Cytotoxic Activity of CD4 T Cells During the Early Stage of Autoimmune Neuroinflammation

Pathogenic CD4+ T cells are capable of initiating neuroinflammation in experimental autoimmune encephalomyelitis (EAE). However, the precise effector mechanism of these autoaggressive CD4+ T cells is not entirely elucidated. Here, we demonstrated that pathogenic CD4+ T cells, upon autoantigen stimulation, developed a cytotoxic phenotype at the onset of EAE. The cytotoxic activity of pathogenic CD4+ T cells was sufficient to explain the initial myelin lesion. Consistently, CD4+ T cells of peripheral blood (PBMCs) and cerebrospinal fluid (CSF) from relapse-remitting multiple sclerosis (RRMS) patients present an enhancement of the cytotoxic profile in comparison with healthy control (HC). Moreover, cytotoxic CD4+ T cells (CD4-CTLs) are restrained in the PBMCs of Natalizumab-treated RRMS patients. Mechanistically, autoaggressive CD4-CTLs matched the majority of the molecular pathways of effector CD8+ T cells. Altogether, our findings point to potential new targets for monitoring MS diagnosis, treatment, and the development of novel therapeutic avenues.


Introduction 14
Experimental autoimmune encephalomyelitis (EAE) is a widely accepted animal model of 15 multiple sclerosis (MS). EAE shares many pathophysiological features with MS, such as chronic 16 neuroinflammation, demyelination, neuronal damage, and is generated by the autoimmune 17 attack on the central nervous system (CNS) (Baxter, 2007;Steinman and Zamvil, 2006). The 18 disease can be induced in susceptible animals by active immunization with self-epitopes of 19 myelin or by the passive transfer of pre-activated lymphocytes (Paterson, 1960;Rivers et al., 20 1933). More specifically, EAE can be induced by the adoptive transfer of CD4+ T cells specific In 2007, Kebir and colleagues showed that the cytotoxic activity of encephalitogenic CD4+ T 38 cells in the EAE model is essential for breaching the blood-brain barrier (Kebir et al., 2007). 39 Furthermore, treating EAE with Serpina3n, a granzyme B (GzmB)-inhibitor, resulted in a 40 significant reduction of EAE severity, although infiltration of T cells into the CNS was not 41 reduced (Haile et al., 2015). Recent  In this context, we investigated the role of CD4-CTLs during the early phase of autoimmune 45 neuroinflammation. We reported that during the early stage of EAE encephalitogenic CD4+ T 46 cells build a cytotoxic profile, which is enhanced after these cells reach the CNS. These  Results 53

Enhancement of the cytotoxic activity of encephalitogenic CD4+ T during EAE 54
In order to evaluate the cytotoxic profile of encephalitogenic CD4+ T cells, we sorted 55 CD3+CD4+ cells from lymph nodes (LN) and central nervous system (CNS) of mice immunized 56 with neuroantigen (MOG35-55), or with non-self-antigen [ovalbumin (OVA)]. Then, we analyzed 57 the expression of classical proinflammatory and cytotoxic-related molecules by qPCR. In 58 10 of Runx3 represses ThPOK and terminates CD4 expression as well (Cheroutre and Husain,142 2013; Setoguchi et al., 2008). In contrast, our data showed a slight increase in expression of 143 ThPOK after CD4+ T cell entry in the CNS (Fig. 2G) as well as no regulation of CD4 molecule 144 expression (Fig. 2L). Besides, we observed that almost all of CD4+Runx3+ T cells were also 145 ThPOK positive ( Fig. 2H and Fig. 2I) Of note, in the context of human cytomegalovirus 146 infection, Th1 CD4-CTLs have been found to enhance the expression of Runx3 in the absence 147 of ThPOK downregulation as well (Serroukh et al., 2018). 148

