Short-chain fatty acid production by gut microbiota from children with obesity is linked to bacterial community composition and prebiotic choice

Pediatric obesity remains a public health burden and continues to increase in prevalence. The gut microbiota plays a causal role in obesity and is a promising therapeutic target. Specifically, the microbial production of short-chain fatty acids (SCFA) from the fermentation of otherwise indigestible dietary carbohydrates may protect against pediatric obesity and metabolic syndrome. Still, it has not been demonstrated that therapies involving microbiota-targeting carbohydrates, known as prebiotics, will enhance gut bacterial SCFA production in children and adolescents with obesity (age 10-18). Here, we used an in vitro system to examine the SCFA production by fecal microbiota from 17 children with obesity when exposed to five different commercially available over-the-counter (OTC) prebiotic supplements. We found microbiota from all 17 patients actively metabolized most prebiotics. Still, supplements varied in their acidogenic potential. Significant inter-donor variation also existed in SCFA production, which 16S rRNA sequencing supported as being associated with differences in the host microbiota composition. Last, we found that neither fecal SCFA concentration, microbiota SCFA production capacity, nor markers of obesity positively correlated with one another. Together, these in vitro findings suggest the hypothesis that OTC prebiotic supplements may be unequal in their ability to stimulate SCFA production in children and adolescents with obesity, and that the most acidogenic prebiotic may differ across individuals. IMPORTANCE Pediatric obesity remains a major public health problem in the US, where 17% of children and adolescents are obese, and rates of pediatric ‘severe obesity’ are increasing. Children and adolescents with obesity face higher health risks, and non-invasive therapies for pediatric obesity often have limited success. The human gut microbiome has been implicated in adult obesity, and microbiota-directed therapies can aid weight loss in adults with obesity. However, less is known about the microbiome in pediatric obesity, and microbiota-directed therapies are understudied in children and adolescents. Our research has two important findings: 1) dietary prebiotics (fiber) cause the microbiota from adolescents with obesity to produce more SCFA, and 2) the effectiveness of each prebiotic is donor-dependent. Together, these findings suggest that prebiotic supplements could help children and adolescents with obesity, but that these therapies may not be one-size-fits-all.


Introduction 57
Approximately 17% of children in the United States have obesity, and the 58 prevalence continues to increase among all ages and populations (1). The prevalence of 59 pediatric obesity is even higher in Hispanic and African American populations in the 60 United States, where rates of severe obesity continue to increase (1). Children with 61 obesity have an increased risk of adverse health events and incur higher healthcare costs 62 (2-4). Despite the severity of the pediatric obesity epidemic, current common treatment 63 strategies centered around lifestyle changes, including behavioral, dietary, and exercise 64 interventions, often fail or have limited success (5). The high prevalence of pediatric 65 obesity, coupled with the low success rate of common interventions, highlights the need 66 for more efficacious, safe strategies to lower BMI in children and adolescents. 67 The human gut microbiome has emerged as a promising therapeutic target in 68 pediatric obesity. Over the past decade, differences in gut microbial community 69 composition and metabolic activity between obese and lean individuals have been 70 observed (6-8). Causal links have also been established; fecal transplantation can 71 transfer the obesity phenotype from obese donors to lean recipients and recapitulate 72 some key metabolic changes in human obesity (9). Multiple mechanisms for this link have 73 been proposed, including increased energy harvest by obese microbiota (10), activation 74 of enteroendocrine signaling pathways by SCFA (11-13), modulation of glucose and 75 energy homeostasis through bile acid signaling (14), and increased local and systemic 76 inflammation caused by a variety of microbial metabolites (15).
