The tissue samples used in this study were obtained from the archives of the Institute of Pathology of Lausanne University Hospital, Switzerland ( Supplementary Table S1 , Supplementary Figure S1A ). LN cells were obtained from the Centro de Investigacion en Enfermedades Infecciosas (CIENI), Instituto Nacional de Enfermedades Respiratorias (INER) in Mexico City, Mexico. All procedures were in accordance with the Declaration of Helsinki and approved by i) the Canton de Vaud-CER-VD, Switzerland for control LN tissues (#2021-01161), ii) the local research consent authorities for LN analysis and the reuse of clinical data from the five SLE patients at CHUV, with oral consent provided by the patients after a thorough explanation of the study by the investigators, and iii) the Research Committee and the Ethics in Research Committee of the National Institute of Respiratory Diseases “Ismael Cosío Villegas,” Mexico City as part of the ‘C71-18’ protocol.

Fresh tissues samples were promptly fixed overnight in formalin, following biopsy, and processed into paraffin embedded (FFPE) blocks using standard procedures. All subsequent tissue processing was carried out in our Institute. The blocks were sequentially cut into 4 μm sections and prepared on Superfrost glass slides (Thermo Scientific, Waltham, MA, USA, Ref. J1800AMNZ), dried overnight at 37°C and stored at 4°C. Before staining, the slides were heated on a metal hotplate (Stretching Table, Medite, Burgdorf, OTS 40.2025, Ref. 9064740715) at 65°C for 20 min. This melting step ensures the proper adherence, deparaffinization and optimal epitope exposure of the tissue section.

Transcriptomic profiling was performed using the commercially available platform GeoMx Digital Spatial Profiling (Nanostring) according to the manufacturer’s instructions. 4 µm FFPE tissue sections from SLE (N=4) and control (N=3) LN were used. Follicular Regions of Interest (ROIs, n=9-12) were identified based on the CD3, CD20, PD-1 in situ staining pattern (active follicles/characterized as CD20-dense areas populated by CD3⁺PD1⁺ cells were selected for analysis) before the probe-hybridization step. The average ROI area was 126336.4 µm² (± 59758.8728, SD) for control and 157725.2281 µm² (± 78722.80994, SD) for SLE ROIs.

For data Processing and quality control (QC), we followed standard GeoMx processing workflows (https://doi.org/10.18129/B9.bioc.GeoMxWorkflows.). The processing of DCC files and quality checks at various levels (segment, probe, and gene) were carried out with the help of the ‘GeoMxWorkflows’(https://doi.org/10.18129/B9.bioc.GeoMxWorkflows.),’GeomxTools’(https://doi.org/10.18129/B9.bioc.GeoMxWorkflows.) and ‘NanoStringNCTools’ (https://doi.org/10.18129/B9.bioc.NanoStringNCTools.). packages in R (version 4.3.2). First, we adjusted all zero expression counts to one to facilitate downstream data transformation. Next, we applied several quality control cut-offs recommended by NanoString, including: a minimum of 1000 reads, 80% trimming, stitching, and alignment, 50% sequencing saturation, a minimum negative control counts of 1, a maximum of 1000 reads observed in NTC wells, and a minimum area of 1000. In addition to the segment quality control, probes with an average expression count across segments below 10% of the average for other probes targeting the same gene were excluded. Likewise, we also removed probes deemed outliers in over 20% of segments, using Grubb’s test (https://doi.org/10.18129/B9.bioc.GeoMxWorkflows.) as the outlier detection method. Finally, we excluded segments where less than 5% of the genes in the panel were detected above the quantification limit (LOQ, calculated as two standard deviations beyond the mean), and genes falling below the LOQ in at least 10% of the segments.

To perform batch correction, we first normalized the raw data using the Trimmed Mean of M-values (TMM) method with the R package “standR” 15, which adjusts for differences in library sizes and composition between RNA-seq samples. We then applied the RUV-4 correction from the same package 16, which removes unwanted variations by identifying negative control genes and calculating scaling factors for batch correction. In our analysis, we identified several negative control genes and set the number of scaling factors to 2.