CD4+ T cells landscape into the CNS 150
In order to identify the CD4-CTLs infiltrated in the CNS, we used t-distributed stochastic 151 neighbor embedding (t-SNE) to evaluate the co-expression level of CD4, Runx3, T-bet, GzmB, 152 VLA-4 (CD49d, α4β1, α4β7), and KLRG1 in CD3+CD4+ T cells population from CNS during the 153 onset of EAE (Fig. 2J). Then, we applied an artificial neural network-based algorithm 154 (FlowSOM) to generate specific populations based on the expression of those markers (Fig.  155   2K). FlowSOM-based nodes were then manually annotated in the CD3+CD4+ t-SNE landscape 156 (Fig. 2K). The CD4-CTLs population was defined by the high expression of Runx3, GzmB, and 157 T-bet. These data were confirmed by the expression intensity of those markers at the single-158 cell level of each annotated population (Fig. 2L). Interestingly, the VLA4 hi population also 159 present a high expression of GzmB, although it was significantly lower in comparison with 160 CD4-CTLs ( Fig. 2K and Fig. 2L). Conventional Th1 (cTh1) population was defined by the high 161 expression of T-bet and low expression of Runx3, GzmB, VLA-4, and KLRG1 ( Fig. 2K and Fig.  162 2L). We identified one population that expresses high levels of KLRG1 and low levels of Runx3, 163 T-bet, GzmB, and VLA-4 ( Fig. 2K and Fig. 2L). KLRG1 is highly expressed in short-lived effector 164 CD8+ T cells (Hamilton and Jameson, 2007). Moreover, the expression of KLRG-1 is a direct 165 consequence of T-bet expression (Joshi et al., 2007).  In EAE, the cytotoxic could be detected in the periphery during the early phase of the disease 190 ( Fig. S1). Therefore, in order to test whether the data generated in the animal model would 191 apply to human disease, we recruited untreated RRMS and matched healthy controls (HC) to 192 evaluate the extension of our data in PBMCs (Table 1). Here, we included RRMS patients 193 assessed at the very early stage of the diagnosis (diagnosis) as well as patients that withdrawal 194 the treatment by therapeutic failure or after reach 24 months of Natalizumab (NTZ)-treatment 195 (washout) ( Table 1). Then, we analyzed the expression of GzmB by CD3+CD4+ or CD3+CD8+ but not CD8+GzmB+ T cells in RRMS patients in relation to HCs (Fig. 3A -3D). Strikingly,there 199 is no correlation between the percentages of CD4+GzmB+ and CD8+GzmB+ T cells, which 200 demonstrates a specific enhancement of the cytotoxic profile by CD4+ T cells (Fig. 3E). not present increased levels of CD4+GzmB+ T cells in relation to HC (Fig. S2B). No significant 207 difference was found for the CD8+ T cells population ( Fig. S2C and S2D). Consistently with our 208 data obtained in the experimental model, the expression of GzmB was almost restricted to 209 Runx3-expressing CD4+ T cells (Fig. 3F). 210 Consistently, qPCR results demonstrated increased expression of Runx3 and GzmB mRNA in 211 CD4+ T cells sorted from RRMS patients in comparison with HC (Fig. 3G). Although no 212 14 significant difference was found in expression levels of T-bet or RORγt, there was a significant 213 positive correlation (R 2 =0.6054) between mRNA expression of Runx3 and T-bet in RRMS 214 patients (Fig. S2E). In contrast, the mRNA expression of Runx3 and RORγt trended (R 2 =0.3689; 215 p=0.1103) to be negatively correlated (Fig. S3E). Taken together, these results reinforce the 216 notion that CD4-CTL exhibits a Th1-like profile promoted by the mutual expression of Runx3 217

