AI-guided use of balanced PPARα/ g dual agonist finetunes 1 macrophage responses in inflammatory bowel disease 2

53 A computational platform, the Boolean network explorer ( BoNE ), has recently been developed 54 to infuse AI-enhanced precision into drug discovery; it enables querying and navigating 55 invariant Boolean Implication Networks of disease maps for prioritizing high-value targets. 56 Here we used BoNE to query an Inflammatory Bowel Disease (IBD)-map and prioritize two 57 nuclear receptors, PPAR a / g . Balanced agonism of PPAR a / g was predicted to impact 58 macrophage processes, ameliorate colitis in network-prioritized animal models, ‘reset’ the gene 59 expression network from disease to health, and achieve a favorable therapeutic index that 60 tracked other FDA-approved targets. Predictions were validated using a balanced and potent 61 PPAR a / g -dual agonist (PAR5359) in two pre-clinical murine models, i.e., Citrobacter 62 rodentium- induced infectious colitis and DSS-induced colitis. Mechanistically, we show that 63 such balanced dual agonism promotes bacterial clearance more efficiently than individual 64 agonists both in vivo and in vitro ; PPAR a is required and its agonism is sufficient to induce 65 the pro-inflammatory cytokines and cellular ROS, which are essential for bacterial clearance 66 and immunity, whereas PPAR g -agonism blunts these responses, delays microbial clearance 67 and induces the anti-inflammatory cytokine, IL10. Balanced agonism achieved controlled 68 inflammation while protecting the gut barrier and ‘reversal’ of the transcriptomic network. 69 Furthermore, dual agonism effectively reversed the defective bacterial clearance observed in 70 PBMCs derived from IBD patients. These findings not only deliver a macrophage modulator 71 for use as barrier-protective therapy in IBD, but also highlight the potential of BoNE to 72 accelerate and enhance the precision of drug discovery.


INTRODUCTION 87
Inflammatory bowel disease (IBD) is an autoimmune disorder of the gut in which diverse 88 components including microbes, genetics, environment and immune cells interact in elusive 89 ways to culminate in overt diseases 1-3 . It is also heterogeneous with complex sub-disease 90 phenotypes (i.e., strictures, fistula, abscesses and colitis-associated cancers) 4,5 . Currently, 91 patients are offered anti-inflammatory agents that have a ~30-40% response-rate, and 40% of 92 responders become refractory to treatment within one year 6,7 . Little is known to fundamentally 93 tackle the most widely recognized indicator/predictor of disease relapse i.e., a compromised 94 mucosal barrier. Homeostasis within this mucosal barrier is maintained by our innate immune 95 system, and either too little or too much reactivity to invasive commensal or pathogenic 96 bacteria, is associated with IBD 8 . Although defects in the resolution of intestinal inflammation 97 have been attributed to altered monocyte-macrophage processes in IBD, macrophage 98 modulators are yet to emerge as treatment modalities in IBD 8 . 99 We recently developed and validated an AI-guided drug discovery pipeline that uses 100 large transcriptomic datasets (of the human colon) to build a Boolean network of gene clusters 101 relevance to this work, the addition of PPARα-agonistic activity to PPARg or d agonists have 187 led to a higher safety profile, leading to their development for use in many diseases, including 188 type 2 diabetes, dyslipidemia and non-alcoholic fatty liver disease 25 . 189 190 An automated target 'report card' for PPARA/G in IBD 191 We next generated an automated target report card for PPARA/G. High levels of both PPARs 192 were sufficient to distinguish healthy from IBD samples, not just in the test cohort that was 193 used to build the IBD-map (ROC AUC of 0.74; Fig 1B; see also Supplementary Fig. 2A-D), 194 but also in four other independent cohorts with ROC AUC consistently above 0.88 (Fig 1C). 195 High levels of both PPARs also separated responders from non-responders receiving  neutralizing mAbs, GSE16879, E-MTAB-7604 or Vedolizumab that block the α4β7 integrin 197 to prevent selective gut inflammatory, GSE73661 (ROC AUC 0.63-0.89, Fig 1D), inactive 198 disease from active disease (two independent cohorts ROC AUC above 0.93; Fig 1D), and 199 quiescent UC that progressed, or not to neoplasia (ROC AUC=1.00 for qUC vs. nUC; Fig 1D). 