Whole transcriptome-sequencing and network analysis of CD1c+ human dendritic cells identifies cytokine-secreting subsets linked to type I IFN-negative autoimmunity to the eye

Background Inflammatory subsets of CD1c+ conventional dendritic cells (CD1c+ DCs) are promoted by type I interferons (IFN), but the molecular basis for CD1c+ DCs involvement in conditions not driven by type I IFNs is unknown. Methods Our objective was to use RNA-sequencing of blood CD1c+ DCs and high-dimensional flow cytometry of two cohorts of autoimmune uveitis patients and healthy donors to characterize the CD1c+ DCs population of type I IFN-negative autoimmune uveitis. Results We report that the CD1c+ DCs pool from patients with autoimmune uveitis (n=45) is skewed towards a transcriptional network characterized by surface receptor genes CX3CR1, CCR2, and CD36. We confirmed the association of the transcriptional network with autoimmune uveitis by RNA-sequencing in another case-control cohort (n=35) and demonstrated that this network was governed by NOTCH2-RUNX3 signaling. Unbiased flow cytometry analysis based on the transcriptional network identified blood CD1c+ DC subsets that can be distinguished by CX3CR1 and CD36 surface expression. A CD36+CX3CR1+CD1c+ DC subset within the novel DC3 population was diminished in peripheral blood of patients, while CD1c+ DCs expressing CD36 and CX3CR1 accumulate locally in the inflamed eye. The CD36+CX3CR1+CD1c+ DC subset showed a differential capacity to produce cytokines, including TNF-alpha, IL-6, and VEGF, but not IL-23. Conclusion These results show that CD1c+ DC subsets defined on the basis of surface expression of CD36 and CX3CR1 are linked to type I IFN-negative human autoimmune uveitis and show a differential capacity to secrete proinflammatory mediators that drive its pathophysiology. Graphical Abstract

CD1c+ DC pool is unlikely to reflect an activation continuum (Fig. 3C). In fact, TLR-189 stimulation (with LPS, LTA, or R848) resulted in a strong upregulation of RUNX3 and 190 downregulation of CD36, which is the opposite of the expression pattern detected in patients. 191 Also, overnight stimulation with cytokines GM-CSF and IFN-alpha that are implicated in the 192 promotion of specific cDC2 subsets (19,20) did not decrease the expression of RUNX3 (Fig.  193   3C). In general, the transcriptional signature of autoimmune uveitis did not resemble the 194 gene signature of in vivo activated cDC2s (termed 'inflammatory' cDC2s' [infcDC2s]) (36). 195 However, we did observe a significant positive enrichment score (NES=1.30, Padj = 0.02) for 196 genes down-regulated in infcDC2s in the transcriptome of IU patients ( Fig. 3A and 3B). 197

