Mediterranean Diet Reduces Monocyte Inflammatory Gene Expression and Influences Social Behavior in Nonhuman Primates

Western diet consumption is associated with inflammation, cardiometabolic disease, and mortality in humans, while Mediterranean diet consumption confers protective effects. One likely pathway for this association is through environmentally induced changes in monocyte function, yet the underlying mechanisms remain elusive. We conducted the first randomized, long-term diet manipulation in a non-human primate model to determine whether Western- or Mediterranean-like diets alter monocyte polarization and health. Monocyte gene expression profiles differed markedly between the two diet groups, with significant differences in over 40% of expressed genes. The Western diet induced a more proinflammatory monocyte phenotype overall and upregulated specific monocyte polarization genes. Diet also disrupted the coexpression of numerous gene pairs, including small RNAs and transcription factors associated with metabolism and adiposity in humans. Diet altered affiliative and anxiety-associated behaviors and mediation analysis showed that the diet-altered behaviors contributed significantly (∼50% of the effect of diet on gene expression) to 25% of the differentially expressed genes, suggesting that diet effects on central mechanisms also modulate monocyte gene expression. Together, these results identify both behavioral and molecular mechanisms underlying the health benefits of a Mediterranean diet regimen. Significance Statement Some of our largest public health burdens are driven by dietary changes associated with industrialization, but we still know very little about the molecular mechanisms underlying this link. Characteristic “Western diets” have been associated with increased risk for diseases related to chronic inflammation, while Mediterranean diets have anti-inflammatory benefits. Here, we identify causal effects of diet on inflammatory gene expression where consumption of the Mediterranean diet reduced inflammatory gene expression in monocytes. Additionally, our diet manipulation induced behavioral changes associated with anxiety and social integration, where Mediterranean-fed animals exhibited more positive affiliative behaviors and reduced anxiety. These behaviors were associated with 25% of the diet-affected genes, suggesting an important behavioral route through which diet can impact immune function.

3 Abstract: Western diet consumption is associated with inflammation, cardiometabolic disease, 33 and mortality in humans, while Mediterranean diet consumption confers protective effects. One 34 likely pathway for this association is through environmentally induced changes in monocyte 35 function, yet the underlying mechanisms remain elusive. We conducted the first randomized, 36 long-term diet manipulation in a non-human primate model to determine whether Western-or 37 Mediterranean-like diets alter monocyte polarization and health. Monocyte gene expression 38 profiles differed markedly between the two diet groups, with significant differences in over 40% 39 of expressed genes. The Western diet induced a more proinflammatory monocyte phenotype 40 overall and upregulated specific monocyte polarization genes. Diet also disrupted the 41 coexpression of numerous gene pairs, including small RNAs and transcription factors associated 42 with metabolism and adiposity in humans. Diet altered affiliative and anxiety-associated 43 behaviors and mediation analysis showed that the diet-altered behaviors contributed significantly 44 (~50% of the effect of diet on gene expression) to 25% of the differentially expressed genes, 45 suggesting that diet effects on central mechanisms also modulate monocyte gene expression. 46 Together, these results identify both behavioral and molecular mechanisms underlying the health 47 benefits of a Mediterranean diet regimen.  Table S1 for a detailed comparison). As previously reported, Western 136 diet significantly increased body weight, caloric intake, body fat, insulin resistance, and 137 hepatosteatosis relative to the Mediterranean diet 38 (Fig. 1C). Behavioral data were collected during the last 6 weeks of the baseline phase and the during months 1-14 of the experimental phase. Body weight measurements reported are from 5 months prior to, and 14 months after the start of the experimental phase. Monocytes were isolated from blood collected 15 months after the start of the experimental phase. B) Experimental diets were isocaloric with respect to macronutrients, but differed in food sources and relative amounts of micronutrients. Orange bars indicate nutrients with higher concentration in the Western diet formulation, while blue bars indicate higher levels of a given nutrient in the Mediterranean diet. See Table S1 for diet compositions. C) Percent change in body weight from baseline after 14 months on the diet (t(23.0) = 3.02, p = 0.0023).

