High-fat diet impact on intestinal cholesterol conversion by the microbiota and serum cholesterol levels

Summary Cholesterol-to-coprostanol conversion by the intestinal microbiota has been suggested to reduce intestinal and serum cholesterol availability, but the relationship between intestinal cholesterol conversion and the gut microbiota, dietary habits, and serum lipids has not been characterized in detail. We measured conserved proportions of cholesterol high and low-converter types in individuals with and without obesity from two distinct, independent low-carbohydrate high-fat (LCHF) dietary intervention studies. Across both cohorts, cholesterol conversion increased in previous low-converters after LCHF diet and was positively correlated with the fecal relative abundance of Eubacterium coprostanoligenes. Lean cholesterol high-converters had increased serum triacylglycerides and decreased HDL-C levels before LCHF diet and responded to the intervention with increased LDL-C, independently of fat, cholesterol, and saturated fatty acid intake. Our findings identify the cholesterol high-converter type as a microbiome marker, which in metabolically healthy lean individuals is associated with increased LDL-C in response to LCHF.


INTRODUCTION
Cholesterol, an amphipathic sterol lipid, is an essential structural component of human and animal cell membranes and serves as a precursor for the biosynthesis of steroid hormones, bile acids and vitamin D. However, hypercholesterolemia, or excess blood cholesterol, is a major risk factor for cardiovascular disease, the leading cause of mortality worldwide. 1Large amounts of cholesterol enter the small intestine every day from exogenous dietary sources, including animal products ($0.3-0.6 g/day), and from endogenous biosynthesis in the liver and secretion with bile acids ($0.7-0.9 g/day). 2 Consequently, hypercholesterolemia treatment can involve limiting dietary cholesterol intake, inhibiting cholesterol biosynthesis, and/or blocking cholesterol uptake from the intestine.Statins, pharmacological inhibitors of the HmG-CoA-dependent cholesterol-generating mevalonate pathway in the liver, represent one of the most successful, widely used and best-selling drug classes worldwide. 3Lowering dietary cholesterol intake alone typically has limited and inconsistent effects, 4 as cholesterol production is tightly controlled via feedback mechanisms that adapt HmG-CoA activity to dietary cholesterol intake and cellular requirements. 5These mechanisms involve insulin, which can increase or decrease HmG-CoA expression in response to dietary carbohydrates 6 and low-carbohydrate diets can reduce circulating cholesterol levels. 7The clinical importance of cholesterol availability in the intestine is demonstrated by the successful combination of statins with small-molecule inhibitors of cholesterol uptake 8 and may be influenced by the gut microbiota. 9Intestinal bacteria can metabolize statins 10 and may contribute to inconsistent and adverse effects of the medication in some individuals. 11,12Thus, a better understanding of the relationship between intestinal cholesterol availability and circulating cholesterol levels in relation to the gut microbiota could identify diagnostic markers for patient stratification, as well as therapeutic targets for the development of new cholesterol-reducing interventions.

ll OPEN ACCESS
The intestinal microbiota can reduce cholesterol to coprostanol, which is unavailable for human absorption, stable under anoxic conditions and excreted in feces. 13Fecal coprostanol is reduced in antibiotically treated animals and humans 14,15 and undetectable in germ-free rats 16 and human newborns. 17Intestinal cholesterol conversion has been attributed to a broad and diverse range of microbial taxa, based on in vitro experiments with bacterial isolates from feces and in silico fecal metagenomic sequence analysis, 18 but few human or animal fecal bacterial isolates with confirmed cholesterol conversion activity are available. 20][21] Yet, whether intestinal cholesterol conversion by the human gut microbiota remains stable over a person's lifetime, to what extent it is affected by dietary changes or other endogenous or exogenous factors, or whether it affects cholesterol availability to the human host remains unclear and has been controversially discussed. 2ow-carbohydrate high-fat (LCHF) diets have become increasingly popular in recent years to induce weight loss and improve blood sugar control in individuals with obesity. 6,22Ketogenic LCHF diets aim to induce a shift of the body's primary dietary energy source from carbohydrates to fats, resulting in a state of ketosis and far-reaching physiological alterations. 23Besides obesity reduction, LCHF diets have been associated with reduced cardiovascular disease risks, improved type 2 diabetes, and other health benefits 24 and, as a consequence, become popular among young, healthy, normal-weight people. 257][28][29] Although the clinical relevance of this elevated LDL-C response to LCHF diets in individuals without insulin resistance has been debated, 30 as well as the benefits of statin therapy to treat it, 31 there is a need to identify affected individuals and avert potential negative consequences of these popular diets, especially in healthy young populations.