Cytotoxic Profile of Myelin-Reactive CD4+ T cells 247
Our data in the experimental model have indicated that the cytotoxic behavior of CD4+ T cells 248 is enhanced upon autoantigen recognition (Fig. 1A, 1B, and Fig. S1A). Therefore, we accessed 249 a publicly available sequencing (RNA-seq) expression dataset of myelin-reactive T cells from and MS tetramer-negative (MS TET-) samples (Fig. 4D). Nevertheless, based on these 264 molecules, hierarchical clustering analysis segregate myelin-reactive MS TET+ cells from MS 265 TET- (Fig. 4D). In order to have a better idea of the functional identity and connectivity of 266 those molecules, we used the STRING network and clustering analysis (https://string-267 db.org/cgi/network.pl?taskId=rgnZ13we8jW). The molecules were grouped by molecular 268 category (Fig. 4E). The analysis demonstrated a highly connected network, which is enriched 269 by Th1 and Th17-related cytokines as well as cytotoxic-related effector molecules and 270 transcription factors. 271 Altogether, our data consistently demonstrated a robust cytotoxic behavior of pathogenic 272 CD4+ T cells upon autoantigen stimulus, which is restricted to patients with MS. Markedly, 273 our analysis showed differential expression of IL2RA, PRDM1 (blimp1) and IRF4 (Fig. 4E)

Single-Cell mRNA Sequence of CSF Cells from MS Patients 289
Our data in the experimental model demonstrated that the cytotoxic profile of CD4+ T cells is 290 enhanced after these cells reach the CNS. Therefore, we again took advantage of publicly 291 available data to evaluate the extension of our data in CD4+ T cells from the CSF of MS 292 patients. For that, we accessed the single-cell mRNA sequence (scRNA-seq) of CSF cells from 293 MS-discordant monozygotic twin pairs (Beltrán et al., 2019a). First, we reconstructed the t-294 SNE leukocytes landscape (Fig. 5A). Interestingly, as in the original analysis, we were not able 295 to segregate CD8+ and CD4+ T cells populations inside the T cell cluster. Instead, we also 296 identify two regions, one housing preferentially CD4+ T cells and another housing CD8+ T cells 297 (Fig. 5B). Therefore, we used the expression of CD3D and CD4 to manually select the CD4+ T 298 cell population in order to evaluate the expression of cytotoxic-related molecules (Fig. 5C). 299 Then, we used Louvain clustering to determine the subpopulations of CD4+ T cells (Fig. S3A). 300 CD4-CTL population (Fig. 5D). The co-expression of those molecules indicates the cluster 3 303 (C3) is the CD4-CTL population (Fig. 5E). The expression of IL2RB, CRTAM, IFNG-AS1, EOMES, 304 and GzmH were significantly higher in C3 in comparison with the remaining CD4+ T cells (Fig.  305   S3B). C3 cluster presents a massive enrichment of CD4+ from patients with MS in comparison 306 with HCs ( Fig. 5F and S3C).  (Fig. S2E). Moreover, 332 ChiP-seq analysis from ENCODE data bank demonstrated direct binding of Runx3 to ITGA4 333 locus at the promoter region (Fig. 6A). In PBMCs from MS patients, the expression of GzmB is 334 almost restricted to CD4+VLA-4+ cells (Fig. 6B). VLA-4 is the target of NTZ, which is a highly 335 effective monoclonal antibody for the treatment of RRMS (Steinman et al., 2012). In this 336 context, we hypothesized that NTZ-treatment might target specifically those CD4-CTLs. 337 Therefore, to evaluate the impact of the treatments, specially NTZ, over CD4-CTL, we enrolled 338 48 treated RRMS patients (Table 1). Then, we evaluated the presence of CD4+GzmB+ and 339

treated RRMS patients present enrichment of CD4+GzmB+ T cells in PBMCs in comparison 342
with the other treatments or untreated patients (Fig. 6C). FTY-treated patients shown a slight 343 increase in the percentage of CD4+GzmB+ T cells in comparison with IFN/GA-treated and 344 untreated patients, which had been previously described (Fujii et al., 2016) (Fig. 6C). NTZ-345 treated patients also shown enrichment of CD8+GzmB+ T in comparison with the other 346 treatments or untreated patients (Fig. 6D).

