Abstract
The redirection of T cells has emerged as an attractive therapeutic principle in B cell non-Hodgkin lymphoma (B-NHL). However, a detailed characterization of lymphoma-infiltrating T cells across B-NHL entities is missing. Here we present an in-depth T cell reference map of nodal B-NHL, based on cellular indexing of transcriptomes and epitopes, T cell receptor sequencing, flow cytometry and multiplexed immunofluorescence applied to 101 lymph nodes from patients with diffuse large B cell, mantle cell, follicular or marginal zone lymphoma, and from healthy controls. This multimodal resource revealed quantitative and spatial aberrations of the T cell microenvironment across and within B-NHL entities. Quantitative differences in PD1+ TCF7− cytotoxic T cells, T follicular helper cells or IKZF3+ regulatory T cells were linked to their clonal expansion. The abundance of PD1+ TCF7− cytotoxic T cells was associated with poor survival. Our study portrays lymphoma-infiltrating T cells with unprecedented comprehensiveness and provides a unique resource for the investigation of lymphoma biology and prognosis.
Similar content being viewed by others
Main
Nodal B cell non-Hodgkin lymphomas (B-NHL) represent a heterogeneous group of indolent and aggressive malignancies. Extensive genetic and transcriptomic profiling has revealed mutational signatures and pathway dependencies, paving the way for molecular therapy1,2,3,4,5,6,7,8. However, in recent years, T cell-engaging immunotherapies, such as bispecific antibodies or chimaeric antigen receptor T cells, have emerged among the leading treatment options for patients with refractory and relapsed B-NHL9,10,11,12. Tailoring these treatment approaches to different B-NHL entities requires systematic investigation of the variety and functions of tumour-infiltrating T cells—analogously to studying the genetic and transcriptomic makeup of tumour cells as a prerequisite for tailoring targeted therapies.
Traditionally, T cell phenotyping studies have been based on immunohistochemistry or flow cytometry13. In recent years, single-cell RNA sequencing (scRNA-seq) emerged as a powerful tool and became an integral part of T cell phenotyping efforts14,15. We and others have pioneered the investigation of transcriptional heterogeneity of lymph node (LN)-derived T cells in B-NHL, but with the limitation of low sample sizes or having focused only on follicular lymphoma (FL, indolent)16,17,18. Stand-alone scRNA-seq studies additionally face the problem to align gene expression profiles with known T cell subsets that have been defined for decades based on surface epitopes and transcription factors19.
In this Resource, we employed cellular indexing of transcriptomes and epitopes by sequencing (CITE-seq), which simultaneously captures transcript and surface epitope abundances at single-cell resolution, and thus enables a multimodal identification of T cell phenotypes20,21. Aiming to create a multimodal reference map of LN-derived T cells in nodal B-NHL, we collected more than 100 LN patient samples and included, besides FL, other B-NHL entities with only little or no prior groundwork: diffuse large B cell lymphoma (DLBCL, aggressive), marginal zone lymphoma (MZL, indolent) and mantle cell lymphoma (MCL, mixed). We identified and quantitated fine-grained T cell subsets and determined their clonality using full-length single-cell T cell receptor (scTCR) sequencing. We further assessed the ability of multicolour flow cytometry and multiplexed immunofluorescence to reproduce these T cell subsets and to localize them within the tumour microenvironment.
Results
Study and sample overview
We collected 101 LN samples (Extended Data Fig. 1a and Supplementary Table 1) from patients with DLBCL (n = 28), FL (n = 30), MZL (n = 15), MCL (n = 15), or from patients without evidence of malignancy (tumour-free/reactive LN (rLN), n = 13). LN samples from 38 patients were collected at initial diagnosis, while 50 LN samples were collected from patients who had received one or more prior lines of systemic treatment (Extended Data Fig. 1a). To minimize potential effects on T cell infiltration patterns, relapse samples were collected at least 3 months after cessation of the preceding treatment. T cell proportions of malignant LN samples determined by flow cytometry showed a broad variation (Extended Data Fig. 1a) but were not significantly associated with pre-treatment status, sex, age or B-NHL entity (Extended Data Fig. 1b–e).
Fourteen multimodally defined T cell subsets
We used CITE-seq to profile T cells from 51 LN patient samples. Surface proteins were detected using 70 oligonucleotide-tagged antibodies (Supplementary Table 2). After quality control and in silico sorting, we obtained data for 74,112 CD3+ T cells with a median of 1,190 T cells per patient sample. Unsupervised clustering based on a weighted combination of transcriptome and epitope similarities (weighted nearest neighbour)22 grouped the T cells into proliferating (Tpr), conventional helper (TH) and follicular helper (TFH), regulatory (Treg), cytotoxic (Ttox) and double-negative T cells (TDN, Fig. 1a). Differentially expressed genes and proteins (Fig. 1b,c) were associated with lineage (CD4 and CD8), functional specialization (for example, FOXP3 and ASCL2), cytotoxicity (for example, GZMA and GZMK) or proliferation (for example, MKI67). These groups could be further partitioned into CD4+ and CD8+ naive T cells, central memory (CM1 and CM2) TH cells, central memory (CM1 and CM2) and effector memory (EM1 and EM2) Treg cells, and effector memory (EM1, EM2 and EM3) Ttox cells. At this level of granularity, differentially expressed markers (Fig. 1b,c) were linked to differentiation (for example, CD45RA, CD45RO and CD62L), homing and migration (for example, KLF2 and CCR7), activation (for example, CD69, CD38 and CD278) and inhibition (for example, PD1, TIM3 and LAG3). This high-granularity classification was supported by a gene regulatory network analysis23, which highlighted differential activities of specific transcription factors (for example, KLF2 (ref. 24), TCF7 (ref. 25), FOXP3 (ref. 26) and ASCL2 (ref. 27), Fig. 1d). On this basis, we compiled profiles of the most discriminating and biologically interpretable surface proteins, genes and transcription factors (Fig. 1e). Further extended profiles are provided in Supplementary Table 3.
Cytometry reproduces multimodally defined T cell subsets
Next, we built gradient boosting classifiers28 to identify the most discriminatory surface markers between multimodally defined subsets (Fig. 2a). While this yielded accurate results for most T cell subsets (Extended Data Fig. 2a), the distinction among Treg cells and the detection of Tpr cells could be improved by additional intracellular markers (Ki67, FOXP3 and IKZF3; Extended Data Fig. 2b). After removal of redundant (for example, CD95 and CD127) and less important markers, we thus compiled a 12- and 13-plex flow cytometry panel (Supplementary Table 4) and established gating strategies supported by the hypergate algorithm29, which enabled classification of all 14 multimodally defined T cell subsets (Supplementary Figs. 1 and 2). We correlated the subset proportions determined by CITE-seq and flow cytometry in a total of 13 LN samples and observed a median Pearson coefficient of 0.92 across all subsets (Fig. 2b). We then applied these panels to an independent cohort of 50 LN samples, which was then used for further quantitative analysis of T cell infiltration patterns.
Quantitative patterns of T cell infiltration in nodal B-NHL
We combined CITE-seq and flow cytometry data, determined the proportion of each subset per sample, and compared each B-NHL entity with tumour-free LN samples (Fig. 3a). B-NHL were characterized by a lack of CD69− CM1 and CD69+ CM2 TH cells, and CD4+/CD8+ naive T cells (Fig. 3a). Conversely, PD1+ TIM3− Ttox EM2 cells and PD1+ TIM3+ EM3 Ttox cells, Tpr cells and CD69+ Treg CM2 cells were significantly enriched in B-NHL (Fig. 3a). FL and MZL were additionally characterized by significant enrichment of TFH and IKZF3+ Treg EM2 cells (Fig. 3a). We also noted a significant increase of CD69+ Treg CM2 cells in MCL, FL and MZL, whereas TFH cells were depleted in DLBCL (Fig. 3a). We observed similar results when using (absolute) subset frequencies instead of (relative) subset proportions (Extended Data Fig. 3). To gain a broader overview of these differences across all B-NHL entities, we used principal component analysis (PCA) on the table of subset proportions and the overall T cell frequency (Fig. 3b). Based on the first two principal components (PCs), we identified three major groups (I–III) represented by tumour-free (I), DLBCL and MCL (II), and FL and MZL LNs (III, Fig. 3b). PC1 (Fig. 3c) and PC2 (Fig. 3d) had high loadings on the characteristic T cell subsets highlighted in Fig. 3a: CD4+ and CD8+ naive T cells, and PD1+ TIM3+ Ttox EM3 cells (PC1), and TFH and IKZF3+ Treg EM2 cells (PC2).
To further explore to what extent T cell composition and frequency is distinctive for different B-NHL entities and tumour-free LN, we built classifiers using LASSO-regularized multinomial logistic regression and estimated classification accuracy using nested leave-one-out cross-validation (Fig. 3e and Extended Data Fig. 4a). Accuracy was best for distinguishing between tumour-free and malignant LNs (balanced accuracy of 74.5%); moreover, DLBCL and MCL could be differentiated with similar accuracy (Fig. 3e). A third, well-distinguishable group was formed by FL and MZL (Fig. 3e). These results indicate that different entities have distinct patterns of T cell infiltration (Extended Data Fig. 4a), although—based on our current data—classification does not provide diagnostic accuracy.