Diet induced major shifts in monocyte gene expression 141
To test how diet affected the phenotypes of circulating monocytes, we used RNA sequencing to 142 measure genome-wide gene expression of purified CD14+ monocytes after 15 months on the 143 experimental diets. Diet had a strong effect on monocyte gene expression: the first principal 144 component of gene expression, which explained 59.2% variance, was significantly associated 145 with diet (t(25.1) = 4.41, p = 1.7 x 10 -4 ; Fig. 2A), and 40% of the 12,240 expressed genes (Table  146 S2A) were significantly differentially expressed between the two diets (n = 4,900 genes, FDR < 147 0.05; Table S2B). The number of diet-responsive genes was roughly balanced between those that 148 were more highly expressed in monkeys fed the Mediterranean diet (n = 2,664; hereafter 149 "Mediterranean genes") and those that were more highly expressed in monkeys fed the Western 150 diet (n = 2,236; hereafter "Western genes"). While balanced in direction, the distributions of 151 effect sizes in these two sets of genes differed significantly (one sided Kolmogorov-Smirnov test, 152 D = 0.33, p = 5.2 x 10 -112 ) and the effect size of diet on Western genes was, on average, 1.6-fold 153 larger than on Mediterranean genes (Mann-Whitney U = 4.1 x 10 6 , p = 6.1 x 10 -117 ; Fig. 2B). 154 Thus, the strongest effects are seen in genes that are either activated by a Western diet or 155 suppressed by a Mediterranean diet. Western genes was 1.6-fold larger than the effect size of diet on Mediterranean genes (Mann-Whitney U = 4.1 x 10 6 , p = 6.1 x 10 -117 ). C) Log2 fold enrichment of proinflammatory (top) and regulatory (bottom) genes in Western genes (orange) and Mediterranean genes (blue). Western genes contained more M1 genes than expected by chance, indicating that the Western diet induced a shift towards a proinflammatory monocyte phenotype.

171
We next conducted a more targeted analysis of monocyte polarization by focusing on genes that 172 were previously reported to be differentially expressed between induced proinflammatory (M1) 173 and regulatory (M2) monocyte polarization 39 (see Table S2C for polarization categories).  associations in the other, suggesting that diet can completely reverse the co-expression 203 relationship between two genes ( Figure 4A). We further identified 16 "hub" genes that exhibited 204 differential correlations with partner genes more so than expected by chance (Fig. 4B, Table  205 S5B). These hub genes were enriched for genes encoding transcription factors (OR = 7.40, FET 206 p = 7.0 x 10 -3 ), including SOX4 (essential for normal insulin secretion and glucose tolerance) 207 and NR4A2 (involved in lipid, carbohydrate, and energy metabolism 48,49 ), suggesting immune 208 and metabolic reprogramming by the diet manipulation. Interestingly, the hub gene involved in 209 the greatest number of differentially-correlated gene pairs was RF00283, aka SCARNA18, a non-210 coding RNA that has been associated with BMI, HDL cholesterol, and aging in human genome-211 wide association studies 50-53 ( Fig. 4B-D), identifying it as a key regulatory RNA that is altered 212 by diet and has a cascading effect on other genes and pathways. were paired with more genes than expected by chance (n = 16 "hub" genes; dotted black line is the maximum number of significant pairs expected by chance). The strongest hub gene was the non-coding RNA RF00283. C) Residual normalized expression of RF00283 is significantly greater in Western-than Mediterranean-fed monkeys (βdiet = 0.507, FDR = 2.3 x 10 -6 ). D) Example of a differential correlation involving RF00283. Residual normalized expression of RF00283 is plotted against expression of KLF11, a differentially-expressed transcription factor that regulates insulin and has been associated with type II diabetes in humans 54 . The two genes were more highly expressed in Western monocytes, and were positively correlated with one another in Western-fed monkeys (r = 0.61, p < 0.005) and negatively correlated in Mediterranean-fed monkeys (r = -0.63, p < 0.01).

