Galectin-9 regulates the threshold of B cell activation and autoimmunity

Despite the mechanisms of central and peripheral tolerance, the mature B cell compartment contains cells reactive for self-antigen. How these cells are poised not to respond and the mechanisms that restrain B cell responses to low-affinity endogenous antigens are not fully understood. Here, we demonstrate a critical role for the glycan-binding protein galectin-9 in setting the threshold of B cell activation and that loss of this regulatory network is sufficient to drive spontaneous autoimmunity. We further demonstrate a critical role for galectin-9 in restraining not only conventional B-2 B cells, but also innate-like B-1a cells. We show that galectin-9-deficient mice have an expanded population of B-1a cells and increased titers of B-1a-derived autoantibodies. Mechanistically, we demonstrate that galectin-9 regulates BCR and distinct TLR responses in B-1a cells, but not B-1b cells, by regulating the interaction between BCR and TLRs with the regulatory molecules CD5 and CD180, respectively. In the absence of galectin-9, B-1a cells are more readily activated and secrete increased titers of autoantibodies that facilitate autoantigen delivery to the spleen, driving autoimmune responses.

12 Taken together, these data demonstrate that Gal9 regulates B-1a cell accumulation and activation at 268 steady-state and implicate B-1a cells in the autoimmune phenotype observed in Gal9-deficient mice. 269 270

Gal9 regulates BCR signaling in B-1a cells 271
We next asked if the expansion of B-1a cells and their increased expression of activation markers in the 272 steady-state in Gal9KO mice is due to altered BCR signaling. To address this, we stimulated PerC B-1 273 cells with titrated concentrations of anti-IgM F(ab')2 for 16 hours, and measured B cell activation by 274 upregulation of CD86. We find that loss of Gal9 leads to enhanced activation of B-1a cells in response to 275 lower concentrations of agonist ( Fig. 5A-C). Notably, this increased activation upon loss of Gal9 is 276 specific to B-1a cells, as we did not observe any difference in B-1b cell responses in the same mouse (Fig. 277 S5A-C). We then asked if Gal9 interactions directly modulate BCR signal transduction in B-1a cells. We 278 stimulated PerC B-1 cells with anti-IgM F(ab')2 for five minutes, followed by fixation and staining for 279 phospho-signalling molecules by flow cytometry. We see that Gal9KO B-1a cells have enhanced 280 phosphorylation of signaling machinery following IgM stimulation (Fig. 5D). To assess if enhanced 281 activation is specifically due to Gal9 at the cell surface, we treated WT PerC cells with the pan-galectin 282 inhibitor lactose to remove all galectins from the cell surface and, alternatively, treated Gal9KO PerC 283 cells with recombinant Gal9 (rGal9) to reconstitute Gal9 expression. Following IgM stimulation, pan-284 galectin inhibition in WT B cells results in increased signal transduction and importantly, reconstitution 285 of Gal9 in Gal9KO cells directly inhibits signalling in B-1a cells (Fig. 5D). In contrast, but consistent with 286 our earlier observations, these treatments have little effect on B-1b cell signalling (Fig. S5D). We 287 wondered if differences in Gal9 expression may explain some of the subset specific discrepancies in B 288 cell activation. Indeed, B-1a cells have higher expression of Gal9 at steady-state compared to B-1b cells 289 (Fig. 5E). Taken together, our results demonstrate a specific role for Gal9 in the regulation of B1a, but 290 not B1b cells. 291 13 292 We have shown that Gal9 regulates B-2 B cell activation through mediating interaction between IgM-293 BCR and negative co-receptors CD22 and CD45 8 ; however, B-1a cells are known to express a unique 294 surface profile lacking these negative regulators. Therefore, Gal9 regulation of B-1a cells must occur 295 through a unique mechanism. To address this, we performed pull-down assays to identify Gal9 ligands in 296 B-1a B cells by incubating protein lysates from sorted B-1a and B-1b PerC B cells or conventional splenic 297 B-2 cells with rGal9-coated beads. We then washed and stained these beads with antibodies targeting 298 various co-receptors and assessed protein enrichment by flow cytometry. We find that Gal9 interacts 299 with IgM-BCR in B1a, B1b, and B-2 B cells, but uniquely interacts with the inhibitory glyco-protein CD5 300 on the surface of B-1a cells (Fig. 5F and Fig. S5 E). Taken together, these data demonstrate that Gal9 301 binds to B-1a cells and regulates their activation, likely through the inhibitory co-receptor CD5. 302