198
Because murine studies underscored a Notch-dependent divergence of cDC2 subsets 199 (24,37), we reasoned that the CD1c+ DC transcriptional signature driven by RUNX3 of 200 patients would be dependent on NOTCH2 signaling. To explore this, we investigated the 201 transcriptome of dendritic cells of notch2∆-CD11c mice (38). In agreement with the 202 expression data of the CD11c-DC-Runx3Δ mice, loss of Notch2 resulted in up regulation of 203 ccr2, cd36, cx3cr1, and decreased expression of runx3 (Fig. 3A), and we detected 204 enrichment for genes upregulated in notch2-KO cDCs2 (Fig. 3B). This supports that 205 NOTCH2 is upstream of RUNX3 and mediates the transcriptomic characteristics of blood 206 CD1c+ DCs of autoimmune uveitis. These findings were further strengthened in reanalysis of 207 previously published transcriptomic data of murine bone marrow progenitors cultured for 7 208 days with OP9 stromal cells that express the NOTCH2 ligand DLL1 or OP-9 cells without 209 DLL1 (39). This analysis revealed that notch2-controlled genes were enriched in the 210 transcriptome of CD1c+ DCs of patients and that notch2-signalling governs the expression of 211 cd36, ccr2, and cx3cr1 in cDC2s ( Fig. 3D and Fig. 3E). To investigate if NOTCH2 was also 212 important at more mature stages of cDC2 development (i.e., in dendritic cells that circulate in 213 blood), we also generated CD1c+ DCs from human CD34+ hematopoietic progenitors in 214 cultures with OP9-DLL1 (NOTCH2-stimulation), in which we inhibited NOTCH2 after CD1c+ 215 DCs normally occur in culture at day 7. We used an anti-NOTCH2 antibody or an inhibitor for 216 ADAM10, a key regulator of the NOTCH2 pathway involved in cDC2 biology (40) and 217 measured the CD1c+ DC numbers at day 14. This analysis showed that late inhibition of 218 NOTCH2 signaling substantially impaired the cell numbers of CD1c+ DCs in culture and 219 supports that NOTCH2 signaling is actively involved in the maintenance of human CD1c+ 220 DCs (Supplemental Fig. S1). Collectively, these observations support that compromised 221 NOTCH2-RUNX3 signaling promotes a transcriptomic signature that closely resembles the 222 gene profile of CD1c+ DCs in human autoimmune uveitis. 223 224 A CD36+CX3CR1+CD1c+ DC3 subset is decreased in autoimmune uveitis 225 We reasoned that the transcriptomic signature of the CD1c+ DC pool in patients may be an 226 impression of changes in the proportions of CD1c+ DC subsets in blood. To allow 227 identification of undefined phenotypes, we first used unbiased flow cytometry data analysis to 228 identify CD1c+ DC clusters in peripheral blood mononuclear cells (PBMCs) samples from 26 229 cases and 11 controls. We designed a panel based on genes from the uveitis gene signature 230 (CD36, CX3CR1, CCR2, and CD180), surface markers previously linked to CD1c+ DC 231 subsets, but that were not DEGs (CD5, and CD163) (17), and classical CD1c+ DC markers 232 (CD1c and CD11c), and a lineage marker (CD3/CD19/CD56). FlowSOM (41) was used on 233 singlets (PBMCs) to cluster cells into a predetermined number of 100 clusters (flowSOM 234 default of 10 x 10 grid) to facilitate unbiased detection of possible CD1c+ DC phenotypes in 235 blood. The analysis with flowSOM clearly distinguished four (Lin-HLA-DR+CD11c+) CD1c+ 236 DC clusters (cluster number 41, 61, 81, and 83) ( Fig. 4A and Supplemental Fig. 2). We 237 extracted the data for the four CD1c+ DC clusters (i.e., clusters 41, 61, 81, and 83) and 238 conducted principal component analysis (PCA). The first two principal components explained 239 98.5% of the variance and the PCA biplot identified CX3CR1 and CD36 as top loadings, 240 indicating that these proteins distinguish CD1c+ DC phenotypes in blood (Fig. 4B). We also 241 repeated this analysis using Lineage-(CD3/CD56/CD19) HLA-DR+ cells as input for 242 flowSOM clustering, which corroborated CD36 and CX3CR1 as key markers that distinguish 243 CD1c+ DC phenotypes (Supplemental Fig. 3). 244 Next, we designed a simple gating strategy for CD1c+ DC subsets based on the expression 245 of CD36 and CX3CR1 (Fig. 4C). Manual gating distinguished four CD1c+ DCs subsets; a 246 CD36-CX3CR1-subset, a CD36+CX3CR1-subset and a rare CD36-CX3CR1+ subset, and 247 a CD36+CX3CR1+ subset of CD1c+ DCs in peripheral blood. The CD36+CX3CR1-and 248 CD36+CX3CR1+ subsets express both CD163 (Fig. 4C) and CD14 ( Fig. 4D and 249 Supplementary Fig. 4). Comparison between patients and controls revealed that the 250 frequency of CD36+CX3CR1+ CD1c+ DCs were decreased in the blood of autoimmune 251 uveitis patients ( Fig. 4E and 4F). 252 We speculated that the decrease in blood CD36+CX3CR1+ CD1c+ DCs was a result from 253 migration of these cells to the eye during autoimmune uveitis. Recent single-cell RNA 254 sequencing data by Kasper and co-workers in eye biopsies of autoimmune uveitis patients 255 show infiltration in the eye by conventional dendritic cells, which are transcriptionally similar 256 to CD1c+ DCs (42). Reanalysis of this data using CLEC10A as a specific marker for CD1c+ 257 DCs in tissues (43) showed that eye-infiltrating CD1c+ DC cells express CD36 and CX3CR1 258 (Fig. 4G) human CD1c+ DC subsets based on the surface expression of CD36 and CX3CR1 (Fig.  265  5A), of which double-positive and double-negative subsets could be sorted from the selected 266 healthy subjects in sufficient numbers for experimental analysis. Since CD36 is a co-receptor 267 for TLR2 (44), we overnight stimulated the subsets with the TLR2 ligand lipoteichoic acid 268 (LTA). Interleukin (IL)-23, a cytokine potently produced by CD1c+ DCs in general, was 269 equally strong secreted by both subsets of CD1c+ DCs (Fig. 5B). To assess the secretome 270 of the CD1c+ DC subsets in more detail, we profiled the supernatants of LTA-stimulated 271 CD1c+ DC subsets for additional soluble immune mediators: The CD1c+ DC subsets could 272 be distinguished based on the secreted protein profile (Fig. 5C), of which the levels of TNF-273 alpha, IL-6, VEGF-A, and TNFR1 showed significant differences between the subsets (Fig.  274   5D). These results show that CD1c+ DC subsets defined on the basis of surface expression 275 of CD36 and CX3CR1 show a differential capacity to secrete pro-inflammatory mediators that 276 participate in the pathophysiology of human autoimmune uveitis. 277 278