200 High level of PPARA/G was also able to distinguish healthy from diseased samples in diverse 201 murine models of colitis ( Fig 1E); but such separation was most effectively noted in some 202 models (Citrobacter infection-induced colitis, adoptive T-cell transfer, TNBS and IL10 -/-), but 203 not in others (DSS, and TNFR1/2 -/-). These findings imply that therapeutics targeting these two 204 genes are best evaluated in the murine models that show the most consistent decrease in the 205 gene expression, e.g., Citrobacter infection-induced colitis, adoptive T-cell transfer, TNBS,206 etc. This was intriguing because the majority (~90%) of the published work on PPARA/G 207 agonists have been carried out in DSS models ( Table 1-2). 208 The expression profile of the target genes in the gut mucosa revealed that PPARA/G 209 are co-expressed at the highest levels in the crypt top epithelial cells and macrophages (Fig 1F;  FDA-approved drugs that have successfully moved through the three phases of drug discovery 216 (i.e., proven efficacy, with acceptable toxicity). A low number is indicative of a high likelihood 217 of success in Phase-III trials. Finally, PPARA/G expression was downregulated to a similar extent in both genders (Fig 1H), predicting that therapeutics targeting them are likely to be 219 effective in both men and women. 220 221

Rationalization of PPARA/G as targets in IBD 222
Because proteins, but not transcripts, are the targets of therapeutic agents, the impact of 223 therapeutics is translated to cellular processes via protein-protein interaction (PPI) networks, 224 a.k.a interactomes. We next asked how dual agonists of PPARa/g might impact cellular 225 pathways and processes. A PPI network visualized using PPARa/g as 'query/input' and the 226 interactive STRING v11.0 database (https://string-db.org/) as a web resource of known and 227 predicted protein-protein interactions curated from numerous sources, including experimental 228 data, computational prediction methods and public text collections. PGC1a (a product of the 229 gene PPARGC1A) was a common interactor between the two PPARs (Fig 2A). We noted that 230 PGC1a also happens to be a major component within the PPARa/g functional network, serving 231 as a central hub for positive feedback loops between the PPARs and their biological function 232 (Fig 2B), i.e., mitochondrial biogenesis, DNA replication and energetics (electron transport 233 chain and oxidative phosphorylation). When we analyzed the functional role of the 234 interactomes of PPARa/g we noted that indeed both interactomes converged on lipid 235 metabolism, mitochondrial bioenergetics and circadian processes (Fig 2C). These findings are 236 consistent with the finding that PPARA/G and PPARGC1A are located within clusters #1-2-3 237 and all of them are predicted to be progressively and simultaneously downregulated in IBD 238 samples (Fig 2D; based on the IBD map, Supplementary Fig. 1). 239

PPARA/G is downregulated in Ulcerative colitis and Crohn's Disease 241
Previous work demonstrated that both PPARa and PPARg are highly expressed in the colon 242 26 . They have also shown that both PPARa/g protein and mRNA are downregulated (by ~60%) 243 in active UC and the expression of PPARg was significantly associated with disease activity 27 . 244 This impaired expression was found in both inflamed and noninflamed areas 28 . 245 Polymorphisms have also been detected in PPARg; while some studies found those to be 246 associated with an increased risk for CD 29,30 , others found no evidence suggesting any form 247 of association with an increased disease risk 31 . We collected endoscopically obtained biopsies 248 from the colons of healthy (n = 7) and IBD (n = 14 and 14 of UC and CD, respectively) patients 249 and assessed the levels of transcripts for PPARA/G and PPARGC1A by qPCR (Fig 2E). We 250 confirmed that all three transcripts were significantly downregulated in UC and CD samples 251 compared to healthy; both PPARG and PPARGC1a were more significantly