303
The nanoscale organization of molecules on the cell surface allow for interactions between receptors 304 and important regulatory co-receptors that enhance or restrain signal transduction 5 . We have shown 305 previously that Gal9 modulates BCR responses in B2 B cells by altering the nano-scale distribution of BCR 306 and its association with inhibitory co-receptors 8,26 . To assess the impact of Gal9 on the nanoscale 307 organization of IgM-BCR and inhibitory co-receptor CD5, we performed dual-color direct stochastic 308 optical reconstruction microscopy (dSTORM) to achieve single molecule resolution of the cell surface. 309 Upon visual inspection of dSTORM images, IgM appears to be more clustered on B-1a cells in the 310 absence of Gal9 (Fig. 5G). Quantification of cluster area demonstrated that IgM clusters are larger in the 311 absence of Gal9 (Fig. 5H). To quantify clustering tendency, we calculated the Hopkins index, which 312 assesses clustering tendency relative to a random distribution (0.5). We find that IgM on the surface of 313 Gal9KO B-1a cells has a higher clustering tendency compared to WT B-1a cells (Fig. 5I). Upon visual 314 inspection of dSTORM images of CD5 we note that it is more highly clustered compared to IgM, however 315 14 we do not observe striking differences between WT and Gal9KO B-1a cells. Quantification of CD5 cluster 316 area and degree of clustering, however, revealed an increase in Gal9KO B-1a cells ( Fig. 5J and K). We 317 hypothesized that Gal9 crosslinks IgM and CD5 to regulate BCR signaling in B1a cells. Dual dSTORM 318 provides the advantage of single-molecule coordinate-based colocalization analysis to examine if there 319 are differences in the colocalization of these molecules on the surface of B-1a cells in the absence of 320 Gal9. Visually, we can discern a loss of colocalized IgM and CD5 clusters in the absence of Gal9 (Fig. 5G). 321 To quantify this, we performed coordinate-based colocalization (CBC) analysis, which ranges from -1 322 (perfectly segregated) to +1 (perfectly colocalized). The frequency distribution of CBC values shifts to the 323 left (i.e. toward negative values) in the absence of Gal9, indicating an increase in the segregation of IgM-324 BCR and CD5, compared to WT B-1a cells (Fig. 5L). This results in a decrease in the mean CBC value per 325 cell (Fig. 5M) and a corresponding increase in the mean nearest neighbour (NN; Fig. 5N). In contrast, but 326 consistent with the lack of effect of Gal9 deficiency on B1b B cell activation, we find no difference in the 327 organization of IgM on the surface of B-1b cells in the absence of Gal9 (Fig. S5F-G). Taken together, 328 these findings identify a unique and specific mechanism of Gal9 regulation of B1a BCR signaling. 329

331
Gal9 regulates TLR4 responses in B-1a cells 332 B-1 cells respond to T cell-independent antigens at barrier sites and rapidly differentiate into effector 333 cells such as ASCs through BCR or TLR stimulation 23 . To assess if Gal9 influences innate-like responses of 334 B-1 cells, and to identify TLRs potentially regulated by Gal9, we stimulated PerC cells from WT or Gal9KO 335 mice with titrated concentrations of various TLR ligands and measured the proportion of cells activated 336 by upregulation of CD86. Interestingly, we find that B-1a cells from Gal9KO mice are more sensitive to 337 LPS and CpG stimulation, but not imiquimod or zymosan stimulation ( Fig. 6A-D). In contrast, but 338 consistent with BCR stimulation, Gal9 appears to be dispensable for TLR activation of B-1b cells . We then asked if enhanced CD86 upregulation in the absence of Gal9 was due to a similar 340 enhancement of TLR signalling in B-1a cells. To address this, we stimulated WT or Gal9KO PerC cells with 341 LPS and assessed total-tyrosine phosphorylation at 10 minutes by phospho-flow cytometry. We find that 342 Gal9KO B-1a cells, but not B-1b cells, have increased total tyrosine phosphorylation at 10 minutes post-343 LPS stimulation, which can be reduced to levels comparable to WT cells by pre-treatment with rGal9 344 (Fig. 6E, S6E). Similarly, pre-treatment of WT B-1a, but not B-1b, cells with lactose, to displace galectins, 345 enhanced total tyrosine phosphorylation upon LPS stimulation (Fig. 6E, S6E). To identify Gal9 ligands in 346 B1a cells that mediate this phenotype, we performed pull down assays with rGal9-coated beads 347 incubated with B-1a or B-1b cell lysates, as previously. We find that Gal9 binds TLR4 and the inhibitory 348 molecule CD180 on B-1 cells, and to a lesser extent on B-2 B cells (Fig. 6 F and G), consistent with 349 previous mass spectrometry analysis of Gal9 ligands in B2 B cells 8 . 350