Recent attention in obesity research has been specifically drawn to the role of 78 microbially-derived short-chain fatty acids (SCFA). SCFAs, primarily acetate, propionate, 79 and butyrate, are produced by enteric microbes as end products of anaerobic 80 fermentation of undigested, microbially-accessible dietary carbohydrates, and serve a 81 variety of important roles in the gut. Of particular interest is the SCFA butyrate, which 82 serves as the primary nutrient source for colonocytes (16) and functions as a histone 83 deacetylase inhibitor (17,18). Through its inhibition of NF-kB signaling in colonocytes, 84 butyrate contributes to barrier integrity maintenance and reduces levels of intestinal 85 inflammation markers (19)(20)(21)(22). Acetate, propionate, and butyrate also each activate G-86 protein coupled receptors (GPR) that modulate key metabolic hormones including peptide 87 YY (PYY) and 23). Consistent with these mechanistic findings, mouse studies 88 have shown that supplementation with acetate, propionate, butyrate, or some mixture of 89 these can protect against weight gain, improve insulin sensitivity, and reduce obesity-90 associated inflammation (24)(25)(26)(27)(28)(29). Given the experimental evidence for SCFA 91 supplementation having an anti-obesogenic effect in a murine system, maintaining high 92 levels of SCFA during a weight loss treatment may improve results (27). 93 If increasing SCFA levels is a potential approach to promote weight loss in 94 children, prebiotic supplementation may provide an effective and low-risk adjunctive 95 therapy. Prebiotics are dietary carbohydrates that are indigestible by human-produced 96 enzymes and thus survive transit to the lower GI tract. Once in the colon, prebiotics serve 97 as carbon sources for bacterial fermentation, which in turn yield SCFAs as metabolic end 98 products (30,31). Multiple types of prebiotics (e.g. fructooligosaccharides (FOS), and inulin-type fructans) have been tested in children with obesity ranging from ages 7-18. In 100 select cases, these treatments have been associated with smaller increases in BMI and 101 fat mass (32), and reductions in body weight z-scores, body fat, and trunk fat (33). Still, 102 other prebiotic trials in overweight children have reported no significant beneficial effects 103 (34). 104 Interpreting the mixed outcomes of prior prebiotic clinical trials in pediatric obesity 105 though is complicated by several challenges. First, in vivo studies in pediatric obesity to 106 date have each used only one prebiotic supplement due to the logistical constraints of 107 clinical trials (32-34). Trials employing testing only a single type of supplement hinder the 108 ability to generalize conclusions regarding the efficacy of prebiotics and also make it 109 challenging to determine whether some prebiotics are inherently more acidogenic than 110 others. Second, in vivo trials in healthy adults have shown substantial inter-individual 111 variation in the single prebiotic effects on stool SCFA concentration (30,31,35). Variation 112 in the primary and secondary outcomes could be due to differences in microbial SCFA 113 production; or differences in host physiology, such as SCFA absorption potential. Third, 114 while SCFA concentrations have been shown to be altered in children with overweight or 115 obesity (36), changes in fecal SCFA during dietary intervention have not been measured 116 in past in vivo studies in pediatric populations. If prebiotics mediate their effects through 117 SCFA (33, 34, 37), directly tracking SCFAs could help determine treatment success. 118 Fourth, in vivo studies in adults, especially those with obesity, may be confounded by the 119 concurrence of chronic disease and the medications a person may be taking to treat 120 chronic disease. 121 In this study, we have taken an in vitro approach to address the limitations of prior 122 human studies. An in vitro approach facilitates more direct comparisons of different 123 prebiotic supplements: the higher-throughput of in vitro experiments allows wider variety 124 of prebiotics to be tested; and, the effects of these supplements can be tested on identical 125 microbiota samples, rather than over time within subjects, which is confounded by 126 microbiota drift over time (38), as well as inconsistencies in dietary composition. Taking 127 an in vitro approach to studying the effects of prebiotics on gut microbiota allows a more 128 direct investigation of microbial SCFA production, as we can study the effects of prebiotic 129 supplementation independent of the effects of host absorption (39, 40). Using a preclinical 130 in vitro fermentation model, and samples from adolescents with obesity who have not 131 developed long-term complications, we pursued three specific lines of inquiry: 1) whether 132 different types of prebiotics lead to differences in SCFA production by gut microbiota from 133 adolescents with obesity; 2) whether the effects of prebiotics are shaped by inter-134 individual differences in gut microbiota structure; and, 3) whether fecal SCFA production 135 is associated with protection from obesity.