Diet altered social behavior 216
There were no differences in behavior during the baseline phase (all p > 0.1; Fig. S1A, B). While 217 on the experimental diets, monkeys fed the Mediterranean diet spent significantly more time in 218 body contact (Mann-Whitney U = 280, Holm-Bonferroni adjusted p (pHB) = 1.2 x 10 -5 ) and 219 resting (U = 267, pHB = 1.6 x 10 -3 ), while those fed the Western diet spent significantly more 220 time alone (U = 48, pHB = 4.7 x 10 -3 ; Fig. 5A). All other measured behaviors did not pass our 221 stringent p-value threshold after multiple hypothesis testing correction (Fig. S1C,D), although 222 two additional behaviors differed at an uncorrected p-value < 0.05 (percent of time attentive and 223 rate grooming self). Therefore, to increase our ability to identify diet-affected suites of behaviors, 224 we leveraged the fact that many behaviors co-occurred (  Table S6A). The first 229 principal component was significantly correlated with dominance rank (Fig. S4, Note S1).  Table S6B). PC2 captured a number of anxiety and social 236 behaviors (Fig. S5, Table S6A). Specifically, body contact is indicative of social integration and 237 was positively correlated with PC2 loading (hereafter, DAB score), which was higher in 238 Mediterranean fed animals. Conversely, behaviors related to social isolation and anxiety 57-62 239 (e.g., percent of time alone, rate of grooming self, rate of scratching) were associated with lower 240 DAB scores, and hence more prevalent in animals fed the Western diet (Fig. 5C). Thus, PC2 241 captured a measure of social integration associated with consuming a Mediterranean-like diet, 242 and social isolation and anxiety associated with consuming a Western-like diet.  Table S6B for correlation between diet and other PCs). D) Six of the 21 behaviors observed are significantly correlated with DAB score (Benjamini-Hochberg adjusted p < 0.05). Here, significant correlations with DAB score in which behaviors are more frequent in Mediterranean diet or Western diet monkeys are indicated with blue or orange points, respectively.

246
Diet-altered behaviors mediate expression of 25% of differentially expressed monocyte genes 247 Given the strong effects of diet on both behavior and monocyte gene expression, we tested if the 248 effect of diet on monocyte gene expression was mediated by the diet-induced changes in 249 behavior. Of the 4,900 diet-affected genes, 29% were also significantly associated with DAB 250 score in a univariate model (n = 1,418, FDR < 0.05). Of these, DAB score significantly mediated 251 the effect of diet on the expression of 1220 genes (25% of all diet-associated genes, p < 0.05; 252 mediated genes were also significantly more likely to be Western genes than Mediterranean 256 genes (n = 741 Western genes, 61%, two-sided binomial test p = 6.3 x 10 -14 ), and were enriched 257 in regulation of inflammatory response (GO:0050727, weighted FET p = 2.9 x 10 -3 ; Table S7A-258 C). Together, this shows that the effect of diet on monocyte gene regulation may partially be due 259 to diet-induced changes in key social behaviors. 260

261
In support of this mediation effect, we compared expression of a well-studied set of social 262 adversity-responsive genes known as the "conserved transcriptional response to adversity" 263 (CTRA) 28 in the Western-and Mediterranean-fed animals in our study. Animals fed a Western 264 diet exhibited significantly higher expression of pro-inflammatory genes included in the CTRA 265 (Mann-Whitney U = 222, p = 0.016) and lower expression of antiviral-and antibody-related 266 CTRA genes (Mann-Whitney U = 82, p = 0.023; Table S2C, Fig. S6). 267

268
We also tested the hypothesis that diet could alter behavior through its changes on peripheral 269 immune cell gene expression. We tested this in the 28% of genes for which monocyte gene 270 expression significant predicted DAB in a univariate model (n = 1,353, FDR < 0.05), and found 271 that gene expression significantly mediated the effect of diet on DAB score in 940 genes (19% of 272 all diet-associated genes, p < 0.05; Fig. 6A). Almost all of these genes (99.5%; 936/940) were 273 significantly mediated by diet-induced changes in DAB. As with DAB score mediating gene 274 expression, the genes that mediated the effect of diet on DAB score were more likely to be 275 Western genes than Mediterranean genes (n = 558 Western genes, 59%, two-sided binomial test