351
To assess the impact of Gal9 on TLR4 nano-scale organization we again performed dual-color dSTORM to 352 identify single molecule localizations of TLR4 and CD180 on the surface of B-1 cells from WT and Gal9KO 353 mice (Fig. 6H). In the absence of Gal9 the mean cluster area and clustering tendency of TLR4 is 354 decreased on B-1a cells ( Fig. 6I-J), but interestingly, Gal9 has no affect on the organization of TLR4 in B-355 1b cells (Fig. S6F-H). Similarly, the clustering tendency of CD180 is decreased in the absence of Gal9 in B-356 1a cells, however, the area of CD180 clusters is larger (Fig. 6H, K-L); whereas deficiency of Gal9 has no 357 effect on the organization of CD180 in B1b cells (Fig. S6F, I-J). To assess whether Gal9 mediates close 358 association of TLR4 with the inhibitory co-receptor CD180, we performed CBC analysis as previously 359 described. The mean CBC value of TLR4 and CD180 is decreased, and mean nearest-neighbour 360 increased, in B-1a cells from Gal9KO mice compared to WT mice ( Fig. 6M- Gal9KO mice and sorted B-1a cells. We stimulated sorted cells ex vivo with 0.5 µg/mL LPS and 0.5 µg/mL 369 anti-IgM and measured ASC differentiation by CD138 expression three days post stimulation. We find 370 that B-1a cells from Gal9KO mice more readily adopt an ASC phenotype following stimulation, which 371 results in more secreted antibody in the culture supernatant ( Fig. 7A-D). 372 373 As B-1a derived autoantibodies are thought to aid in clearance of cellular debris 22 , we therefore asked if 374 enhanced sensitivity of B-1a cells to LPS could have an impact in vivo. To address this, we injected mice 375 with a low (0.05 µg/mL) or high dose (1 µg/mL) of LPS intraperitoneal (IP) and assessed B cell activation 376 24 hours post injection. Following low dose LPS injection, upregulation of CD86 is enhanced in Gal9-377 deficient B-1a cells, but not B1b cells (Fig. S7A). We then asked if these activated B-1a cells are secreting 378 autoantibodies into circulation. Indeed, we find a greater fold increase in both PC-specific IgM and IgG in 379 Gal9KO mice 24 hours post low dose LPS injection (Fig. S7B). 380

