The overall study design, methods, diet descriptions and a comprehensive metabolic characterization has been published previously (Li et al., 2018a,b). All experiments were pre-approved by the Danish Animal Experimental Inspectorate and complied with the “European Convention for the Protection of Vertebrate Animals Used for Experiments and Other Scientific Purposes.” In brief, retired female C57BL/6J (Janvier, France) breeder mice aged 8–9 months were randomized to either chronic chow diet, low-protein/high-carbohydrate diet, high-protein/low-carbohydrate diet, HFD or a periodized low-protein/high-carbohydrate diet regimen, switching between pLPHC diet and HFD every 14 days. The chow [SF14-162, 63/19/18 E% Carbohydrate(c)/protein(p)/fat(f)], LPHC (SF09-048, 74/5/21 E% c/p/f), and HPLC diets (SF09-069, 29/40/31 E% c/p/f), all from Speciality Feeds, Australia are described in Supplementary Table S1 in Li et al. (2018b). The high fat diet provided 20/20/60 E% c/p/f (Cat.: D12492) was from Research Diet Inc.

At the end of the study, the mice were killed following 2–3 h fasting from 9 a.m. to collect tissues and plasma. Approximately 200 mg of cecum content was collected from a cut made on apex ceci and was then snap frozen in liquid nitrogen until future analysis.

Worth noting, our study did not include a pre-intervention control measurements of cecal microbiome to validate the randomization. Previous studies indicate that this is not a concern if properly randomized (Anhe et al., 2015; Le et al., 2018; Montonye et al., 2018). However, since we in principle cannot know if the GM changed due to the intervention, we have worded the descriptions and conclusions throughout to indicate rather than show changes over time.

Approximately 50–120 mg of cecal content was from each sample weighed into Powersoil tubes on a Radwag model AS 220.R2 (Bracka, Poland). DNA was extracted with DNeasy PowerSoil Kit (Cat. No. 12888-100, Qiagen, Germany) by following manufactures instructions, however, the volume of elution buffer was decreased from 100 to 40 μL to increase DNA concentration of the isolates. FastPrep-24 (MP Biomedicals, United States) was applied for the physically disruption of bacterial cells, and Hermle Z216MK (Hermle, Germany) for all centrifugation steps. DNA concentration and purity were assessed using a NanoDrop 1000 spectrophotometer (Thermo Fisher Scientific, United States). DNA samples were subjected to tag-encoded 16S rRNA gene NextSeq-based high throughput sequencing (Illumina, United States). The V3 region (∼190 bp) of the 16S rRNA gene was amplified using primers compatible with Nextera Index Kit (Illumina): NXt_388_F: 5′- TCGTCGGCAG CGTCAGATGT GTATAAGAGA CAGACWCCTA CGGGWGGCAG CAG -3′ and NXt_518_R: 5′- GTCTCGTGGGC TCGGAGATGTG TATAAGAGAC AGATTACCGC GGCTGCTGG -3′ (Integrated DNA Technologies; Leuven, Belgium). PCR reactions containing 12 μl AccuPrimeTM SuperMix II (Life Technologies, CA, United States), 0.5 μl of each primer (10 μM), 5 μl of normalized genomic DNA (∼20 ng/μl), and 2 μl nuclease-free water to a total volume of 20 μl were run on a SureCycler 8800. Cycling conditions applied were: 95°C for 2 min; 33 cycles of 95°C for 15 s, 55°C for 15 s, and 68°C for 30 s; followed by final step at 68°C for 5 min. Agarose gel electrophoresis was performed to ensure that a ∼250 bp PCR product was obtained for each sample, except negative control. To incorporate primers with adapters and indexes, PCR reactions contained 12 μl Phusion High-Fidelity PCR Master Mix (Thermo Fisher Scientific, United States), 2 μl corresponding P5 and P7 primer (Nextera Index Kit), 2 μl PCR product and nuclease-free water for a total volume of 25 μl. Cycling conditions were: 98°C for 1 min; 13 cycles of 98°C for 10 s, 55°C for 20 s and 72°C for 20 s; and 72°C for 5 min. The amplified fragments with adapters and tags were purified using AMPure XP beads (Beckman Coulter Genomic, United States). Pooling and purification were performed with Biomek 4000 pipetting robot (Beckman Coulter). Prior to library pooling clean constructs were quantified using a Qubit Fluorometer (Invitrogen, United States) and mixed in approximately equal concentrations to ensure even representation of reads per sample. Each sample was represented with approximately 40 ng DNA. Subsequently 150 bp pair-ended NextSeq (Illumina, United States) sequencing was performed according to the instructions of the manufacturer.

Plasma samples collected within 5 min post-mortem were shipped to Olink (Uppsala, Sweden) for analyses on their Mouse exploratory panel, using single measurements on 1 μl of plasma. A full list of the 92 proteins on this panel is available at https://www.olink.com.