Western diet induces mosaic response 283
Western diet induced substantial variation in multiple phenotypes, including body weight, gene 284 expression, and behavior; consistent with previous studies demonstrating that some individuals 285 may be more resistant (or susceptible) to the effects of a Western diet 63 , presumably due to 286 genetic variation or past environmental exposures. However, we were unable to identify any 287 20 consistencies in individual responsiveness across the phenotypes (Fig. S7). For instance, 288 monkeys that exhibited a strong gene regulatory response to the Western diet did not exhibit a 289 large increase in body weight or a strong negative DAB score (all p > 0.2). Furthermore, change 290 in body weight did not significantly predict the expression of any genes at an FDR < 20%. 291 Western diet fed individuals thus exhibited a mosaic response to diet across multiple phenotypes, 292 presumably involving interactions between diet, environment, and the genome. in the central nervous system by altering gut microbiota which alters vagal input to the brain 78 . 355 We previously showed in these NHPs that diet had a strong effect on the gut microbiome 79 , and 356 that compared to the Mediterranean group, Western diet NHPs had lower parasympathetic 357 (vagal) activity at the time the monocyte transcriptome was assessed 77 . Taken together these 358 observations suggest that diet-induced changes in vagal tone in the gut-brain axis may be one 359 pathway through which diet impacted brain function, potentially affecting behavior. 360

361
We also observed that for some genes (19%), diet-induced changes in monocyte gene expression 362 significantly mediated the effect of diet on behavior (DAB). This observation suggests 363 underlying mechanisms which first impact peripheral monocyte function, which in turn impacts 364 brain function. Western diet may disrupt the blood-brain barrier, increasing infiltration of 365 Western-diet induced cytokines, chemokines, and myeloid cells from the periphery 80,81 . Once in 366 the brain these molecules can alter BDNF production, neurotransmitter systems, and Center (Winston-Salem, NC) as previously described 38 . Briefly, the monkeys were socially 403 housed in groups of 3-4 and consumed standard monkey chow (Table S1) during an eight-month 404 baseline phase, after which pens were assigned to receive either the Western (5 groups, n = 21) 405 or Mediterranean (6 groups, n = 22) diet, balanced on pretreatment characteristics that reflected 406 overall health, including body weight, body mass index, and plasma triglyceride concentrations 407 ( 38 ; Fig. 1A). Two monkeys did not tolerate the experimental diet, and were switched to standard 408 monkey chow, three animals died during the course of the study, and three samples were Experimental diets (Table S1) were formulated to be isocaloric with respect to protein, fat, and 417 carbohydrates, and identical in cholesterol content (~ 320mg / 2000 kilocalories (Cals)/day) as 418 previously described 38 . The Western diet was formulated to be similar to that consumed by 419 American women age 40-49 as reported by the US Dept. Agriculture, with protein and fat 420 derived mainly from animal sources. The Western diet was relatively high in saturated fat and 421 sodium, and low in monounsaturated fat and n-3 fatty acids. The Mediterranean diet was 422 formulated to mimic key aspects of the traditional Mediterranean diet, with an n-6:n-3 fatty acid 423 ratio similar to a traditional hunter-gatherer type diet 12,86,87 . Protein and fats were derived mainly 424 from plant sources, fish and dairy, and monounsaturated fatty acids were relatively high. 425 Mediterranean diet contained more complex carbohydrates and fiber, and less sodium and 426 refined sugars than Western diet. Key ingredients included English walnut powder and extra-427 virgin olive oil which were the primary components provided to participants in the PREDIMED 428 study, a landmark dietary intervention study that illustrated the role of the Mediterranean diet in 429 cardiovascular disease prevention 88 . 430 431

Behavioral Characterization 432
Behavioral data were collected weekly during two 10-minute focal observations, randomly 433 ordered and balanced for time of day, for 6 weeks during the baseline phase (2 hours/monkey 434 total) and for 14 months during the experimental phase (17.7 hours/monkey total). Behaviors 435 were collected as previously described 89 , and combined into summary behaviors (e.g., 436 "aggression" was a combination of all total, noncontact, contact aggressive events). No 437 significant differences in behavioral variables were observed between the diet groups which

