Article Caloric restriction disrupts the microbiota and colonization resistance https://doi.org/10.1038/s41586-021-03663-4 Reiner Jumpertz von Schwartzenberg1,2,3,13, Jordan E. Bisanz4,13, Svetlana Lyalina5, Peter Spanogiannopoulos4, Qi Yan Ang4, Jingwei Cai6, Sophia Dickmann1, Marie Friedrich1, Received: 28 November 2017 Su-Yang Liu7, Stephanie L. Collins6, Danielle Ingebrigtsen8, Steve Miller8, Jessie A. Turnbaugh4, Accepted: 21 May 2021 Andrew D. Patterson6, Katherine S. Pollard5,9,10,11,12, Knut Mai1,2, Joachim Spranger1,2,3,14 ✉ & Peter J. Turnbaugh4,14 ✉ Published online: 23 June 2021 Check for updates Diet is a major factor that shapes the gut microbiome1, but the consequences of diet-induced changes in the microbiome for host pathophysiology remain poorly understood. We conducted a randomized human intervention study using a very-low-calorie diet (NCT01105143). Although metabolic health was improved, severe calorie restriction led to a decrease in bacterial abundance and restructuring of the gut microbiome. Transplantation of post-diet microbiota to mice decreased their body weight and adiposity relative to mice that received pre-diet microbiota. Weight loss was associated with impaired nutrient absorption and enrichment in Clostridioides difcile, which was consistent with a decrease in bile acids and was sufcient to replicate metabolic phenotypes in mice in a toxin-dependent manner. These results emphasize the importance of diet–microbiome interactions in modulating host energy balance and the need to understand the role of diet in the interplay between pathogenic and benefcial symbionts. Very-low-calorie diets (VLCDs) involve extreme calorie restriction Data Fig. 1g). Glucose regulation reverted after CONVD despite the (approximately 800 kcal per day, typically in liquid form2). Gut micro- maintenance of lower adiposity; both phenotypes were unchanged biome composition is altered during weight loss induced by VLCDs3,4; in the control group. however, the consequences of this alteration for health and disease remain unclear. To address this knowledge gap, we performed an exploratory analysis in a trial of 80 post-menopausal women who were Diet restructures the gut microbiome overweight or obese and were randomized to either a medically super- Consistent with the hypothesis that sustained caloric restriction during vised weight loss program or to a control group instructed to maintain the VLCD could have pronounced impacts on the gut microbiota—anal- a stable weight for 16 weeks (Extended Data Fig. 1a). The primary out- ogous to the effects of long-term fasting6,7 or parenteral nutrition8—the comes—changes in muscle mass and myocellular insulin sensitivity—are VLCD significantly decreased the level of gut microbial colonization reported elsewhere5. The weight loss program included 8 weeks of measured both by faecal DNA content (Extended Data Fig. 2a) and 16S VLCD (800 kcal per day), 4 weeks of a conventional low-calorie diet rRNA gene copies (Fig. 1b). Next, we performed 16S ribosomal RNA (CONVD), and 4 weeks of weight maintenance (MAINT; Extended Data (rRNA) gene sequencing on 207 samples from 70 individuals (Supple- Fig. 1b). Both groups had similar baseline characteristics (Supplemen- mentary Table 3). The number of denoised 16S rRNA amplicon sequence tary Table 1). The intervention group showed a weight reduction of variants (ASVs) was increased after the VLCD but returned to baseline 13.6 ± 4.0% (mean ± s.d.) relative to baseline, which corresponds to a loss levels during the CONVD (Fig. 1c). The VLCD also induced a marked, but of 12.5 ± 3.9 kg after 12 weeks (Fig. 1a). Total caloric and macronutrient reversible, shift in gut microbial community structure. Quantification intake was lower during the VLCD, CONVD, and MAINT phases than at of community distances from baseline for each individual identified a baseline (Extended Data Fig. 1c, d). The VLCD lowered proportional fat significant perturbation following the VLCD that reverted during sub- intake from baseline, whereas relative macronutrient intake was com- sequent diet phases (Extended Data Fig. 2b, c). Multivariate comparison parable to baseline in the CONVD and MAINT phases (Extended Data also showed that the VLCD induced significant differences from the Fig. 1e, Supplementary Table 2). The VLCD led to decreased adiposity baseline microbiota (P < 0.01; ADONIS with Bray–Curtis dissimilarity). (Extended Data Fig. 1f) and improved glucose regulation (Extended Reproducible shifts were apparent via principal coordinates analysis 1Charité Universitätsmedizin Berlin, Department of Endocrinology and Metabolic Diseases, Berlin, Germany. 2Berlin Institute of Health (BIH), Berlin, Germany. 3DZHK (German Centre for Cardiovascular Research), partner site Berlin, Center for Cardiovascular Research (CCR), Berlin, Germany. 