Multimodal profiling identifies CD4+CXCR5+PD-1– Tfh cells as prognostic and predictive biomarkers in diffuse large B-cell lymphoma

To the Editor: Diffuse large B-cell lymphoma (DLBCL) is characterized by high heterogeneity. Elucidation of the underlying cause of the biological mechanism of DLBCL, is clinical. Here, we utilized single-cell RNA sequencing (scRNA-seq), DNA sequencing, and multiplex immunohistochemistry (mIHC) to investigate the cellular composition of primary and relapsed/refractory DLBCL (R/R DLBCL) ). We conducted a series of validations on published single-cell data and bulk sequencing data, performed in vivo validations in mouse models, and used a deep generative network model for prediction and identification. Detailed information is presented in Supplementary Methods, Supplementary Figure 1, and Supplementary Table 1, https: //links. lww. com/CM9/C618. The study protocol was approved by the Ethics Committee of Sichuan Cancer Hospital (SCCHEC-02-2022-009) and the West China Hospital Animal Ethics Committee (2020361A). Given the retrospective nature, the informed consent was waived. Samples collected from 9 patients diagnosed with DLBCL, whose information is presented in Supplementary Table 2, https: //links. lww. com/CM9/C618, were used to investigate the cellular composition of DLBCL (performed on 8 of 9 samples). After quality control procedures Supplementary Table 3, https: //links. lww. com/CM9/C618, 77, 344 single cells were acquired in scRNA-seq. We also obtained an available single-cell dataset from SE131907, SE182436, and Roider et al1 with 72, 351 cells to validate our results. After clustering, three main cell clusters were identified. Further clustering revealed 21 main cell types, including 3 types of B cells, 10 types of T/NK cells, and 8 types of myeloid cells. Cells from different sources exhibited overlap, indicating the robustness of our single-cell methods. The heatmap of canonical gene expression and the correlation analysis of variable gene further confirmed the cell annotation. For DLBCL profiles, the top three cell counts were normal B cells (Bₙor), malignant B cells (Bₘal), and CD8+ exhausted T cells (CD8+TEX) Supplementary Figure 2, https: //links. lww. com/CM9/C618. The B-cell clusters distributions and immune cell proportions differed between primary and R/R DLBCL, with CD4+ Tfh, CD4+ TN, and CD8+ TEMRA cells enriched in primary DLBCL, and Bₘal cells enriched in relapsed DLBCL. Besides, we found that most enrolled subjects harbored all of the main cell types. The ratio of observation to expectation (RO/E) revealed specific cell types associated with relapsed and primary DLBCL. Compared to primary tumors, the proportions of various cells in relapsed tumors, including CD4+ T cells, CD8+ T cells, and NK cells, were significantly lower, indicating an immune dysfunctional microenvironment Supplementary Figure 3, https: //links. lww. com/CM9/C618. The B-cell cluster was further divided into 19 clusters, and the calculated IGKC/ (IGKC2+IGLC) value was displayed. Most clusters were previously defined as Bₙor or malignant Bₘal. 1 The remaining unannotated clusters were defined as malignant-like B cells (Bₘal-like). Combined dataset with 4 datasets was used to identify this method. After B-cell definition, all normal lymph nodes harbored Bₙor, whereas all Bₘal and Bₘal-like were discovered in the DLBCL group. The type-specific genes of these B cells are listed in Supplementary Table 4, https: //links. lww. com/CM9/C619. Besides, a distinct categorization of 6 Bₙor, 6 Bₘalₗike, and 7 Bₘal clusters was identified and the Bₘal cell is the dominant cell type in the relapsed group. The Bₘal and Bₘal-like cell types highly expressed genes involved in the naive stage of B-cell differentiation. 2 DEG analysis of Bₘal and Bₘal-like and the Bₙor revealed that the DEG were enriched in several cancer-associated hallmark pathways. scTour identified three sets of DEGs along the primary-to-relapsed trajectory. Hypergeometric tests were performed on specific genes for each cluster along the trajectory via hyperR for C5 gene sets from MSigDB human collections. Cluster 1, consisting of leukocyte-mediated immunity (TNFRSF1B, C1QA), decreased along the trajectory, whereas the other two sets, consisting of GOCC-ribosomes (RPL11, RPS8, RPL5), increased. Collectively, the results demonstrated that B cells exhibited reduced leukocyte-mediated immunity and enhanced proliferation ability Supplementary Figure 4, https: //links. lww. com/CM9/C618 and Supplementary Table 5, https: //links. lww. com/CM9/C620. Bₘal and Bₘal-like cells have a higher CNA burden than Bₙor; in relapsed DLBCL, Bₘal has a higher burden than Bₘal-like. Bₘal and Bₘal-like share similar CNA patterns across 18 chromosomes. Thus, we hypothesized an upstream transcription factor (TF) as a key regulator, so used scRNA-seq to identify key TFs among DEGs. As a positive control, BATF, a well-known lymphoma transcription factor, was identified as a key regulator in activated