To correct for batch effects, we first normalized the dataset using the Trimmed Mean of M-values (TMM) method, implemented via the ‘standR’ package in R (https://doi.org/10.18129/B9.bioc.standR). This method adjusts for discrepancies in library size and composition between RNA-seq samples. Following this, we applied the RUV-4 correction (27), which uses negative control genes to calculate scaling factors and remove unwanted variability due to batch effects. In this study, we identified 300 negative control genes and set the number of scaling factor to 3 for the RUV-4 correction.

For the differential gene expression analysis, we employed the limma-voom workflow (28, 29). That is, we built a linear model using a design matrix that accounted for the treatment variable and RUV-4 correction scaling factors as covariates. We then identified differentially expressed genes between treatments based on an adjusted p-value cut-off of less than 0.05. Finally, we performed gene set enrichment analysis (GSEA) on the differentially expressed genes to identify biological pathways of interest (30). We applied the ‘fry’ method from the R package ‘limma’ (28) and used canonical pathway gene sets from the Reactome (https://doi.org/10.18129/B9.bioc.msigdb.). The results were then analyzed and visualized using the R package ‘vissE’ (https://doi.org/10.18129/B9.bioc.vissE).

To perform cell deconvolution on the GeoMx data, we followed the ‘SpatialDecon’ R package pipeline (31). We used lymph node single-cell RNA-seq data from HIV patients as our reference dataset. We focused on cells classified as T cells (CD4, CD8, T_FH), B cells and dendritic cells. Likewise, we only considered genes expressed in immune and stromal cells (genes used in ‘safeTME’ cell profile matrix from ‘SpatialDecon’) and some additional immune cell-specific genes if needed. To avoid either different population sizes across or genetic variability within cell types to interfere with the deconvolution methods, we first clustered each cell type into smaller populations with well-defined gene signatures, using a k-means algorithm with a total of 15 centers (32). Then, using these new annotations, we generated the lymph-node specific cell profile matrix with ‘SpatialDecon’, and perform the deconvolution on the GeoMx data (31). To study the effects of deconvolution-derived cell proportions on gene expression, we designed several modelling strategies to address the relationship between the expression of specific genes and cell populations. Preliminary tests of the association between gene expression and deconvoluted data generated ΄noisy΄ results, mainly because our deconvolution approach yielded unsatisfactory outcomes in terms of successfully distinguishing the relevant T cell subsets. For this purpose, we merged all T cell populations together, considering that our imaging results proved that most T cells in our ROIs (follicles) are T_FH. We used the R package ‘DESeq2’ (33) to model gene counts with a negative binomial regression as a function of the RUV-4 scaling factors, segment area, treatment, and the percentage for a given cell population. We applied two models: a first model without the treatment effect and a slope for the cell percentage, and a second model with the treatment effect and treatment-specific slopes for the cell percentage. These models allowed us to understand the extent to which the presence of a particular cell population in each segment is related to the expression levels of specific genes.

Cells from viremic HIV LNs were used for the generation of a data set to support the GeoMx cell deconvolution analysis. Briefly, FASTQ files were uploaded to Cell Ranger (10X Genomics cloud), and no depth normalization was carried out. The generated filtered count matrix was further analyzed using the Seurat package in R. Doublet cells were removed from the analysis using the DoubletFinder package in R. Cell annotations were performed using SingleR and the reference expression dataset was derived from the Monaco Immune Data atlas from the cell dex R package. Differentiation gene expression was assessed using the MAST R package. Sequencing data were deposited in the GEO database under the accession ID GSE288212.