Enrichment analyses 492
Gene ontology (GO) enrichment analyses were conducted using Fisher's Exact Tests and the 493 weight01 algorithm to test for enrichment implemented in the R package topGO 99 . For a more 494 targeted analysis of M1 and M2 specific genes, we identified a set of differentially expressed 495 genes in our data set that were previously found to be involved in monocyte polarization 39 (638 496 proinflammatory and 138 regulatory), which we used to explore monocyte polarization in the 497 current study. We calculated the proportion of genes more highly expressed in the 498 Mediterranean-and Western-fed animals in each polarization category and tested for 499 significance using a permutation test (n = 100,000 permutations). 500 501

Transcription Factor Binding Site Analysis 502
We tested for enrichment of transcription factor binding motifs within 2 kb (upstream or 503 downstream) of the transcription start sites of differentially expressed "Western genes" or 504 "Mediterranean genes" (FDR < 0.05) using the program HOMER 100 and equivalent regions 505 around the transcription start sites of all genes expressed in these data as the background set for 506 enrichment testing. We searched for known vertebrate transcription factor binding motifs and 507 report the TF motifs passing a threshold of FDR < 0.05. 508 509

Gene-gene co-expression analysis 510
In addition to testing whether diet led to mean differences in gene expression between Western 511 and Mediterranean animals, we also tested whether diet impacted the correlation structure among 512 expressed genes (i.e., gene co-expression). Specifically, we used 'correlation by individual level 513 product' (CILP) 47 , to test whether diet affected the magnitude or direction of pairwise gene 514 expression correlations among the top 140 most differentially expressed genes (n = 9730 gene-515 gene pairs tested, equivalent to 140C2). To test whether a given pair of genes was differentially 516 co-expressed as a function of diet, we first obtained a vector of products for each gene pair by 517 multiplying the normalized gene expression values for two genes together. Normalization was 518 performed by scaling expression values to mean 0 and unit variance within Mediterranean and 519 Western subsets of the data respectively, to ensure that distributional differences between sample 520 groups did not bias our results, following previously described procedures 47 . Each of these 521 vectors of products were used as the outcome variable in a linear mixed effects model 522 implemented in the R package EMMREML 96 , which included a fixed effect of diet and a random 523 effect to control for genetic relatedness. To assess significance, we extracted the p-value 524 associated with the diet effect for all 9730 gene pairs. We then repeated each linear mixed effects 525 model 100 times after permuting diet, extracted the p-value associated with the diet effect, and 526 used these values to calculate an empirical FDR distribution 30 . 527 528 Using this approach, we identified 445 gene pairs that were significantly differentially co-529 expressed as a function of diet at a 20% empirical FDR. Next, we performed two follow up 530 analyses to understand their biological import. First, we tested for the existence of 'hub genes', 531 defined as genes that displayed differential co-expression to their tested partner genes more so 532 than expected by chance. To define the null distribution for identifying hub genes, we randomly 533 sampled 445 gene pairs from the set of all 9730 tested gene pairs 1000 times and calculated the 534 number of partners a focal gene had in each sample; we considered a gene to be a significant 535 'hub gene' if it fell outside the 95 th percentile of this distribution, which was equivalent to a focal 536 gene that displayed significant differential co-expression with 13 or more of its tested partner 537 genes. Second, we asked whether the set of 'hub genes' we identified were enriched for 538 transcription factors, relative to the background set of all 140 genes tested for differential co-539 expression. We performed this analysis because many of the proposed mechanisms to generate 540 large scale changes in gene co-expression patterns involve changes in transcription factor 541 function or activity 45,46 . To implement the enrichment analysis, we used the TRRUST database 542 of known mammalian transcription factors for annotation 101 paired with hypergeometric tests. 543 544

Mediation 545
To explore relationships between DAB score and differential gene expression, we conducted 546 mediation analyses using a bootstrapping approach involving 10,000 bootstrap iterations of two 547 models: (Model 1) the expression of each gene as a function of diet, and (Model 2) the 548 expression of each gene as a function of diet and DAB score 102 . For each bootstrap iteration, we 549 then calculated the mediation effect (i.e., the indirect effect) of DAB score as the difference 550 between the effect size of diet in Model 1 (βdiet) and Model 2 (β'diet). We considered there to be a 551 mediation effect when the 90% confidence interval for the indirect effect (βdiet-β'diet) did not 552 include zero.       Note S1. Regarding rank and RNA integrity (RIN). 588