B-cell-like diffuse large B-cell lymphoma, which was enriched in primary Bₘal/Bₘal-like cells. Other TFs (e. g. , BCL2) were enriched in primary Bₙor cells. Compared with Bₙor cells, Bₘal/Bₘal-like cells presented increased NFATC1 activity. This TF was also enriched in relapsed Bₘal/Bₘal-like cells but was largely absent in primary Bₘal/Bₘal-like and any B-nor cells. Consistent with the CNA findings, the NFATC1 gene is located on chromosome 18 Supplementary Figure 5, https: //links. lww. com/CM9/C618. The T/NK cell groups were further categorized into 19 cell types. We applied T-cell functional genes to explain the functional states of T cells and NK cells. As most T cells differentiated and matured, the naive T-cell signal decreased, while signals for regulatory T-cell (Treg) and exhausted T-cell increased. However, these observations are based on specific gene expression patterns and may not reflect the overall trend. Cell fraction analysis found that CD4+ Tregs, which contributed to an immunosuppressive TME, were significant higher in relapsed DLBCL than in primary DLBCL, while subsets of CD8+ T cells and NK cells did not significantly differ between the two groups. In the combined dataset, we observed a similar pattern. Further exploration of CD4+ T cells (excluding the Tregs) found that CD4+ Tfh1 cells constitute an independent cell cluster. DEGs (primary vs. relapsed groups) were identified in these CD4+ T cells, and the top 10 stable genes are displayed. Besides, several published lymphoma-associated genes (such as TCL1A and BCL11A) were recognized. The primary DLBCL group overexpressed some innate immunity-associated genes (such as S100A8 and PFKFB3). PFKFB3 gene was highly expressed in CD4+ Tfh1 cells. Intriguingly, the canonical immune checkpoint gene PDCD1 exhibited low and cell-specific expression in CD4+ Tfh1 cells. Moreover, the gene expression signatures associated with innate immune system gene sets were significantly enriched in the primary DLBCL group. This evidence suggests a deficiency in the innate immune microenvironment in relapsed DLBCL Supplementary Figure 6, https: //links. lww. com/CM9/C618 and Supplementary Table 6, https: //links. lww. com/CM9/C621. Trajectory analysis was performed using CD4+ TN cells as the starting point. Two branches and 2 cell fates were identified. Cell fate 2 showed higher cell density and a greater percentage of CD4+ Tfh1 cells. Branch analysis was performed at the branchpoint of the trajectory for exploring the mechanism of disease relapse. The Cluster 2 genes, including innate immune system-related genes like CXCL13 and HLA-DB, were highly expressed during cell fate 2. Gene enrichment analysis found that Cluster 1 genes were significantly enriched in TNF-α signaling via NF-κB. Along the pseudotemporal trajectory of cell fate 2, this signaling activity gradually decreased. The classification of myeloid cells in the single-cell profile of DLBCL patients is displayed. Taken together Supplementary Figure 7, https: //links. lww. com/CM9/C618, the absence of CD4+ Tfh1 cells with innate immune function in the microenvironment may be associated with DLBCL relapse. Communication analysis explored the communication between CD4+ Tfh1 cell receptors and B cell ligands in DLBCL. 3 The interaction frequency between CD4+ Tfh1 cells and Bₙor cells is lower than that between CD4+ Tfh1 cells and. Cell membrane proteins interact to facilitate cells communication. The liana R package was used to identify membrane-bound interactions from OmniPath. We applied these pairs to investigate the further interactions between target cells. The top 20 specific membrane-bound interactions highlight the crucial role of HLA as a vital ligand. APLP2 and LILRB1 may play inhibitory roles in innate immune responses via HLA class I molecules, consistent with previous results. 4 Intriguingly, among the top 20 specific interactions, the PTPRJ-LAT pair, which may negatively regulate TCR signaling in CD4+ Tfh1 cells, is specific to Bₘal and Bₘal-like cells and is nearly absent in Bₙor cells Supplementary Figure 8, https: //links. lww. com/CM9/C618. The genomic data of tumor samples from matched patients Supplementary Table 7, https: //links. lww. com/CM9/C622 were analyzed to study the activation mechanisms of the TNFα-NF-κB pathway. The top 20 somatic genomic signature genes for lymphoma from the COSMIC database show that most patients harbored mutations in KMI2D and CREBBP and TNFAIP3 mutations were enriched in replased DLBCL. These findings suggest that the TNF-α-associated pathway might play an important role in the communication between CD4+ Tfh1 cells and tumor cells, as well as in tumor relapse. Gene set enrichment analysis (GSEA) revealed activated TNF-α signaling in Bₘal or Bₘal-like cells compared to Bₙor cells. B-cell trajectory analysis validated the expression of two key genes (TNF and TNFAIP8) in the TNF-α signaling pathway, which increased along the trajectory Supplementary Figure 9, https: //links. lww. com/CM9/C618. Moreover, the