Multiplex antibody staining was performed on the Ventana Discovery Ultra Autostainer (Roche Diagnostics) as previously described (34). Briefly, the procedure is consisted of consecutive rounds of antigen retrieval, antibody blocking steps (using the Opal blocking/antibody diluent solution) for non-specific binding of antibodies, staining with primary antibodies (details on antibodies, clones and panel are listed in Supplementary Tables S2, S3 ), incubation with secondary HRP-labeled antibodies for 16 min, then detection with optimized fluorescent Opal tyramide signal amplification (TSA) dyes (Opal 7-color Automation IHC kit, from Akoya, Ref. NEL821001KT). Repeated antibody denaturation cycles were introduced. Tissue sections stained with Alexa-Conjugated antibodies (panel 5) were incubated with the primary Ab for 90 min and with the secondary Ab (if needed) for 90 min at RT. Just for panel 5, a cycling staining approach was followed. After staining with Alexa-conjugated Abs, the Ab complexes were dissociated using a citric acid-based buffer and a second staining cycle followed with Opal-coupled Abs. The images were aligned using SimpleITK (35) as an Imaris extension (Imaris software version 9.9.0, Bitplane) using a common marker (specifically a nuclear dye). The samples were counterstained with Spectral DAPI from Akoya Biosciences (NEL821001KT) for 4 min or SYTO45 (1/10000 dilution in TBS-T, CatNo10297192, ThermoFischer Scientific) for 35 min, rinsed water with soap and mounted using DAKO mounting medium (Dako/Agilent, Santa Clara, CA, USA, Ref. S302380-2).

Multispectral images (MSI) were acquired using the Vectra Polaris imaging system from Akoya or the Leica Stellaris 8 SP8 confocal system. Images (512x512 and 1024x1024 resolution) were acquired using 0.75× optical zoom and a 20× objective (NA) (unless otherwise specified) for all the images used for quantification. Frame averaging or summing was never used while acquiring the images. At least 80% of each section was imaged, to ensure an accurate representation and minimize selection bias. For images acquired with Vectra Polaris dye unmixing was conducted using inForm image analysis software, version 2.4.8 (Akoya Biosciences, Marlborough, MA 01752, USA) and for those acquired with Leica Stellaris 8 we used Leica LAS-AF Channel Dye Separation module which was included in LAS-X (Leica Application Suite X (LAS-X)-4.6.1.27508 software). All the tissues stained for the same panel were imaged using the same platform to be able to process them all together and quantify their cell densities (normalized cell counts per mm²) and/or frequencies.

For images acquired using the Vectra Polaris the Phenochart 1.0.12 software (Akoya Biosciences, Marlborough, MA 01752, USA), a whole-slide contextual viewer with annotation capability was used for navigation around slides and for identification of Regions of Interest (ROIs). MSI were analyzed using the inForm image analysis software, version 2.4.8 from Akoya. Firstly, the images were unmixed and were segmented using CD20, PD-1, KI67 and DAPI as components for training into specific tissue ROIs (GC, non-GC CD20-enriched area, low CD20 area). Individual cells were segmented using the counterstained-based adaptive cell segmentation algorithm, with the help of nuclear (DAPI and BCL-6) and membrane (CD4) markers. Following tissue and cell segmentation, the phenotyping configuration was used, by assigning around 100 cells to the positive phenotype for each marker, while selecting additional 100 cells characterized as “other” for the negative phenotype, choosing across several images. The quantification was based on PhenoptrReports from Akoya Biosciences, an automated R-script platform, where separated merged cell segmentation data, retaining the same tissue segmentation, were created for each phenotyped marker and were consolidated afterwards. The different combinations of phenotyped populations were defined and the analysis was run creating reports, which contained the number of analyzed fields (slide summary), cell counts, cell percentages, and cell densities.

Quantitative data was generated from images captured with Leica Stellaris using Histo-cytometry analysis (given the low abundance of PD1^hiCD57^hi GATA3^hi cells, images from panel 3 were analyzed using both approaches as a cross validation step), as previously reported (34, 36). In brief, the Surface Creation module of Imaris software (version 9.9.0 Bitplane) was used to generate 3-dimensional segmented surfaces (based on the nuclear signal) of unmixed images. Data generated from Histo-cytometry, such as average voxel intensities for all channels, in addition to the volume and sphericity of the 3-dimensional surfaces, were exported in Microsoft Excel format. The files were converted to comma separated value (.CVS) files, and the data were imported into FlowJo (version 10) to be further analyzed and quantitated. Well-defined areas devoid of background staining were included in the analysis, and the data were quantified either as relative frequencies or as cell counts normalized to the imaged follicular area. The area and other morphological characteristics (circularity, solidity) of individual follicles were calculated using FIJI software (37). Optimal z-stack settings were applied in all collected images. Maximum Intensity Projections (MIPs) are presented throughout the manuscript.