In this study, we examine the relationship between intestinal cholesterol-to-coprostanol conversion, gut microbiota, and serum lipid profiles, in the context of LCHF interventions.We identify the stratification of individuals into cholesterol high and low-converter types as a conserved feature of the fecal microbiome in two human cohorts with distinct geographic (Germany/Norway) and metabolic (with/without obesity) study parameters.Across both cohorts, we find cholesterol conversion to be strongly correlated to the fecal relative abundance of E. coprostanoligenes and to be increased in previous low-cholesterol converters on (ketogenic/non-ketogenic) LCHF diets without affecting circulating cholesterol levels.Finally, we report the cholesterol high-converter type to be associated with adverse circulating lipid profiles in metabolically healthy lean individuals (increased triacylglycerides [TAG] and decreased high-density lipoprotein cholesterol [HDL-C]), who also respond to ketogenic LCHF with increased LDL-C levels.Thus, while the cholesterol high-converter type may not be associated with reduced serum cholesterol levels, it may be indicative in metabolically healthy lean individuals of adverse blood lipid profiles and increased LDL-C responses to LCHF diets.
To compare cholesterol high and low-converter type fractions among individuals from independent cohorts and with different metabolic health backgrounds, fecal sterol and stanol concentrations were also characterized in 145 individuals with obesity from the CARBFUNC study, a Norwegian 2-year randomized controlled dietary intervention trial (ClinicalTrials.gov:NCT03401970).The main characteristics of both study cohorts are listed in Table S1.Individuals with obesity from the CARBFUNC study exhibited increased concentrations of fecal coprostanol (Table S1, p < 1e-3, Wilcoxon rank-sum test), but otherwise similar sterol or stanol levels (p > 0.05) and compositional profiles (Figure 1E), including a negative correlation of coprostanol to cholesterol (R = À0.47,q < 1e-9) and a bimodal distribution into cholesterol high (61% of CARBFUNC study participants) and low (21% of CARBFUNC study participants) converter types based on the fecal coprostanol/cholesterol ratios (Figure 1F).Stratification of individuals into larger cholesterol high-converter and smaller cholesterol low-converter type fractions therefore appears to be a conserved feature of the human fecal microbiome in individuals with and without obesity.

Distinct microbiota associations with fecal cholesterol and coprostanol
To better understand the intra-intestinal relationship of cholesterol-to-coprostanol conversion with the gut microbiota, fecal taxonomic microbiota, and metabolite profiles were compared between cholesterol high and low-converters from the KETO and CARBFUNC studies.Fecal samples from both cohorts were independently analyzed using different 16S rRNA gene amplicon sequencing protocols, presenting a potential confounding factor for the direct comparison of taxonomic microbiota profiles from both cohorts but also allowing for the identification of robust associations with cholesterol conversion.
