Association of the Gut Microbiota with Weight-Loss Response within a Retail Weight-Management Program

Retail programs offer popular weight-loss options amid the ongoing obesity crisis. However, research on weight-loss outcomes within such programs is limited. This prospective-cohort observational study enrolled 58 men and women between ages 20 and 72 years from a retail program to assess the influence of client features on energy-restriction induced weight-loss response. DESeq2 in R-studio, a linear regression model adjusting for significantly correlating covariates, and Wilcoxon signed-rank and Kruskal–Wallis for within- and between-group differences, respectively, were used for data analyses. An average 10% (~10 kg) reduction in baseline-weight along with lower total-, android-, gynoid-, and android:gynoid-fat were observed at Week 12 (all, p < 0.05). Fifty percent of participants experienced a higher response, losing an average of 14.5 kg compared to 5.9 kg in the remaining low-response group (p < 0.0001). Hemoglobin-A1C (p = 0.005) and heart rate (p = 0.079) reduced in the high-response group only. Fat mass and A1C correlated when individuals had high android:gynoid fat (r = 0.55, p = 0.008). Gut-microbial β-diversity was associated with BMI, body fat%, and android-fat (all, p < 0.05). Microbiota of the high-response group had a higher baseline OTU-richness (p = 0.02) as well as differential abundance and/or associations with B. eggerthi, A. muciniphila, Turicibacter, Prevotella, and Christensenella (all, p/padj < 0.005). These results show that intestinal microbiota as well as sex and body composition differences may contribute to variable weight-loss response. This highlights the importance of various client features in the context of real-world weight control efforts.


Introduction
The burgeoning field of microbiota has revealed how inner environment can shape human health and wellbeing. Gut bacterial compositions may also contribute to and help predict physiological dysfunctions in patients [1][2][3]. In particular, pre-clinical and clinical studies have reported links between gut bacteria and obesity [4][5][6][7]. Experiments in germ-free mice established a direct contribution of the gut bacteria composition and metabolism on body weight outcomes [8,9]. Microbiomes in obese mice and obese humans showed enhanced energy harvest capacity from non-digestible carbohydrates by way of short chain fatty acid (SCFA) production. Furthermore, germ-free mice and rats with low SCFA-levels are protected from diet-induced obesity. Paradoxically, SCFA-intervention protected mice from diet-induced obesity. It is also known that SCFAs modulate fatty acid synthesis and oxidation to favorably impact body fat and weight. However, molecular mechanisms underlying the apparent paradox related to SCFAs as well as other aspects of microbiota mediated host energy metabolism are not fully understood [8,[10][11][12][13][14].
In parallel to mechanistic research endeavors, there is a growing interest in the manipulation potential of the gut microbiota for achieving effective weight-loss. While it was proposed that microbiota

Study Design, Setting, and Participants
Institutional Review Board for Human Subject Research at South Dakota State University approved (IRB-1608010-EXP) this prospective cohort, 12-week, observational study. Incoming clients in a weight-reduction program from its Brookings county retail location in South Dakota were recruited through flyers between November 2016 and February 2018. Fourteen men and 44 women-between ages 20 and 72 years and predominantly Caucasian due to geographical location-joined the study on a first-come-first-serve basis after going through eligibility screening and providing informed consent in writing. Despite our best effort, a low number of males signed up for the study. This experience is similar to another study reporting that men are less likely than women to participate in retail weight-loss programs [30]. Inclusion criteria included being generally healthy (i.e., absence of major illnesses including diabetes), any ethnicity and either sex, willing to undergo planned weight-loss by adhering to the guidelines of the retail-program, and willingness to comply with the research study protocol. Exclusion criteria were pregnancy; lactation; probiotic-, prebiotic-, and antibiotic-usage; being on any special diet in the previous 12-months; and cancer, diabetes, immune compromised state, and other conditions that would affect the ability to provide informed consent or comply with the study protocol. Nineteen out of 58 recruited participants (32.8%) dropped out of the program before Week 12 and were excluded from the study. There was no case where a participant continued on the retail program but discontinued with the study (Figure 1). This led to a longer than anticipated recruitment phase so that sufficient number of participants could still complete the study to allow 85% probability of detecting a weight-loss at a 0.05 significance level. Of note, as high as 58% dropout by 13th week into a well-known commercial weight-loss program has been previously reported by other researchers [29]. This experience is similar to another study reporting that men are less likely than women to participate in retail weight-loss programs [30]. Inclusion criteria included being generally healthy (i.e., absence of major illnesses including diabetes), any ethnicity and either sex, willing to undergo planned weight-loss by adhering to the guidelines of the retail-program, and willingness to comply with the research study protocol. Exclusion criteria were pregnancy; lactation; probiotic-, prebiotic-, and antibiotic-usage; being on any special diet in the previous 12-months; and cancer, diabetes, immune compromised state, and other conditions that would affect the ability to provide informed consent or comply with the study protocol. Nineteen out of 58 recruited participants (32.8%) dropped out of the program before Week 12 and were excluded from the study. There was no case where a participant continued on the retail program but discontinued with the study (Figure 1). This led to a longer than anticipated recruitment phase so that sufficient number of participants could still complete the study to allow 85% probability of detecting a weight-loss at a 0.05 significance level. Of note, as high as 58% dropout by 13th week into a well-known commercial weight-loss program has been previously reported by other researchers [29].