SCFA production capacity 138
To measure SCFA production by gut microbiota, we adapted the in vitro approach 139 of Edwards et al. 1996 (41). This method was specifically designed to study fermentation 140 of starch in the human lower GI tract, and has since been used to measure metabolite 141 production from human stool samples when exposed to prebiotic fiber (42)(43)(44). In brief, 142 we homogenized previously frozen feces in reduced phosphate buffered saline (pH 7.0 143 ± 0.1) to create a fecal slurry with a final concentration of 100g/L ( Figure 1). These fecal 144 slurries were then supplied with each of five prebiotic carbon sources, and a carbon-free 145 control, and allowed to ferment at 37°C in anaerobic conditions for 24 hours, to 146 approximate colonic transit time (45). Following the incubation period, the concentrations 147 of SCFA in the samples were measured by gas chromatography. To control for 148 differences in overall cell viability or stool slurry nutrient content between donors, we 149 corrected measurements of SCFA concentration by dividing the treatment SCFA 150 concentration by the control SCFA concentration. 151 To validate our assay, we ran a series of experiments using feces from validation 152 sample sets. We verified that our control-corrected SCFA production data was not 153 influenced by bacterial abundance (p = 0.38, r = 0.14, Spearman correlation; Figure S1). 154 Absolute (not relativized to control) SCFA concentrations are supplied in the supplement 155 ( Figures S2 and S3). As our fermentation experiments used previously frozen fecal 156 samples, we verified that total SCFA production was strongly correlated between fresh 157 samples and twice freeze-thawed samples (p < 0.0001, r = 0.75, Spearman correlation; Figure S4A). Since we elected to not provide our fermentation reactions with nutrients in 159 excess of what was contained in the fecal slurries, we verified that there existed strong 160 correlation in total SCFA production between PBS-grown and colonic medium-grown 161 cultures (46), both when supplied with dextrin and inulin (Dextrin: p = 0.001, r = 0.68; 162 inulin: p = 0.02, r = 0.51; Spearman correlations; Figure S5). We found that total SCFA 163 production over control was positively correlated with the pH of starting fecal slurries (p = 164 0.003, r = 0.46; Spearman correlation; Figure S6A). A weaker correlation may exist 165 between SCFA production and the final pH of the fermentation vessels (p = 0.067, r = 166 0.29, Spearman correlation; Figure S6B). 167 We subsequently applied our assay to fecal microbiota from a cohort of 17 children 168 ranging in age from 10 -18, Tanner stages 2 -5, and body-mass index (BMI) from 25.9 169 -75.3 (Table 1). We found all 17 individuals demonstrated a net gain of SCFA relative to 170 the control in at least one prebiotic treatment, which led us to conclude that all tested 171 cultures were viable and metabolically active ( Figure 2). 172 173

Donor and prebiotic both impact SCFA production in vitro 174
We next tested the hypothesis that different prebiotics equally promote the 175 production of SCFA by performing statistical analysis of SCFA production as a function 176 of the prebiotic type and individual identity. Our analysis revealed heterogeneity in the 177 efficacy of prebiotic supplements (two-way ANOVA, p < 0.001; Table S1; Figure 2a), 178 ranging from inulin, which resulted in a 2.35 mean fold change in total SCFA, to GOS, 179 which resulted in 3.55 mean fold change in total SCFA. Frequently, only two or three of the five tested prebiotics resulted in increased total SCFA production within an individual. 181 Our statistical testing also revealed consistent patterns between individuals' gut 182 microbiota in terms of SCFA production (two-way ANOVA, p < 0.001; Table S1; Figure  183 2b), with mean fold change in SCFA over control ranging from 2.37 to 6.12. Within 184 individuals, the average fold change in SCFA concentration in the prebiotic treatments 185 often appeared to be driven by a few strongly acidogenic prebiotics. Last, our analysis 186 indicated a significant interaction between prebiotic type and individual identity (two-way 187 ANOVA, p< 0.001; Table S1; Figure  If inter-individual differences in gut microbiota mediated responses to prebiotic 194 treatment, we would expect that specific bacterial taxa, which varied between individuals, 195 could also be associated with SCFA production. To evaluate this hypothesis, we used the 196 R package stray (47) to create a Bayesian multinomial logistic normal linear regression 197 (pibble) model that tested for correlations between in vitro SCFA production in response 198 to each prebiotic and 16S rRNA community composition of