Next, a generalized linear mixed model (GLMM) was used to identify shared linear associations between specific members of the fecal microbiota and fecal cholesterol and coprostanol concentrations across both the KETO and CARBFUNC studies combined (Table S2).Whereas coprostanol detected in feces should originate exclusively from microbial cholesterol reduction, fecal cholesterol could have both endogenous and exogenous origins. 2 Fecal coprostanol and cholesterol concentrations were therefore queried for associations with the centered log-ratio (clr)-transformed relative abundances of specific bacterial taxa both independently and in combination, while controlling in the model for cohort and gender-specific effects (Figures 2C and S1, and Table S2).Significant associations from the GLMMs (q < 0.1) were independently assessed by Spearman's rank correlation analysis (Figures 2D, 2E, and S1).Only two bacterial taxa, i.e., E. coprostanoligenes.group and Ruminococcaceae.UCG.014,showed consistent associations with cholesterol-to-coprostanol conversion across all comparisons and analyses, i.e., positive and negative correlations to fecal coprostanol and cholesterol levels, respectively (Figures 2C, 2D, and S1, Spearman's rank correlation FDR ), and a positive association with the coprostanol/cholesterol ratio (Figure S1).In addition, Lachnoclostridium was positively and Clostridiales.XIII.AD3011 negatively correlated to fecal cholesterol concentrations (Figures 2E and S1, Spearman's rank correlation FDR ).Detection of Clostridiales.XIII.AD3011 was limited to the CARBFUNC cohort (Figure S1), indicating obesity, technical or otherwise study-related link of this genus to fecal cholesterol.
To also test for non-linear microbiota associations with the cholesterol high and low converter types, a random forest classifier was trained on clr-transformed microbiota compositions.This classifier performed well at identifying cholesterol high-converters (84.82% precision, 94.06% recall), but lacked sensitivity for the detection of low-converters (76.92% precision, 54.05% recall).In line with the GLMM, leaveone-out cross-validation (LOOCV) identified E. coprostanoligenes.group as the most important bacterial taxon for this classification, whereas several others detected by LOOCV, such as Christensenellaceae.R.7.group, were only correlated to fecal cholesterol levels by the GLMM and flagged with singularity fit warnings, indicating potential overfit of the linear model (Figures 2F and 2G).In summary, linear and non-linear models consistently linked E. coprostanoligenes to intestinal cholesterol-to-coprostanol conversion across both cohorts.However, other bacterial taxa may also be involved in the process.
Next, fecal cholesterol and coprostanol levels were compared to short-chain fatty acid (SCFAs) and branched-chain fatty acid (BCFAs) concentrations in stool samples.SCFAs are mainly produced by microbial fermentation of non-digestible dietary fiber in the colon, whereas BCFAs predominantly result from microbial protein fermentation (Wolter et al. 33 ).Fecal cholesterol but not coprostanol was positively correlated across both studies to the SCFAs acetate (R = 0.38, q < 1e-4), propionate (R = 0.52, q < 1e-9) and butyrate (R = 0.33, q < 1e-3) (Figure 3A, Spearman's rank correlation FDR ).In contrast, fecal coprostanol but not cholesterol, showed a positive correlation to the BCFA isobutyrate (Figure 3B, R = 0.25, q = 0.005, Spearman's rank correlation FDR ).An association of the cholesterol converter type with fecal SCFA or BCFA levels was only identified in study participants with obesity (CARBFUNC), including decreased fecal propionate (p < 1e-4) and increased isobutyrate (p = 0.023) concentrations in high-converters (Wilcoxon rank-sum test).Thus, our findings are in agreement with previous reports of increased SCFA secretion in cholesterol low-converters. 34They indicate distinct intra-intestinal associations of cholesterol and coprostanol with specific microbial taxa and metabolites, which may be influenced by obesity or other cohortspecific parameters.