downregulated in 252 CD compared to UC (Fig 2F). These findings are in keeping with the network-based 253 predictions that these genes should be downregulated invariably in all IBD samples, regardless 254 of disease subtype (see individual disease maps; Supplementary Fig. 4-5). While both PPARA 255 and PPARG are in cluster #2 in the UC map, PPARG and PPARA are in separate clusters, 256 clusters 2 and 6, respectively, in the CD map (Supplementary Fig. 4-5). Reactome pathway 257 analyses implied that in the case of UC, the two nuclear receptors may co-regulate similar 258 cellular homeostatic processes associated with cluster #2, i.e., mitochondrial biogenesis and 259 translation initiation, infectious disease and detoxification of ROS (see Supplementary Fig.  260   4). By contrast, in the case of CD, they may independently regulate diverse cellular processes 261 that maintain cellular homeostasis; while PPARG is associated with cellular metabolism (TCA 262 cycle) and inhibition of NFkB signaling, PPARA is associated with transcriptional activity of 263 nuclear receptors, cholesterol biosynthesis and Met/Ras signaling (see Supplementary Fig.  264 5). Taken together, these findings demonstrate that both targets are downregulated in IBD and 265 that they may regulate key pathophysiologic processes that are vital for cellular homeostasis. 266 267

Synthesis and validation of PAR5359, a potent and specific PPAR α/g dual agonist 268
We noted that all commercially available PPAR α/g dual agonists lack 'balanced' agonistic 269 activities ( Table 3) 32,33 . Drugs that have fallen aside due to safety concerns also lack balanced 270 agonism; most of them are more potent on PPARγ than on PPARα by a log-fold ( Table 3). All 271 these PPAR α/g dual agonists have been withdrawn due to safety concerns 25 , but the cause of 272 the 'unsafe' profile remains poorly understood. Saroglitazar, the drug that is the only active 273 ongoing Phase-III trial (NCT03061721) in this class, has ~3 log-fold more potency on PPARa 274 than PPARg 34 . Because our AI-guided approach suggested the use of simultaneous and 275 balanced agonism, we favored the use of the only balanced and yet, specific PPARa/g agonist 276 described to date, PAR5359 35,36 (see Table 4). In the absence of commercial sources or well-277 defined methods on how to synthesize this molecule, we generated PAR5359 in 4 synthetic steps 278 (see details in Methods) and confirmed its specificity and the comparable agonistic activities 279 using pure single PPARa [GW7647 37 ] or PPARg [Pioglitazone 38 ] agonists as controls 280 ( Supplementary Fig. 6). With these potent and specific compounds as tools, and their doses 281 adjusted to achieve the same potency, we set out to validate the network-based predictions 282 using pre-clinical models. 283

PAR5359 ameliorates C. rodentium-induced colitis, enhances bacterial clearance 285
We next sought to assess the efficacy of individual and dual agonists of our compounds in 286 murine pre-clinical models. PPARa/g's role (or the role of their agonists) in protecting the gut 287 barrier has been evaluated primarily in DSS-induced colitis (Table 1, 2). However, BoNE 288 prioritized other models over DSS, many of which accurately recapitulate the 289 PPARa/g-downregulation that is observed in the barrier-defect transcript signature in human 290 IBD (Fig. 1E). Among those, we chose C. rodentium-induced infectious colitis, a robust model   to achieve equipotent agonistic activities (Supplementary Fig. 6). Fecal pellets of individual 299 mice were collected to determine the number of live bacteria present in the stool. As 300 anticipated, the bacterial burden in all mice increased from day 5, reaching a peak on day 7, 301 forming a plateau until day 11 before returning to pre-infection baseline by day 18 (Fig. 3B). presumably infarcts (arrows, Supplementary Fig. 7C, 7E). Notably, the spleens of mice 318 treated with PPARα-alone agonist, GW7647, showed a significant increase in spleen length 319 ( Supplementary Fig. 7D, 7F). 320 Taken together, these findings indicate that PPARa/g dual agonist PAR5359 is superior 321 in ameliorating C. rodentium-induced colitis than either PPARa or PPARg agonist used alone. Taken together, these findings suggest that dual agonists of PPARa/g are sufficient to 346 either resist network shift and/or reverse the disease network in the setting of colitis. They also 347 offer clues suggestive of epithelial and macrophage processes, two key cellular components of 348 innate immunity in the gut lining as major mechanisms. These transcriptome wide impacts suggest that PPARa/g dual agonist PAR5359 is superior in restoring colon homeostasis in C. 350 rodentium-induced colitis than either PPARa or PPARg agonist used alone. 351

