cells ( Figure 5C ). Plotting the signature genes of T_FK cells to all clusters in FL, DLBCL, and tonsils revealed the presence of them in tissues of each origin ( Figure 5D ). Interestingly, in FL and more pronounced in DLBCL the small cluster of proliferating CD4⁺ T cells (c12) acquired a similar cytotoxic phenotype. Taken together, despite lower CXCR5 and BCL6 levels than in T_FH cells, tonsils, FL and DLBCL harbored cytotoxic CD4⁺ T cells with a T_FK phenotype, confirming the T_FH-like phenotype of cytotoxic CD4⁺ T cells found in FL (26).

Next, we directly compared c1/GZM⁺ between the different origins ( Figure 6A ). Contrasted to non-malignant tonsils, the level of T_FK marker gene expression was usually higher in at least one of the lymphomas. Only CCL4, CCR4 and TBX21 (encoding T-BET) were more pronounced in tonsillar T_FK cells compared to FL and DLBCL. The follicular phenotype defined by CXCR5 and BCL6 was best observed in FL coinciding with LAG3, LITAF, PRDM1, EOMES, CCR5, CXCR3, and CRTAM upregulation. Of note, GZMK was expressed in a high percentage of all cytotoxic CD4⁺ and T_FK cells, yet the GZMK expression level per cell was clearly chief in FL. c1 cells from DLBCLs exposed their cytotoxic transcriptome with comparably highest expression of GZMB, GZMA, GZMH, PRF1, IFNG, CCL5 and NKG7, HAVRC2 (encoding TIM3) and KLRG1 ( Figure 6A ). The differences were consistent with the appearance of c1 in the UMAPs of FL, DLBCL, and tonsils, defining the dominant phenotype of FL-T_FK cells as GZMK^hi and c1/GZM⁺ as either GZMK⁺ or GZMK⁺GZMB^hi enriched in DLBCL ( Figure 6B ).

Our previous study demonstrated an oligoclonal nature of T_FK cells (15), a finding corroborated by Spasevska et al. through TCR sequencing (32). Clonal expansion was particularly evident in c1/GZM⁺, showing moderate expansion in FL and more prominent in DLBCL ( Figure 6C ). Notably, some of the enriched TCRs in c1 could be found in the T_FH clusters ( Figures 6D-E ). A comparison between the two T_FH clusters indicated elevated CXCR4, ZNP331, TNFAIP3, and MAP3K8 transcript levels in c5/T_FH (1) ( Supplementary Figure S6 ), which is consistent with GC-T_FH phenotypes (15). Interestingly, FL-c1/GZM⁺ cells were more likely to share TCRs with GC-T_FH (c5) cells than with c8, although c8 was augmented in FL compared to tonsils ( Figure 5C ). One clone defined by a common TCR, appeared in c1, c5, and c8 of DLBCL. Amazingly, the DLBCL-c1-enriched TCR clone found in T_FH cells were also present in activated and resting Tregs ( Figure 6E ).

This clonal overlap suggests potential antigen-driven associations, reminiscent of GZMK⁺CD4⁺ T cell behavior in IgG-related diseases (35). However, despite the abundance of GZMK⁺TIA-1⁺ T_FK and T_FR/Treg cells in FL, TIA-1⁺ T_FK cells were rarely detected in direct cell-cell contact with FOXP3⁺ Treg cells ( Figures 6F-G ). Meanwhile, interactions between Tregs and B cells seemed inevitable ( Supplementary Figures S7A-F ). Nevertheless, spatial proximity testing by imaging mass spectrometry found activated Treg/T_FR cells to rather interact with other T cells and macrophages than the malignant B cells, while indeed the cytotoxic T_FH cells are closer to the malignant B cells (32). These findings collectively indicate that while T_FK cells are oligoclonal and share TCRs with T_FH cells, their activity in lymphoma appears independent of direct Treg-mediated suppression.

To measure the cytotoxic potential within the CD4⁺CXCR5⁺PD-1⁺ T_FH population, we employed the ‘cytokine-independent activation-induced marker’ (AIM) method, which identifies Ag-specific GC-T_FH cells (15, 36). Here, SEB activates CD4⁺ T cells in an MHCII-restricted manner. Following four days of SEB stimulation, we observed a significant induction of CD107a surface expression in FL and DLBCL cultures, indicating recent degranulation of cytotoxic granules (37) ( Figures 7A, B ). SEB stimulation not only increased the frequency of T_FH cells within CD4⁺FOXP3^– Tconv cells, but also significantly expanded GZMK⁺TIA-1⁺ and GZMB⁺TIA-1⁺ T_FK cells ( Figure 7C ). In FL, the percentage of GZMK⁺TIA-1⁺ and GZMB⁺TIA-1⁺ T_FK cells within the CD4⁺CXCR5⁺PD-1⁺ T_FH population was markedly increased. Interestingly, the dominance of GZMK over GZMB in FL-T_FK cells was lifted upon SEB stimulation. To ensure a specific T_FK phenotype, we examined the expression of CD49b and LAG3, which are uniquely co-expressed by CD4⁺ T_R1 cells—a subset also capable of cytotoxicity (38). Only a minor percentage of T_FK cells co-expressed CD49b and LAG3 ( Figure 7D ). In sum, GZMK⁺TIA-1⁺ and GZMB⁺TIA-1⁺ T_FK cells can degranulate and release their cytotoxic cargo upon MHCII-restricted stimulation.

To assess the cytotoxic potential of T_FK cells, we performed a killing assay using FACS-sorted CD19⁺ B cells (predominantly malignant lymphoma cells) and co-cultured with either CD4⁺PD-1⁺CXCR5⁺TIA-1^– T_FH or CD4⁺PD-1⁺CXCR5⁺TIA-1⁺ T_FK FACS-sorted from FL and DLBCL samples ( Supplementary Figures S8A, B ). While isolated GC-B cells (being mostly HLA-DR⁺; Supplementary Figure S8C ) exhibit poor survival in vitro, the presence of T_FK cells significantly increased the frequency of Annexin V⁺ B cells. This effect was not observed when B cells were co-cultured with T_FH cells ( Figure 7E ). These findings suggest that T_FK cells may exhibit direct cytotoxic activity against lymphoma cells in vivo.