Diet, Anthropometrics, and Other Data Collection during Participant Visits
All data and samples were collected at two time points: Week 0 (baseline), i.e., right after enrollment in the program, but before starting the plan-diet, and at Week 12, i.e., after continuing to follow the program-diet for 12 weeks. The 12-week time period was previously observed as an optimal average time window when majority of program participants tend to be within 50% of their target weight (personal communication from McCormack-group, Brookings, SD, USA). Furthermore, clients were required to consume pre-formulated, ready-to-eat food items sold by the program during the initial months for maximizing weight-loss prior to transitioning to regular food. This provided greater uniformity in dietary ingredients and patterns among participants at Week 12 data collection, potentially enhancing quality and consistency of the reported outcomes. Dietary-intake data were collected in-person by nutrition science and dietetics students under supervision of a registered dietitian nutritionist using 24-h recalls and were entered in the Nutritionist Pro software connected to USDA nutrition information database (Axxya Systems, Redmond, WA, USA) [31]. Recall data were collected during the scheduled study-visits for two non-consecutive days (specific days for an individual were decided during the interview) from recent past of which one was a weekday and the other a weekend-day. Each participant was provided with a wallet-size WebMD portion size guide for reference (2012 version by Kathleen Zelman, MPH, RD, LD). For packaged food items, nutrition data were obtained from labels.

Diet, Anthropometrics, and Other Data Collection during Participant Visits
All data and samples were collected at two time points: Week 0 (baseline), i.e., right after enrollment in the program, but before starting the plan-diet, and at Week 12, i.e., after continuing to follow the program-diet for 12 weeks. The 12-week time period was previously observed as an optimal average time window when majority of program participants tend to be within 50% of their target weight (personal communication from McCormack-group, Brookings, SD, USA). Furthermore, clients were required to consume pre-formulated, ready-to-eat food items sold by the program during the initial months for maximizing weight-loss prior to transitioning to regular food. This provided greater uniformity in dietary ingredients and patterns among participants at Week 12 data collection, potentially enhancing quality and consistency of the reported outcomes. Dietary-intake data were collected in-person by nutrition science and dietetics students under supervision of a registered dietitian nutritionist using 24-h recalls and were entered in the Nutritionist Pro software connected to USDA nutrition information database (Axxya Systems, Redmond, WA, USA) [31]. Recall data were collected during the scheduled study-visits for two non-consecutive days (specific days for an individual were decided during the interview) from recent past of which one was a weekday and the other a weekend-day. Each participant was provided with a wallet-size WebMD portion size guide for reference (2012 version by Kathleen Zelman, MPH, RD, LD). For packaged food items, nutrition data were obtained from labels. Height, waist (at navel), and hip (at the widest part) were recorded to the nearest 0.5 cm while body weight (BW) was measured to the nearest 0.1 kg in light clothes or scrubs and no shoes (electronic scale, Seca GmbH & Co., Hamburg, Germany) following established protocols [32]. Waist:hip (WH) and body mass index (BMI) are two widely used markers in obesity control programs due to simplicity of measurements and low-cost tools. However, since these were reported to be less useful for metabolic risk assessment [33][34][35], we additionally assessed body composition features using dual energy X-ray absorptiometry (DXA) whole body scan (Hologic QDR Discovery, Waltham, MA, USA) that included-total fat mass, body fat percentage (BF%), android fat, gynoid fat, and amount and distribution of corresponding fat-free masses (to indicate lean masses) [36][37][38][39]. In presenting WH, BMI, and body composition data side-by-side, we pointed out when one measure may not serve as a surrogate for the other.
Glycated hemoglobin (A1C), an indicator of glucose metabolism and diabetic status, was measured using A1C Now ® + kit (PTS Diagnostics, Indianapolis, IN). Individuals with an A1C of ≥6.5% were considered diabetic range and excluded from the study [40]. The retail plan is administered differently to diabetics which could have potentially introduced variability in response data. Blood pressures and resting heart rates were measured using a digital sphygmomanometer (Greater Goods, LLC, St. Louis, MO, USA) using standard procedures [32].