Circulating blood lipids in cholesterol high and low-converters
To determine if intestinal cholesterol-to-coprostanol conversion types were reflected in altered circulating cholesterol levels, we compared fecal cholesterol and coprostanol concentrations with the serum levels of total cholesterol and other lipids.Lean cholesterol high and lowconverters showed comparable total cholesterol and LDL-C concentrations before the intervention (Figure 4A, KETO cohort, p > 0.05, Wilcoxon rank-sum test), but TAG levels were increased (p = 0.04) and HDL-C levels decreased (p = 0.04) in lean cholesterol high-converters (Figure 4A, Wilcoxon rank-sum test).Cholesterol high and low-converters with obesity showed no difference in serum total cholesterol, TAG, HDL-C, and LDL-C (Figure 4B, CARBFUNC cohort, p > 0.05, Wilcoxon rank-sum test).However, CARBFUNC study participants had increased serum TAG (p < 1e-7), LDL-C (p = 0.03) and reduced serum HDL-C (p < 1e-13) levels compared to KETO study participants and showed an increased TAG/HDL-C ratio (p < 1e-11), a marker for insulin resistance, 35 consistent with generally adverse health profiles in the cohort with obesity (Table S1).In line with this notion, CARBFUNC study participants had increased blood glucose and insulin levels (Table S1), but no difference was observed for either of these parameters between cholesterol high and low-converters from both studies (KETO: blood glucose p = 0.092, insulin p = 0.77; CARBFUNC: blood glucose p = 0.66, insulin p = 0.14, Wilcoxon rank-sum test), and neither glucose nor insulin were significantly correlated with fecal coprostanol or cholesterol (q > 0.05, Spearman's rank correlation FDR , Figure S2).However, compared to cholesterol low-converters, high-converters from the CARBFUNC study had elevated serum b-hydroxybutyric acid (BHB) levels (low: 39.2 G 33.8 vs. high: 61.6 G 59.2, p = 0.013, mean G sd, Wilcoxon rank-sum Test), indicative of increased ketosis in the cholesterol high-converters with obesity.
Neither fecal coprostanol nor cholesterol levels were significantly correlated with serum total cholesterol, TAG, LDL-C, or HDL-C in either the CARBFUNC or KETO cohorts, or individuals from both studies combined (Figure S2, q > 0.05, Spearman's rank correlation FDR ).Similarly, no significant associations between fecal microbial taxa and serum lipid levels were identified by the GLMM after false discovery rate correction (Table S2, q > 0.1).Our findings therefore provide no indication for a link between increased intestinal cholesterol-to-coprostanol conversion and reduced circulating cholesterol levels.(G) Increased relative abundance of E. coprostanoligenes.group and Christensenellaceae.R.7.group in cholesterol high-converters.Dashed lines indicate pseudocount values (0.0001% relative abundance) of samples with zero taxon counts.Benjamini-Hochberg (BH) corrected: q > 0.1 ns, q < 0.1 *, q < 0.05 **, q < 0.01 ***, q < 0.001 ****.Pooled data are represented as mean G SD.

Article Diet impact on cholesterol-to-coprostanol conversion
To identify associations between dietary habits and cholesterol-to-coprostanol conversion, lean KETO study participants were compared based on available food frequency questionnaire data.Caloric intake from fats, fiber, carbohydrates, protein, or cholesterol was comparable between lean cholesterol high and low-converters (Table 1, p > 0.05, Wilcoxon rank-sum test) and both converter types exhibited similar fecal fatty acid profiles, in terms of chain length and saturation level (Figure S3, p > 0.05, Wilcoxon rank-sum test).To estimate the ratio of animal to plant-derived fat intake, fecal coprostanol and stigmastanol levels were compared, as the former should originate mostly from conversion of cholesterol from animal origins (besides endogenous sources) and the latter mostly from conversion of the phytosterol b-sitosterol from plant origins. 2 While no difference was detected between lean high and low-converters (Figure 5A, KETO cohort, p > 0.05, Wilcoxon rank-sum test), the fecal coprostanol/stigmastanol ratio was increased in cholesterol high-converters with obesity (Figure 5A, CARBFUNC cohort, p < 1e-5, Wilcoxon rank-sum test), indicating that, compared to low-converters, high-converters with obesity obtained a larger fraction of their fat intake from animal sources.
