Reprogramming the Human Gut Microbiome Reduces Dietary Energy Harvest

The gut microbiome is emerging as a key modulator of host energy balance1. We conducted a quantitative bioenergetics study aimed at understanding microbial and host factors contributing to energy balance. We used a Microbiome Enhancer Diet (MBD) to reprogram the gut microbiome by delivering more dietary substrates to the colon and randomized healthy participants into a within-subject crossover study with a Western Diet (WD) as a comparator. In a metabolic ward where the environment was strictly controlled, we measured energy intake, energy expenditure, and energy output (fecal, urinary, and methane)2. The primary endpoint was the within-participant difference in host metabolizable energy between experimental conditions. The MBD led to an additional 116 ± 56 kcals lost in feces daily and thus, lower metabolizable energy for the host by channeling more energy to the colon and microbes. The MBD drove significant shifts in microbial biomass, community structure, and fermentation, with parallel alterations to the host enteroendocrine system and without altering appetite or energy expenditure. Host metabolizable energy on the MBD had quantitatively significant interindividual variability, which was associated with differences in the composition of the gut microbiota experimentally and colonic transit time and short-chain fatty acid absorption in silico. Our results provide key insights into how a diet designed to optimize the gut microbiome lowers host metabolizable energy in healthy humans.

paradigm of quantitative bioenergetics (NCT02939703) 2 (Extended Data Fig. 1a-b). The intervention included a highly digestible control Western Diet (WD) and a Microbiome Enhancer Diet (MBD). The MBD was 49 designed to maximize the availability of dietary substrates to the gut microbiome and included these four 50 drivers: dietary fiber, resistant starch, large food particle size, and limited quantities of processed foods 51 (Extended Figure 1a). Our design provided equivalent metabolizable energy and total macronutrients (fat, 52 protein, carbohydrates) based on classic principles and equations of food digestibility 13 . Diets were prepared in To avoid the confounding effects of energy imbalance on host and microbial metabolism, the diet intervention 57 maintained each participant in energy balance. Energy balance, evaluated by real-time energy intake and 58 energy expenditure (measured via whole-room indirect calorimetry), was maintained within our target of +/-50 59 kcals per 6-day calorimeter stay (WD 4.1 ± 5.1 kcal/day; MBD 5.4 ± 2.8 kcal/day; p = 0.8) (Extended Data Fig.   60 2a). Weight stability was a secondary criterion for evaluating energy balance, and we previously reported that 61 weight was stable during the 6-day calorimetry assessment period whilst the primary endpoint was measured; 62 the study team members were blinded to the diet assignment 2 .

64
Surveillance of adverse events revealed minimal gastrointestinal or other side effects (Extended Data Table 1). 1b; P < 0.0001), which equates to an additional 116 ± 56 kcals daily channeled to feces ( Fig. 1c; P < 0.0001). 98 These data align with the preclinical literature showing that the quantitative impact of the gut microbiome on 99 host energy balance is primarily via its critical roles on energy harvest from the diet 8,9 . 100 101 Diet reprogramed the gut microbiome 102 Given our primary finding that diet produced a clinically significant change in host metabolizable energy, we 103 next evaluated the microbial phenotype associated with host energy balance. Mean daily fecal weight was 104 higher on the MBD (P < 0.0001; Extended Data Fig. 3a), and a proportion of this additional weight was due to a 105 significant increase in 16S rRNA genes (P < 0.0001; Fig. 2a), an indication of fecal bacterial biomass increase 106 since the MBD produced 19.6 ± 3.5 gCOD/d of microbial biomass compared to 9.4 ± 1.2 gCOD/d on the WD.  To further explore the compositional changes in the microbiome associated with diet-induced changes in host 118 metabolizable energy, we used metagenomic sequences to evaluate microbial taxonomic differences and 119 derived regression coefficients describing each microbe's association with diet using Maaslin2's compound metabolizable energy intake based strictly on existing food digestibility paradigms. These paradigms do not 148 account specifically for the microbial biomass or microbial energy harvest 13 .