The distance between relevant cell subsets (CD20^hi/dimKi67^hi, PD1^hiCD57^hi, PD1^hiGATA3^hi) was calculated with Python 3.10.9 using the SciPy library (38). X and Y coordinates were used to create the matrix interaction for each cell phenotype, and the median distance was also extracted. Additionally, to calculate the probability of observing different patterns of cellular distribution across ROIs and individuals, we studied the curves generated from the Ripley’s G function and the theoretical Poisson curve using pointpats 2.3.0 (https://doi.org/10.5281/zenodo.7706219). The area between the empirical and theoretical Poisson curve was extracted using the NumPy library (39). ROIs harboring at least 20 positive cells for each cell subset under investigation were analyzed.

The Mann-Whitney test and simple linear regression analysis were used to analyze the imaging data. Analyses and graphs were generated using GraphPad Prism 8.3.0 software. Regarding the ROI measurements (Area, Circularity, Solidity), we applied a Mixed Effect Model using a python script comprising both fixed effects corresponding to various follicular areas and random effects originating from individual donors. A p-value < 0.05 was considered statistically significant.

To investigate the unique characteristics of SLE follicular landscape, we collected LN sections from treatment-naive active SLE patients exhibiting lymph node involvement and from non-autoimmune, cancer- and HIV-free control donors harboring hyperplastic, active follicles ( Supplementary Table S1 ) ( Supplementary Figure S1A ). Despite the different etiology of follicular activation/maturation, our selected non-autoimmune tissues represent a strict control group, providing a powerful evaluation for the SLE follicular profiling. To start gaining insights on the complex molecular profile of SLE follicles, we performed spatial transcriptomic analysis using the GeoMX Digital Spatial Profiler platform. Secondary mature follicles were defined as CD20^hi/dim-dense microanatomical structures populated by CD3^hiPD1^hi (T_FH) cells ( Supplementary Figure S1B ). Principal component analysis (PCA) of batch-corrected sequencing data revealed that SLE follicles exhibit a transcriptionally distinct profile compared to non-autoimmune reactive controls ( Figure 1A , Supplementary Figure S1C ). Volcano analysis showed that several inflammation-related genes (STAT1, CXCL9, CXCL11, IRF1 etc.) were upregulated in SLE over non-autoimmune control follicles, while IL4-related (IL4R, FCER2) and cellular oxidation-related (TXNIP) genes were upregulated in control follicles ( Figure 1B ). Notably, most of the top upregulated Differentially Expressed Genes (DEGs) in SLE follicular areas were interferon stimulated- (ISGs like IRF7, IFI6, IFI44L, ISG20 etc) or plasma cell-related (PRDM1, IRF4, IGHG1 etc) genes, a profile independent of the gender of SLE individuals. ( Figure 1C ). In line with the DEGs profile, we detected significantly higher expression of type I IFN and immunoglobulin complex pathways in SLE compared to control follicles ( Figure 1D ). Compared to SLE, control follicles were characterized by significant enrichment of T_FH cell differentiation and IL1b-related pathways ( Figure 1D ).

To further validate the potent transcriptomic type I IFN signature of SLE follicles a mIF assay, using an antibody against IFN-α2 (panel 4, Supplementary Tables S2, S3 ), was applied ( Figure 1E ). To analyze and quantify IFNα2^hi cells in follicles (defined as CD19-dense areas), we employed the histo-cytometry approach (40) ( Supplementary Figure S2A ). Contrary to extrafollicular area, a clear trend (p=0.0952) towards increased levels, both as frequency or normalized counts, of IFN-α2^hi cells in SLE compared to control follicles was found ( Figure 1F , Supplementary Figure S2B ), further supporting our transcriptomic findings. Altogether, our findings revealed a highly inflammatory follicular microenvironment, dominated by a type I IFN signature, that could affect the development of T_FH/GC-B cell responses in SLE.