381
We then asked if B-1a cell-derived immune complexes that form following LPS injection carry 382 autoantigens into secondary lymphoid organs. To address this, we IP injected mice with low dose LPS 383 followed by CFSE-labelled apoptotic bodies (ApoBs) by intravenous injection 24 hours later. At 12 hours 384 post-transfer, we looked for ApoB capture by subcapsular sinus (SCS) macrophages in the spleen of WT 385 or Gal9KO mice. We find that ICs formed following LPS injection are capable of capturing CFSE labelled 386 ApoB and recruiting them to SCS macrophages in the spleen (Fig. S7C-E). To determine if autoantigen 387 delivery to SCS macrophages was directly driven by increased antibody secretion by B-1a cells, we 388 isolated secreted antibodies from sorted B-1a and B-1b cultures stimulated ex vivo and injected these 389 antibodies into mice every three days over a two-week period (Fig. 7E). Following two weeks of antibody 390 transfer, we detect ICs in circulation as well as in the spleens of mice receiving B-1a derived antibodies 391 in both WT and Gal9KO recipient mice ( Fig. 7F-K). allowing for B cell responses against low affinity antigens. We therefore asked if loss of Gal9 would lead 408 to more rapid disease onset in the NZB/W F1 model system for lupus-like disease in mice. Indeed, we 409 find that Gal9 deficient mice produce anti-dsDNA IgG antibodies earlier in life compared to WT 410 18 littermate controls (Fig. 8 A). Furthermore, mice lacking Gal9 expression develop accelerated proteinuria 411 compared to WT controls (Fig. 8 B). Consistent with these observations, Gal9 deficient mice have a 412 larger population of splenic GC B cells early in disease progression compared to WT littermates ( Fig. 8 C  413 and D). We see a similar increase in the proportion of splenic antibody secreting cells in Gal9 deficient 414 mice early in disease onset (Fig. 8 E and F). These ASCs skew towards long-lived plasma cells, and 415 consistent with this, we see enhanced immune complexes in circulation in mice lacking Gal9 (Fig. 8 G  416 and H). We then asked if loss of Gal9 would lead to an expanded pool of B-1a cells in the peritoneal 417 cavity, as we observed in the C57bl/6 background. Indeed, NZB/W F1 mice deficient in Gal9 have a 418 larger population of peritoneal B-1a cells (Fig. 8 I). If these expanded B-1a cells are contributing to 419 disease progression we would expect to see signs of their activation such as migration to secondary 420 lymphoid tissues. Indeed, there is an increased proportion of B-1 cells in the spleen in Gal9 deficient 421 mice ( Fig. 8 J), and this population is enriched for B-1a cells (Fig. 8 K). Furthermore, loss of Gal9 leads to 422 enhanced production of B-1a-derived PC-specific IgM antibodies detectable in the serum (Fig. 8 L). 423 Together, these data demonstrate that Gal9 regulates the onset and progression of lupus-like disease in 424 mice. Loss of Gal9 leads to accelerated B cell driven autoimmunity, marked by GC B cell and ASC 425 formation as well as B-1a cell expansion, migration, and antibody secretion. 426

DISCUSSION 428
Our data indicate that Gal9 regulates IgM signal transduction and thereby sets a threshold of agonist 429 required for productive activation. Our previous work demonstrated that Gal9 binds to co-receptors of 430 BCR signalling and alters the nano-scale organization of the plasma membrane 8 . Herein, we show that 431 loss of this regulation leads to spontaneous autoimmunity in mice, consistent with an expanded 432 population of B-1a cells. Interestingly, we identify Gal9 as a regulator of B-1a cell activation through 433 interactions with IgM-BCR, CD5, TLR4, and CD180. We find that loss of Gal9 leads to decreased 434 colocalization between receptors IgM/TLR4 and their cognate regulators, CD5 and CD180, on the surface 435 of B-1a cells. This coincides with enhanced signal transduction and a reduced threshold for activation. 436 We demonstrate that B-1a derived antibodies promote development of autoimmunity in mice, likely 437 due to their capacity to bind and transport autoantigens to secondary lymphoid tissues. 438