150
One of the gaps in prior human studies was the lack of a precise quantitation of the entire energy balance 151 equation. In addition to our evaluation of energy intake (Extended Data Table 2) and fecal energy loss to derive 152 host metabolizable energy (Fig. 1 a-c), we measured energy expenditure with whole room indirect calorimetry 153 over 6 days and found no diet difference in sleep metabolic rate (in kcal/day) by diet (P = 0.15; Fig. 3d), despite 154 being able to detect a posteriori a 26.5 kcal/day difference 2 . This suggests that, under conditions of fixed energy 155 intake, the main quantitative contribution of the gut microbiome to host energy balance was through its effect 156 on energy harvested from the diet, particularly when sufficient substrates were available for fermentation, as  The relationships among diet composition, gut microbes, and colonic transit time (CTT) are complex, multi-160 directional, and vary within individuals over time and between individuals 27 . Given the potential importance of 161 CTT on the microbiota-driven host response to dietary manipulations, we evaluated whole-gut transit using a 162 pH-sensing radiotransmitter device. We did not find a statistically significant difference in CTT by diet (39.2 ± 163 6.2 hours on WD vs. 29.7 ± 4.4 hours on MBD; P = 0.14; Fig. 3e). Gastric emptying evaluated by 164 acetaminophen appearance in the blood after a fixed liquid meal also was not different by diet (Extended Data 165 Fig. 4a). The pH of the colon can be an indicator of microbial fermentation activity. Neither median pH (which 166 reflects both fermentation and the impact of food mixing in the colon) nor the median pH within a 1-hour 167 window of the ileocecal passage (which is impacted primarily by microbial fermentation products) 28 differed by 168 diet (P = 0.11 and 0.23, respectively; Fig. 3f; Extended Data Fig 4b). The lack of statistically significant effects 169 likely was due to the substantial amount of interindividual variability in CTT, gastric emptying and colonic pH, 170 confirming the complex and individualized relationships among these parameters, which may be critical to 171 understanding the host-microbiota axis within individuals 27 . 172 We hypothesized that the MBD might decrease appetite relative to the WD via the inclusion of high-fiber foods 174 and production of metabolites through gut microbial fermentation 29 . This hypothesis was rejected (Extended 175 Data Fig. 4c-h). Thus, the observed negative energy balance and small changes in body composition on the 176 MBD did not trigger a compensatory change in appetitive behaviors or food intake compared to the WD.