Both the KETO and CARBFUNC studies involved interventions with low-carbohydrate high-fat (LCHF) diets, based on R75 energy percent [E%] fat and %10 E% carbohydrate intake.Lean KETO study participants followed a 6-week ad libitum LCHF diet, which resulted in increased urinary and blood ketone bodies, as well as other hormonal and metabolic changes indicative of ketosis, as previously described elsewhere in detail. 32,36CARBFUNC study participants with obesity were restricted to a normocaloric (males: 2,500 kcal, females: 2,000 kcal) LCHF diet (https://clinicaltrials.gov/ct2/show/NCT03401970), which was accompanied by at least transient ketosis at three months of the intervention A B Figure 3. Distinct associations of fecal cholesterol and coprostanol concentrations with short and branched-chain fatty acids (A and B) Positive correlation of fecal cholesterol with the concentrations of the SCFAs acetate, propionate and butyrate (A) and of fecal coprostanol with the BCFA isobutyrate (B) in fecal samples from lean KETO study participants (N = 28) and in individuals with obesity from the CARBFUNC study (N = 145) before the dietary intervention.Spearman's rank correlation, BH-corrected: q > 0.05 ns, q < 0.05 *, q < 0.01 **, q < 0.001 ***.based on increased serum BHB levels (PRE: 62.68 mM +/À 68.67 vs. 3 months: 264.01 mM +/À 251.74, p = 0.00031; PRE vs. 6 months: 163.26 G 274.60, p = 0.26; mean G SD, paired Wilcoxon signed-rank test).Fecal and serum samples were collected after six weeks (KETO study) and three and six months (CARBFUNC study) and used for microbiota analysis and lipid profiling.As both studies included individuals with different metabolic health backgrounds and involved variable time spans, we first tested whether comparable taxonomic compositional microbiota alterations could be detected in both cohorts after LCHF diet intervention.A strong and consistent shift in microbiota compositions was detected across both cohorts (Figure 5B; Table S2, GLMM), including changes in the relative abundance of bacterial genera, such as Bifidobacterium (mean reduction: À4.23% +/À 2.48), previously reported to be altered by ketogenic diet. 37Our microbiota analysis therefore demonstrates reproducible, temporally stable LCHF diet-induced compositional microbiota changes in individuals with and without obesity.
To test whether intestinal cholesterol-to-coprostanol conversion could be dietarily modulated, we compared fecal cholesterol and coprostanol levels in KETO and CARBFUNC study participants in response to the LCHF diets using GLMMs (Figure 5C).These models identified cohort-specific effects of the LCHF diets on both fecal cholesterol and coprostanol concentrations but not their ratios (Figure 5D; Table S2).In cholesterol low-converters from both cohorts, LCHF diet increased the cholesterol-to-coprostanol conversion, as illustrated by decreased fecal cholesterol (q < 1e-7), increased fecal coprostanol (q < 1e-3) levels, and increased coprostanol/cholesterol (q < 1e-9) ratios (Figure 5D; Table S2, GLMM-based estimated marginal means [EMM]).The increased cholesterol conversion in low-converters on the LCHF diets was accompanied by a higher fecal relative abundance of E. coprostanoligenes (q = 0.003, Figure 5E; Table S2, GLMM-based EMMs).Cholesterol high-converters responded to the LCHF diets with reduced cholesterol conversion, at least based on increased fecal cholesterol (q = 0.08) concentrations and a decreased coprostanol/cholesterol ratio (q = 0.03), although fecal coprostanol levels were not altered (q > 0.1) (Figure 5F; Table S2, GLMM-based EMMs).No difference was detected in serum BHB levels between cholesterol high and low-converters from both studies (p = 0.067, paired Wilcoxon signed-rank test), nor positive associations between ketosis and cholesterol-to-coprostanol conversion (D BHB vs. D coprostanol, R = 0.17, p = 0.39, Spearman's rank correlation), indicating that the LCHF-induced increase in cholesterol conversion in previous low-converters was not ketosis-dependent.
Thus, LCHF diets consistently increased cholesterol-to-coprostanol conversion in low-converters from the KETO and CARBFUNC cohorts, despite different metabolic health backgrounds and underlying fecal cholesterol and coprostanol concentrations.