Nucleic Acid Extraction and 16S Ribosomal RNA Amplicon Sequencing
Participants were provided with collection hat, gloves, and Omni-Gut stool collection kit (DNA Genotek, Ontario, ON, Canada). The samples were received within 24 h after collection, processed, aliquoted, and stored at −80 • C. DNA was isolated with MagAttract PowerMicrobiome Kit following manufacturer's guidelines (Qiagen, Valencia, CA, USA) and quantified with Qubit ® Quant-iT High Sensitivity Kit (Invitrogen, Life Technologies, Grand Island, NY, USA). 16S V4 rRNA region was enriched using PCR amplification using primers designed against the surrounding conserved regions, followed by ligation of adapters and indexing barcodes (Illumina, San Diego, CA, USA). PCR products were quantified by PicoGreen (Life Technologies, Grand Island, NY, USA) and prepared equimolar for sequencing step.
Sequencing (250 cycles, paired-end) using Illumina Miseq platform (San Diego, CA, USA) was carried out by Second Genome (South San Francisco, CA, USA). Sequenced reads were quality filtered and resulting unique sequences were clustered at 97% by UPARSE (de novo OTU clustering) and a representative consensus sequence per de novo OTU was determined. The UPARSE clustering algorithm comprises a chimera filtering and discards likely chimeric OTUs. Resulting sequences were searched against Greengenes reference database (closed reference operational taxonomic unit (OTU) picking). The longest sequence from each OTU was then assigned taxonomic classification via Mothur's Bayesian classifier, trained against the Greengenes database clustered at 99%. Further bioinformatics and data analyses (except Piphillin) were carried out in our laboratory (next sections). We also carried out BLAST search to cross reference with closest hits from NCBI 16S rRNA database with >95% query cover, >87% identity, and <0.01 E value for unidentified OTUs in Greengenes. The raw sequences are deposited in NCBI sequence read archive (SRA, accession number SRP237387), belonging to BioProject number PRJNA595387.

Sub-Grouping of Participants for Data Analyses
All data are presented for 36 participants (ALL) as well as after grouping them based on baseline differences in age, sex, BMI, and android:gynoid fat ratio (AG) to assess any role of these participant characteristics in weight-loss. Participants are also grouped based on the extent of body-weight reduction (weight-loss response) at Week 12. Data are presented as: within group differences from baseline to Week 12 and between group differences at baseline and Week 12. The subgroups are defined as follows: (1) young adults (YA, n = 20, average age 33 years), and older adults (OA, n = 16, average age 62 years; (2) males (n = 9) and females (n = 27); (3) higher BMI (HI BMI , n = 24, average BMI 38 kg/m 2 ) and lower BMI (LO BMI , n = 12, average BMI 27 kg/m 2 ), where the BMI cut off was determined based on normal + overweight (<29.9 kg/m 2 ) and obese ranges (>30 kg/m 2 ); (4) higher AG (HI AG , n = 23, average AG 0.63) and lower AG (LO AG , n = 13, average AG 0.43), where lower AG is generally deemed healthier, but, unlike BMI, currently there is no reference value available, thus the arbitrary cut-off for our mixed sex cohort was taken as 0.5, partially based on a study that reported 50th percentile of AG for 50 years and older Caucasians as 0.46 (female) and 0.71 (male) [39]; and (5) the higher response (HI res , n = 18) and lower response (LO res , n = 18) groups were based on a 10 kg weight-loss cut off. Significant difference in the respective defining criterion between the subgroups was present for: age (YA/OA), BMI (HI BMI /LO BMI ), AG (HI AG /LO AG ), and weight-loss response (HI res /LO res ).