To explore the potential role of patient-inherent characteristics, we fit multivariate linear models using sex, age, treatment status and cell-of-origin (only DLBCL) as covariates, and the proportion of each T cell subset as dependent variable (Extended Data Fig. 4b). We found that pre-treatment (Fig. 3f, P < 0.001) and higher age (Fig. 3g, P = 0.04) were associated with a lower proportion of naive CD4+ T cells. Pre-treatment was also linked to a lower proportion of CD69+ TH CM2 cells (Fig. 3h, P < 0.001) and a higher proportion of CD69+ Treg CM2 cells (Fig. 3i, P < 0.001), while we observed no statistically significant effect on the T cell composition for sex and cell-of-origin (Extended Data Fig. 4b). Larger sample sizes might be necessary to detect less strong associations, but overall, patient characteristics had only a moderate effect on the T cell composition compared with entity-specific differences.
Entity-specific clonality of CD4+ and CD8+ T cell subsets
To investigate whether the enrichment of T cell subsets results from their clonal expansion, we performed full-length scTCR sequencing alongside 5′ scRNA-seq in a representative subset of 17 patient samples (Extended Data Fig. 1a and Supplementary Table 1). After quality control, 5′ scRNA and full-length TCR data were available for 66,896 T cells with a median of 3,045 cells per patient sample. We mapped the 5′ scRNA onto the CITE-seq reference data above, indicating high consistency of the inferred subset proportions between both modalities with a median correlation coefficient of R = 0.92 (Extended Data Fig. 5a). Then, we compiled the scTCR data to clonotypes on the basis of their complementarity-determining regions30 and projected them onto the reference uniform manifold approximation and projection (UMAP; Fig. 4a). Clonally expanded Ttox EM1 cells were present across all entities and in tumour-free LNs (Fig. 4a), while clonally expanded PD1+ TIM3+ Ttox EM3 cells were limited to DLBCL, FL and MZL (Fig. 4a). Clonality of Ttox cells was not restricted to CD8+ but included also CD4+ Ttox cells (Extended Data Fig. 5b,c). In addition, TFH and IKZF3+ Treg EM2 cells were clonally expanded exclusively in FL and MZL (Fig. 4a). Consequently, the TCR diversity was substantially reduced in DLBCL, FL and MZL samples compared with tumour-free and MCL samples (Extended Data Fig. 5d–f). To further support our findings, we quantified the proportion of clonally expanded T cells per subset and compared each entity with tumour-free samples. While Ttox EM1 cells were clonally expanded to a similar extent across all sample types, Ttox EM3 cells, TFH cells and Treg EM2 cells had significantly enriched proportions of clonally expanded T cells only in entities specified above (Fig. 4b). We further used scTCR data to track the original identity of Tpr cells, that is at the time of sample collection temporarily masked by an S or G2M phase-dependent gene expression signature31. Apart from MCL and tumour-free LN, which both harboured very low proportions of Tpr cells (Fig. 4a,c), we identified groups of shared clonotypes in DLBCL, FL and MZL predominantly between Tpr cells and PD1+ TIM3+ Ttox EM3 cells (Fig. 4c). To a lower extent, TCR clonotypes were also shared between Tpr cells and TFH cells, and between Tpr cells and IKZF3+ Treg EM2 cells in FL and MZL (Fig. 4c). Overall, this analysis highlights that an altered T cell microenvironment results from active proliferation and differential expansion of specific T cell subsets.
PD1+ TCF7− Ttox cells converge into exhausted T cells
To unveil the process of T cell exhaustion across B-NHL entities in detail, we performed a trajectory analysis32 of Ttox cells based on the CITE-seq data. We identified two paths: (I) one from naive to PD1+ TIM3− Ttox EM2 cells, and (II) another one from naive to PD1+ TIM3+ Ttox EM3 cells (Fig. 5a). Apart from TIM3, both expression (Fig. 5b) and inferred transcription factor activity23 of TCF7 was an important discriminator between both trajectories (Fig. 1d). TCF1 (encoded by TCF7) is a hallmark of stemness and longevity and its presence (trajectory I) or absence (trajectory II) indicates maintained or impaired self-renewal capacity, respectively33, thereby suggesting that only trajectory II converges into terminally exhausted T cells.
To further resolve gradual changes along trajectory II, we applied pseudotime analysis32 and ranked the cells starting from naive Ttox cells (Fig. 5a). We found that pseudotime was strongly linked to a continuously increasing expression of both differentiation and activation markers, such as CD45RO, CD69, CD38 and ICOS, and inhibitory molecules, such as PD1, TIM3, LAG3, TIGIT and CD39 (Fig. 5c). Cells with the highest levels of inhibitory receptors had reduced RNA expression levels of effector molecules, particularly of GZMA and GZMH, and reduced protein levels of CD244 (ref. 34) (Fig. 5c). Likewise, the inferred activity23 of the transcription factors PRDM1, BATF, IRF4 and EOMES, which have previously been associated with T cell exhaustion35,36,37,38, were strongest in Ttox cells at the end of the trajectory (highest pseudotime), whereas the inferred transcription factor activity of TCF7 was lowest in these cells (Fig. 5c). On this basis, we established a signature profile (Supplementary Table 5), to facilitate the identification of terminally exhausted T cells in scRNA-seq data (Extended Data Fig. 6a). While the initial trajectory and pseudotime analysis was performed on complete Ttox cells, scoring gave similarly high values when applied separately to CD4+ or CD8+ Ttox cells (Extended Data Fig. 6b).
Plotting the proportion of Ttox cells by sample and pseudotime revealed that terminally exhausted T cells were most abundant in DLBCL and FL, most variable in MZL, and lowest in MCL (Fig. 5d). Clonality analysis based on scTCR data supported this finding by demonstrating that the expansion of PD1+ TIM3+ Ttox EM3 cells was a key feature of the tumour microenvironment of DLBCL, FL and MZL, while MCL and tumour-free LNs were predominantly characterized by clonal PD1− Ttox EM1 cells (Fig. 4a).
T cell exhaustion is linked to adverse prognosis in B-NHL
To investigate if T cell exhaustion is associated with clinical outcome in B-NHL, we extracted a transcriptional signature from terminally exhausted T cells of our data and applied digital cytometry39 to bulk RNA data from two large independent retrospective DLBCL cohorts2,4. We found that a higher proportion of terminally exhausted Ttox cells was associated with inferior progression-free survival in both cohorts (Fig. 5e, P = 0.003, Fig. 5f, P = 0.011). Moreover, the cohort from Schmitz et al.4 harboured higher proportions of exhausted T cells in ABC- than GCB-subtype DLBCL (Extended Data Fig. 6c), which is in line with a recent flow cytometry-based study40. However, there was no difference between ABC- and GCB-subtype DLBCL in our data (Extended Data Fig. 4b) or in the cohort from Chapuy et al.2 (Extended Data Fig. 6d). Neither of the genetic subtypes defined in these cohorts were associated with the proportion of exhausted Ttox cells (Extended Data Fig. 6c,d). We also evaluated individual somatic mutations (for example, MYD88), amplifications (for example, BCL2), deletions (for example, 17p) and structural variants (for example, BCL6). After correction for multiple testing, none of the genetic aberrations was associated with the proportion of exhausted Ttox cells (Extended Data Fig. 6e–h). We performed a similar analysis using bulk RNA-seq data of a prospective FL cohort of the GALLIUM trial (NCT01332968)41,42, and again a higher proportion of exhausted T cells was associated with inferior survival (Fig. 5g, P = 0.04).
IKZF3+ Treg EM2 cells are clonally related to TFH cells
In FL and MZL, not only Ttox cells but also TFH and Treg EM2 cells were clonally expanded (Fig. 4a) and significantly enriched (Fig. 6a). While TFH cells are well characterized and known to promote the growth of malignant B cells in FL43, the role of LN-derived Treg cells in nodal B-NHL has not been investigated systematically. Aiming to characterize Treg EM2 cells, we compared protein and gene expression profiles of Treg EM2 cells and Treg EM1 cells, and found that Treg EM2 cells were characterized by high protein levels of CD69, ICOS, CD38, PD1 and TIGIT (Fig. 6b). At the gene expression level, Treg EM2 cells showed high expression of IKZF3, CXCL13 and ASCL2, but low expression of KLF2 and IKZF2 (Fig. 6c). We used flow cytometry to confirm the presence of IKZF3 on protein level (Extended Data Fig. 7a) and the enrichment of IKZF3+ Treg cells in MZL and FL using an independent cohort of 24 LN samples (Extended Data Fig. 7b). IKZF2, alias Helios, is well studied as a marker of natural Treg cells44, but only few studies have explored the role of IKZF3, alias Aiolos, in Treg cells. A previous study suggested that IKZF3+ Treg cells usually lack IKZF2 and represent an inducible rather than natural Treg cell subset with potent suppressive capacity45. On this basis, we intended to identify potential populations related to Treg EM2 cells using scTCR data. We found that Treg EM2 cells share a substantial proportion of clonotypes with TFH cells (Fig. 6d), whereas this was not the case for other Treg cell populations (Extended Data Fig. 7c). Based on these aspects, IKFZ3+ Treg cells could resemble follicular regulatory T cells46.