Given the potential expression of IFNAR by several follicular cell types (41), we aimed to investigate the main cellular targets of the observed type I interferon transcriptomic signature in SLE. To this end, we applied a cell deconvolution pipeline using a single-cell dataset from human LNs as a reference. As an internal validation of our pipeline, we found that selected T (e.g. CD3, TRAC) and B (e.g. CD19, AICDA, MS4AI) cell genes were correctly assigned to the corresponding cell types ( Supplementary Figure S2C ). The approximate calculated relative proportions of T and B cells showed a dominance by B cells (≈50%) with T cells occupying a ≈25% of total deconvoluted cells as expected ( Figure 2A ). We then sought to determine the contribution of T, B and dendritic cell types on interferon stimulated gene (ISG) expression. First, using the full dataset, consisted of both control and SLE ROIs, we observed a higher contribution of T and dendritic cells, compared to B cells, in the observed ISG expression profile ( Figure 2B ). By dividing the dataset into control and SLE ROIs, we found that DCs have a higher contribution to the increased expression of ISGs in the SLE group, a finding consistent with previous single-cell RNA sequencing studies in blood suggesting that innate immune cells have a stronger IFN fingerprint than adaptive immune cells in SLE (42) ( Figure 2C ). Furthermore, for SLE ROIs, a pattern favoring T cell correlation with increased ISG expression compared to B cells was observed for most of the ISGs while B cell correlation was higher only in one case (IRF9) ( Figure 2C ). Therefore, our deconvolution approach suggests that follicular T cells are presumably more responsive to type I IFN signaling compared to B cells, despite the expression of IFNAR by both cell types.

The above-mentioned spatial transcriptomic profile of SLE follicles raises the possibility of dysregulated follicular immune dynamics, in particular T-cell dynamics. To this end we developed imaging panels allowing for the in-situ detection, phenotyping, and quantitative analysis of different T- and B- cell subsets ( Supplementary Tables S2, S3 ). First, assessment of geometrical characteristics of follicular areas (identified based on the density of CD19^hi/dim B cells) ( Figure 3A ) revealed that SLE follicles tend to be larger (p=0,08) and significantly less solid (more irregular boundaries) (p=0.008) compared to controls ( Figure 3B ) while a similar circularity was found between SLE and control LNS ( Supplementary Figure S2D ). T_FH cells represent the main follicular T cell population (40). Their indispensable role for the activation, maturation of B cells and the generation of high affinity antigen-specific antibodies are well established (42). The use of PD-1 as a T_FH cell biomarker ( Figure 3A ), showed comparable cell densities (cell counts normalized to mm²) of CD4^hiPD1^hi T_FH cells between SLE and control follicles ( Figure 3C ). T_FH cells were further analyzed based on the expression of Ki67 and Bcl6 ( Figure 3D ). In line with previous reports (18, 43), only a subset of T_FH cells exhibited a proliferative capacity ( Figure 3E , left panel). No significant differences were observed between control and SLE proliferating T_FH cells ( Figure 3E , left panel). However, the cell density of T_FH cells expressing Bcl6, a master regulator of T_FH cells (43) was significantly downregulated(p<0,05) in SLE compared to control LNs ( Figure 3E , right panel). Therefore, the significantly altered morphology of SLE follicles is associated with a dysregulated differentiation of T_FH cells characterized by the significantly reduced prevalence of Bcl6hi T_FH cells.

The T_FH cell compartment is characterized by phenotypical and functional heterogeneity. A relatively high expression of GATA-3, a transcription factor associated with IL-4 production, has been described in human follicular areas by T_FH cells (19, 44, 45). Moreover, CD57 expression of T_FH cells has been associated with a unique position, molecular and functional profile (18, 46, 47). Furthermore, CD57^hi T_FH cells were found to be potent producers of IL-4 compared to CD57^lo T_FH cells, at least in vitro ( 18). Therefore, we investigated the in situ phenotype of T_FH cells with respect to CD57 and GATA-3 expression (panel 3, Supplementary Tables S2, S3 ) ( Figures 4A, B , Supplementary Figure S3A ) by employing histo-cytometry analysis ( Supplementary Figure S3B ). In line with the data generated from our panel 1 ( Figures 3C, E ), we measured similar cell densities for PD1^hi, PD1^hi Ki67^hi cells between SLE and control LNs ( Figure 4C ). However, a clear trend (p=0.06) for reduced PD1^hiCD57^hiGATA3^hi T_FH cell densities was revealed in SLE compared to control follicles ( Figure 4C ). A similar trend between control and SLE follicles was found when the relative frequencies of CD57^hiGATA3^hi T_FH cells (% of PD1^hi cells) were plotted ( Figure 4C , right panel).