PAR5359 ameliorates DSS-induced colitis 353
It is well known that no single mouse model recapitulates all the multifaceted complexities of 354 IBD 43,44 . Because almost all studies evaluating PPARa/g-modulators have been performed on 355 the DSS-induced colitis model (Table 1-2), we asked whether the PPARa/g dual agonist 356 PAR5359 can ameliorate colitis in this model. Mice receive intrarectal DMSO vehicle control 357 or PAR5359 while receiving DSS in their drinking water (Supplementary Fig. 9A). Disease 358 severity parameters, i.e., weight loss, disease activity index, shortening of the colon and histology 359 score were significantly ameliorated in the PAR5359-treated group (Supplementary Fig. 9A-E). 360 These findings show that the PPARa/g-dual agonist, PAR5359, is also effective in DSS-361 induced colitis. It is noteworthy that the PAR5359 dual agonist offered protection in the DSS-362 model, because prior studies using the same model have demonstrated that PPARa-agonists 363 worsen 45,46 , and that the PPARg-agonists ameliorate colitis 47-49 (see Table 1-2). 364 365 PAR5359 promotes bacterial clearance with controlled production of ROS and 366 inflammation in peritoneal macrophages 367 Because the intestinal macrophages play a crucial role in maintaining the integrity of the gut 368 epithelial barrier and in the control of pathogen invasion by triggering an appropriate immune 369 response, the fine-tuning macrophage is essential in maintaining homeostasis and, potentially, 370 the development of IBD. To explore the mechanism(s) by which dual, but not individual, PPAR 371 agonists may alter macrophage response to microbes, we incubated macrophages treated or not 372 with the drugs and challenged them with CD-associated adherent invasive E. coli (AIEC)-373 LF82; this strain, originally isolated from a chronic ileal lesion from a CD patient 50 . As for the 374 source of macrophages, we isolated metabolically active primary murine peritoneal 375 macrophages using Brewer thioglycolate medium using established protocols 51,52 . These 376 macrophages are known to have high phagocytic activity 51 (Fig. 4A). Thioglycolate-induced 377 peritoneal macrophages (TG-PMs) were lysed, and viable intracellular bacteria were counted 378 after plating on an agar plate. Pre-treatment with 1 µM PAR5359 and an equipotent amount of 379 GW7647 (PPARa agonist) promoted bacterial clearance and reduced the bacterial burden 380 when compared to vehicle control (Fig. 4B). By contrast, pre-treatment with Pioglitazone 381 (PPARg agonist) inhibited bacterial clearance; notably, bacterial burden was significantly higher at both 3 h and 6 h after infection (Fig. 4B). Reduced clearance of microbes in the latter 383 was associated also with reduced cellular levels of reactive oxygen species (ROS) (Fig. 4C); 384 oxidative burst and induction of ROS is key component for effective bacterial killing 53,54 . 385 PAR5359 did not interfere with the production of microbe-induced ROS, and the PPARa 386 agonist (GW7647) was permissive to ROS induction (in fact, even induces it over bacteria-387 alone control) during initial time points after infection (Fig. 4C). 388 These patterns of microbial clearance and cellular ROS were associated also with the 389 expression of cytokines, as determined by qRT-PCR analyses (Fig. 4D). As expected, infection 390 of TG-PM with AIEC-LF82 induced il1b, il6, tnfa and il10. PAR5359 significantly and 391 selectively suppressed the expression of the pro-inflammatory cytokines il1b, il6 and tnfa (but 392 not the anti-inflammatory cytokine, il10) (Fig. 4D). By contrast, the PPARg-specific agonist 393 pioglitazone significantly and indiscriminately suppressed all the cytokines, while there was 394 no effect of the PPARa specific agonist GW7647 (Fig. 4D). ELISA studied on the supernatant 395 media further confirmed these findings (Fig. 4E), demonstrating that the effects in gene 396 expression were also translated to the levels of secreted cytokine protein released by the 397 macrophages in the supernatant. 