439
Our data indicate that Gal9 modulates antigen affinity discrimination in B cells, a critical regulatory step 440 in peripheral tolerance 18 . Unchecked signal transduction leads to autoimmunity in mice 10 . In fact, 441 aberrations in CD19 or BTK signaling, both of which are known to play important roles in defining this 442 threshold of activation, lead to autoantibody production 10 . Additionally, it is well established that 443 negative regulators such as CD45 and CD22 play central roles in restraining B cell activation and are 444 critical for antigen affinity discrimination in vivo 10 . We have previously shown that Gal9 binds directly to 445 IgM-BCR and the negative regulators CD45 and CD22, thereby dampening BCR signal transduction. Here, 446 we demonstrate that by fine-tuning these signal transduction networks, Gal9 coordinates affinity 447 discrimination of B cells and that loss of this regulation leads to the development of autoimmunity in 448 mice. 449 450 Our data indicates Gal9 suppresses B cell responses to low affinity autoantigens; yet the role of Gal9 in 451 autoimmunity may be cell type specific and multifaceted. For example, administration of rGal9 to 452 NZB/W F1 mice, which develop spontaneous autoimmunity with age, prior to disease manifestation led 453 to a reduction in disease severity 27 . In this work, the authors elucidated a role for Gal9 in restraining 454 TLR7 and TLR9 responses in pDCs, important drivers of type I IFNs and the inflammatory profile that 455 supports autoimmunity 27 . Consistent with this, we show that loss of Gal9 leads to accelerated 456 spontaneous autoimmunity in NZB/W F1 mice. Mechanistically, we show that Gal9 regulates TLR4 and 457 TLR9 responses specifically in B-1a cells, but not B-1b cells, although unlike in pDCs, we see no effect of 458 20 Gal9 on TLR7 responses in B-1 cells. This cell type disparity in the requirement or role of Gal9 in 459 regulating receptor signaling and therefore cellular activation may be due to differential expression of 460 Gal9 and protein glycosylation between different cell types and subsets. For example, Gal9 seems to be 461 essential for development of autoimmune disease in the pristane-induced model, where Gal9KO mice 462 have significantly reduced disease severity 28 . This may result, in part, from impaired DC activation, which 463 is essential for disease onset in pristane-induced autoimmunity 29 . In this case, intracellular Gal9 interacts 464 with the cortical actin cytoskeleton to facilitate phagocytosis and DC activation 30 , and therefore loss of 465 Gal9 may be detrimental for DC-dependent models of autoimmunity. The growing compendium of data 466 suggest that Gal9 may have pleotropic effects that result in context-dependent and cell type specific 467 regulation 31 . 468 469 B-1a cells are expanded in several models of autoimmunity, however their role in disease pathology 470 remains controversial 22 . Their autoreactive repertoire and heightened capacity for T cell activation 471 suggest these cells may be drivers of autoimmunity. Here we demonstrate that Gal9 regulates activation 472 of B-1a cells specifically through regulating IgM and TLR4 signal transduction by increasing the 473 colocalization of these receptors with inhibitory co-receptors CD5 and CD180, respectively. 474 Furthermore, we show that Gal9 restrains ASC differentiation of B-1a cells and antibody production. 475 Transfer of apoptotic bodies into mice is sufficient to drive autoantibody production, likely through 476 increasing the density of autoantigens in secondary lymphoid organs 32 . Likely through a similar 477 mechanism, mutations that impair efferocytosis of circulating immune complexes can lead to 478 autoimmunity 33 . Here, we demonstrate that enhanced production of B-1a derived antibodies increases 479 circulating immune complexes in mice, and that these immune complexes transit to secondary lymphoid branching enzyme GCNT2 in GC B cells, but not naïve and memory B cells, with a concomitant decrease 498 in Gal9 binding 9 . How these varied factors may be altered in the context of autoimmunity to modify the 499 glycan profile of BCR, TLR, and key regulatory co-receptors, and thus galectin binding, has yet to be 500 elucidated. 501

502
In summary, the present work demonstrates that Gal9 plays a critical role in tuning the outcome of 503 receptor-ligand interactions. Loss of Gal9 leads to robust activation of B cells to low-affinity and low-504 density antigens. Additionally, Gal9 regulates BCR and unique TLR driven B-1a cell activation through a 505 unique molecular mechanism, important in the regulation of autoantigen transit to secondary lymphoid 506 22 organs. Collectively, these data demonstrate that Gal9 may serve as a therapeutic target to help 507 mitigate the onset of autoimmunity.

Lysozyme Isolation 527
Lysozymes were isolated from egg whites by ion exchange. Egg whites were diluted in PBS and filtered 528 through 30 kDa filter to restrict protein size (Amicon). Fraction <30 kDa was added to 100 mM 529 ammonium acetate buffer (pH 9.0) and incubated with carboxymethylcellulose (CMC, Sigma Aldrich). 530 23 Unbound proteins were washed with ammonium acetate buffer, lysozymes were then eluted with 400 531 mM ammonium carbonate buffer (pH 9.2). Isolated lysozymes were diluted in PBS and concentrated 532 using a centrifugation column (Amicon). Purified lysozymes were then mono-biotinylated (Fleire & 533 Batista, 2009)  allowed to settle onto OP9 cells. Cells were fixed using 2% paraformaldehyde, and permeabilized on ice 569 in 100% methanol at -20°C. Cells were washed 3x in PBS and then stained for total tyrosine 570 phosphorylation (clone 4G10, Sunnybrook), followed by secondary antibodies (Invitrogen) as well as 571 anti-B220-Pacific Blue (BioLegend). 572