Cholesterol converter type-specific dietary impact on serum lipids
To determine if the cholesterol converter type affected serum lipid responses to LCHF diets, total cholesterol, TAG, HDL-C, and LDL-C concentrations were compared in high and low-converters with and without obesity.The LCHF diet-induced increase in cholesterol conversion in previous low-converters from both cohorts was not accompanied by altered blood lipid concentrations (Figure 6A, q > 0.1, GLMM-based EMMs).However, both cholesterol high and low-converters responded to the LCHF diets with a reduction in serum TAG levels (Figure 6B, q [low] = 0.081, q[high] < 1e-3).This cholesterol converter type-independent effect was apparent even when controlling for cohort-specific differences in serum lipids (q < 0.1, Table S2).However, cholesterol high-converters from the lean KETO cohort responded to the LCHF diet with increased serum LDL-C levels (Figure 6C; Table S2, q = 0.015, GLMM-based EMMs).This effect was not explained by different total fat or saturated, mono-unsaturated, or poly-unsaturated fatty acid intake (E%) between lean cholesterol high and low-converters (Table 1, p > 0.05, Wilcoxon rank-sum test).Lean cholesterol high-converters from the KETO study even consumed less cholesterol during the LCHF diet than low-converters (Table 1, p = 0.016, Wilcoxon rank-sum test).Neither cholesterol high nor low-converters with obesity from the CARBFUNC study exhibited increased serum LDL-C concentrations on the LCHF diet (Table S2, q > 0.05), despite increased saturated fatty acid consumption (30 %E) during the intervention (https://clinicaltrials.gov/ct2/show/NCT03401970), which has previously been suggested to increase LDL-C levels. 38n summary, the LCHF diet-induced increase in intestinal cholesterol conversion had no discernible effect on circulating serum lipids but in lean individuals, the cholesterol high-converter type was associated with increased LDL-C levels in response to LCHF diet, independently of fat, cholesterol, and saturated fatty acid intake.

DISCUSSION
The prospect of reducing cholesterol availability via microbial conversion of endogenous and exogenous cholesterol to non-absorbable coprostanol in the intestine is conceptually appealing as a potential non-pharmacological treatment option for hypercholesterolemia. 9 However, associations between intestinal and serum cholesterol levels in relation to the gut microbiota, dietary habits, and the metabolic state of the human host, have not been fully resolved.To gain new insights into these relationships and identify robust associations, the present study compared data from two vastly different and geographically independent human cohorts (KETO, CARBFUNC), which are characterized by distinct metabolic backgrounds (individuals with/without obesity), low-carb/high-fat diet interventions (ketogenic/non-ketogenic LCHF diets) and microbiota analysis methods (variable protocols for fecal metagenomic DNA isolation and 16S rRNA gene amplification).
The stratification of healthy human populations into reproducible fractions of high and low cholesterol-to-coprostanol converters has previously been documented for different European countries, 39,40 to be stable over the course of several days, 41 and to shift toward decreased fractions of cholesterol low-converters among older males. 39,40The present study expands on these findings by demonstrating cholesterol conversion to represent an obesity-independent organizational feature of the human fecal microbiome, characterized by conserved proportions of cholesterol high and low-converters, despite adverse metabolic health profiles in CARBFUNC study participants with obesity compared to metabolically healthy lean KETO study participants (increased serum TAG and LDL-C, and decreased HDL-C levels).Cholesterol converting activity has been attributed to a broad and taxonomically diverse group of microbes, including isolates from the genera Bacteroides, 42 Bifidobacterium, 43 Eubacterium, 44 and Lactobacillus. 