398 It is noteworthy that for the most part, the qPCR (Fig. 4D) and ELISA (Fig. 4E) studies 399 matched, except il10; although pioglitazone appeared to suppress il10 mRNA, it did not 400 suppress the levels of the il10 protein, suggesting that PPARg-agonist is sufficient for an overall 401 anti-inflammatory phenotype. Similarly, although GW7647 appeared to not affect il10 mRNA, 402 it suppressed the levels of the il10 protein, suggesting that PPARa agonist is sufficient for an 403 overall pro-inflammatory phenotype. Similar findings were also observed in the case of another 404 enteric pathogen, S. enterica, i.e., unlike the dual agonist, neither PPARa nor PPARg agonist 405 could enhance bacterial clearance with a modest induction of pro-inflammatory cytokines 406 (significantly lower than control) and, yet, had no impact on anti-inflammatory IL10 407 production (Supplementary Fig. 10). 408 Taken together, these results show that-(i) PPARg-agonism induces 'tolerance' by 409 suppressing inflammation, inhibiting ROS production and delaying bacterial clearance; (ii) 410 PPARa-agonism enhances the induction of inflammation and ROS, and promotes bacterial 411 clearance; and (iii) PPARa/g-dual agonism strikes a somewhat balanced response. The latter 412 suppresses proinflammatory cytokines without suppressing anti-inflammatory cytokine il10, 413 and is permissive to inflammation and ROS induction that is optimal and sufficient to promote 414 bacterial clearance.

PPARa, but not PPARg is required for the induction of inflammatory cytokines and ROS 416
To further dissect which nuclear receptors are responsible for the balanced actions of the dual 417 agonist, we next used a set of highly specific and potent PPARa/g-inhibitors (Table 4). We 418 pre-treated TG-PMs with PPARa and PPARg inhibitors, either alone, or in combination, 419 followed by stimulation with bacterial cell wall component LPS (Fig. 5A). As expected, LPS 420 induced the cellular levels of ROS (Fig. 5B) and inflammatory cytokines (Fig. 5C-D) in TG-421 PMs significantly higher than in untreated control cells. Inhibition of PPARa suppressed the 422 induction of cellular ROS and inflammatory cytokines, both at the level of gene and protein 423 levels ( Fig. 5B-D). By contrast, inhibition of PPARg did not interfere with either response (Fig.  424   5B-D). Simultaneous inhibition of both PPARa and PPARg mimicked the cellular phenotypes 425 in the presence of PPARa-inhibitors ( Fig. 5B-D), indicating that inhibition of PPARa is 426 sufficient to recapitulate the phenotype of dual inhibition. Taken together, these findings 427 indicate that PPARa is required for the proinflammatory response of macrophages. 428 429

PPARa/g dual agonist PAR5359 promotes bacterial clearance in patient-derived PBMCs 430
In search of a pre-clinical human model for testing drug efficacy, we next assessed microbial 431 handling by PBMCs derived from patients with IBD and compared them with that in age-432 matched healthy volunteers. We enrolled both male and female patients and both CD and UC 433 (Table 5). Consecutive patients presenting for routine care to the UC San Diego IBD clinic 434 were enrolled into the study; the only exclusion criteria were failure to obtain informed consent 435 for the study or active infections and/or disease flare. Peripheral blood collected in the clinic 436 was freshly processed as outlined in Fig. 6A to isolate PBMCs. Pre-treatment for 30 min with 437 vehicle or PAR5359 was followed by infection for 1h. Subsequently, the cells were treated 438 with gentamicin for 60 min to kill extracellular bacteria to assess intracellular bacterial burden 439 at 1 and 6 h after the gentamicin wash. 440 Two observations were made: First, CD but not UC patient-derived PBMCs when 441 infected with AIEC-LF82 showed an increased