Autoantibody detection 584
HEp-2 cells were grown on glass bottom dishes and fixed with 2% paraformaldehyde. Cells were then 585 stained with sera isolated from mice (1:50) at 4°C. Cells were then washed and autoantibody binding 586 was detected using secondary anti-mouse IgM or IgG (Jackson ImmunoResearch). Cells were imaged by 587 spinning disc confocal microscopy (Quorum Technologies) consisting of an inverted fluorescence 588 microscope (DM16000B; Leica) equipped with a 63x /1.4 NA oil-immersion objective and an electron-589 multiplying charge-coupled device (EMCCD) camera (ImagEM; Hamamatsu). Z-section images were 590 collapsed into a Z-projection and mean fluorescence intensity (MFI) was determined by dividing the 591 total intensity of the field of view by the number of nuclei using ImageJ. Molecules were fitting assuming the same intensity. Super resolution images were rendered with a pixel 645 size of 20 nm. 646

Processing and Analysis 647
Reconstructed images were processed using the built-in method in the Thunderstorm plugin. Duplicate 648 localizations of a single molecule in a given frame were removed based on uncertainty radium of 649 28 localization. Localizations with an uncertainty >20 nm were filtered out. Isolated localizations with fewer 650 than 2 neighbors in a 50 nm radius were removed. The images were drift corrected using fiducial 651 markers, and molecules that appear within 20 nm in multiple frames were merged. Fiducial markers 652 were used to align the two channels post processing using the ImageJ plugin TurboReg, as necessary. For 653 each cell a 3 x 3 µm region was selected in the middle of the cell for analysis. Cluster area was calculated 654 using the density-based spatial clustering of applications with noise (DBSCAN) algorithm in the SMoLR 655 package in R. The Hopkins index was calculated using the Spatstat package in R. Colocalization was 656 assessed using the coordinate-based colocalization (CBC) analysis tool in ThunderSTORM. Each 657 localization is assigned an individual colocalization value based on individual distribution functions of 658 that species and weighted by the distance of nearest neighbours in the local environment. For IgM and 659 CD5 CBC analysis a search radius of 60 nm was used, and for TLR4 and CD180 CBC analysis a search 660 radius of 50 nm was used, based on the radius of IgM and TLR4 nanoclusters, respectively. CBC values 661 then reflect the degree of colocalization between 2 channels on a scale between +1 (perfectly 662 colocalized) to -1 (perfectly excluded). 663

Transfer of B-1a derived antibodies 675
Mice were transferred 100 µg of total antibody by intraperitoneal injection every three days over a two-676 week period. At day, 14 mice were euthanized and a single cell suspension was prepared from the 677 spleen using enzymatic digestion (DNAse/Collagenase). Cells were stained with fluorescently labelled 678 antibodies and measured by flow cytometry. Additionally, mice were assessed 4 weeks following the last 679 transfer for signs of autoimmunity as described above. 680

Generation of apoptotic bodies 681
Primary mouse thymocytes were isolated and formed into a single cell suspension using a 70 µm cell 682 strainer. Cells were labeled with 10 µm of CFSE and washed as described above. Cells were then 683 cultured for two days in RPMI1640 containing 100 U/mL penicillin and streptomycin (Gibco), and 50 µM 684 2-mercaptoethanol (Amresco) and 5 µg/mL puromycin (Thermo Fisher Scientific) to induce apoptosis. 685 Cultures were collected and washed with PBS; intact cells were pelleted by gentle centrifugation at 300 686 x g for five minutes. Cell supernatant containing ApoB were transferred into mice by intravenous 687 injection. 688

LPS activation in vivo 689
Mice were injected with 1 µg/mL or 0.05 µg/mL of LPS by intraperitoneal injection to activate B-1 cells in 690 the peritoneal cavity. Mice were euthanized 24 hours after injection and B-1 cells were analysed by flow 691 cytometry. Additionally, mice received CFSE labelled ApoB derived from 2 million initial thymocytes at 692 12 h after injection. Mice were then analyzed for ApoB staining on subcapsular sinus macrophages in the 693 spleen, as described above, by flow cytometry.