patient stool used in the 199 fermentations, at the genus level. This analysis revealed that SCFA production from 200 prebiotics was correlated with the relative abundances of 18 different bacterial genera 201 (95% credible interval not covering 0, Figure 3). Of the 13 genera positively associated 202 with SCFA production, 9 are known or likely fiber degraders (48-52), Akkermansia, is 203 often observed to increase in abundance after prebiotic treatment (53), and one, 204 Methanobrevibacter, an archaeon hydrogenotrophic methanogen, is known to increase 205 the efficiency of carbohydrate metabolism by the microbiota (54) ( Table 3). Most genera 206 identified by stray were associated with SCFA production in a limited set of prebiotic 207 treatments. One genus, Lactobacillus, is positively associated with SCFA production on 208 XOS, but were negatively associated with SCFA production on GOS. Overall, the 209 presence of specific associations between bacterial taxa and different prebiotics supports Finally, we tested the hypothesis that in vitro SCFA production would be 215 associated with obesity-related phenotypes. We compared clinical metadata from 216 individuals, which included BMI, insulin, and HbA1c, with average total SCFA production 217 across prebiotics and found no significant correlations in our population (Spearman 218 correlation; Table 2). Fecal microbial SCFA production capacity may not be directly 219 associated with obesity though because rates of host SCFA uptake likely vary, and this 220 variance may influence host intestinal physiology (55-57). Indeed, in support of the idea 221 that SCFA absorption rate (which was not measured in this study) shape metabolic 222 homeostasis and host health, we observed a negative association between fecal SCFA 223 concentrations and in vitro SCFA production across the range of tested prebiotics ( Figure 4). Furthermore, if SCFA absorption efficiencies varied by individual, residual fecal SCFA 225 concentrations may not directly reflect the complete effect of bacterial metabolism on 226 obesity. Consistent with this notion, no significant relationships were apparent between 227 concentrations of SCFA in patient stool and clinical markers of obesity measured at 228 enrollment, including BMI, insulin levels, and HbA1c (Table 2), although this may also be 229 explained by uncontrolled patient parameters. 230

Discussion 231
In this study we found that the microbiota of all tested adolescents with obesity 232 increased total SCFA production when exposed in vitro to at least one prebiotic. Both 233 donor and prebiotic were significant factors in determining SCFA production in vitro, as 234 was their interaction. Our modeling revealed distinct associations between specific 235 microbial taxa and SCFA production on different prebiotics. We interpret this result as 236 suggesting that the associated bacteria play a role in the fiber fermenting capacity of the 237 community. We observed no correlations between either stool SCFA concentrations or in 238 vitro acidogenic capacity of communities and any metrics of obesity (Table 2). 239 We have recapitulated previous findings that both donor and prebiotic are 240 important in determining the SCFA production from in vitro prebiotic supplementation (31, 241 50, 58), and we found that not all prebiotics appear equally acidogenic (50). Since our in 242 vitro system removes the host as a potential source of variation, our data support a gut 243 microbial role for inter-donor variation in fecal SCFA production. In addition, the strength 244 of the interaction between donor and prebiotic strongly suggests that prebiotics are not 245 one-size-fits-all; rather, inconsistent results from prior studies of prebiotics in pediatric 246 obesity (32, 34, 59) may be due to variation in the SCFA production capacity of 247 individuals' gut microbiota across the tested prebiotics. Future therapeutic efforts 248 involving prebiotics in patients with obesity may benefit from stratified or personalized 249 treatments. Nutritional therapies that are personalized to individuals' microbiota are 250 already in development (60). 