Bioinformatics and Statistical Data Analyses
All data analyses were carried out using R-studio and/or Sigma Plot Software (Systat Software Inc., San Jose, CA, USA) and data presented as mean ± SD unless otherwise stated. Normality was assessed using Shapiro-Wilk test followed by t-test or Mann-Whitney U-test for analysis of anthropometric and metabolic features. Statistical significance was considered at p ≤ 0.05. A p value greater than 0.05 and less than 0.08, when shown, indicates approaching significance.
For metagenomics data, multiplexed sequence reads were converted to taxonomic and phylogenetic profiles using QIIME2 (Quantitative Insights Into Microbial Ecology). Alpha diversity i.e., OTU richness and Shannon diversity were calculated by summing unique OTUs found in each sample and combining richness with the relative abundance data, respectively. For beta diversity, dissimilarity score was determined by comparing in a pairwise fashion using a distance matrix. Abundance-weighted difference was calculated using Bray-Curtis and binary dissimilarity were identified using Jaccard index [41,42]. Statistical testing of alpha-diversity was carried out using Linear Regression, Wilcoxon, and Kruskal-Wallis tests. Permutational Analysis of Variance (PERMANOVA) testing utilized the sample-to-sample distance matrix directly, and not a derived ordination or clustering outcome, to find the differences among discrete, categorical, or continuous variables by randomly reassigning the samples to various sample categories. Correlations were determined using Pearson's coefficient. Univariate differential abundance of OTUs was calculated using DEseq2 package with default settings in R-studio [43]. The package uses a negative binomial noise model for the overdispersion and Poisson process intrinsic to the data [44]. Additionally, the package also takes both technical and biological variability between experimental conditions into account and provides the difference in terms of Log 2 fold-change. Furthermore, the p value was adjusted (p adj ), when appropriate, using Benjamini-Hochberg correction for false discovery rate (FDR) inherent to large number of dependent variables [45]. The functional capacity of the metagenome from OTU counts and representative sequence of each OTU was assessed using Piphillin version 6.0 (Second Genome Inc., South San Francisco, CA, USA). The closest matched 16S rRNA sequence above the identity cut-off at 95% was considered as the inferred genome for that OTU. In the case of multiple nearest neighbor genomes with equal identities, the count is equally split to sum the inferred genome content. The content is expressed as orthologs searched against pathways counts in Kyoto Encyclopedia of Genes and Genomes (KEGG) genome database [46].

Energy-Intake in Study Participants
At Week 12, there was an average 21.8% lower daily calorie intake than at baseline in ALL (Table 1, p = 0.000003). Energy intake from carbohydrates was lower while that from proteins was higher at Week 12 (both, p ≤ 0.01). Average fiber intake-critical for nourishing the gut microbiota-did not change between the microbial data collection time points in the study, i.e., from baseline to Week 12. Intakes were similar between HI res and LO res at baseline and at Week 12 (all, p > 0.05). Furthermore, Microorganisms 2020, 8, 1246 6 of 20 HI res and LO res experienced a similar average reduction of 531.7 and 485.8 kcal/day, respectively, over the 12-week period (p > 0.05).