To further investigate the perturbed differentiation of T_FH cells observed in SLE follicles, the gene expression of molecules that could mediate T_FH cell differentiation (IL21, CD200, MAFF, ICOSLG, Bcl6b) were analyzed using our spatial transcriptomic data. These genes, which can be expressed by more than one cell type, were downregulated in SLE compared to control follicles ( Figure 4D ). Taking into consideration that IL-4 is expressed mainly, if not exclusively, from T cells we reasoned that the reduced presence of potential IL4-producing T_FH cells would have an impact on IL4-related genes of SLE follicles. Indeed, SLE follicular areas exhibited a downregulated IL-4 signaling gene signature compared to controls, further supporting our imaging findings ( Figure 4E ). Furthermore, pathway analysis revealed that IL-4 related pathways were enriched in control compared to SLE follicles ( Figure 4F ). In addition to their capacity for secretion of critical cytokines (e.g. IL-21, IL-4), spatial positioning of T_FH cells is crucial for their optimal interaction with neighboring GC-B cells. To this end, the distribution profile (judged by the ‘G function’ parameter (48) of T_FH cell subsets was calculated. Follicles with at least 20 cells for each of the cell types were investigated, precluding thus the analysis of PD1^hiCD57^hiGATA3^hi T_FH cells. However, PD1^hiCD57^hi and PD1^hiGATA3^hi T_FH cells showed a significantly higher dispersed distribution in SLE compared to control follicles ( Figure 4G ) that could affect their interaction with B cells within the follicular areas. In conclusion, reduced cell densities of PD1^hiCD57^hiGATA3^hi cells in follicles could lead to impaired IL4-related responses, which may affect the development and the maturation of GC B cells.

Then we focused our investigation on analyzing relevant B cell subsets (panel 1, Supplementary Tables S2, S3 ) in follicular Regions of Interest (ROIs) identified based on the expression pattern of CD20 and Ki67 (CD20^hi/dimKi67^lo-follicular enriched in Mantle Zone, hereafter F-non-GC and CD20^hi/dimKi67^hi/dim-follicular enriched in LZ/DL, hereafter F/GC) ( Figure 5A , Supplementary Figure S4A ). Contrary to SLE, a consistently higher number of CD20^hi/dim B cells in the F/GC compared to F/non-GC follicular area was monitored in control LNs ( Figure 5B ). Within the F/GC area, however, no significant differences of bulk CD20^hi/dim or CD20^hi/dimKi67^hi B cell density were observed between control and SLE LNs ( Figure 5C , upper panel, Supplementary Figure S4B ). Notably, a clear trend (p=0.0635) for lower cell density of CD20^hi/dimBcl6^hi B cells was found in SLE to control F/GCs ( Figure 5C , upper panel). Proliferating CD20^hi/dim B cells expressing Bcl6 (CD20^hi/dimBcl6^hiKi67^hi, dominating the Dark Zone) exhibited similar cell densities whereas their non-proliferating counterparts (CD20^hi/dimBcl6^hiKi67^lo, mainly found in the Light Zone) were significantly decreased in SLE compared to control F/GCs ( Figure 5C , lower panel).

Given the mutual regulation between T_FH and GC B cells (49), we asked whether the counts of these two immune cell types are correlated in our tissue cohort. A significant correlation between T_FH and GC-B cells was found only in control LNs further supporting our hypothesis of deregulated GC-responses in SLE ( Figure 5D ). As a surrogate of T/B cell proximity, presumably reflecting the possibility for their interaction too, we measured the minimum Euclidean distances between T_FH and CD20^hi/dimKi67^hi B cells (enriched in dark zone). The minimum number of cells required per F/GC for this analysis excluded PD1^hiCD57^hiGATA3^hi T_FH cells from this comparison. However, a significantly greater minimum distance between PD1^hiCD57^hi or PD1^hiGATA3^hi T_FH and CD20^hi/dimKi67^hi B cells was found in SLE compared to control F/GCs ( Figure 5E ) suggesting a lower possibility for T_FH-B cell interaction in SLE. Therefore, SLE F/GCs are characterized by a concomitant dysregulated dynamics of both T_FH and B cells.

