Regulation of blood – brain barrier integrity and cognition by the microbiome-1 associated methylamines trimethylamine-N-oxide and trimethylamine 2 3

Communication between the gut microbiota and the brain is primarily mediated via soluble microbe-derived metabolites, but the details of this pathway remain poorly defined. Methylamines produced by microbial metabolism of dietary choline and L-carnitine have received attention due to their proposed association with vascular disease, but their effects upon the cerebrovascular circulation have not hitherto been studied. Here we use an integrated in vitro/in vivo approach to show that physiologically relevant concentrations of the dietary methylamine trimethylamine N- oxide (TMAO) enhanced and protected blood-brain barrier (BBB) integrity, acting through the tight junction regulator annexin A1. In contrast, the TMAO precursor trimethylamine (TMA) impaired BBB function and disrupted tight junction integrity. Moreover, we show that long-term exposure to TMAO has beneficial effects upon cognition in mice, improving visual recognition memory. Our findings demonstrate a direct interaction of microbiome-associated metabolites with the mammalian BBB, with consequences for cerebrovascular and cognitive function.

were precipitated by incubation with 60% trichloroacetic acid, and dye content of 242 resulting supernatants was detected using a CLARIOstar spectrophotometer (BMG 243 Labtech GmbH, Germany) alongside a standard curve of defined concentrations of 244 Evans blue in the same buffer. Brain Evan's blue content was expressed as µg of dye 245 per mg of brain tissue, normalized to circulating plasma concentrations. 246 247

Long-term LPS and TMAO treatments 248
To assess the long-term impact of both LPS and TMAO on cognitive performance, 249 mice were divided into four groups (n=8 per group): 1) Water + PBS; 2) Water + TMAO; 250 3) LPS + PBS; 4) LPS + TMAO. C57Bl/6 mice were administered phosphate-buffered 251 saline (PBS) or LPS (Escherichia coli O55:B5, Sigma-Aldrich, UK) via intraperitoneal 252 (i.p.) injection (0.5 mg/kg/wk) for 8 weeks 43 . A final LPS treatment was administered 253 the day before sacrifice for nine total injections. Body weights were recorded prior to 254 each injection. TMAO was provided in the drinking water (500 mg/L) and water bottles 255 replaced every other day. Drinking volumes were recorded before bottle change. 256

Processing and analyses of RNAseq data 258
Mice were transcardially perfused with 0.9% saline at 4 °C to remove circulating blood, 259 and brains were removed and collected into RNAlater (Thermofisher Scientific Ltd., 260 UK) prior to storage at -20 °C for later analysis. Whole brain total RNA was extracted 261 using a PureLink RNA Mini Kit (Thermofisher Scientific Ltd., UK) and quantified using 262

Behavioural analyses 278
Behavioural tests were performed in the order they are introduced below. Apparatus 279 was cleaned using 70 % ethanol upon completion of each trial, eliminating any residual 280

odour. 281
Open field test (OFT) was conducted as previously described 46 . Briefly, mice were 283 placed in the centre of the OFT, a grey 50 x 50 x 50 cm apparatus illuminated with low 284 lux (100 lux) lighting. Total travel distance and time spent in the centre of the field was 285 determined at 5 min with a video tracking system (Smart 3.0 tracking software, Panlab, 286 The novel object recognition (NOR), a measure of recognition memory, was performed 289 as described previously 47,48 , with slight modifications. Briefly, on day 1 mice were 290 habituated in a grey 50 x 50 x 50 cm apparatus illuminated with low lux (100 lux) 291 lighting, mice were placed into the empty maze and allowed to move freely for 10 min. 292 On day 2, mice were conditioned to a single object for a 10 min period. On day 3, mice 293 were placed into the same experimental area in the presence of two identical objects 294 for 15 min, after which they were returned to their respective cages and an inter-trial 295 interval of 1 h was observed. One familiar object was replaced with a novel object. 296 Mice were placed back within the testing area for a final 10 min. Videos were analysed 297 for a 5 min period, after which if an accumulative total of 15 s with both objects failed 298 to be reached, analysis continued for the full 10 min or until 15 s was achieved. Those 299 not achieving 15 s were excluded from the analysis 49 . A discrimination index (DI) was 300 calculated as follows: DI = (TN−TF)/(TN+TF), where TN is the time spent exploring 301 the novel object and TF is the time spent exploring the familiar object. 302 303 Y-maze spontaneous alternation test, a measure of spatial working memory, was 304 performed on the final day of behavioural testing as previously described 50 . Briefly, the 12 maze and allowed to explore freely for 7 min while tracking software recorded zone 308 transitioning and locomotor activity (Smart 3.0 tracking software, Panlab, Kent, UK).  Right hemi-brains were stored at -80 °C until further analysis; cerebellums were 325 processed immediately for the sodium fluorescein extravasation assay. Cleared 326 volume of sodium fluorescein that passed from the plasma into the brain was 327 calculated as described previously 43 . 328 329