Baseline Body Composition: BMI versus AG
A growing body of research points to BF%, android region fat, and AG (relative measure of android fat) being more relevant than frequently utilized measures of BW and BMI for assessing chronic disease risk [47][48][49][50][51][52][53][54][55]. Overall, two participants had BMI < 25 kg/m 2 at baseline, while the rest were overweight, obese, or severely obese (BMI range 24.42-51.84 kg/m 2 ). Males started with an average 19 kg higher BW than females (p = 0.04), although the BMI was similar between the groups (Table 2). However, males had a higher AG than females (p = 0.001). In addition, total body fat was 10% lower, while fat-free mass was higher by approximately 20 kg (both, p < 0.001) in males than females ( Table 2). The observations support that BMI, but not AG, may be influenced by lean-mass and/or height. Thus, AG may be more relevant than BMI in deciding weight-reduction goals. Furthermore, measures of metabolic functions such as A1C (p = 0.014) as well as systolic and diastolic blood pressures (both, p < 0.001) were higher in the HI AG than the LO AG group, but this distinction was absent between the BMI subgroups ( Table 2). These observations indicate that AG is superior to BMI for indicating body fat content and distribution as well as for assessing possible metabolic implications of increased fat in the central region of the body. Of note, the average age difference of 12 years was significant between the AG groups (p = 0.025) but such a distinction was not observed between the BMI groups. Central obesity tends to increase with age when overall metabolism tends to slow down. This may explain higher age of the HI AG group (Table 2). Our observations are in line with previous reports that body fat and its distribution pattern independently associate with higher risk of chronic disease and mortality [34,35,53,54]. When possible, retail programs may consider body composition measurement as an option for determining weight-loss and metabolic wellbeing goals.

Anthropometric and Metabolic Changes over 12-weeks
Seventeen participants lost ≥10%, fifteen participants lost 5-10%, three participants lost ≤5%, and one participant gained approximately 2% of their respective baseline BW. Average weight lost in 12 weeks was~10 kg (Tables 2 and 3) and the BW reduction was significant in all sub-groups (Tables 3 and 4). Along with body-weight, the following measures were reduced: blood pressure, total fat and lean masses, BF%, android fat and lean masses, android fat%, gynoid fat and lean masses, and AG (all, p ≤ 0.05, Table 4). An association between reduction in fat mass and A1C (r = 0.49, p = 0.0028) was observed in ALL that was further augmented in HI BMI and HI AG (both, r ≥ 0.55, p < 0.01) but absent (p > 0.3) in corresponding low-groups ( Figure 2). There was no apparent relationship between participant's age and how they responded to energy reduction for weight-loss (Tables 2 and 3).      weeks was ~10 kg (Tables 2 and 3) and the BW reduction was significant in all sub-groups (Tables 3  and 4). Along with body-weight, the following measures were reduced: blood pressure, total fat and lean masses, BF%, android fat and lean masses, android fat%, gynoid fat and lean masses, and AG (all, p ≤ 0.05, Table 4). An association between reduction in fat mass and A1C (r = 0.49, p = 0.0028) was observed in ALL that was further augmented in HIBMI and HIAG (both, r ≥ 0.55, p < 0.01) but absent (p > 0.3) in corresponding low-groups ( Figure 2). There was no apparent relationship between participant's age and how they responded to energy reduction for weight-loss (Tables 2 and 3).
Response variability not explained by anthropometric characteristics: Exactly half of the participants experienced a higher degree of weight and fat loss than the rest at Week 12-an average reduction of 13.2% (14.5 kg) versus 6.8% (5.9 kg) of baseline BW; this corresponded to an average 10% versus 5.2% lower BF% (both, p < 0.05) at Week 12 than baseline in HI res and LO res , respectively. In addition, HI res had a greater (6.5%, p = 0.005) reduction in A1C than LO res (3.2%, p = 0.072) at Week 12 (Table 4). However, with the exception of a higher starting BW in HI res (p = 0.003), the two response-groups had similar age, body-composition types (BF% and AG), A1C, blood pressures, and heart rates at baseline (all, p > 0.05, Table 2). Notably, changes in energy intake over the 12-week period were similar between the two response groups (Table 1).
Mean OTU richness increased from 512 at baseline to 543 at Week 12 in ALL (p = 0.017) (Figure 3). The two groups who lost most weight, i.e., most responsive to weight-loss, started with significantly higher OTU richness at baseline-HI res (551) and males (567)-than LO res (473, p = 0.02) and females (493, p = 0.03), respectively. Post weight-loss, the richness in the HI res (566) and males (556) remained more or less unchanged compared to large increases in LO res (521) and females (539), respectively.
Four response groups were similar in age ( Table 2). The observations supports a previous report that a greater existing alpha diversity predicts an increased resistance to further diversification of the microbiome [58]. Shannon diversity ranged from 4.09 to 4.15 across all samples and did not differ significantly between sub-groups or from baseline to Week 12.