251 by acetate, propionate, and butyrate, increases satiety and insulin sensitivity, while 253 decreasing adipogenesis (12, 23, 61), yet, we did not observe associations between fecal 254 SCFA levels and metrics of obesity. The effects of SCFA on obesity may be masked by 255 uncontrolled patient factors, such as differences in caloric intake and variation in 256 individual nutrient harvest and utilization. In order to observe the effects of SCFA on 257 obesity, it would be necessary to control for these variable physiological and lifestyle 258 parameters, which we did not attempt. These patient factors may also have influenced 259 our inability to observe an association between acidogenic capacity of microbiota and 260 fecal SCFA concentrations. However, this may also be explained by the potential 261 uncoupling of fecal SCFA production and fecal SCFA concentration. In vitro, increased 262 luminal concentrations of butyrate have been shown to upregulate the sodium-coupled 263 monocarboxylae transporter SLC5A8 (55), and addition of physiological mixtures of 264 SCFA has been shown to upregulate the monocarboxylate transporter SLC16A1 (62), 265 both of which uptake acetate, propionate, and butyrate from the lumen. Since gut epithelia 266 have the capacity to absorb up to 95% of SCFA before excretion (63), increased host 267 SCFA uptake (triggered by increased gut bacterial production) could, therefore, lead to 268 constant or even decreased fecal SCFA concentrations. This complex relationship could 269 explain the absence of positive correlations we observed between stool SCFA levels and 270 the acidogenic capacity of gut microbiota. It may be necessary to delve further upstream 271 of fecal SCFA concentration by measuring proxies for host SCFA uptakes, such as the expression of SCFA transporters (SLC5A8 and SLC16A1) and SCFA receptors (GPR43, 273 GPR41, and GPR109A) (55). 274 The primary limitations of this study involve constraints common to in vitro culture 275 studies. First, many factors affecting bacterial SCFA production in vivo are difficult to 276 replicate in vitro, including the availability of nutrients such as nitrogen, the starting 277 concentration of SCFA, the redox state of the environment, and the efficiency of cross-278 feeding interactions (64, 65). Different metabolic results between prebiotics may have 279 occurred if we provided alternative co-metabolites or nutrients, in addition to the tested 280 prebiotics. We chose our culture conditions, namely a media-free approach that does not 281 add any nutrients beyond what is present in the stool, in an effort to avoid inducing artificial 282 selective conditions within our cultures. Prior experimental digestion studies have shown 283 that prebiotic response patterns can be recapitulated across varying culture conditions 284 (42, 44). Indeed, we found strong correlation in SCFA production between cultures grown 285 with our media-free approach and those grown in a more conventional medium containing 286 added nitrogen, vitamins, minerals, and acetate. Further, this approach allowed us to 287 minimize the influence of the host on measurements of microbiota production of SCFA. 288 We did observe shifts in community composition during the 24 hour fermentations (Figure  289 S7); however, we remained able to find statistical associations between SCFA production 290 capacity and pre-fermentation community composition. A second set of limitations in this 291 study involves our reliance on patient collection of stool. Inter-donor variation in prebiotic 292 response could have originated in technical variation between how patients exposed stool 293 to aerobic conditions (66) or how they froze their samples (67), which in turn could have affected the fraction of viable microbial cells in stool samples. Still, we found a significant 295 correlation between in vitro total SCFA production from fresh stool and stool that had 296 been frozen and thawed twice. Variation in donor prebiotic response could also have 297 biological origins due to physiological differences between people (e.g. efficiency of food 298 digestion, consistency of stool (68)) or differences in diet, which can lead to variation in 299 stool microbial load and nutrient content (69). Rather than control for a myriad of different 300 sources of variation whose origins we did not measure, we chose the straightforward 301 approach of standardizing donor samples by employing a consistent concentration of 302 stool slurry (5% w/v stool in PBS) in our experiments. 303 Future work to address these limitations could test multiple stool samples per 304 subject to confirm whether the observed variation in prebiotic response is durable 305 between individuals over time. Future studies could also examine the correlation between 306 the metabolic effects of prebiotic supplementation in vitro and in vivo using randomized 307 human trials that couple human prebiotic supplementation, in vivo measurement of SCFA 308 production, and in vitro tests of microbiota metabolic activity. It would also be useful for 309 such studies to explore the impact of prebiotic supplementation on host physiology, both impacts of prebiotic supplementation, as well as explain why fecal SCFA concentrations 314 may not mirror the metabolic capacity of gut microbiota.