number of internalized viable bacteria when 442 compared to healthy PBMCs (Fig. 6B, 6E), indicative of either defective clearance and/or 443 increased permissiveness to bacterial replication within the cells is limited to the CD. Second, 444 pre-treatment with PAR5359 could improve clearance significantly (Fig. 6C-D, 6F-G). These 445 results indicate that bacterial clearance is delayed in PBMCs of patients with CD and that 446 PPARa/g dual agonism with PAR5359 can reverse that defect. The possibility that such 447 reversal could be due to any direct bacteriostatic/-cidal effect of PAR5359 agonist was ruled out (see bacterial viability assay in Supplemental Fig. 11). Our findings demonstrate that 449 bacterial clearance is delayed primarily in CD and not UC are in keeping with the fact that 450 delayed bacterial clearance from inflamed tissues (up to ~4-fold) is uniquely observed in CD 451 55 . These findings are also in keeping with our own observation that the downregulation of 452 PPARG/PPARGC1A was more prominent in patients with CD (Fig. 2E-F). In fact, delayed 453 clearance is one of the major reasons for persistent inflammation and disease progression 454 among patients with CD 55,56 . 455

456
DISCUSSION 457 Barrier-protection/restoration is the treatment endpoint for all clinical trials in IBD 458 therapeutics; however, despite much success in the development of anti-inflammatory 459 therapies 7,57 , barrier-protective therapeutics in IBD have been slow to emerge 58 . Here we 460 report the discovery of an effective barrier-protective therapeutic strategy in IBD identified 461 using an AI-guided navigation framework (summarized in Fig 7). First, a network-based drug 462 discovery approach 9 was used to identify, rationalize and validate dual and balanced agonism 463 of PPARa/g (but not one at a time) is necessary for therapeutic success. The impact of using such an approach is 4-fold: (i) Because the network approach used 485 here relies on the fundamental invariant Boolean implication relationships between genes, and 486 their patterns of changes in expression between healthy and IBD samples, such 'rule of 487 invariant' implies that any given relationship and/or change in expression pattern annotated 488 within the network must be fulfilled in every IBD patient. By that token, targets/drugs 489 prioritized based on this network is expected to retain efficacy beyond inbred laboratory mice, 490 into the heterogeneous patient cohorts in the clinic. (ii) This AI-guided approach not just helped 491 compute pre-test probabilities of success ("Therapeutic Index"), but also helped pick models 492 that are most insightful and appropriate to demonstrate therapeutic efficacy (e.g., Citrobacter 493 rodentium infection-induced colitis) and to pinpoint the cell type and mechanism of action 494 (microbial clearance by macrophages). This is noteworthy because the conventional approach 495 in studying PPARs has been limited to the use of DSS-induced colitis (see Table 1-2), which 496 has often given conflicting results (see Table 2 'target report card', like the one shown here, is a project navigation tool that is geared to 506 streamline decision-making (i.e., which genes, which animal models, which cell type/cellular 507 process, what is the likelihood of success, etc.), which in turn should reduce attrition rates, 508 waste and delays; the latter are well-recognized flaws in the current process of drug discovery. 509 Second, regarding IBD therapeutics, our studies demonstrate that single or unbalanced 510 combinations of PPAR agonists are inferior to dual/balanced agonists. Conventional and 511 reductionist approaches have inspired numerous studies with single PPAR agonists over the 512 past decade (Tables 1-2). However, given the devastating side effects of most single or 513 unbalanced PPARα/g agonists (Table 3), translating to the clinic beyond a Phase II trial 20,62,63 514 has not been realized. Because the therapeutic index for the dual PPARα/g agonists matches that of other FDA-approved targets/drugs, it is predicted that barring unexpected side effects, 516 dual PPAR agonists are likely to be effective as barrier-protective agents. As for side effects, 