178
The mammalian gut senses nutrients and microbial fermentation products and is part of the larger  Fig 3h), with a significantly higher AUC at breakfast and lunch and a trend towards a higher 192 AUC at dinner (P = 0.02, 0.04 and 0.08, respectively) on the MBD compared with the WD. Pancreatic 193 Polypeptide (PP) iAUC was significantly increased on the MBD (Fig 3i). GLP-1 and PP decrease food 194 intake 33 . Therefore, the short-term negative energy balance within our experimental paradigm did not trigger the  199 Given the robust response to our diet intervention by the gut microbiome and host, we sought to determine the 200 quantitative role of the gut microbiome on energy harvest from the diet versus the impact driven solely by food 201 digestibility 11 . We tested the hypothesis that methane production by methanogenic archaea contributes to a net 202 negative energy balance. We developed and validated a first-in-human method to quantify 24-hour methane 203 production in a whole room calorimeter at part-per-billion resolution 34 . The range of methane measured within 204 our study was 0.28-1613 ml/day, translating to 0.002-14 kcals lost per day. While this negative energy balance  This led us to hypothesize that the variability in host energy balance could be associated with the repertoire of 218 gut microbes in the colon. To test this hypothesis, we asked whether the quantitatively important variability in 219 host metabolizable energy on the MBD could be related to a unique microbial signature. To identify those 220 microbial signatures, we derived regression coefficients describing each microbe's association with the 221 independent variable of host metabolizable energy using Maaslin2's compound Poisson regression model 18 . In 222 total, host metabolizable energy was associated with 16 species (Extended Data Fig. 5a-b). The significant 223 microbes with the largest effect size (Q < 0.05; effect size ≥ 2) were Clostridium bolteae, Streptococcus 224 parasanguinis, Streptococcus australis, and Erysipelatoclostridium ramosum. All were inversely associated 225 with host metabolizable energy, indicating that reduced energy availability to the host may increase substrate 226 availability for the growth of these specific microbes (Fig. 4a). absorbed by the host due to microbial fermentation in the colon and the associated biomass. We applied this 233 model to predict the host metabolizable energy we measured in our study by inputting actual energy intake 234 components and fecal energy in grams COD/day. Our previously published model used a fixed CTT of 48 235 hours, which is a reasonable population-level estimate for healthy adults 37 . With a fixed CTT, the modeled host 236 metabolizable energy for participants on the WD was 95.2 ± 0.001% and for MBD was 92.4 ± 0.001% (Fig.   237 4b). This is similar to the mean host metabolizable energy we measured on the WD and the MBD (95.4 ± 238 0.21% and 89.5 ± 0.73%, respectively; Fig. 1b). However, the variability we saw experimentally on the MBD 239 was not reproduced by the mathematical model. We hypothesized that we could improve the model's predictive 240 ability by adding measured CTT since it is a key modulator of microbial composition, fermentation, and host 241 energy balance 27 . When we included measured CTT, the modeled range of metabolizable energy on the MBD 242 was 84.6-92.9%, which was very similar to the measured range of 84.2-96.1%; furthermore, systematic and 243 proportional bias was minimized (Extended Data Fig. 5c-d). Thus, using the CTT explained some of the 244 variability in host metabolizable energy.

Microbes contributed to energy balance
A significant proportion of the reduced metabolizable energy on high-fiber diets is due to colonic microbial 247 fermentation of fiber and resistant starch into absorbable SCFA 38 . Our model predicted that more total energy 248 (g COD) as SCFAs was absorbed by the host on the MBD, compared to the WD (72.3 ± 13 gCOD/d on the 249 MBD vs. 36.4 ± 4.3 gCOD/d of microbially-derived SCFAs; P < 0.00001; Fig. 4d). When we adjusted the 250 SCFA absorption for energy intake, we found a nearly 2-fold greater absorption of energy as SCFAs on the 251 MBD as compared to the WD (P < 0.00001; Fig. 4e). Therefore, despite less total energy being absorbed by the 252 host on the MBD, a larger proportion was derived from SCFAs. Consistent with our experimental data, our 253 model strongly supports a significant microbial contribution to host metabolizable energy and, therefore, the 254 overall energy balance. The reduction in energy harvest from the diet on the MBD relative to the WD was not accompanied by a 274 reduction in energy expenditure or an increase in hunger or ad libitum energy intake. However, the significant focus on whole, minimally processed foods resets the integrated sensing mechanisms known to affect food 280 intake and body energy stores. One or more of these mechanisms or other unknown mechanisms might be 281 responsible for the population associations between a diverse human gut microbiome and lower body mass 1 .

282
The slightly greater reduction in weight and body fat on the MBD, compared to the WD, over the inpatient 283 period despite daily titration of energy requirements to match calorimetry-derived measures of energy 284 expenditure, suggests that the use of a diet that adequately feeds colonic microbes and increases microbial 285 fermentation products (i.e., short-chain fatty acids) will not lead to additional absolute energy availability to the 286 host. In contrast, diets such as the MBD promote additional fecal energy loss and an increase in host uptake of 287 SCFAs from the colon, despite the overall decrease in host uptake of energy. Future microbiome-focused 288 research should delve into these systems for controlling body weight.