Cohort 317
Stool was collected from human donors under a protocol approved by the Duke 318 Health Institutional Review Board (Duke Health IRB Pro00074547) for a prospective 319 longitudinal cohort study and biorepository. Participants whose samples were used in 320 this study were treatment-seeking adolescents with obesity who were newly enrolled in 321 a multi-disciplinary weight management program. All subjects received family-based 322 intensive lifestyle modification. Based on clinical necessity, some participants also were 323 placed on a low-carbohydrate diet, medications to facilitate weight loss, or underwent 324 weight loss surgery (Table S2). Due to the low number of patients assigned to each 325 treatment arm, we did not attempt to base any analyses on patient treatment plan. 326 Patients were aged 10-18, with BMI ≥ 95 th percentile. None had prior antibiotic use in 327 the 1 month prior to enrolment, used medications known to interfere with the intestinal 328 microbiome, and did not have other significant medical problems. Stool samples used in 329 this study were from enrollment, 3-month, 4.5-month, and 6-month follow-up visits 330 (Table S2). The clinical metadata used for correlations was collected at enrollment, 3 331 months, and 6 months. The metadata collected nearest to the stool sample collection 332 date was used in our analyses. 333 334

Stool Collection 335
Patients collected intact stool samples in the clinic or at home using a plastic stool 336 collection container (Fisher Scientific: 02-544-208) and were asked to immediately store 337 this container in their home freezer. Patients then returned the sample by either bringing 338 it to the study team or scheduling a home pickup within 18 hours of stooling. Stool was 339 transported frozen in an insulated container with an ice pack. Upon receipt in the lab, 340 samples were placed on dry ice until transferred to a -80°C freezer for long term storage. 341 All patient samples were frozen at -80°C within 19 hours of stooling (range, 0.08hr -342 18.83hr; median, 11.42hr) except one which was stored 44.03hr after stooling. The time 343 between stooling and freezing at -80°C did not have a significant effect on average SCFA 344 production (p = 0.58, r = -0.15, Pearson correlation). Stool samples for analysis were 345 processed by removing containers from -80°C storage and thawing on ice in a biological 346 safety cabinet until soft enough to aliquot. Thawed containers of stool were opened to 347 atmosphere for a maximum of 10 minutes while samples were aliquoted. After primary 348 aliquoting, the remaining stool was transferred to an anaerobic chamber (COY Laboratory 349 Products, 5% hydrogen, 5% CO2, 90% Nitrogen) and further portioned into approximately 350 2g aliquots for this study. These aliquots were then stored as solid stool pellets at -80°C 351 until used for this study. PBS. The resulting fermentation conditions where therefore 5% fecal slurry with 0.5% 379 prebiotic (w/v). A 5% fecal slurry was selected because its fermentative capacity has been 380 previously demonstrated to be insensitive to small variations in concentration and is feasible to work with using this method (42). A 0.5% final concentration of prebiotic in the 382 context of a 5% fecal slurry is analogous to an average adult consuming 20g of dietary 383 fiber per day, assuming an average daily stool mass of 200g (73). Fermentation reactions 384 were carried out in an anaerobic chamber at 37°C for 24 hours. Following fermentation, 385 1mL media was taken from each reaction vessel for SCFA quantification. During our 386 validation experiments, a separate 1mL aliquot was taken for pH measurement. 387 388

Simulation of Freeze/Thaws Experienced by Study Samples 389
To test the effects of freeze/thaw cycles on in vitro SCFA production, we collected 390 fresh, whole fecal samples from four healthy adults who were not patients in the study 391 cohort. Informed consent was obtained from volunteers and the protocol was approved 392 by the Duke Health Institutional Review Board. Samples were brought into an anaerobic 393 chamber after voiding. Once in anaerobic conditions, these samples were divided into 394 three aliquots. One aliquot was processed immediately following the same in vitro 395 fermentation protocol used in our study. transferred to -80C storage. After a minimum of 396 24 hours, one of these two aliquots was removed from the freezer and thawed at room 397 temperature for 2 hours, before being returned to -80C for an additional minimum of 24 398 hours. Each of these frozen aliquots was thawed and processed following the same in 399 vitro fermentation protocol. This allowed direct comparison of samples that had been used 400 in fermentations immediately after voiding to those that had been frozen and thawed one 401 and two times. 402

Media Preparation 404