4Department of Microbiology & Immunology, University of California San Francisco, San Francisco, CA, USA. 5Gladstone Institutes, San Francisco, CA, USA. 6Center for Molecular Toxicology and Carcinogenesis, Department of Veterinary and Biomedical Sciences, The Pennsylvania State University, University Park, PA, USA. 7Department of Pathology, University of California San Francisco, San Francisco, CA, USA. 8Department of Laboratory Medicine, University of California San Francisco, San Francisco, CA, USA. 9Department of Epidemiology & Biostatistics, University of California San Francisco, San Francisco, CA, USA. 10Institute for Human Genetics, University of California San Francisco, San Francisco, CA, USA. 11Institute for Computational Health Sciences, University of California San Francisco, San Francisco, CA, USA. 12Chan Zuckerberg Biohub, San Francisco, CA, USA. 13These authors contributed equally: Reiner Jumpertz von Schwartzenberg, Jordan E. Bisanz. 14These authors jointly supervised this work: Joachim Spranger, Peter J. Turnbaugh. ✉e-mail: [email protected]; [email protected] 272 | Nature | Vol 595 | 8 July 2021 a b c 110 VLCD CONVD MAINT Intervention Control Intervention Control P = 0.94 P = 0.15 P = 0.98 P = 0.76 P = 0.32 P = 1.0 P = 0.99 P = 0.94 P = 0.010 P = 0.88 P = 1.0 100 P = 0.99 P = 0.88 P < 0.001 P = 0.96 12 120 90 11 ght (% of baseline) 80 Control 10 [16S rRNA (copies per g)] 80 Intervention Observed ASVs 10 Body wei –4 Pgroup = 5.2 × 10 9 40 –16 log Pgroup:time < 2.2 × 10 0481216 0812 16 01216 0812 16 01216 Time (weeks) Time (weeks) Time (weeks) e VLCD CONVD MAINT GT77 Baseline VLCD (8 weeks) CONVD (12 weeks) MAINT (16 weeks) GT84 GT10 CBM40 GH91 d GH111 Bacterial chemotaxis Pentose phosphate pathway GH104 GH4 Fructose and mannose GT39 CBM51 10 ) TS GH25 (P Galactose GH39 5 s GH94 m te CBM34 s 0 y Ascorbate and aldarate GH68 s Signal GH70 e s transduction CBM36 (fold-change) –5 a 2 r GH11 fe Cell s Pyruvate CBM77 log –10 n CAZyme family GH59 a motility r t CBM72 o Carbohydrates Propanoate CBM2 h p CBM78 s CBM76 o h Photosynthesis CBM23 P Environmental M GH74 e CBM79 information t h CBM65 processing a GH44 n Energy e CBM27 Membrane metabolism transport f Intervention Control Metabolismlog Q = 0.07 10,000 Q = 0.30 Q = 0.41 2 Q = 0.01 1 (fold-change) Q = 0.40 ) Q = 0.06 Q = 0.30 –1 Q = 0.46 Nucleotides Sulfur Q = 0.85 0 Lipids g μ Q = 0.30 –1 Amino Q = 0.07 KO (FDR < 0.1) acids 1,000 Purines Q = 0.08 Q = 0.85 VLCD CONVD Q = 0.11 Q = 0.85 A Xenobiotics B versus versus C baseline VLCD t Cofactors ra n and s Tryptophan po vitamins 100 rte Phenylalanine rs Phe, Tyr and Trp biosynthesis Thiamine Terpenoid backbone biosynthesis Faecal concentration ( Effect of VLCD –log10(FDR) Decreased 1.5 Increased 2.5 e e e e e Acetat Butyrat Valerate Acetat Butyrat Valerate Isobutyrate Propionate Isobutyrat Propionate SCFA Fig. 1 | Very low-calorie diets alter microbiota composition and activity. significantly between VLCD and baseline shows that they reverted in the a, Diet participants lost significantly more weight over time than controls over CONVD phase (Extended Data Fig. 2i). e, Heat map of diet-induced changes to the combined 12-week intervention (0.84% per week (0.68–1.0 95% confidence CAZymes compared to baseline (FDR Q < 0.1, two-sided paired t-test VLCD −16 interval (CI)), P < 2.2 × 10 , LMM; ncontrol = 40, nintervention = 40 participants). b, The versus baseline) demonstrates that changes revert during CONVD and MAINT VLCD decreased overall gut microbial colonization (qPCR-based quantification phases (Supplementary Table 5). f, The SCFAs acetate, butyrate, and valerate of 16S rRNA gene copies per gram wet weight; P < 0.001, LMM; nintervention = 39, are significantly reduced in stool samples during during VLCD (nintervention = 18, ncontrol = 35 participants). c, Richness of 16S rRNA ASVs increased following the ncontrol = 10 participants per time point; FDR Q < 0.1, two-sided Wilcoxon consumption of VLCD (P = 0.010, negative binomial generalized LMM; signed-rank test; downward triangles represent a concentration below −1 nintervention = 37, ncontrol = 32 participants). d, Microbiome functional capacity 10 μg g ). Statistical analysis carried out using linear mixed effects models (measured by shotgun metagenomic sequencing) is altered by VLCD (LMM) with participant as random effect and Tukey all-pair two-sided (Supplementary Table 4; FDR Q < 0.1, Limma and mROAST). The tree comparison unless otherwise noted. In boxplots: centre line, median; box, first demonstrates a functional hierarchy; inner nodes and outer tips represent and third quartiles; whiskers, 1.5× interquartile range (IQR) with outliers pathways and modules, respectively. Inset, a heatmap of KOs that differed individually plotted. (Extended Data Fig. 2d). To identify bacterial markers of VLCD, we fit of the predictive genera showed that the VLCD increased the growth a random forest classifier of baseline versus VLCD in the intervention of bacteria capable of foraging on host glycans (Akkermansia) at the group as a function of genus-summarized abundances (Extended Data expense of bacteria specialized for the metabolism of plant polysac- Fig. 2e). Tenfold cross-validation showed that the best classifier was charides (Roseburia, Ruminococcus, Eubacterium)9,10.