We further investigated whether the proportion of Treg EM2 cells was associated with clinical parameters. Lower-grade (1/2) FL had a higher proportion of Treg EM2 cells (Fig. 6e, P = 0.002), whereas higher-grade FL (3A) had a higher proportion of TFH cells (Fig. 6f, P = 0.03). We applied digital cytometry39 on the GALLIUM cohort41,42 to estimate the proportion of Treg EM2 cells and TFH cells, and to investigate whether these subsets were associated with progression-free survival. Indeed, there was a trend towards inferior prognosis in patients with higher proportions of TFH cells (P = 0.05, Extended Data Fig. 7d), while no association was found for Treg EM2 cells (P = 0.17, Extended Data Fig. 7e).
CODEX captures the tumour and T cell microenvironment
To localize T cell subsets in their spatial context, we used highly multiplexed immunofluorescence in formalin-fixed paraffin-embedded (FFPE) LN tissues using co-detection by indexing (CODEX)47. We established a panel of 50 antibodies (Supplementary Table 6) and imaged 35 FFPE biopsy cores from 19 patient samples (Extended Data Fig. 1a and Supplementary Table 1). Finally, we identified B cells, T cells, natural killer (NK) cells, NK T cells, mast cells, plasma cells, dendritic cells, follicular dendritic cells (FDCs) and stromal cells, as well as macrophages and granulocytes (Extended Data Fig. 8a–c). Among a total of around 5.5 million processed cells, we detected a median of approximately 45,000 T cells per tissue core (Extended Data Fig. 8d), which we further divided into eight subpopulations including naive CD4+ and CD8+ T cells (CD45RA+), TFH cells (CD45RO+, PD1+, CXCR5+), Treg cells (FOXP3+), memory CD4+ cells (CD45RO+), memory CD8+ Ttox cells (CD45RO+), exhausted Ttox cells (CD45RO+, PD1+, TIM3+) and Tpr cells (Ki67+, Extended Data Fig. 8a,c). A high-granularity classification of T cell subsets, as possible with the CITE-seq and flow cytometry, was hampered by lower signal-to-noise ratio (for example, CD62L and CD69) or reduced sensitivity of available antibodies (for example, IKZF3). Still, we aligned the low-granularity T cell subpopulations detected by multiplexed immunofluorescence with the 14 high-granularity T cell subsets identified by CITE-seq (Extended Data Fig. 8a), resulting in a median Pearson coefficient of 0.71 across all subpopulations (Extended Data Fig. 8e).
B-NHL generates entity-specific microenvironmental patterns
In situ mapping of the above-mentioned cell types and T cell subpopulations enabled an intuitive visualization of tumour-free or malignant LN structure (Fig. 7a and Extended Data Fig. 9). To capture spatial organization systematically, we identified the 25 nearest neighbours of each cell by a sliding window approach and tabulated the frequencies of cell types and T cell subsets in each window. We used k-means clustering on the neighbour frequency tables to identify ten recurrent neighbourhoods (N1–N10; Fig. 7b) and PCA to identify the cell types most important for distinguishing N1 to N10 (Fig. 7c). Seven out of ten neighbourhoods captured important elements of intact LN architecture, including B cell follicles (N1) with TFH-rich and FDC-rich germinal centres (N2), follicle-surrounding mantle zones (N5), inter-follicular T cell zones (N6, N7 and N9) and sinuses (N10) harbouring predominantly stromal cells, macrophages, mast cells, granulocytes and NK cells (Fig. 7a–c), whereas N3, N4 and N8 could not be assigned to physiological LN regions.
The described pattern was preserved in all tumour-free tissue cores (Extended Data Fig. 10a) but largely disrupted in B-NHL LNs (Fig. 7d and Extended Data Fig. 10b–e). LN-derived tissue cores infiltrated by DLBCL had the least degree of structure and exhibited a diffuse excess of a neighbourhood (N4) harbouring exhausted T cells, macrophages and tumour cells, whereas neighbourhoods (N7 and N8) rich in naive and memory TH cells and Ttox cells were absent (Fig. 7d–f and Extended Data Fig. 10b). Tissue cores from LNs infiltrated by FL were characterized by expansion of germinal centre-like areas (N2) containing high numbers of (clonal) TFH cells and FDC, surrounded by areas (N6 and N9) containing Treg, memory TH cells, amd memory and exhausted Ttox cells, but also B cells (Fig. 7d–f and Extended Data Fig. 10d). In tissue cores infiltrated by MCL, we found a significant predominance of follicle-like B cell areas (N1, Fig. 7f), where TFH cells and germinal centres were absent (Fig. 7d,e and Extended Data Fig. 10c). In contrast to DLBCL and FL, B cell areas in MCL were well separated and barely infiltrated by T cells (Fig. 7d and Extended Data Fig. 9), resulting in only little contact surface in the transition areas (N6, Fig. 7d,e). To identify interaction partners of B cells on the basis of spatial proximity, we determined the nearest neighbour of each B cell and ranked them by frequency (Fig. 7g). This analysis suggested strikingly different interaction partner across entities: While B cells derived from FL, MCL and tumour-free LNs were in closest contact to varying T cell subsets, for instance Treg cells and TFH cells in FL, we observed that B cells derived from DLBCL were mostly surrounded by macrophages and stromal and exhausted T cells (Fig. 7g).
Discussion
Our study provides an in-depth and systematic reference map of LN-derived T cells in DLBCL, FL, MCL, MZL and tumour-free samples. Across all entities, malignant LNs were characterized by loss of CD4+ and CD8+ naive T cells, as well as CD69− TH CM1 cells and CD69+ TH CM2 cells, but to variable extents harboured clonally expanded CD4+ and CD8+ PD1+ TIM3+ Ttox cells. The latter were part of a cellular trajectory that continuously converged into terminally exhausted T cells and had lost TCF7 transcription factor activity48. Higher proportions of these cells were associated with inferior survival in FL and DLBCL, which is in line with previous studies in DLBCL49 or FL50. PD1+ TIM3+ Ttox cells were located within several spatial neighbourhoods, thereby allowing for contact with various types of immune cell. Specifically in LNs infiltrated by DLBCL, PD1+ TIM3+ Ttox cells were strongly co-localized with tumour cells and macrophages, which have recently been shown to be attracted by exhausted T cells and inversely, to reinforce T cell exhaustion51. This observation might explain why DLBCL harboured the highest proportions of PD1+ TIM3+ Ttox cells and why higher numbers of macrophages are associated with inferior outcome in DLBCL52,53.
A second trajectory was characterized by expression of PD1, absence of TIM3 and other inhibitory receptors, but maintained transcription factor activity of TCF7. Previous work suggested that only TCF7+ PD1+ T cells can be reinvigorated upon checkpoint inhibition, whereas terminally exhausted TCF7− T cells cannot be restored54,55. This observation is of particular interest because immune checkpoint blockade is remarkably ineffective in nodal B-NHL56,57, which might be—in the light of our study—due to the predominance of terminally exhausted T cells. We speculate that patients with B-NHL harbouring high levels of PD1+ TIM3− Ttox cells could represent a subgroup that benefits most from combining T cell-engaging immunotherapies and immune checkpoint blockade.
Beyond clonal Ttox cells, we found that both FL and MZL were characterized by clonal TFH cells and IKZF3+ Treg EM2 cells, and overall had a similar pattern of T cell infiltration. Whereas TFH cells have been extensively studied and are known to support the growth of malignant B cells in FL43, an enrichment of IKZF3+ Treg EM2 cells has, to our knowledge, neither been described in FL nor in MZL. We observed higher proportions of IKZF3+ Treg EM2 cells in patients with low-graded FL, suggesting that this T cell subset could modulate the proliferation capacity of malignant B cells. IKZF3+ Treg EM2 cells are suggested to bear strong suppressive capacity and represent an induced Treg phenotype45. Indeed, we demonstrated that IKZF3+ Treg EM2 cells and TFH cells carry a substantial proportion of identical TCR clonotypes, which implies that IKZF3+ Treg EM2 cells most likely derive from TFH cells. More mechanistically oriented studied are needed to clarify the role of IKZF3+ Treg cells.
In summary, our work refines previous knowledge of lymphoma-infiltrating T cells by employing recent technological advances and offers a different perspective on various B-NHL entities. This broader yet more detailed view revealed that B-NHL entities shape their T cell microenvironment in distinct manners, which could not be readily detected in studies investigating only single entities.