290
The quantitative contributions of gut microbes to host energy balance were addressed in two forms. First, the 291 energy in feces increased by 40.9 ± 4.6 g COD/d (116 ± 56 kcals kcal/day) for participants on the MBD, even 292 though their total metabolizable energy intake was the same. Second, the microbial community increased in 293 size (biomass) and fermentation processes that were reflected by increased fecal and serum SCFAs on the MBD 294 as compared to the WD. Thus, the host's energy intake shifted towards microbially produced SCFAs and away from proximally digested and absorbed carbohydrates in the food. While the quantitative contribution of 296 microbially generated SCFAs was overshadowed by the additional loss of microbial biomass in the feces, the 297 uptake of more microbially produced SCFAs was associated with increased GLP-1 and pancreatic polypeptide 298 concentrations.

300
We also found a taxonomic signature that was in alignment with the expected impacts of the substrates 301 available to the gut microbes on the two diets. First, many of the species detected at higher abundance on the 302 MBD were fiber degraders and/or butyrate producers. Second, our data reveal that, when the gut microbiome Host metabolizable energy was highly variable on the MBD. Given our tight control of energy intake and 311 energy expenditure, this suggests that the microbial contribution to this variability was greater in some hosts 312 than others. Indeed, with a proportionally equivalent input of substrates for microbes, fecal energy losses varied 313 over an ~6-fold range. Understanding the mechanisms by which the microbial communities in the human colon 314 modulate energy harvest and their interaction with host factors such as CTT will provide valuable quantitative 315 data to drive personalized strategies to optimize host-microbiota-diet interactions and prevent or treat obesity.

317
Host metabolizable energy was associated with a unique microbial profile on the MBD, with 4 microbial 318 species whose relative abundance increased in association with decreasing host metabolizable energy. One of 319 those species, Streptococcus australis, transiently increases after weight loss due to bariatric surgery as 320 compared to normal weight controls 44 . Hungatella hathewayi and Erysipelatoclostridium ramosum were more abundant in germ-free mice colonized with feces from a human that underwent caloric restriction with a 322 concomitant phenotype characterized by lower adiposity 45 . Clostridium bolteae, in addition to being a lactic-323 acid producing bacterium 46 , has recently been reported to bind phenylalanine, tyrosine, or leucine amino acids 324 to microbially deconjugated bile acids. While the clinical effects of these microbially transformed bile acids are 325 unclear, bile acids are known to play an important role in microbial energy extraction 47 . Overall, these findings 326 make it plausible that the variability in host metabolizable energy on the MBD is related to a specific microbial 327 signature and the metabolic processes driven by the relationships between host and microbes.

329
We also investigated, in silico, the factors that might be contributing to hose metabolizable energy variability 330 and found that colonic transit time was an important driver. Host metabolizable energy prediction with 331 measured CTT more closely captures the variability seen in measured host metabolizable energy. Our 332 mathematical model, which generated outputs consistent with the clinical data describing the metabolizable 333 energy of participants consuming WD and MBD, allowed us to determine the important role of CTT and to 334 quantify that a host on the MBD produced feces containing 19.6 ± 3.5 gCOD/day of microbial biomass (about 335 10 gCOD/day more than WD) and led to 36.4 ± 4.3 gCOD/day more uptake of microbial derived SCFAs. We 336 believe these factors, and others that may be revealed in future studies, could capitalize on the adaptability of 337 the gut microbiome as a target for personalized medicine 48 .