Changes in Beta Diversity
Samples from same subject clustered together with strong inter-individual differences. Abundance of Firmicutes (p = 0.052) and Actinobacteria (p = 0.03) decreased while that of Tenericutes (p = 0.06) and Euryarchaeota (p = 0.05) as well as that of family Porphyromonadaceae (p = 0.00011) increased at Week 12, in ALL. PERMANOVA of distance matrices showed general association of beta diversity with age, heart rate, total fat mass, BF%, android fat mass and android fat% (all p < 0.05), indicating contribution of these factors to the overall microbiome differences among individuals. These relations did not change from baseline to Week 12 and, hence, were likely not influenced by weight-loss or the calorie-restricted diet. In ALL, Turicibacter sp. (r = −0.55, p = 0.001) and Christensenella sp. (r = −0.61, p = 0.012) negatively correlated with BF% at baseline but not at Week 12. Christensenella minuta-the first identified species in this genus-was previously associated with healthy, low-BMI individuals and short chain fatty acid production in previous studies from our and other groups [59,60].

Microbiome Differences in Response Groups
Prevotella copri increased in HIres but decreased in LOres from baseline to Week 12 (both padj < 0.014). Two OTUs of Ruminococcus were differentially abundant between HIres and LOres groups at baseline only (both, padj < 0.005, Table 5). One OTU of Akkermansia muciniphila showed more than 10fold higher abundance in LOres compared to HIres consistently at baseline and at Week 12 (both, padj < 0.003). In addition, an OTU of Bacteriodes eggerthi showed over 50-fold higher abundance at baseline in HIres than LOres (padj = 0.005) and associated with weight-and BMI-change in HIres (r = 0.60, p = 0.008) but not in LOres. Abundance of Bacteriodes plebius and Eubacterium biforme were consistently lower in HIres and males at baseline (all, padj ≤ 0.03). Interestingly, unlike various differences in individual OTUlevel response between HIres and LOres, fewer distinctions were observed between males and females over the 12-week period (Tables 5 and 6). The data from HIres and LOres show potential contribution of microbiome differences to weight-loss variations. However, sex may influence weight-loss in a

Changes in Beta Diversity
Samples from same subject clustered together with strong inter-individual differences. Abundance of Firmicutes (p = 0.052) and Actinobacteria (p = 0.03) decreased while that of Tenericutes (p = 0.06) and Euryarchaeota (p = 0.05) as well as that of family Porphyromonadaceae (p = 0.00011) increased at Week 12, in ALL. PERMANOVA of distance matrices showed general association of beta diversity with age, heart rate, total fat mass, BF%, android fat mass and android fat% (all p < 0.05), indicating contribution of these factors to the overall microbiome differences among individuals. These relations did not change from baseline to Week 12 and, hence, were likely not influenced by weight-loss or the calorie-restricted diet. In ALL, Turicibacter sp. (r = −0.55, p = 0.001) and Christensenella sp. (r = −0.61, p = 0.012) negatively correlated with BF% at baseline but not at Week 12. Christensenella minuta-the first identified species in this genus-was previously associated with healthy, low-BMI individuals and short chain fatty acid production in previous studies from our and other groups [59,60].

Microbiome Differences in Response Groups
Prevotella copri increased in HI res but decreased in LO res from baseline to Week 12 (both p adj < 0.014). Two OTUs of Ruminococcus were differentially abundant between HI res and LO res groups at baseline only (both, p adj < 0.005, Table 5). One OTU of Akkermansia muciniphila showed more than 10-fold higher abundance in LO res compared to HI res consistently at baseline and at Week 12 (both, p adj < 0.003). In addition, an OTU of Bacteriodes eggerthi showed over 50-fold higher abundance at baseline in HI res than LO res (p adj = 0.005) and associated with weight-and BMI-change in HI res (r = 0.60, p = 0.008) but not in LO res . Abundance of Bacteriodes plebius and Eubacterium biforme were consistently lower in HI res and males at baseline (all, p adj ≤ 0.03). Interestingly, unlike various differences in individual OTU-level response between HI res and LO res , fewer distinctions were observed between males and females over the 12-week period (Tables 5 and 6). The data from HI res and LO res show potential contribution of microbiome differences to weight-loss variations. However, sex may influence weight-loss in a manner that may or may not be linked to microbiome differences.