Methods
LN samples
Our study was approved by the Ethics Committee of the University of Heidelberg (S-254/2016). Informed consent of all patients was obtained in advance. As collection of LN samples was part of routine diagnostic procedures, participants were not compensated. LN patient samples were processed and frozen until further analysis58. In brief, whole LNs were cut into small pieces of approximately 1–2 mm size. LN cells were gently washed out by rinsing the pieces several times. After centrifugation, cells were frozen until further analysis. We did not include LN samples that showed obvious contamination of red blood cells after isolation. Also, samples from patients after allogeneic stem cell transplantation, chimaeric antigen receptor T cell or bispecific antibody therapy were not used in this study to minimize treatment-associated effects on the T cell microenvironment. For the same reason, samples were collected earliest 3 months after cessation of the last treatment.
Single-cell 3′ RNA-seq and epitope expression profiling
Cells were thawed and immediately washed to remove dimethyl sulfoxide. To prevent entity-associated batch effects, samples were processed in batches of four to five containing at least three different entities. After thawing, we applied a dead cell removal kit (Miltenyi Biotec, 130-090-101) to all samples to reach a viability of at least 85–90%. Samples not reaching a viability above 85% were excluded. Then, 5 × 105 viable cells were stained by a pre-mixed cocktail of oligonucleotide-conjugated antibodies (Supplementary Table 2) and incubated at 4 °C for 30 min. Cells were washed three times with ice-cold washing buffer and each time centrifuged at 4 °C for 5 min. After completion, cells were counted and viability was determined again. Samples not reaching a viability above 85% were excluded. The preparation of the bead–cell suspensions, synthesis of complementary DNA and single-cell gene expression and antibody-derived tag (ADT) libraries were performed using a Chromium single-cell v3.1 3′ kit (10x Genomics, 1000269) according to the manufacturer’s instructions.
Single-cell 5′ RNA-seq and TCR repertoire profiling
Apart from epitope staining, sample processing was identical to 3′ scRNA-seq. The preparation of the bead-cell suspensions, synthesis of complementary DNA and single-cell gene expression and TCR libraries were performed using a Chromium single-cell v2 5′ and human TCR amplification kit (both 10x Genomics, 1000265, 1000252) according to the manufacturer’s instructions.
Single-cell library sequencing and data processing
The 3′ gene expression and ADT libraries were pooled in a ratio of 3:1 aiming for 40,000 reads (gene expression) and 15,000 reads per cell (ADT), respectively. Sequencing was performed on a NextSeq 500 (Illumina). 5′ gene expression libraries were sequenced on a NextSeq 2000 (Illumina) aiming for 50,000 reads per cell. TCR libraries were sequenced on a NextSeq 500 (Illumina) aiming for a minimum of 5,000 reads per cell. After sequencing, the Cell Ranger (10x Genomics, v6.1.1) function cellranger mkfastq was used to demultiplex and to align raw base-call files to the reference genome (hg38). For 3′ gene and epitope expression libraries, the obtained FASTQ files were counted by the cellranger count command, whereas cellranger multi was used for 5′ gene expression and TCR libraries. As reference for TCR libraries, the VDJ ensembl reference (hg38, v5.0.0) was used. If not otherwise indicated, default settings were used for all functions.
Analysis and integration of CITE-seq data
The R package Seurat (v4.1.0) was used to perform data quality control, filtering, and normalization. Gene counts per cell, ADT counts per cell and percentages of mitochondrial reads were computed using the built-in functions. PCA, shared nearest neighbour-based clustering and UMAP were done on the basis of the combined transcriptome and epitope data. After mapping the CD3 and CD19 epitope expression, non-T-cell clusters and doublets were removed. For data integration across the different preparation batches, we used the IntegrateData function of the Seurat package. For multimodal clustering based on gene and epitope expression, the weighted nearest neighbour approach was used22. To estimate the proportion of positive cells for each surface marker, we calculated the denoised protein expression using the totalVI Python package59.
Inferring transcription factor activity on the basis of single-cell gene expression
We used the SCENIC (Single-Cell rEgulatory Network Inference and Clustering) package23 to infer gene regulatory networks and transcription factor activity on the basis of scRNA-seq data. Functions were used according to publicly available vignettes.
Surface and intracellular flow cytometry staining
LN-derived cells were thawed and stained for viability using a fixable viability dye e506 (Thermo Fisher Scientific, 65-0866-14) and for different surface markers depending on the experimental set-up. The following surface antibodies were used: anti-CD3-PerCP-Cy5.5, anti-CD4-PE-Dazzle, anti-CD8-APC-Cy7, anti-CD45RA-FITC, anti-CD25-BV421, anti-CD31-BV605, anti-CXCR5-BV711, anti-TIM3-BV711, anti-CD278-BV605, anti-PD1-PE-Cy7, anti-CD69-AF700 and anti-CD244-BV421 (all BioLegend, Supplementary Table 4). For subsequent intracellular staining, cells were fixed and permeabilized with the intracellular fixation/permeabilization buffer set (Thermo Fisher Scientific, 88-8824-00) and stained with anti-Ki67-BV785, anti-FOXP3-AF647, anti-IKZF3-PE or adequate isotype controls (Thermo Fisher Scientific, BD Biosciences, Supplementary Table 4). Then, cells were analysed using an LSR Fortessa (BD Biosciences) and FACSDiva (BD Biosciences, version 8). For analysis and gating of flow cytometry data, FlowJo (v10.8.0) was used.
Multinomial classification of T cell subpopulations
First, we evaluated whether multimodally defined T cell subsets can be distinguished in general by using surface markers only. Therefore, we trained gradient boosting models (‘xgbTree’)28 on the basis of surface marker expression of single-cell data. To reduce data load, only 30% of all cells were used. Tenfold cross-validation was employed to optimize the model. Since surface marker were not sufficient to reach sufficient accuracy, additional models were trained using surface marker plus gene expression of MKI67, IKZF3 and FOXP3, since these genes were differentially expressed between T cell subsets that could not be sufficiently predicted using only surface proteins. Finally, markers were ranked and selected by their variable importance to build flow cytometry panels. In case two or more markers deliver similar information (for example, CD95 and CD45RA), only one was selected. Gating strategies were built using the R package hypergate29 and optimized in an iterative process. The final gating strategy for all multimodally defined T cell subsets is illustrated in Supplementary Figs. 1 and 2.
Prediction of B-NHL entity from T cell proportions
To assess feasibility of predicting B-NHL entity or tumour-free condition based on the proportions of all T cell subsets and overall T cell frequency, we trained classifiers based on multivariate regression with L1 penalty (LASSO) implemented in the R package glmnet (v4.1)60. The hyperparameter lambda was determined using cv.glmnet with balanced folds and weights inversely proportional to class size. The confusion matrix was computed using leave-one-out cross-validation.
Analysis of TCR diversity
To estimate the diversity of the TCR repertoire on the basis of scTCR profiling, we employed the R package immunarch (v0.8.0; https://immunarch.com/) on the output files of the cellranger pipeline. TCR diversity across samples was compared using a rarefraction analysis. Therefore, the function repDiversity with method = ‘raref’ was applied.
Mapping of 5′ scRNA-seq data onto CITE-seq reference data
To evaluate scTCR data in the context of multimodally defined T cell subsets, 5′ scRNA-seq data were mapped onto the CITE-seq reference data, using the built-in functions FindTransferAnchors and MapQuery of the Seurat package (v4.1.0). The mapping accuracy was evaluated by comparing the T cell subset proportion of 5′ and 3′ data of the identical patient sample.
Pseudotime analysis and exhaustion signature
Pseudotime analysis based on gene expression profiles of Ttox cells was performed using the monocle3 package32. In brief, the Seurat object was converted into a cell dataset. Trajectory and pseudotime analysis were performed using the functions learn_graph and order_cells, respectively. As root cells, naive CD8+ T cells were selected. Minimal branch length was set to 10; otherwise, default settings were used.
To define a transcriptional module for T cell exhaustion, differentially expressed genes of terminally exhausted T cells, meaning Ttox cells with highest levels of inhibitory receptors and decreasing expression of effector molecules (Fig. 5c), were determined (Supplementary Table 5). The UCell package (v1.3.1)61, which is based on the Mann–Whitney U statistic, was then applied to calculate an exhaustion score for each cell.
Deconvolution of bulk RNA sequencing data
Deconvolution of bulk RNA-seq data was performed using the interactive web application (https://cibersortx.stanford.edu/) developed by Newman and colleagues39. First, a signature matrix was created on the basis of the scRNA-seq data and the cluster annotation as cell types. Minimum expression was set to zero, and 200 replicates were used. Otherwise, default settings were applied to create a signature matrix. Second, cell fractions were imputed using bulk RNA-seq data and signature matrix as input. S-mode batch correction and absolute mode was used for each analysis.
Survival analysis
Survival data were obtained only from previously published data or studies2,4,41. Analysis of the progression-free survival probability was performed in combination with the estimated cell type proportions using deconvolution of bulk RNA-seq data. To divide the data into two groups (low and high), we determined a cut-off based on the maximized P value of a log-rank test using the maxstat R package (v0.7.25). Kaplan–Meier curves were drawn using the survminer R package (v0.4.9).