45However, most of these bacterial isolates, besides the well-described E. coprostanoligenes strain ATCC 51222, 44,46,47 have been lost 2 and other isolates of the same species failed to show similar activities.Moreover, available genome sequences from these species tested negative for the presence of homologs to the recently identified cholesterol Data were collected over seven consecutive days 1-2 weeks before (PRE) and during the last week of dietary intervention (LCHF) and used to calculate average daily intakes (+/À standard deviations [SD]) and determine significant differences between before and on LCHF time points (paired Wilcoxon signed-rank test) and between high/low-converter types (Wilcoxon rank-sum), N(low) = 6, N(high) = 17, p < 0.05 *, p < 0.01 **, p < 0.001 ***, Abbreviations: SFA = saturated fats, MUFA = mono-unsaturated fats, PUFA = poly-unsaturated fats.(B) LCHF diets induced consistent taxonomic microbiota alterations in the KETO and CARBFUNC cohorts, based on a combined GLMM analysis (N PRE = 173, N LCHF = 62).Bacterial taxa with significant changes in relative abundance (q < 0.1, BH-corrected Tukey's test, see horizontal blue line) and a positive or negative dehydrogenase-encoding ismA gene from E. coprostanoligenes, 18 suggesting strain-specific variations in the ability for cholesterol reduction.In this study, in spite of the heterogeneous background of the KETO and CARBFUNC cohort data, cholesterol-to-coprostanol conversion showed a robust positive association with the fecal relative abundance of E. coprostanoligenes.Possibly, other bacterial taxa, including Ruminococcaceae.UCG.014 and Lachnoclostridium, which positively correlated with fecal coprostanol and cholesterol levels, respectively, also contribute to intestinal cholesterol conversion and not all E. coprostanoligenes strains may possess the ability to reduce cholesterol.However, the detection of E. coprostanoligenes in most individuals (97.6%) and samples (96%) from both cohorts suggests the capacity for cholesterolto-coprostanol conversion to represent a common and widespread trait of the human gut microbiome, at least in the studied populations.In consequence, cholesterol conversion may be more dependent on intestinal cholesterol availability or diet than on microbiota-related factors, such as the presence of ismA-encoding bacteria. 18cross the two studied cohorts, intestinal cholesterol-to-coprostanol conversion increased in previous low-converters after transition to LCHF diet, independently of the underlying metabolic background (with/without obesity) or the degree of LCHF diet-induced ketosis.This finding demonstrates the feasibility of modulating intestinal cholesterol conversion via dietary intervention, with potential consequences for intestinal cholesterol availability and absorption.However, there is only limited experimental evidence to support a cholesterol-reducing effect of intestinal cholesterol-to-coprostanol on the host.An inverse relationship between serum cholesterol and the fecal coprostanol/ cholesterol ratio was reported in a cohort of 13 hospitalized patients in Japan. 48Germ-free rats 49 and antibiotically treated spontaneously hypercholesterolemic ApoE À/À mice 50 exhibited elevated serum cholesterol levels.Similarly, fecal microbiota transplantation from humans with excess serum cholesterol to microbiota-depleted ApoE À/À mice produced similar effects, 50 whereas feeding rabbits with cholesterolconverting E. coprostanoligenes reduced circulating cholesterol levels. 51However, similar effects could not be induced by feeding E. coprostanoligenes to laying hens 52 or germ-free mice, 47 despite at least transient intestinal colonization and cholesterol-to-coprostanol activity.Associations of human cholesterol metabolism and gut microbiome, which have previously been reported, 53 could be mediated by other mechanisms than intestinal cholesterol-to-coprostanol conversion. 9For instance, the microbiota is a major determinant of intestinal (primary and secondary, conjugated and unconjugated) bile acid metabolism, 54 which represents a major route for cholesterol excretion from the body. 55Dysbiosis of the gut microbiome could therefore be reflective of a dysregulated human cholesterol or bile acid metabolisms.In the present study, cholesterol high compared to low-converters showed no reduction in circulating cholesterol levels (total cholesterol, LDL-C, HDL-C) and no reduction in serum cholesterol levels was observed in previous low-converters that increased cholesterol conversion in response to LCHF diet.Thus, findings from this study indicate that increasing intestinal cholesterol-to-coprostanol conversion in low-converters with LCHF diet is insufficient as a therapeutic strategy to reduce circulating cholesterol levels.