517 we noted is that balanced PPARg/α agonists are rare; while all dual PPARα/g agonists that have 518 been discontinued due to side effects happen to be either single (only PPARg) or 'unbalanced' 519 (PPARg >> PPARα agonistic activity), the newer generation formulations that are currently in 520 the clinical trial have a reversed agonistic potency (PPARa >> PPARg agonistic activity) (see 521   Table 3). Because macrophage responses require finetuning (discussed below), our studies 522 show how unopposed agonism of either PPARg or PPARα is harmful and can 523 impair/dysregulate the way macrophages respond when microbes breach past the gut barrier. 524 It is possible that many of the side effects of the discontinued thiazolidinediones are due to 525 their inability to achieve that 'optimal' spectrum of macrophage function. 526 Third, when it comes to macrophage biology, this work sheds some unexpected and 527 previously unforeseen insights into the role of the PPARs in the regulation of macrophage 528 processes. Extensively studied for over ~3 decades, PPARs are known to regulate macrophage 529 activation in health and disease 64 . Targeting PPARs as a host-directed treatment approach to 530 infectious/inflammatory diseases appears to be a sound strategy because they regulate 531 macrophage lipid metabolism, cholesterol efflux, inflammatory responses (ROS and cytokine 532 production), apoptosis, and production of antimicrobial byproducts 65 . We found that 533 unopposed PPARg activation suppresses bacterial clearance and blunts the induction of 534 proinflammatory (but not anti-inflammatory, IL10) cytokines and ROS in response to infection 535 both in vivo and in vitro. In other words, and consistent with prior reports, PPAR g activation 536 suppressed inflammation at the cost of impairing immunity. Our findings are in keeping with 537 the findings of a systematic review and meta-analysis of 13 long-term randomized controlled 538 trials that involved 17,627 participants (8,163 receiving PPARg agonists and 9,464 receiving 539 control drugs) 66 . Long-term (~1-5.5 y) use of PPARg agonists increases the risk of pneumonia 540 or lower respiratory tract infection significantly, some of which result in hospitalization, 541 disability, or death 66 . In the case of PPARa, unopposed activation-induced ROS and 542 proinflammatory cytokines and accelerated bacterial clearance. Inhibitor studies further 543 confirmed that PPARa was required for these responses (Fig 5). These findings are in keeping 544 with others' showing that PPARa, but not PPARg is required for NADPH-induced ROS 545 formation both in human and murine macrophages 67 . PPARa agonists induce the expression 546 of NADPH oxidase subunits p47(phox), p67phox, and gp91phox, which are all essential 547 functional components of NADPH complex 67 . Dual and balanced PPARα/g agonism enhanced 548 bacterial clearance with only a moderate induction of proinflammatory cytokines or ROS. Such 549 a response ensures that the macrophage functions within a 'goldilocks' zone, mounting 550 inflammation that is just sufficient for microbial clearance and immunity. In our analysis, the 551 only other PPAR-related gene within the IBD network, i.e., PGC1a, and its role within the 552 PPARα/g axis suggests that the intricate network of forward feedback loops orchestrated by 553 PGC1a may be critical for achieving the critical balance between immunity and inflammation, 554 which is a key outcome of the dual PPARα/g agonists. 555 Because previous studies using cell-specific gene depletion have indicated that the 556 barrier-protective role of PPARg may be mediated via cells other than the macrophages 48 , 557 namely, the T cells 68 and the epithelial cells 69 , it is possible that the dual PPARα/g agonists 558 also act on those cells, promoting bacterial clearance and balancing cellular bioenergetics, ROS 559 and cytokine production, in manners similar to that we observe in macrophages.  826  827  828  829  830  831  832  833  834  835  836  837  838  839  840  841  842  843  844  845  846  847  848  849  850  851  852 853 854

881
Significance was tested using two-way/one-way ANOVA followed by Tukey's test for multiple comparisons.