To validate our methods, namely our use of a 5% fecal slurry in PBS, without 405 supplementation of other nutrient components, we compared SCFA production with our 406 methods to SCFA production when stool was instead resuspended in a medium 407 designed to simulate the large intestine. We used a slightly modified medium derived 408 from Gamage et al. 2017 (46). The media contained, per liter: peptone 0.5g, yeast 409 extract 0.5g, NaHCO3 6g, hemin solution (0.5% (w/v) hemin and 0.2% (w/v) NaOH) The SCFA concentration of fecal slurries and fermentation vessels was determined 417 following a protocol adapted from Zhao,Nyman,and Jönsson (74). First, a 1mL aliquot 418 of either 10% fecal slurry in PBS or the fermentation vessel contents was obtained. To 419 this, 50 µL of 6N HCl was added to acidify the solution to a pH below 3. The mixture was 420 vortexed, centrifuged at 14,000rcf for 5 minutes at 4°C to remove particles. Avoiding the 421 pellet, 750 µL of this supernatant was passed through a 0.22µm spin column filter. The 422 resulting filtrate was then transferred to a glass autosampler vial . 423 Filtrates were analyzed on an Agilent 7890b gas chromatograph (GC) equipped 424 with a flame-ionization detector (FID) and an Agilent HP-FFAP free fatty-acid column (25m x .2mm id x .3µm film). A volume of 0.5µL of the filtrate was injected into a sampling 426 port heated to 220°C and equipped with a split injection liner. The column temperature 427 was maintained at 120°C for 1 minute, then ramped to 170°C at a rate of 10°C/min, then 428 maintained at 170°C for 1 minute. The helium carrier gas was run at a constant flow rate 429 of 1mL/min, giving an average velocity of 35 cm/sec. After each sample, we ran a one 430 minute post-run at 220°C and a carrier gas flow rate of 1mL/min to clear any residual

Identifying Sequence Variants and Taxonomy Assignment
We used an analysis pipeline with DADA2 (78) to identify and quantify sequence 447 variants, as previously published by Silverman et al. (79). To prepare data for denoising 448 with DADA2, 16S rRNA primer sequences were trimmed from paired sequencing reads 449 using Trimmomatic v0.36 without quality filtering (80). Barcodes corresponding to reads 450 that were dropped during trimming were removed using a custom python script. Reads 451 were demultiplexed without quality filtering using python scripts provided with Qiime v1.9 452 (81). Bases between positions 10 and 150 were retained for the forward reads and 453 between positions 0 and 140 were retained for the reverse reads. This trimming, as well 454 as minimal quality filtering of the demultiplexed reads was performed using the function 455 fastqPairedFilter provided with the DADA2 R package (v1.8.0). Sequence variants were 456 inferred by DADA2 independently for the forward and reverse reads of each of the two 457 sequencing runs using error profiles learned from all 20 samples. Forward and reverse 458 reads were merged. Bimeras were removed using the function removeBimeraDenovo 459 with default settings. Taxonomy was assigned using the function assignTaxonomy from 460 DADA2, trained using version 123 of the Silva database. 461 462

Modeling Microbial Composition Data 463
To associate microbial genera to SCFA production on different prebiotics, the 464 sequence variant table was amalgamated to the genus level using the R package 465 phyloseq (81). Genera that were observed with at least 3 counts in at least 3 samples 466 were retained. This filtering step retained 99.3% of sequence variant counts and a total 467 of 97 genera. 468 To associate microbial composition to SCFA production on different prebiotics we 469 made use of Bayesian Multinomial Logistic-Normal linear regression implemented in the 470 R package stray as the function pibble (82). We chose this method to account for 471 uncertainty due to counting, and compositional constraints as motivated in Silverman,472 Durand (79)  prior assumption that, on average, the association between each prebiotic and each taxon 480 is zero. 481 We set the hyperparameters Gamma to be the matrix 482 ⎣ ⎢ ⎢ ⎢ ⎡ 5 0 0 0 0 0 2 . 6 . 6 . 6 0 . 6 2 . 6 . 6 0 . 6 . 6 2 . 6 0 . 6 . 6 . 6 2 ⎦ ⎥ ⎥ ⎥ ⎤ 483 which was chosen to reflect the following prior information: (1) the relative scale of 484 33 to 44 (for ∈~2, … , 5}) implies that we have little knowledge regarding 485 the mean composition between individuals but that we conservatively expect that the 486 association between butyrate production and microbial composition is small 487 comparatively.
(2) the value of 0.6 state that, on average across genera, we assume that the effects each prebiotic are correlated with an average correlation of 0.3, (3) in concert 489 with our prior choices for Xi and upsilon (below), the scale of Gamma represents our 490 assumption that the technical noise in our community measurements is smaller (by a 491 factor of ≈ 5 ) than the magnitude of the biological variation between samples. This later 492 prior regarding technical versus biological variation was informed by Silverman,Durand 493 (79). All prior choices were further investigated using prior predictive checks (84). To 494 reflect a weak prior assumption that the absolute abundance of each taxon is uncorrelated    Table 1: Demographic characteristics of participants in this study. One patient 813 provided samples used in all analyses but was lost to follow-up before providing clinical 814 metadata; that patient is only counted in the total column.