Discussion 361
Here, we demonstrated that autoaggressive encephalitogenic CD4+ T cells present a cytotoxic 362 phenotype during the early phase of EAE and RRMS. Robust and concordant data support the 363 critical role of CD4-CTLs during initial autoimmune neuroinflammation.

CD4+ T cells presents a population that expresses high levels of VLA-4 and intermediary levels 379
of GzmB (VLA-4 hi ). VLA-4 is an adhesion receptor essential to the transmigration of circulating 380 leukocytes into the CNS in EAE (Yednock et al., 1992). Thus, the VLA-4 hi population is probably 381 the CD4+ T cells that just infiltrated the CNS. In this case, the differential expression of GzmB 382 by this population is in agreement with Kebir's data (Kebir et al., 2007). 383  was normalized to that of a housekeeping gene (GAPDH or HRP). The data were generated 477 mice or unimmunized mice) were sorted with a flow cytometer. Sorted cells were incubated 502 with slices (5x10 5 cells/slice) for 6 hours in Hanks' Balanced Salt solution (Sigma-Aldrich, USA) 503 supplemented with 1% of penicillin/streptomycin solution (Thermo-Fisher Scientific, USA). 504

The conversion of Th17 cells into IFNγ-producing cells is dependent on T-bet and Runx
After the incubation period, brain slices were placed in Tissue-Tek OCT compound (Sakura 505 Finetek, USA) and frozen. 506 507

Confocal Microscopy 508
In situ expression of cleaved caspase-3 in the brain from EAE and from control animals 509 (ovalbumin immunized mice or unimmunized mice) was performed following the protocol for 510 immunofluorescence staining. Briefly, frozen brains in Tissue-Tek OCT compound (Sakura 511 Finetek, USA) were sliced to produce 10 μm sections using a cryostat and submitted to 512 staining. For this, the tissue was blocked to avoid nonspecific reactions with donkey serum 513 (Sigma-Aldrich, USA) and permeabilized with Triton X-100 (Sigma-Aldrich, USA). Then, a 514 primary antibody against cleaved caspase-3 (Cell Signaling Technology, USA) was added 515 (1:200) and incubated overnight. The following day, slices were washed and incubated with 516 secondary antibodies (Alexa 488) (Cell Signaling Technology, USA) (dilution 1:500) plus DAPI 517 (Sigma-Aldrich, USA). For confocal controls, primary antibodies were omitted from the 518 staining procedure and were negative for any reactivity. All images were obtained using 519 confocal microscopy at INFABIC-UNICAMP. 520 521

RNA-seq data analysis 522
We accessed publicly RNA-Seq expression data deposited in the NCBI's Gene Expression  Correlations were determined by Pearson or Spearman rank tests as appropriate. Bootstrap 541 resampling (100,000x) and permutation (BsRP) were used in the qPCR analysis of human 542 samples to minimize statistical interference due to the small sample size. The results of BsRP 543 have expressed p-value (nH0+1/99999+1). Hierarchical cluster analysis was performed using 544 the Euclidean distance metric. A P-value of less than 0.05 was considered significant in all tests. 545 t-distributed stochastic neighbor embedding (t-SNE) (iteration = 3,000; perplexity = 30; 546 learning rate = 252; gradient algorithm = Barnes-Hut) was performed using Flowjo R studio-547 based plugin. 548 549 Software 550 Basic statistical analysis was performed using GraphPad Prism software. Bootstrap resampling 551 and permutation were performed using Rstudio software. Heat map and hierarchical cluster 552 analysis were performed using the MORPHEUS web-based tool 553 (https://software.broadinstitute.org/morpheus/). All images obtained were analyzed using 554 the Image J. FACS data was analyzed using the FlowJo 10.6 software package (Tri-Star, USA). 555 Single-cell data analysis was done using scOrange 3.24 software.