Disturbed B cell differentiation/maturation can lead to the generation of pathogenic B cell subsets like Age-associated B cells (50). ABCs (CD19^hi/dimCD11c^hiTbet^hi) are crucial mediators of SLE autoreactive humoral responses (51). Our mIF assay (panel 4, Supplementary Tables S2, S3 ) ( Figure 6A , Supplementary Figure S4C ) showed increased prevalence (both as frequencies and cell density) of CD11c^hi/dimTbet^hi B cells in SLE follicular (CD19-dense areas) and especially extrafollicular areas (p<0,05), compared to control LNs, regardless the gender of SLE individuals ( Figure 6B ). Furthermore, we found a positive correlation between extrafollicular ABC and extrafollicular or follicular IFNα2^hi cell densities in SLE LNs ( Figure 6C ). Notably, the strongest positive correlation was observed between the follicular IFNα2^hi and extrafollicular ABC cell densities ( Figure 6C ). Therefore, the dominant follicular type I IFN signature and the increased generation of atypical B cell subsets in the extrafollicular area may represent molecular and cellular mechanisms contributing to the described dysregulated development of GC-B cell responses in SLE follicles.

Given the transcriptomic profile dominated by type I IFN in SLE follicles, we sought to analyze the expression of critical soluble mediators as well as innate immunity subsets in our tissue cohort. First, a mIF assay allowing for the analysis of CD20, FDC, IL-21 and CXCL13 (panel 5, Supplementary Tables S2, S3 ) was applied ( Figure 7A ). Our gating strategy allowed us to analyze IL2-1^hi and CXCL13^hi cells in both follicular and extrafollicular areas ( Figure 7B ). Contrary to FDC associated IL21 ( Supplementary Figure S5A , left panel), follicular IL21^hiFDC^lo positive cells were more abundant in SLE (p=0.057) compared to control follicles ( Figure 7C , left panel). A similar but less evident profile (p=0.1143) was found for the extrafollicular IL21^hi cells ( Figure 7C , right panel). No differences were found when the cell densities of CXCL13^hi cells were analyzed either in extrafollicular ( Figure 7D ) or follicular ( Figure 7D , Supplementary Figure S5 A right panel) areas of SLE and control LNs.

Follicular responses could also be directly or indirectly affected by innate immune cells and CD8^hi T cells (52–56). Analysis of bulk CD11c (panel 5, Supplementary Tables S2, S3 , Supplementary Figure S5B ) revealed a trend for higher CD11c^hi cells in the T cell zone (defined as CD4-dense extrafollicular areas) of control LNs ( Supplementary Figure S5C ). Analysis of CD14 and CD16 cell subsets (panel 4, Supplementary Table S2 , Supplementary Figures S5D, E ) showed similar cell densities for CD14^hi cells, in follicular and extrafollicular areas, among SLE and control LNs ( Supplementary Figure S5F ). Notably, non-classical CD14^loCD16^hi monocytes were increased in SLE follicles ( Figure 7E , left panel). In the extrafollicular areas, CD14l°CD16^hi monocytes were also elevated in SLE compared to control LNs, however without reaching statistical significance ( Figure 7E , right panel).

Next, the cell densities of bulk and effector CD8^hi T cells (panel 2, Supplementary Tables S2, S3 ) were analysed ( Supplementary Figure S5G , upper panel). Similar cell densities of bulk and potential CTLs (GrzB^hiPrf^hi CD8^hi T cells) were measured in the extrafollicular areas between SLE and control LNs ( Supplementary Figure S5G , lower panel). However, a clear trend (p=0.057) for higher cell density of proliferating Ki67^hiCD8^hi T cells was found in SLE compared to control LNs ( Supplementary Figure S5G , lower panel). Collectively, these results indicate that SLE follicles exhibit alterations in IL-21-fuelled GC-reactivity and increased infiltration of inflammatory non-classical monocytes.