Tissue microarray and coverslip preparation
Representative tumour or tumour-free LN areas in archival FFPE tissue blocks from 19 patients (Extended Data Fig. 1a and Supplementary Table 1) were selected by board-certified pathologists of the Tissue Bank of the National Center for Tumor Diseases and Institute of Pathology at the University Hospital Heidelberg. Tissue microarrays (TMAs) containing two 4.5-mm cores per patient were generated. TMA sections (4 μm) were mounted onto Vectabond-precoated 25 × 25 mm coverslips and coated in paraffin for storage until staining62.
Antibody conjugation, validation and titration
Multicolour immunofluorescence was performed using the CODEX approach47. Antibodies used for CODEX experiments are summarized in Supplementary Table 6. Purified, carrier-free antibodies (50–100 μg per reaction) were reduced with tris(2-carboxyethyl)phosphine and conjugated at 1:2 weight/weight ratio to maleimide-modified CODEX DNA oligonucleotides, which were purchased from TriLink Biotechnologies and deprotected via retro-Diels–Alder reaction. Conjugated antibodies were first evaluated in CODEX singleplex stains on tonsil and/or lymphoma tissue by comparison with online databases (The Human Protein Atlas, Pathology Outlines), immunohistochemical reference stains and/or published literature under the supervision of a board-certified pathologist. Staining patterns were validated in multiplex experiments in the presence of positive and negative control antibodies and the appropriate dilution of each antibody was titrated starting from 1:100 to optimize signal-to-noise ratio.
Multiplex tissue staining and fixation
Coverslips were deparaffinized, rehydrated and submitted to heat-induced epitope retrieval at pH 9 (Dako target retrieval solution, S236784-2, Agilent) and 97 °C for 10 min in a Lab Vision PT module (Thermo Fisher). After blocking of non-specific binding with CODEX FFPE blocking solution, coverslips were stained overnight with the full antibody panel at the dilutions given in Supplementary Table 6 in CODEX FFPE blocking solution63 in a sealed humidity chamber at 4 °C on a shaker. Coverslips were then fixed with 1.6% paraformaldehyde, followed by methanol and BS3 fixative (Thermo Fisher) before storage in CODEX buffer S4 until imaging63.
Multicycle imaging
Stained coverslips were mounted onto custom acrylic plates (Pololou Corporation) with mounting gaskets (Qintay), thereby creating a flow cell with a surface area of 19 × 19 mm above the tissue for fluid exchange. Acrylic plates were inserted into a Keyence BZ-X710 inverted fluorescence microscope equipped with a CFI Plan Apo λ 20×/0.75 objective (Nikon) using custom adapters. For each core an area of 7 × 7 fields of view (30% overlap between tiles) and an adequate number of z planes (10–14) required to capture the best focal plane across the imaging area were selected. Multicycle imaging was performed using a CODEX microfluidics device and CODEX driver software v1.29.0.1 (Akoya Biosciences). Exposure times and assignment of markers to imaging cycles and fluorescent channels are provided in Supplementary Table 6. After completion of multicycle imaging, coverslips were stained with haematoxylin–eosin, and the same areas were imaged in brightfield mode.
Image processing
Raw TIFF images were processed using the RAPID pipeline64 in MATLAB (version R2020a) with the following settings: nCyc = 51, nReg = number of regions imaged (depending on TMA), nTil = 49, nZ = number of z-planes imaged (depending on TMA), nCh = [1,4], nTilRow = 7, nTilCol = 7, overlapRatio = 0.3, reg_range = 1:nReg, cyc_range = 1:nCyc, til_range = 1:nTil, cpu_num = depending on computer system used, neg_flag = 1, gpu_id = depending on number of GPUs available, cyc_bg = 1. After deconvolution (two iterations), best focal plane selection, lateral drift compensation, stitching of individual images and background subtraction, processed images were concatenated to hyperstacks. All tissue cores were checked visually for staining quality of each antibody in ImageJ/Fiji (version 1.53q).
Cell segmentation and cell type annotation
Individual nuclei were segmented on the basis of the Hoechst stain (cycle 1), derived nuclear masks were dilated, and cellular marker expression levels were quantified using a modified version of the Mask region-convolutional neural network-based CellSeg software65 run on the full-resolution RAPID stitched images with the following parameters: GROWTH_PIXELS_PLANE = 1.0, output_adjacency_quant = True, BOOST = 1, OVERLAP = 80, MIN_AREA = 80, INCREASE_FACTOR = 4.0, AUTOBOOST_PERCENTILE = 99.98. A threshold based on the intensity of the nuclear markers Hoechst and DRAQ5 was used to exclude non-cellular events and remove cells from tissue areas of low image quality. Marker expression levels compensated for lateral spillover were used for further analysis and the range of each marker was z normalized per imaging run. A total of n = 5,690,284 cells were submitted to an initial round of Leiden-based clustering (n_neighbors = 10, resolution = 2) on key phenotypic markers (CD11b, CD11c, CD14, CD15, CD16, CD163, CD20, CD206, CD21, CD25, CD3, CD31, CD34, CD38, CD4, CD45, CD5, CD56, CD57, CD68, CD7, CD79a, CD8, CD90, FOXP3, HLA-DR, kappa light chain, lambda light chain, MCT, PAX5 and PDPN) using the scanpy Python package66,67. Each of the resulting clusters (n = 55) was assessed for purity of its cell type composition on the basis of marker expression and overlays of the cells in each cluster onto image hyperstacks using CODEX scripts for ImageJ/Fiji (available at ref. 68). Clusters were merged, split and/or further subclustered as appropriate to define broad cell types (for example, T cells, B cells and so on). This process was repeated within these cell types using additional markers as appropriate to annotate more granular cell subsets. Briefly, marker combinations used for cell typing of non-T cells include CD16, CD68, CD163, CD206 and HLA-DR (macrophages); CD11c, CD68 and HLA-DR (DC); CD15 (granulocytes); MCT and GRZB (mast cells); CD34, CD31, CD90 and PDPN (stromal cells); PDPN and CD21 (FDC); CD56 (NK and NK T cells); CD38, CD31, and kappa and lambda light chain (plasma cells); PAX5, CD20 and CD79a (B cells). Further, T cell subsets were derived on the basis of key subset markers (CD45, CD45RA, CD45RO, CD3, CD5, CD7, CD4, CD8, FOXP3, CXCR5, CXCL13, PD1, TIM3, CD31 and Ki67) and were validated using expression of additional T cell markers in the panel and overlays onto image hyperstacks.
Neighbourhood and nearest neighbour analysis
Neighbourhood analysis was modified based on a previously described approach47. For each cell of the joint highly multiplexed immunofluorescence data, the 25 nearest neighbours were determined on the basis of their Euclidean distance of the x and y coordinates, resulting in one ‘window’ of cells per individual cell. Next, these windows were grouped using k-means clustering based on the cell type proportions within each window. Finally, each cell was annotated by the neighbourhood of its surrounding ‘window’. k = 10 was selected on the basis of the overlays of the neighbourhood assignments with the original fluorescent and haematoxylin–eosin-stained images. Higher values of k did not result in an improved biologically interpretable number of neighbourhoods.
Interactive browsing of highly multiplexed immunofluorescence data
All tissue cores imaged in this study including staining of 52 different markers are available for interactive browsing at http://45.88.80.143:8501/.
Statistics and reproducibility
Statistical analysis and data illustration was performed in R (v4.2.1) using the R packages glmnet (v4.1-2), ggplotify (v0.1.0), maxstat (v0.7-25), R.oo (v1.24.0), rstatix (v0.7.0), viridis (v0.6.2), dplyr (v1.0.10), tidyverse (v1.3.1), FNN (v1.1.3), Matrix (v1.5-1), ggraph (v2.0.6), survival (v3.2-13), R.methodsS3 (v1.8.1), ggpubr (v0.4.0), viridisLite (v0.4.1), purrr (v0.3.4), future.apply (v1.8.1), immunarch (v0.7.0), igraph (v1.3.5), survminer (v0.4.9), readxl (v1.4.1), ggrepel (v0.9.1), SeuratObject (v4.0.4), readr (v2.1.2), future (v1.23.0), data.table (v1.14.2), ggrastr (v1.0.1), ggridges (v0.5.3), caret (v6.0-90), matrixStats (v0.61.0), Seurat (v4.1.0), tidyr (v1.2.1), pamr (v1.56.1), dtplyr (v1.2.2), ggtext (v0.1.1), cowplot (v1.1.1), lattice (v0.20-45), scales (v1.2.1), forcats (v0.5.1), tibble (v3.1.8), cluster (v2.1.2), rmdformats (v1.0.4), ggalluvial (v0.12.3), R.utils (v2.11.0), patchwork (v1.1.2), RColorBrewer (v1.1-3), stringr (v1.4.1) and ggplot2 (v3.3.6). The computational codes to exactly reproduce all analysis steps and figures is provided below (see ‘Code availability statement’). No statistical method was used to determine sample size before data collection. No data were excluded from the analyses. The experiments were not randomized. The investigators were not blinded to allocation during experiments and outcome assessment.
Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
Data availability
RNA-seq, epitope and TCR data that support the findings of this study have been deposited in the Gene Expression Omnibus (GEO) under accession codes GSE252608 and GSE252455. Highly multiplexed immunofluorescence images will be available in the BioStudies database (https://www.ebi.ac.uk/biostudies/) under accession number S-BIAD565 (ref. 69). Flow cytometry data have been deposited in figshare under https://doi.org/10.6084/m9.figshare.24915633. Source data are provided with this paper.
Code availability
The computational codes, in the form of Rmarkdown documents, for reproducing all figures are available on GitHub (https://github.com/Huber-group-EMBL/CITEseqLN-Tcells).
References
Bea, S. et al. Landscape of somatic mutations and clonal evolution in mantle cell lymphoma. Proc. Natl Acad. Sci. USA 110, 18250–18255 (2013).
Chapuy, B. et al. Molecular subtypes of diffuse large B cell lymphoma are associated with distinct pathogenic mechanisms and outcomes. Nat. Med. 24, 679–690 (2018).
Okosun, J. et al. Integrated genomic analysis identifies recurrent mutations and evolution patterns driving the initiation and progression of follicular lymphoma. Nat. Genet. 46, 176–181 (2014).
Schmitz, R. et al. Genetics and pathogenesis of diffuse large B-cell lymphoma. N. Engl. J. Med. 378, 1396–1407 (2018).
Spina, V. et al. The genetics of nodal marginal zone lymphoma. Blood 128, 1362–1373 (2016).
Morin, R. D. et al. Genetic landscapes of relapsed and refractory diffuse large B-cell lymphomas. Clin. Cancer Res. 22, 2290–2300 (2016).
Yi, S. et al. Genomic and transcriptomic profiling reveals distinct molecular subsets associated with outcomes in mantle cell lymphoma. J. Clin. Invest. 132, e153283 (2022).
Martinez, N. et al. Whole-exome sequencing in splenic marginal zone lymphoma reveals mutations in genes involved in marginal zone differentiation. Leukemia 28, 1334–1340 (2014).
Budde, L. E. et al. Single-agent mosunetuzumab shows durable complete responses in patients with relapsed or refractory B-cell lymphomas: phase I dose-escalation study. J. Clin. Oncol. 40, 481–491 (2022).
Hutchings, M. et al. Glofitamab, a novel, bivalent CD20-targeting T-cell-engaging bispecific antibody, induces durable complete remissions in relapsed or refractory B-cell lymphoma: a phase I trial. J. Clin. Oncol. 39, 1959–1970 (2021).
Locke, F. L. et al. Axicabtagene ciloleucel as second-line therapy for large B-cell lymphoma. N. Engl. J. Med. 386, 640–654 (2022).
Roider, T. et al. An autologous culture model of nodal B-cell lymphoma identifies ex vivo determinants of response to bispecific antibodies. Blood Adv. 5, 5060–5071 (2021).
Scott, D. W. & Gascoyne, R. D. The tumour microenvironment in B cell lymphomas. Nat. Rev. Cancer 14, 517–534 (2014).
Papalexi, E. & Satija, R. Single-cell RNA sequencing to explore immune cell heterogeneity. Nat. Rev. Immunol. 18, 35–45 (2018).
Suva, M. L. & Tirosh, I. Single-cell RNA sequencing in cancer: lessons learned and emerging challenges. Mol. Cell 75, 7–12 (2019).
Roider, T. et al. Dissecting intratumour heterogeneity of nodal B-cell lymphomas at the transcriptional, genetic and drug-response levels. Nat. Cell Biol. 22, 896–906 (2020).
Andor, N. et al. Single-cell RNA-seq of follicular lymphoma reveals malignant B-cell types and coexpression of T-cell immune checkpoints. Blood 133, 1119–1129 (2019).
Han, G. et al. Follicular lymphoma microenvironment characteristics associated with tumor cell mutations and MHC Class II expression. Blood Cancer Discov. 3, 428–443 (2022).
Nathan, A. et al. Multimodally profiling memory T cells from a tuberculosis cohort identifies cell state associations with demographics, environment and disease. Nat. Immunol. 22, 781–793 (2021).
Nathan, A. et al. Multimodal single-cell approaches shed light on T cell heterogeneity. Curr. Opin. Immunol. 61, 17–25 (2019).
Stoeckius, M. et al. Simultaneous epitope and transcriptome measurement in single cells. Nat. Methods 14, 865–868 (2017).
Hao, Y. et al. Integrated analysis of multimodal single-cell data. Cell 184, 3573–3587 e29 (2021).
Aibar, S. et al. SCENIC: single-cell regulatory network inference and clustering. Nat. Methods 14, 1083–1086 (2017).
Sebzda, E. et al. Transcription factor KLF2 regulates the migration of naive T cells by restricting chemokine receptor expression patterns. Nat. Immunol. 9, 292–300 (2008).
Kim, C. et al. The transcription factor TCF1 in T cell differentiation and aging. Int. J. Mol. Sci. 21, 6497 (2020).
Fontenot, J. D., Gavin, M. A. & Rudensky, A. Y. Foxp3 programs the development and function of CD4+CD25+ regulatory T cells. Nat. Immunol. 4, 330–336 (2003).
Liu, X. et al. Transcription factor achaete-scute homologue 2 initiates follicular T-helper-cell development. Nature 507, 513–518 (2014).
Friedman, J. H. Greedy function approximation: a gradient boosting machine. Ann. Stat. 29, 1189–1232 (2001).
Becht, E. et al. Reverse-engineering flow-cytometry gating strategies for phenotypic labelling and high-performance cell sorting. Bioinformatics 35, 301–308 (2018).
Pai, J. A. & Satpathy, A. T. High-throughput and single-cell T cell receptor sequencing technologies. Nat. Methods 18, 881–892 (2021).
Tirosh, I. et al. Dissecting the multicellular ecosystem of metastatic melanoma by single-cell RNA-seq. Science 352, 189–196 (2016).
Qiu, X. et al. Reversed graph embedding resolves complex single-cell trajectories. Nat. Methods 14, 979–982 (2017).
Hanna, B. S. et al. Interleukin-10 receptor signaling promotes the maintenance of a PD-1int TCF-1+ CD8+ T cell population that sustains anti-tumor immunity. Immunity 54, 2825–2841 e10 (2021).
Garni-Wagner, B. A. et al. A novel function-associated molecule related to non-MHC-restricted cytotoxicity mediated by activated natural killer cells and T cells. J. Immunol. 151, 60–70 (1993).
Quigley, M. et al. Transcriptional analysis of HIV-specific CD8+ T cells shows that PD-1 inhibits T cell function by upregulating BATF. Nat. Med. 16, 1147–1151 (2010).
Man, K. et al. Transcription factor IRF4 promotes CD8+ T cell exhaustion and limits the development of memory-like T cells during chronic infection. Immunity 47, 1129–1141 e5 (2017).
Shin, H. et al. A role for the transcriptional repressor Blimp-1 in CD8+ T cell exhaustion during chronic viral infection. Immunity 31, 309–320 (2009).
Li, J. et al. High levels of eomes promote exhaustion of anti-tumor CD8+ T cells. Front. Immunol. 9, 2981 (2018).
Newman, A. M. et al. Determining cell type abundance and expression from bulk tissues with digital cytometry. Nat. Biotechnol. 37, 773–782 (2019).
Roussel, M. et al. Functional characterization of PD1+TIM3+ tumor-infiltrating T cells in DLBCL and effects of PD1 or TIM3 blockade. Blood Adv. 5, 1816–1829 (2021).
Marcus, R. et al. Obinutuzumab for the first-line treatment of follicular lymphoma. N. Engl. J. Med. 377, 1331–1344 (2017).
Bolen, C. R. et al. Treatment dependence of prognostic gene expression signatures in de novo follicular lymphoma. Blood 137, 2704–2707 (2021).
Mintz, M. A. & Cyster, J. G. T follicular helper cells in germinal center B cell selection and lymphomagenesis. Immunol. Rev. 296, 48–61 (2020).
Thornton, A. M. et al. Expression of Helios, an Ikaros transcription factor family member, differentiates thymic-derived from peripherally induced Foxp3+ T regulatory cells. J. Immunol. 184, 3433–3441 (2010).
Scarlata, C. M. et al. Differential expression of the immunosuppressive enzyme IL4I1 in human induced Aiolos+, but not natural Helios+, FOXP3+ Treg cells. Eur. J. Immunol. 45, 474–479 (2015).
Sage, P. T. & Sharpe, A. H. T follicular regulatory cells. Immunol. Rev. 271, 246–259 (2016).
Schurch, C. M. et al. Coordinated cellular neighborhoods orchestrate antitumoral immunity at the colorectal cancer invasive front. Cell 182, 1341–1359 e19 (2020).
Blank, C. U. et al. Defining ‘T cell exhaustion’. Nat. Rev. Immunol. 19, 665–674 (2019).
Zhang, T. et al. Genetic mutations of Tim-3 ligand and exhausted Tim-3+ CD8+ T cells and survival in diffuse large B cell lymphoma. J. Immunol. Res 2020, 6968595 (2020).