Discussion 279
In this study of 80 autoimmune uveitis patients and controls, we identified and replicated a 280 core disease transcriptional network in CD1c+ DCs. We were able to track back the network 281 to a cytokine-producing CD36+CX3CR1+ CD1c+ DC subset that was diminished in 282 peripheral blood of patients with autoimmune uveitis. 283 Using data from genetic models, we show that reciprocal expression of the gene network 284 associated with autoimmune uveitis relies on transcription factors NOTCH2 and RUNX3.  Other preceding studies into human CD1c+ DCs revealed functionally distinct subsets 292 termed "DC2" and "DC3", with the DC3 showing both transcriptomic features reminiscent of 293 cDC2s and monocytes -such as elevated CD36 levels (16). DC3s also have distinct 294 developmental pathways and transcriptional regulators compared to DC2 (16)(17)(18)20). 295 Recently, Cytlak and associates revealed that lower expression of IRF8 is linked to DC3 (20), 296 a transcription factor that was also decreased in autoimmune uveitis. Dutertre and co-297 workers (17) showed that the phenotype of peripheral blood CD1c+ DCs can be further 298 segregated according to the expression of CD163 and CD5, with "DC3" cells being 299 characterized as CD5-CD163+ cells and "DC2" as CD5+CD163 cells. We show that CD1c+ 300 DCs can also be segregated based on surface expression of CD36 and CX3CR1 and that 301 DC3s (CD1c+CD5-CD163+) are found both in the CD36+CX3CR1-and CD36+CX3CR1+ 302 subsets. In other words, CD36 and CX3CR1 surface expression defines phenotypically 303 discrete DC3 subsets in peripheral blood. The CD36-CX3CR1-subset was CD5+CD163-304 and similar to DC2 cells (12). Patients with Systemic lupus erythematosus (SLE) display 305 accumulation of CD5-CD163+ DC3s in blood (13), while this population of DC3 cells (CD5-306 CD163+CD1c+ DCs) was decreased in autoimmune uveitis patients (Fig. 4). The differences 307 between autoimmune uveitis and SLE may be related to distinct (i.e., opposite) 308 immunopathological mechanisms; Type I interferons drive the maturation of cDC2s into 309 "inflammatory cDC2s" (infcDC2s) (36) and can induce CD1c+ DCs to express a distinct set of 310 surface-receptors (19). The type I interferon (IFN)-α drives immunopathology of SLE (21) and 311 administration of type I interferon therapy can induce lupus-like disease (22,23). In favor of 312 attributing the seemingly contrasting observations in blood CD1c+ subsets between SLE and 313 autoimmune uveitis to distinct biology is the fact that, in contrast to elevated IFN-α in patients 314 with SLE, in autoimmune uveitis patient's disease exacerbations correlate with reduced 315 blood IFN-α concentrations (21). In addition, we demonstrated that the transcriptional 316 signature of CD1c+ DCs in autoimmune uveitis was not positively enriched for transcriptomic 317 features of IFN-driven cDC2 subset (Fig. 3B). 318