Raw NextSeq reads were processed by the IMNGS (Lagkouvardos et al., 2016) pipeline (version 1.0 Build 1808) which is based on UPARSE (Edgar, 2013), and USEARCH (Edgar, 2010). Allowed mismatches was set to 2, minimum fastq quality score for trimming of unpaired reads to 3, max number of expected errors in paired sequences to 3, and the minimum relative OTU abundance to 0.25%. The analysis delivered 9,012,489 quality and chimera-checked reads (average 219,816 ± 72,910 reads per sample). OTU’s were clustered at 97% sequence identity. Alignment of IMNGS generated OTU-sequences with SINA version 1.2.11 (Pruesse et al., 2012) (default settings) was performed to obtain the highest possible level of taxonomy of each OTU. The R-based RHEA pipeline (Lagkouvardos et al., 2017) (version 1.1.1.) were used for data analysis yielding rarefaction curves, correlations, group significance (taxonomy and meta-data), and taxonomic binning. Bray–Curtis and Weighted UniFrac (Lozupone and Knight, 2005) represented the phylogenetic beta-diversity analysis and the Shannon index determinations were used for alpha-diversity analysis (see Supplementary Figure S1 for alpha-diversity: Faith phylogenetic diversity, Richness. Beta-diversity: weighted and unweighted Unifrac). Normalization of sequencing depth was done by rarefaction based on 74583 reads via division by the sum of sequences in a given sample and multiplication by the minimum sum across all samples. See Supplementary Figure S3 for rarefaction curves for each individual sample. Data files processed by RHEA were imported to Quantitative Insight Into Microbial Ecology 2 (Caporaso et al., 2010) (2018.4 build 1525276946) for visualization and for group significance of the alpha and-beta-diversity. QIIME 2 is an open source software package for Oracle Virtual Box (Version 5.2.10). The non-parametric Kruskal–Wallis and ANOSIM were used to calculate group significance for, respectively, alpha- and beta-diversity. Taxonomic heatmap was generated with QIIME 1.9.17 (make_otu_heatmap.py) based on binned taxonomy table from RHEA. The volcano plots in Figure 3 were made using OriginPro 2017 with the cut-off set to p < 0.05 and fold change >1.5. The p-values used were computed using Student’s t-test in Perseus 1.6.1.3 (Tyanova et al., 2016).

Initially, we compared the effects of our different 13.5-week diet and diet+exercise training interventions on the GM alpha-diversity in the cecum. Consistent with previous findings, we found that HFD diet mice displayed lower Shannon-diversity compared to chow diet (Figure 1A). Conversely, mice on the pLPHC diet regimen exhibited significantly (p = 0.0004) higher Shannon-diversity compared to HFD (Figure 1A). Chronic LPHC diet had a more pronounced additional effect on Shannon-diversity compared to pLPHC diet. The HPLC diet resembled the HFD group (p = 0.64) and was significantly different from the other groups (p < 0.025). Similar observations were made for the number of the bacterial Richness and Faith phylogenetic diversity in Supplementary Figure S1.

For beta-diversity, i.e., the difference in the taxonomic abundance profiles between our interventions, a principal coordinate analysis (PCoA) clearly indicated that the pLPHC groups were pushed in the direction of the chow and chronic LPHC diet groups (Figure 1B). Overall, our statistical analyses revealed a significant (p ≤ 0.002) separation between the intervention groups for chow vs. HFD, chow vs. pLPHC diet, HFD vs. pLPHC diet, and pLPHC vs. chronic LPHC diet (Figure 1B and Supplementary Figure S1). Of particular relevance, the pLPHC diet group was significantly and strongly separated from the HFD group (R = 0.929, p = 0.001) and separated from the chow (R = 0.476, p = 0.001) and chronic LPHC group (R = 0.580, p = 0.001). Thus, the higher R-value supported the PCoA analysis of a greater separation of pLPHC vs. HFD than pLPHC vs. chow and chronic LPHC. HPLC was significantly different from the HFD (R = 0.519, p = 0.001) and pLPHC (R = 0.784, p = 0.001) with regards to beta-diversity (Figure 1C). In contrast, no effect of exercise training on either alpha or beta-diversity measures were observed (Figure 1C and Supplementary Figure S1). Additional weighted and unweighted UniFrac metrices were applied and produced the same overall tendencies (see Supplementary Figure S1). Thus, the bacterial composition of the gut seems strongly influenced by diet and a pLPHC diet regimen seems sufficient to significantly change this parameter.