Yang, Z. Z. et al. Expression of LAG-3 defines exhaustion of intratumoral PD-1+ T cells and correlates with poor outcome in follicular lymphoma. Oncotarget 8, 61425–61439 (2017).
Kersten, K. et al. Spatiotemporal co-dependency between macrophages and exhausted CD8+ T cells in cancer. Cancer Cell 40, 624–638 e9 (2022).
Li, Y. L. et al. Tumor-associated macrophages predict prognosis in diffuse large B-cell lymphoma and correlation with peripheral absolute monocyte count. BMC Cancer 19, 1049 (2019).
Nam, S. J. et al. An increase of M2 macrophages predicts poor prognosis in patients with diffuse large B-cell lymphoma treated with rituximab, cyclophosphamide, doxorubicin, vincristine and prednisone. Leuk. Lymphoma 55, 2466–2476 (2014).
Siddiqui, I. et al. Intratumoral Tcf1+PD-1+CD8+ T cells with stem-like properties promote tumor control in response to vaccination and checkpoint blockade immunotherapy. Immunity 50, 195–211 e10 (2019).
Miller, B. C. et al. Subsets of exhausted CD8+ T cells differentially mediate tumor control and respond to checkpoint blockade. Nat. Immunol. 20, 326–336 (2019).
Armand, P. et al. Efficacy and safety results from CheckMate 140, a phase 2 study of nivolumab for relapsed/refractory follicular lymphoma. Blood 137, 637–645 (2021).
Ansell, S. M. et al. Nivolumab for relapsed/refractory diffuse large B-cell lymphoma in patients ineligible for or having failed autologous transplantation: a single-arm, phase II study. J. Clin. Oncol. 37, 481–489 (2019).
Roider, T., Brinkmann, B. J. & Dietrich, S. Processing human lymph node samples for single-cell assays. STAR Protoc. 2, 100914 (2021).
Gayoso, A. et al. Joint probabilistic modeling of single-cell multi-omic data with totalVI. Nat. Methods 18, 272–282 (2021).
Friedman, J., Hastie, T. & Tibshirani, R. Regularization paths for generalized linear models via coordinate descent. J. Stat. Softw. 33, 1–22 (2010).
Andreatta, M. & Carmona, S. J. UCell: robust and scalable single-cell gene signature scoring. Comput Struct. Biotechnol. J. 19, 3796–3798 (2021).
Economou, M. et al. Proper paraffin slide storage is crucial for translational research projects involving immunohistochemistry stains. Clin. Transl. Med 3, 4 (2014).
Black, S. et al. CODEX multiplexed tissue imaging with DNA-conjugated antibodies. Nat. Protoc. 16, 3802–3835 (2021).
Lu, G. et al. A real-time GPU-accelerated parallelized image processor for large-scale multiplexed fluorescence microscopy data. Front. Immunol. 13, 981825 (2022).
Lee, M. Y. et al. CellSeg: a robust, pre-trained nucleus segmentation and pixel quantification software for highly multiplexed fluorescence images. BMC Bioinform. 23, 46 (2022).
Wolf, F. A., Angerer, P. & Theis, F. J. SCANPY: large-scale single-cell gene expression data analysis. Genome Biol. 19, 15 (2018).
Hickey, J. W. et al. Strategies for accurate cell type identification in CODEX multiplexed imaging data. Front. Immunol. 12, 727626 (2021).
CODEX-fiji-scripts. GitHub https://github.com/bmyury/CODEX-fiji-scripts
Ellenberg, J. et al. A call for public archives for biological image data. Nat. Methods 15, 849–854 (2018).
Acknowledgements
T.R. was supported by a fellowship of the German Federal Ministry of Education and Research (BMBF) and a physician scientist fellowship of the Medical Faculty of University Heidelberg. M.A.B. was supported by a Career Development award of the International Myeloma Society (IMS). D.F. was supported by the PhD programme of the European Molecular Biology Laboratory (EMBL). H.V. was supported by a fellowship of the German Federal Ministry of Education and Research (BMBF). N.L. was supported by a Heidelberg School of Oncology (HSO2) fellowship from the National Center for Tumor Diseases (NCT) Heidelberg. O.W. is supported by an Else-Kröner Excellence Fellowship from the Else-Kröner-Fresenius Stiftung (Project-ID 2021_EKES.13). Martina S. was supported by a grant of the Deutsche Forschungsgemeinschaft (DFG). S.D. was supported by a grant of the Hairy Cell Leukemia Foundation, the Heidelberg Research Centre for Molecular Medicine (HRCMM) and an e:med junior group grant of the German Federal Ministry of Education and Research (BMBF). For the data management we thank the Scientific Data Storage Heidelberg (SDS@hd) which is funded by the state of Baden-Württemberg and a DFG grant (INST 35/1314-1 FUGG). We thank C. Kolb (University Hospital Heidelberg) and the EMBL flow core facility, and the DKFZ Single-Cell Open Lab (scOpenLab) for their excellent (technical) assistance. The GALLIUM study (NCT01332968) was sponsored by F. Hoffmann-La Roche.
Funding
Open access funding provided by European Molecular Biology Laboratory (EMBL).
Author information
Authors and Affiliations
Contributions
Conceptualization, T.R., M.A.B., W.H. and S.D.; software, H.V.; validation, T.R., M.A.B., L.L.-C., C.M.S. and Martina S.; formal analysis, T.R., M.A.B., D.F., H.V., F.C., A.M., V.P. and W.H.; investigation, T.R., M.A.B., D.F., M.K., B.F., M.H., T.L. and Marc S.; resources, P.-M.B., N.L., A.B., G.M., C.M.-T., O.W., G.P.N., W.H. and S.D.; writing—original draft, T.R., M.A.B., B.J.B., W.H. and S.D.; writing—review and editing, all; visualization, T.R. and W.H.; supervision, T.R., C.M.S., G.P.N., W.H. and S.D.; project administration, T.R., W.H. and S.D.; funding acquisition, G.P.N., W.H. and S.D.
Corresponding authors
Ethics declarations
Competing interests
C.M.S. is a scientific advisor to, has stock options in and has received research funding from Enable Medicine, Inc. G.P.N. is a co-founder and stockholder of Akoya Biosciences, Inc. and inventor on patent US9909167 (On-slide staining by primer extension). The remaining authors declare no competing interests.
Peer review
Peer review information
Nature Cell Biology thanks Ralf Küppers, Guoji Guo and the other, anonymous, reviewer(s) for their contribution to the peer review of this work.
Additional information
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Extended data
Extended Data Fig. 1 Basic patient characteristics.
a) Overview showing entity, pre-treatment status, sex, assay availability and the overall T-cell proportion of a total 101 lymph node samples used in this study. b-e) Box or scatter plots illustrating the associations of patient characteristics in n = 101 biologically independent patient samples with the overall T-cell proportion determined by flow cytometry. P values and/or correlation coefficients were calculated using the two-sided Wilcoxon-test (B, C), Pearson’s linear correlation (D) or Kruskal-Wallis-test (E). Box plots: centre line, median; box limits, first and third quartile; whiskers, 1.5 x interquartile range.
Extended Data Fig. 2 Confusion tables comparing flow cytometry and CITE-seq.
a-b) Gradient boosting classifiers were trained and tested to evaluate whether T-cell subsets identified by CITE-seq can be predicted on basis of surface markers (A) or surface markers and differentially expressed intracellular markers accessible by flow cytometry (B, FoxP3, IKZF3, Ki67). Shown is the proportion of correctly predicted cells per T-cell subset across the complete CITE-seq data set (n = 51 bvles). TPr: Proliferating T-cells. TH: Helper T-cells. TFH: Follicular helper T-cells. TREG: Regulatory T-cells. TTOX: Cytotoxic T-cells. TDN: Double negative T-cells. CM: Central memory. EM: Effector memory.
Extended Data Fig. 3 Overall frequencies of T-cell subsets determined by CITE-seq or flow cytometry.
Overall frequencies of T-cell subsets determined by CITE-seq or flow cytometry are illustrated in box plots (n = 101). Outliers are shown as individual dots. Each entity and subset were tested versus rLN using a two-sided Wilcoxon-test. P values were corrected for multiple testing using the Benjamini-Hochberg procedure. Only p values ≤ 0.05 are shown. Dashed lines indicate the median of rLN. rLN: Tumor-free lymph nodes. TPr: Proliferating T-cells. TH: Helper T-cells. Box plots: centre line, median; box limits, first and third quartile; whiskers, 1.5 x interquartile range. TFH: Follicular helper T-cells. TREG: Regulatory T-cells. TTOX: Cytotoxic T-cells. TDN: Double negative T-cells. CM: Central memory. EM: Effector memory.