Predicted Metabolic Functions
Functional inference of the metagenomic shifts between baseline and Week 12 was carried out [46]. The Piphillin-generated hits against KEGG Orthologs (KO) abundance data showed potential modulation of 297 functional pathways based on gene content predictions for protein-coding genes. Several pathways showed subtle but statistically significant changes in KO abundance at Week 12 in ALL (Table 8). This included lower KO abundance for starch and sucrose metabolism as well as primary and secondary bile acids (all, p < 0.05). Notably, participants consumed fewer dietary carbohydrates at Week 12 compared to baseline (Table 1). This section may be divided by subheadings. It should provide a concise and precise description of the experimental results, their interpretation, and the experimental conclusions that can be drawn.

Discussion
National guidelines for weight-loss and chronic disease prevention recognize that referral outside of the clinical setting to weight-loss programs, including retail programs, is needed for effective weight management and improved public health [24]. However, reports of client experience in terms of weight-loss outcomes in commercial real-life settings are relatively scarce. As obesity rates continue to rise, it is critical to dedicate research into factors relating to effectiveness of such programs. This study is among the first to present a detailed evaluation of gut microbiota features in the context of weight-loss response-outcome within a commercial weight-management program. Simultaneous evaluation of the gut microbiota and an extensive array of body composition parameters helped assess the relationship of these outcomes with weight-change response in these participants. The main observations that emerged are: (1) All but one participant achieved significant weight-loss over 12 weeks. However, the extent of weight-loss was non-uniform; half of them lost more than twice as much body weight as the rest despite a similar level of energy restriction and same formula diet between the response groups. Individuals with higher response had higher OTU richness (alpha diversity measure) and taxa level differences (beta diversity measures) compared to the low response individuals. The results indicate that differences in the gut microbiota of participants may relate to the response to weight-management interventions. (2) A1C associated with fat mass and was attenuated in the high-response group but not in the low-response group. It is possible that a larger weight-loss (>10% of initial weight in this cohort) may be necessary when a significant A1C reduction is desired such as in type 2 diabetics. Ironically, losing weight is particularly challenging for individuals with type 2 diabetes [63]. (3) Aging is associated with metabolic alterations that can lead to excess weight gain and slower weight-loss [64,65].
Thus, we hypothesized that participant's age may influence weight-loss response. However, we did not observe a differential weight-change between young and older adults.
Adherence of participants is one of the many factors that may contribute to the impact of a program on community health and fitness. A low initial weight-loss may reduce motivation to adhere in low-responders and can lead to up to 27% reported dropout rate by end of first four weeks [29,66]. The observed extent of variability in weight-loss response in as many as half of our participants even after 12 weeks was striking from that context. As a next step, our results indicate a potential relationship between baseline gut microbiota and the variability in weight-loss response. Educating incumbent program participants regarding potential response-variability due to their own baseline gut features may help set realistic expectations and improve adherence. Furthermore, observational research such as this contributes to the groundwork needed for mechanistic research on how to precisely predict response-level in patients based on their gut signature. We anticipate that, with increased application of microbiome research knowledge in practice, routine metagenomics profiling in retail participants would be logical in the future. On a side note, the high cost of retail programs could also lead to low-adherence [24]. A third of the recruited participants chose to discontinue the retail program and were dropped from the study. Common reasons for quitting were: (i) high cost of proprietary food items; and (ii) dislike for low-calorie formula diet. In contrast, reasons to continue were: (i) convenience of ready-to-eat foods; and (ii) reinforcement through personalized coaching. While quantitative assessments of these responses were beyond the scope of this research, they align with a prior report of even higher dropout (58%) from another well-established commercial program due to costs and other barriers [24,29]. Therefore, validation of strategies to facilitate adherence must be a priority.