At the phylum level, the B/F ratio (Figure 2A) was significantly (p = 0.0009) lower in the HFD group compared to chow (0.35 and 1.89, respectively). Both pLPHC and chronic LPHC diet had B/F ratios in-between the chow and HFD groups (0.79 and 1.38, respectively). The HPLC diet had a slightly elevated B/F-ratio compared to chow at 1.96. Concurrent exercise training caused a numerically higher (non-significant) difference in B/F ratio in the pLPHC groups (no-exercise: 0.79 vs. exercise: 1.25) compared to the HFD groups (no-exercise: 0.35 vs. exercise: 0.38). The prevalence of the phylum of Proteobacteria seemed increased (p < 0.05) in the HPLC and HFD group compared to any other group (Figure 2A).

The overall genus level changes are illustrated in Figure 2B. Significant differences between HFD vs. pLPHC diet were observed, with higher Akkermansia and Bacteroides in the pLPHC groups and lower abundance of Blautia, Oscillibacter, Oscillospira, and Desulfovibrionaceae in the pLPHC groups (Figure 2C and Supplementary Figure S2). No significant effects of concurrent exercise were found. Thus, pLPHC diet appears sufficient to elicit changes in the bacterial composition of the gut which have previously been associated with health-benefits to the host.

To test the influence of pLPHC diet and exercise training on plasma proteins, particularly immune function-related proteins, we applied an exploratory 92 mouse protein multiplex biomarker panel to plasma samples from our different intervention groups. We detected 89 of 92 proteins in our plasma samples and these are shown as volcano plots divided by either exercise or diet in Figure 3A,B, with Supplementary Table S1 listing the 19 significantly different proteins found. Since we measured plasma, we chose to focus only on known secreted proteins (Supplementary Table S2). This resulted in deregulation of a total of 8 known plasma proteins including a number of immune function biomarkers, specifically higher Interleukin 17f (IL-17f) with pLPHC diet vs. HFD and lower IL-17f with exercise vs. non-exercise, higher V-Set And Immunoglobulin Domain Containing 2 (Vsig2), and Chemokine (C-C motif) ligand 20 (Ccl20) with pLPHC diet vs. HFD, and lower Interleukin 6 (IL-6) and Chemokine (C-C motif) ligand 3 (Ccl3) with exercise vs. non-exercise. In addition, Follistatin (Fst) was lower and Ghrelin (Ghrl) higher on pLPHC diet vs. HFD, whereas Neurotrophin 3 (Ntf3) decreased significantly with exercise vs. non-exercise. The significant proteins from Supplementary Table S2 are presented as bar graphs in Figure 3B. Overall, pLPHC diet tended and exercise training appeared to significantly lower the classical proinflammatory cytokine IL-6. In addition, these interventions significantly seemed to change a number of other proteins of potential relevance to the pLPHC phenotype.

Next, we compared our current analysis of the GM to our previous published meta-data on a number of health-related parameters in the same mice (Li et al., 2018a). To briefly sum up our published findings, pLPHC diet mice vs. HFD controls showed improved whole-body glucose tolerance, HOMA2-IR, increased total energy intake but also increased adaptive thermogenesis leading to overall reduced feed efficiency and weight gain-protection, resulting from significant loss of fat mass and tendency to loss of lean body mass. Concurrent running wheel access blunted pLPHC-induced FGF21 secretion and tended to decrease both the body and fat mass gain on HFD and loss on pLPHC diet. These correlations are summed up as a heatmap in Figure 4 with selected statistically significant individual Pearson correlations shown in Supplementary Table S3. As expected, fat mass correlated tightly and positively with total body mass, lean body mass and insulin-resistance (HOMA2-IR) (Figure 4). Fat mass also tended to show a positive correlation with the proinflammatory cytokine IL-6 (r = 0.39, p = 0.15), the chemokine CCL-3 (r = 0.49, p = 0.023) and an inverse correlation (r = -0.41, p = 0.067) with plasma FGF21 (Figure 4). This is consistent with HFD increasing body mass and inflammation whereas pLPHC diet upregulates FGF21 to decrease obesity and insulin resistance. Fat mass and HOMA2-IR showed a moderate positive correlation to Proteobacteria and Desulfovibrionaceae and moderate negative correlation to the alpha-diversity (Shannon index) (Supplementary Table S3). Akkermansia was strongly and negatively associated with HOMA2-IR (Figure 4). Phylum Bacteroidetes showed a significant negative and positive association with HOMA2-IR (r = -0.58, p = 0.005) and alpha-diversity (Shannon index, r = 0.49, p = 0.025), respectively. In summary, this indicates that the obesity-protected, insulin-sensitive, less inflamed phenotype of pLPHC diet mice is tightly associated with higher gut bacterial diversity, increased Bacteroidetes and Akkermansia and lower Proteobacteria and Desulfovibrionaceae. This suggests that changes in GM may have contributed to the generally healthier phenotype of mice on the pLPHC diet.