Statistical analyses 330
Sample sizes were calculated to detect differences of 15 % or more with a power of 331 0.85 and α set at 5 %, calculations being informed by previously published data 2,31 . In 332 vitro experimental data (except those for in vitro microarray experiments) are 333 expressed as mean ± SEM, with a minimum of n = 3 independent experiments 334 performed in triplicate for all studies. In all cases, normality of distribution was 335 established using the Shapiro-Wilks test, followed by analysis with two-tailed 336 Student's t-tests to compare two groups or, for multiple comparison analysis, 1-or 2-337 way ANOVA followed by Tukey's HSD post hoc test. Where data were not normally 338 distributed, non-parametric analysis was performed using the Wilcoxon signed rank 339 test. A P value of less than or equal to 5 % was considered significant. Differentially 340 expressed genes were identified in microarray data using LIMMA (Ritchie et al, 2015); (False Discovery Rate); a P value of less than or equal to 10 % was considered 343 significant in this case; n = 5 for control, TMAO and TMA groups. Significantly 344 differentially expressed genes (PFDR<0.1) in RNAseq data (Supplementary Table 11 was apparent. In contrast, TMA had no effect upon TEER at any concentration studied, 361 while TMAO enhanced TEER by approximately 65%, an effect that was notably dose-362 independent (Fig. 1B). 363

364
The physical barrier that the BBB provides is only one aspect by which it separates 365 the brain parenchymal environment from the periphery, equally important is the 366 immunological barrier that it represents. To model this, we employed a simple system 367 in which adhesion of CMFDA-labelled U937 monocytic cells to TNFa-activated 368 (10 ng/ml, 16 h) hCMEC/D3 monolayers was quantified in response to TMA or TMAO 369 treatment. Treatment with a physiologically relevant concentration of TMA (0.4 µM 42 , 370 24 h post-TNFa) had no effect on the density of adherent U937 cells, but exposure of 371 hCMEC/D3 monolayers to physiological levels of TMAO (40 µM 42 , 24 h post-TNFa) 372 significantly reduced U937 cell adhesion by approximately 50% compared to cultures 373 stimulated with TNFa alone (Fig. 1C). 374

375
The endothelial cells of the BBB express numerous efflux transporter proteins that 376 serve to limit entry of endogenous and exogenous molecules into the parenchyma, 377 with BCRP and P-glycoprotein being two of the most important. Consequently, we 378 examined whether treatment with TMA or TMAO affected function or expression of 379 either of these two transporters. Using commercially available in vitro assays, neither 380 range (TMA: 4.9 nM to 10.8 µM; TMAO 0.5 µM to 1.08 mM) (Suppl. Fig. 2A-D).  Table 4). No pathways were shown to be activated or inactivated by 401 the 49 TMA-affected genes (data not shown). 402 with two principal groupings being particularly evident, namely those associated with 416 aspects of cellular metabolism and with regulation of actin cytoskeletal dynamics 417 ( Figure 2E). Finally, we compared the 19,309 genes represented on the microarray 418 with a set of 203 genes 2 known to be associated with the BBB. While TMA treatment 419 had no significant effects on expression of these genes (Supplementary Table 9), 420 TMAO significantly (PFDR<0.1) upregulated expression of four genes from this set 421 associated with transporter proteins and barrier integrity (  The fundamental role of the BBB is to protect the brain, preserving its homeostatic 501 environment; damage to BBB integrity is therefore detrimental, and is believed to 502 directly contribute towards cognitive impairment 57-59 . Having shown TMAO to exert a 503 protective effect upon BBB function/integrity in response to acute inflammatory insult, 504 we next examined whether a similar effect held true for chronic conditions, and 505 whether this protection extended to cognition. TMAO was administered to male 506 C57Bl/6J mice through drinking water (0.5 mg/ml) over 2 months, in combination with 507 chronic low-dose LPS administration (0.5 mg/kg/week, i.p.) to model a mild 508 inflammatory stress known to impact cognitive behaviour 43 . There were no differences 509 in volumes of water drunk or, where relevant, final consumption of TMAO between any 510 groups ( Table 2). The serum inflammatory cytokines TNFa and IL-1b were both 511 nominally elevated in response to LPS treatment, although not reaching statistical 512 significance, indicating a sub-clinical inflammatory response; TMAO had no effect on effect reversed by TMAO treatment (Fig. 6A). Treatment with LPS increased 516 cerebellar FITC extravasation, an effect that was prevented by TMAO treatment, 517 although this did not reach statistical significance on post hoc analysis (Fig. 6B). To 518 corroborate these findings, we investigated a second marker of impaired BBB integrity, 519 confocal microscopic detection of brain perivascular IgG deposition. In comparison 520 with sham-treated animals, exposure to LPS caused a significant accumulation of IgG 521 in the perivascular compartment, an effect prevented by TMAO treatment (Fig. 6C). 522

523
The OFT confirmed neither LPS nor TMAO treatment affected motor function, with 524 movement speed and distance travelled comparable across treatment groups (Fig.  525 6D-E). Similarly, no effect was apparent on the proportion of time animals spent in the 526 centre of the field, suggesting limited effects upon anxiety (Fig. 6F). Working memory, 527 however, determined via NOR indicated a significant reduction in performance in 528 animals exposed to LPS, a behavioural deficit notably prevented in animals co-treated 529 with TMAO (Fig. 6G). In contrast, no effect of either LPS or TMAO treatment was 530 apparent in the Y-maze spontaneous alternation task (Fig. 6H) or in distance travelled 531 during this task (Fig. 6I), indicating no differences in spatial memory. 532 The relationship between the BBB and cognitive behaviour is complex and far from 537 being fully understood, but it is clear from both human and animal studies that deficits 538 in barrier integrity can exert a profound and deleterious effect upon memory, language 539 and executive function 60-63 . Indeed, BBB impairment is among the first events to occur 540 in the course of Alzheimer's disease, and may aggravate the pathological processes 541 that underlie the condition 64 . Strategies to promote BBB function may thus have 542 significant value in helping to protect the brain from progressive neurological diseases 543 such as dementia. In this study we identify novel and distinct roles for the microbiome-