Extended Data Fig. 4 Multinomial and multivariate models to explain varying T-cell subset proportions.
a) Beta coefficients derived from a LASSO-regularized multinomial regression model predicting rLN, MZL, FL, MCL, or DLBCL based on the subset and overall T-cell proportions. b) Multivariate linear models were fit using sex, age, treatment status, COO (only DLBCL) as covariates, and the proportion of each T-cell subset as dependent variable in a two-sided fashion. Models were fit for the total dataset and for each entity separately, as indicated. Diameter and color of the dots indicate –log10 p value and size of the coefficient, respectively. P values were corrected using the Benjamini-Hochberg procedure and significant p values are indicated by +/−. A-B) Analysis is based on n = 101 biologically independent samples. TPr: Proliferating T-cells. TH: Helper T-cells. TFH: Follicular helper T-cells. TREG: Regulatory T-cells. TTOX: Cytotoxic T-cells. TDN: Double negative T-cells. CM: Central memory. EM: Effector memory. COO: Cell-of-origin.
Extended Data Fig. 5 5′ scRNA alongside full-length TCR sequencing and TCR diversity.
5′ scRNA alongside full-length TCR repertoire data from n = 17 biologically independent lymph node samples were mapped to the CITE-seq reference data. a) T-cell subset proportions determined by 5′ scRNA or CITE-seq reference data were correlated across 17 patient samples and 14 multimodally defined T-cell subsets. R values represent Pearson’s linear correlation coefficients. b-c) CD4 and CD8 gene expression of TTOX EM1 and TTOX EM3 cells with clone size samller (B) or greater / equal 3 (C). Dots are jittered in x and y direction. Numbers indicate proportions of CD4+, CD8+, CD4+ CD8+, or CD4−CD8− T-cells. d) TCR diversity was esimated for each LN patient sample using a rarefraction analysis (D). e-f) TCR diversity is exemplarily illustrated showing the 10 % most abundant clonotypes per sample. TPr: Proliferating T-cells. TH: Helper T-cells. TFH: Follicular helper T-cells. TREG: Regulatory T-cells. TTOX: Cytotoxic T-cells. TDN: Double negative T-cells. CM: Central memory. EM: Effector memory. TCR: T-cell receptor.
Extended Data Fig. 6 T-cell exhaustion associated with genetic features.
a) An exhaustion score was calculated based on a transcriptional signature of terminally exhaustion T-cells (see Method section for details) and then projected onto the reference UMAP plot. b) Locally estimated scatterplot smoothing (loess) of pseudotime versus exhaustion score (as shown in panel A) separately for CD4+, CD8+, CD4+CD8+, and CD4−CD8− TTOX cells. c-h) Bulk RNA-seq data from DLBCL patients from the Schmitz et al. cohort (C, n = 190) or the Chapuy et al. cohort (D-H, n = 137) were deconvoluted based on a gene expression signature of terminally exhausted T-cells. Box plots (C-D) show associations between T-cell exhaustion and cell-of-origin or genetic subtypes. Conditions were tested for significance using the Kruskal-Wallis-test. Wilcoxon-test was used as post-hoc test (C). Heatmaps (E-H) illustrate the mean estimated proportion of exhausted T-cells for somatic variants (E), amplifications (F), deletions (G), or structural variants (H), as indicated, based on deconvoluation of the bulk RNA-seq of the Chapuy et al. cohort (n = 137). Conditions were tested for significance using the two-sided Wilcoxon-test and corrected for multiple testing using the Benjamini-Hochberg procedure. After correction, all p values were > 0.05. To highlight the strongest differences, uncorrected p values ≤ 0.05 are shown to the right of each panel. Box plots: centre line, median; box limits, first and third quartile; whiskers, 1.5 x interquartile range. ABC: Activated B-cell subtype. GCB: Germinal center B-cell subtype. CN: Copy number.
Extended Data Fig. 7 TREG EM2 express IKZF3 at protein level and have no impact on overall survival in FL.
a) Representative pseudocolor and density plots showing the IKZF3 (Aiolos) protein expression in CD4+ FoxP3+ cells determined by flow cytometry. Numbers indicate percentage of positive cells based on gates as inidicated. Grey shaded histogramm represent fluorescence-minus-one control. b) IKZF3 protein expression in FoxP3+ T-cells determined by flow cytometry in n = 24 biologically independent B-cell lymphoma patient samples. Each entity was tested versus tumour-free samples (rLN) using the two-sided Wilcoxon-test. Only p values ≤ 0.05 are shown. c) 5′ scRNA alongside full-length TCR repertoire data from 11 biologically independent samples were mapped to the CITE-seq reference data. Lines connect all TREG CM1, TREG CM2 or TREG EM1 cells with any other cell given that both T-cells have the same TCR clonotype. d-e) Bulk RNA-seq data from patients with FL were deconvoluted based on gene expression signature of TREG EM2 (D) or TFH cells (E). Kaplan-Meier plots with p values of corresponding log-rank test. Box plots: centre line, median; box limits, first and third quartile; whiskers, 1.5 x interquartile range. TCR: T-cell receptor. TPr: Proliferating T-cells. TH: Helper T-cells. TFH: Follicular helper T-cells. TREG: Regulatory T-cells. TTOX: Cytotoxic T-cells. TDN: Double negative T-cells. CM: Central memory. EM: Effector memory.
Extended Data Fig. 8 Profiles of cell types and T-cell subpopulations identified by multiplexed immunofluorescence.
a-b) Overview of T-cell subsets (A) and other cell types (B) identified by highly multiplexed immunofluorescence including their marker profiles. c) Heatmap illustrating the mean expression of key marker proteins that were used to annotate the 18 cell types and T-cell subsets using highly multiplexed immunofluorescence. Expression values were scaled between 0 and 1. d) Shown are the absolute numbers of T-cells per subset, tissue core, and patient. A maximum number of two cores per patient were imaged. e) The low-granularity T-cell subpopulations detected by multiplexed immunofluorescence were aligned with the 14 high-granularity T-cell subsets identified by CITE-seq. Shown is the correlation of proportions for each of the T-cell subpopulations across n = 19 biologically indepdendent LN samples that were analyzed by both approaches. Pearson’s correlation coefficient is given for each panel (R). mIF: Multiplexed immunofluorescence. TPr: Proliferating T-cells. TH: Helper T-cells. TFH: Follicular helper T-cells. TREG: Regulatory T-cells. TTOX: Cytotoxic T-cells. CM: Central memory. EM: Effector memory. FDC: Follicular dendritic cells.
Extended Data Fig. 9 Representative lymph node tissue cores colored by T-cell subpopulation or non T-cell cell types.
Shown are five representative LN tissue cores for all entities investigated. Each cell was colored by T-cell subsets or non T-cell cell types as indicated. FDC: Follicular dendritic cell. TPr: Proliferating T-cells. TH: Helper T-cells. TFH: Follicular helper T-cells. TREG: Regulatory T-cells. TTOX: Cytotoxic T-cells. FDC: Follicular dendritic cells.
Extended Data Fig. 10 Qualitative and quantitative patterns of cellular neighbourhoods.
a–e) Representative lymph node-derived tissue cores from three or two different patients per entity. Each cell was colored according to its neighbourhood. F) Box plots showing the proportions of selected neighborhoods of each tissue core across n = 19 biologically indepdent samples. Each entity and neighborhood was tested versus rLN using the two-sided Wilcoxon-test. P values were corrected for multiple testing using the Benjamini-Hochberg procedure. Only p values ≤ 0.05 are shown. Dashed lines indidcate the median of rLN. Box plots: centre line, median; box limits, first and third quartile; whiskers, 1.5 x interquartile range. rLN: Tumor-free lymph node. TPr: Proliferating T-cells. TFH: Follicular helper T-cells. TREG: Regulatory T-cells. TTOX: Cytotoxic T-cells. FDC: Follicular dendritic cells.
Supplementary information
Supplementary Information
Supplementary Figs. 1 and 2.
Supplementary Table 1
Supplementary Tables 1–6.
Source data
Source Data Fig. 1
Statistical source data.
Source Data Fig. 2
Statistical source data.
Source Data Fig. 3
Statistical source data.
Source Data Fig. 4
Statistical source data.
Source Data Fig. 5
Statistical source data.
Source Data Fig. 6
Statistical source data.
Source Data Fig. 7
Statistical source data.
Source Data Extended Data Fig. 1
Statistical source data.
Source Data Extended Data Fig. 2
Statistical source data.
Source Data Extended Data Fig. 3
Statistical source data.
Source Data Extended Data Fig. 4
Statistical source data.
Source Data Extended Data Fig. 5
Statistical source data.
Source Data Extended Data Fig. 6
Statistical source data.
Source Data Extended Data Fig. 7
Statistical source data.
Source Data Extended Data Fig. 8
Statistical source data.
Source Data Extended Data Fig. 9
Statistical source data.
Source Data Extended Data Fig. 10
Statistical source data.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Roider, T., Baertsch, M.A., Fitzgerald, D. et al. Multimodal and spatially resolved profiling identifies distinct patterns of T cell infiltration in nodal B cell lymphoma entities. Nat Cell Biol 26, 478–489 (2024). https://doi.org/10.1038/s41556-024-01358-2
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1038/s41556-024-01358-2
This article is cited by
-
Mapping T cell landscapes in B cell lymphomas
Nature Cell Biology (2024)