The presented metagenomics observations support the increasingly recognized contribution of the gut microbiome toward human health, metabolism, and treatment response [7,67]. B. eggerthii was shown to correlate with body fat and inflammation [68,69]. This corroborates with our observation of higher abundance of this species in the high-response individuals who had higher baseline android fat. However, we also observed a hitherto unknown direct association of B. eggerthii with weight-reduction selectively in high-responders. Whether higher abundance of this species on the one hand indicates increased body fat, but on the other hand enhances responsiveness to energy-reduction warrants follow-up research. Further, Christensenella minuta has been linked with healthy body composition and Turicibacter with reduction in high-fat diet induced inflammation [59,60,69,70]. Our data showing negative correlation of Christensenella and Turicibacter with body-fat while undergoing weight-loss support these prior observations.
A favorable role of Prevotella abundance on weight-loss, especially in the context of a high fiber diet, has been reported [71]. We observed an inverse association of the Prevotella genus with android fat post weight-loss in older individuals, and an increased abundance of Prevotella copri in high-response group. Our results also indicate an overall metabolic shift post weight-loss that included changes in carbohydrate metabolism. More recently, it was proposed that Prevotella enterotype patients with genetically low starch digestion capability may produce higher levels of SCFA that contributes to greater weight-loss [72]. The concept of microbial enterotypes attempts to simplify patient stratification based on complex gut signature, although consensus is lacking among researchers regarding the extent of its usefulness [73][74][75][76]. Nevertheless, our observations along with prior evidence direct to the possibility that SCFA-producing Prevotella spp. may at least partially drive the metabolic shift needed for weight-loss. Furthermore, for additional insights on bacteria derived metabolites (e.g., SCFAs) associated with weight-loss response, fecal metabolomics profiling may be considered in the future.
Dao et al. reported association of A. muciniphila with a healthier metabolic status in calorie-restricted overweight and obese adults [18]. We observed a higher abundance of this species in low responders who had lower BMI than high responders. A lower BMI suggests a healthier metabolic status. However, Dao et al. reported no difference in weight-loss between high and low baseline A. muciniphila groups. Another trial reported no reduction in visceral fat and BMI after three months of A. muciniphila supplementation [77]. Therefore, it is possible that the observed differential A. muciniphila levels in our response groups is more indicative of the differences in their metabolic status than the differential response to weight-loss. A larger body of independent research would help determine the precise role of A. muciniphila in weight-loss versus maintenance of healthier metabolism.
The study is not without limitations: (a) This is not an intervention/mechanistic study and does not establish whether the observed relationships are causal or consequential in nature. However, being among the first human studies presenting the relationship of the gut microbiome with weight-loss response within a real-life retail program setting, this exploratory research contributes to the groundwork needed for larger observational as well as mechanistic research. (b) Our observational and small study cohort, although adequate from statistical perspective, was racially homogeneous. It mostly represented Caucasians owing to the geographical location of the study site. The findings may not directly apply to other races as host-genetics may influence microbiome characteristics [78]. (c) Species in the mucus layer (e.g., A. muciniphila) differ from those in the intestinal lumen. Stool samples-a practical and non-invasive way to assess gut microbiota in human subjects-represent more luminal species and may not fully reflect the entire gut microbiota [79]. (d) Since the retail program has everyone undergoing energy restriction, an eucaloric control arm was not feasible. However, we argue that the response sub-groups consumed same diet and underwent similar level of calorie-restriction; therefore, differences in weight-loss outcomes are likely contributed by client-features such as the gut microbiota profile. Furthermore, we believe that understanding real-life outcomes despite logistical challenges has its merits.

Conclusions
With the obesity epidemic showing no sign of decline, it is increasingly recognized that retail weight-management programs can play important role in mitigation efforts. We examined the extent of weight-loss variation in response to planned energy-restriction among participants from a commercial program and if such variability is related to participant characteristics, including the profile of their gut microbiota. Although subtle, clear differences in gut microbiome signature existed between the weight-loss response groups. Our results suggest contribution of the gastrointestinal microbiota as well as sex and body composition differences toward differential weight-reduction within a retail setting. Additional research-providing independent validation in different cohorts as well as exploring how such knowledge may help customize improved weight-loss approaches-will be key.