339
Given the size and scope of the global obesity epidemic and its continued increase, new solutions are needed.

340
The scientific community has recently reoriented itself towards population interventions that promote small 341 changes in energy intake and expenditure as a means of preventing weight gain 49 . This study demonstrates the 342 potential to enact this "small changes" principle through the consumption of a simple whole food intervention     high-range digestion vials followed by a colorimetric assay (HACH, Loveland, CO; Product # 2125925). To 536 ensure that fecal energy was accurately reflective of 24-hour fecal production, we utilized the non-absorbable, 537 non-digestible fecal marker polyethylene glycol (PEG). Participants consumed 1.5g/day of PEG of molecular 538 weight 3350 g/mol (PEG3350). The PEG3350 was procured by a compounding pharmacy that prepared 0.5g capsules (percent error = 2.8%) (Pharmacy Specialists, Altamonte Springs, FL). The details of the PEG assay 540 are below. Fecal energy was measured in 6-day composites of feces collected in our calorimeters. We 541 normalized fecal energy produced to the weight of all feces produced in those 6-days and then to PEG recovery.

542
Fecal energy loss was converted to host metabolizable energy by calculating the percentage of energy that was 543 lost in feces (in g COD) relative to total energy intake (in g COD). The conversion from energy in COD to kcals 544 lost in feces per day (non-metabolizable kcals) was calculated by multiplying total EI in kcals by the percent 545 host metabolizable energy.

547
Polyethylene glycol assay. We utilized a method that is slightly modified from the initial published method by

555
The assay is linear as evidenced by the R 2 of the calibration curve (0.9987). The linear range of the assay was 556 from 0.1 uM to 20 uM with PEG3350 recovery ranging from 96.2-104.5%. The relative standard deviation of 557 the assay was 1.8%. There was no co-elution of analyte with expected excipients or related compounds in 558 chromatograms demonstrating the assay is specific for PEG3350.

563
Calibration curves using 7 data points were generated on each run using plasmids with 16S rRNA genes, and 564 adding a plasmid concentration to achieve copy numbers in the range from 10 1 to 10 9 per reaction. Reaction were trimmed using TrimGalore 11 . DNA sequences were aligned to Hg38 using bowtie2 12 and RNA sequences 588 = � � �� * 6.022 * 10 23 ℎ ( ) * 10 9 * 660 were aligned to Hg38 using STAR 13 . DNA and RNA sequences were then analyzed for taxonomic composition 589 with MetaPhlAn3 14 , using standard parameters.

591
Species Alpha-and Beta-Diversity. All calculations and analyses were conducted in R 15 . Taxonomic 592 composition output from MetaPhlAn3 was processed for beta-diversity analysis using the "phyloseq" R 593 package 16 . A rarefaction curve was created using the "vegan" R package 17 to determine the optimal count-depth 594 for rarefaction. Once the optimal count-depth was determined, rarefaction was performed using phyloseq.

595
Alpha-diversity metrics were calculated using the "microbiome" R package 18 . After samples were rarified, each 596 sample had 3,578,445 sequences. Bray-Curtis and Jaccard distance matrices were calculated on the rarefied 597 count data using vegan. The distance matrices were tested for significance by PERMANOVA using vegan. Differential Abundance. Differential abundance testing by diet and host metabolizable energy was carried out 605 using the output of MetaPhlAn3 in the "MaAsLin2" R package 21 . Taxonomic counts were filtered with a 25% 606 prevalence cut-off. Compound Poisson multivariate linear models were used to account for zero-inflated data 21 .

607
In the diet analysis the dependent variable was microbial abundance, the fixed variables were diet, period, and 608 period sequence, and participant ID was a random factor. In the host metabolizable energy analysis, the 609 dependent variable was microbial abundance, and the fixed independent variable was host metabolizable 610 energy. Appetite. Subjective ratings of appetite were determined using visual analog scales (VAS) administered at -30,

619
-15, +30, +60, +120, and +180 min pre/post each meal. Breakfast was fixed at 500 kcals and lunch and dinner Acknowledgements: We thank our study participants, without whom this work would not have been possible.   Figure   941 shows all significant associations with Q < 0.05 and effect size ≤ 2.    inpatient days where all 3 meals were consumed on-site, and no changes were made to the feeding for testing.

022
All data reported as mean ± s.e.m. N=17 per diet for both panels.