319
This also indicates that unlike SLE (45), changes of the CD1c+ DC peripheral blood pool in 320 autoimmune uveitis are not driven by increased type I interferon signaling (21), but rather by 321 yet unspecified pathological molecular pathways. This is supported by the fact that in our 322 stimulation experiments, IFN alpha did not induce the uveitis gene signature in CD1c+ DCs 323 ( Fig. 3) and that type I interferon therapy inhibits autoimmune uveitis (21). However, an 324 argument against this is the observation that in IU patients we detected enrichment for genes 325 down-regulated in infcDC2s (20,36). Indeed, although not all statistically significant, genes 326 such as ccr2, cd36, cx3r1 show a relative decrease in expression, while runx3 shows a 327 relative increase in expression in the transcriptome of infcDC2s versus "non-inflammatory" 328 cDC2s (based on GSE149619). In other words, this suggest that the profile of infcDC2s 329 within the CD1c+DC pool was 'decreased' in IU patients, which is in line with the decrease in 330 the frequency of CD5-CD163 DC3s as a whole or the TNF-alpha and IL-6-secreting 331 CD36+CX3CR1+ DC3 subset in patients with autoimmune uveitis. One possible explanation 332 for the decrease in this subset may be that in the systemic condition SLE 'inflammatory' 333 DC3s 'accumulate' in blood, while in autoimmune uveitis this population exits the circulation 334 to infiltrate secondary lymphoid tissues and ocular structures to mediate eye inflammation. 335 Indeed, emerging single cell analysis of ocular fluids of patients support the infiltration of 336 CD1c+ DC subsets in the eye during autoimmune uveitis (42), and we show that these cells 337 express CD36 and CX3CR1 (Fig. 4G). However, ideally longitudinal data are used to follow 338 the dynamics of these populations in patients in relation to disease activity, which is a 339 limitation of the current study. In conclusion, we discovered a CD36+CX3CR1+CD1c+ DC subset that decreased in the 368 peripheral blood of patients with autoimmune uveitis. The fact that this population secretes 369 high levels of TNF-alpha, is decreased in the circulation of patients, while CD1c+ DCs 370 expressing CD36 and CX3CR1 accumulate locally in autoimmune uveitis patients may not 371 only explains the therapeutic benefit of TNF inhibition for autoimmune uveitis, it also opens 372 new avenues for therapeutic targeting to prevent blindness due to autoimmune uveitis. 373

Patients and patient material 376
This study was conducted in compliance with the Helsinki principles. Ethical approval was 377 requested and obtained from the Medical Ethical Research Committee in Utrecht. All patients 378 signed written informed consent before participation. We collected blood from a discovery 379 cohort of 29 and a replication cohort of 22 adult patients ( Table 1)

Power analysis 507
We conducted power analysis of the discovery cohort using the PROPER R package 508 (version 1.22.0)(56) with 100 simulations of the build-in RNA-seq count data from antigen 509 presenting (B) cells from a cohort of 41 individuals (i.e., large biological variation as expected 510 in our study) (57). Simulations parameters used the default of 20,000 genes and an 511 estimated 10% of genes being differentially expressed. We detected 0.8 power to detect 512 differentially expressed genes (P<0.05) at a log 2 (fold change)>1 for the smallest patient 513 group (9 cases) and we considered the sample size reasonable for analysis. 514 515

Differential gene expression and statistical analysis 516
Quality check of the raw sequences was performed using the FastQC tool. Reads were 517 aligned to the human genome (GRCh38 build 79) using STAR aligner (58) and the Python 518 package HTSeq was used to count the number of reads overlapping each annotated gene 519 (59). We aligned the reads of the RNA sequencing data sets to 65,217 annotated Ensemble 520 Gene IDs. Raw count data were fed into the DESeq2 (60) to identify differentially expressed 521 genes (DEGs) between the four disease groups (AU, IU, BU, HC). Using DESeq2, we 522 modelled the biological variability and overdispersion in expression data following a negative 523 binomial distribution. We then used Wald's test to identify DEGs in each pair-wise 524 comparison and used likelihood ratio test to identify DEGs considering multiple disease 525 groups. We constructed co-expression gene networks (β = 6) with the WGCNA R package 526 (61) using the cumulative differentially expressed genes (P<0.05) from all pairwise and group 527 comparisons. We calculated the intersect between the modules constructed from the two 528 cohorts and used Fisher's exact test to identify modules that exhibited significant overlap in 529 genes. Pathway enrichment analysis was done with ToppGene Suite (BMI CCHMC, 530 Cincinnati, OH, USA)(62). 531 Module specific regulator-target networks were generated using the RegEnrich R package 532