crucial transcription factor NFATC1 is associated with TNF-α gene expression. To validate the clinical prognostic significance of CD4+ Tfh1 cell types, three DLBCL bulk sequencing datasets (GSE181063; GSE31312; NCICCR) were included. A high ssGSEA score of CD4+ Tfh1 cells was linked to better prognosis, consistent with scRNA-seq finding. Genes related to TNF-α were altered at both genomic and transcriptomic levels in relapsed DLBCL. Furthermore, the TNFα-NF-κB signaling pathway was activated in Bₘal or Bₘal-like cells, which may suppress the innate immune response of HLA to CD4+ Tfh1 cells, inhibiting antitumor immunity and promoting tumor recurrence Supplementary Figure 10, https: //links. lww. com/CM9/C618. To examine whether CD4+ Tfh1 cells correlate with R-CHOP treatment response in DLBCL patients, we established a multiplexed immunofluorescence panel to characterize the interactions among neoplastic cells, T cells, and CD4+CXCR5+PD-1– Tfh cells in the TME. We enrolled 23 R-CHOP responders and 19 nonresponders for mIHC analysis and clinical evaluation Supplementary Table 8, https: //links. lww. com/CM9/C618. Nonresponders had more PD-1+ cells and fewer CXCR5+ cells than responders, despite similar levels of total CD8+ and CD4+ T cells. Further assessment of the clinical relevance of CD4+CXCR5+PD-1– Tfh cells found higher levels of these cells to be linked to longer overall survival. To better understand the spatial interaction, we detected the distribution of CD8+ T cells within a radius of 100 μm from the nuclear center of the CD4+CXCR5+PD-1– Tfh cells and observed that, in responders, CD8+ T cells were significantly closer to CD4+CXCR5+PD-1– Tfh cells than in nonresponders, and shorter distances correlated with improved survival. Moreover, we collected and analyzed five pairs pre-treatment and relapse specimens to track the dynamic changes of immune cell within the TME. The CD8+ T cells, CD4+ T cells, and PD-1+ cells in tissues were decreased, while CD4+CXCR5+PD-1– Tfh cells increased after R-CHOP treatment, suggesting that R-CHOP treatment could affect PD-1 expression in immune cells. Besides, similar results, greater intratumoral CD4+CXCR5+PD-1– Tfh cell infiltration in the responders than in the nonresponders, were obtained in animal experiments. This evidence suggests that CD4+CXCR5+PD-1– Tfh cells (CD4+ Tfh1 cells) exist in the DLBCL TME and could be favorable indicators for R-CHOP treatment Supplementary Figure 11, https: //links. lww. com/CM9/C618. We identified CD4+ Tfh1 cells as a predictive biomarker for R-CHOP treatment response in DLBCL patients. However, detecting CD4+ Tfh1 cells relies on mIHC images, which are challenging to obtain. We developed an end-to-end conditional generative adversarial network (GAN) for predicting mIHC images from DAPI-stained images. The model was trained on approximately 32, 000 paired patch images and performed well. Further evaluation demonstrated an accuracy of 0. 86 and a recall score of 0. 68 in identifying CD4+ Tfh1 cells. The generated mIHC images effectively detected CD4+ Tfh1 cells. These findings indicate that our model can efficiently generate mIHC images that are capable of accurately identifying CD4+ Tfh1 cells Supplementary Figure 12, https: //links. lww. com/CM9/C618. We comprehensively analyzed the cellular landscape of DLBCL from the primary diagnosis to the relapsed state following R-CHOP treatment. We identified CD4+CXCR5+PD-1– T Tfh cells as a key cell subtypes involved in DLBCL relapse within the TME. We confirmed that CD4+CXCR5+PD-1– Tfh cells are a prognostic and predictive biomarker for DLBCL patients response to R-CHOP treatment. This approach characterizes tumor cells and their immune microenvironments, providing insights into cellular and molecular networks changes during disease progression and treatment response of DLBCL. Besides, we established a GAN to predict mIHC images from DAPI-stained images and validated it. Further discussion, see the Supplementary File, https: //links. lww. com/CM9/C618. In summary, we characterized the shared and distinct features of primary and relapsed DLBCL at single-cell resolution. The deficiency of CD4+CXCR5+PD-1– Tfh cells with innate immune functionality and the activation of TNF-NF-κB signaling in malignant B cells are vital cellular mechanisms underlying the TME in R/R DLBCL. Funding This work was supported by a 1. 3. 5 project for disciplines of excellence from West China Hospital of Sichuan University (ZYYC24005), the National Natural Science Foundation of China (82073158 and 32200532), the National Key Research and Development Program of China (2022YFC3401600), Leading Innovative and Entrepreneur Team Introduction Program of Zhejiang (2022R01002), and Key research and development program of Zhejiang Province. Conflicts of interest The authors H. J. , P. W. , S. Y. , H. K. and X. Z. are coinventors on patents for the methods described in this study, which were filed by Force Biotech Ltd. H. J. is a cofounder of Force Biotech Ltd.