LCHF diets can induce elevated LDL-C levels in a subset of lean individuals, an observation termed the lean mass hyper-responder (LMHR) phenotype. 56The clinical relevance of this phenomenon, which has been controversially debated 30 and explained with homeostatic adaptations to increased dietary fat intake, 57 remains incompletely understood.However, the lean mass hyper-responder phenotype received widespread attention on social media and outside of the scientific community, 58 where LCHF diets continue to be popular among healthy normalweight individuals. 25Therefore, there is a need to identify individuals that are affected by the increased LDL-C response to LCHF diets and to better understand the responsible mechanisms and associated cardiovascular disease risks.In the present study, cholesterol high-converters from the KETO cohort responded to the ketogenic LCHF diet intervention with an increase in serum LDL-C levels.These individuals exhibited increased serum TAG (87.65 mg/dL +/À 47.42, mean G SD) and decreased HDL-C (69.65 mg/dL+/À 11.46, mean G SD) before the intervention.While these concentrations fall within the ranges recommended as optimal by the U.S. Centers for Disease Control and Prevention, based on Grundy et al., 59 they may indicate a subclinical, adverse metabolic risk profile of the cholesterol high compared to the low-converter types from the lean KETO cohort.In addition, LDL-C levels in lean high-converters on the ketogenic LCHF diet (119 mg/dL +/À 36.05,mean G SD) stayed below those that were originally described for the lean mass hyper-responder phenotype (LDL-C R200 mg/dL 58 ).The clinical importance of the observed association between LDL-C and cholesterol conversion should therefore be independently confirmed and individuals with the lean mass hyper-responder LDL-C phenotype tested for cholesterol conversion, in order to assess the diagnostic value of this microbiome trait for personalized response predictions to LCHF diets.
In summary, our findings are in agreement with a model that explains individual intestinal cholesterol-to-coprostanol converter types as a result of long-term dietary or other habits.They do not indicate a hypocholesterolemic effect of intestinal microbial cholesterol (D) Increased cholesterol-to-coprostanol conversion in low-converters from both cohorts on the LCHF diets (N PRE = 37, N LCHF = 11), as evidenced by reduced fecal cholesterol and increased fecal coprostanol levels and increased coprostanol/cholesterol ratios.(E) The increased cholesterol-to-coprostanol conversion in low-converters on LCHF diets was accompanied by an increased fecal relative abundance of E. coprostanoligenes.group(N PRE, LOW = 37, N LCHF, LOW = 11), resulting in similar relative abundances in high and low-converters on the LCHF diets (N PRE, HIGH = 106, N LCHF, HIGH = 43).(F) Decreased cholesterol-to-coprostanol conversion in high-converters from both cohorts after LCHF diet intervention (N PRE = 106, N LCHF = 43), at least based on increased fecal cholesterol levels and an increased coprostanol/cholesterol ratio.Individuals were classified as cholesterol high/low-converters based on preintervention time points, with symbol colors indicating the classification during the LCHF diet.Significance determined by GLMM and post-hoc Tukey's test (Benjamini-Hochberg-corrected): q > 0.1 ns, q < 0.1 *, q < 0.05 **, q < 0.01 ***, q < 0.001 ****.Pooled data are represented as mean G SD. conversion but point to an adverse response of increased LDL-C levels to ketogenic diet in lean cholesterol high-converters.Our data suggest potential relevance of the cholesterol converter type as a personalized microbiome marker for metabolic health and response to dietary intervention.

Limitations of the study
The present study has several limitations: The total numbers of individuals in both cohorts, especially of cholesterol low-converters that represented minor fractions compared to high-converters, were small, providing limited statistical support for comparative analyses among cohort subgroups.In addition, as laid out previously, the two studied cohorts varied with respect to several parameters, including the metabolic background of study participants, the type of diet intervention and the microbiota analysis methods.The diet interventions also involved different durations of six weeks (KETO) and six months (CARBFUNC), although for the latter cohort no difference was detected in fecal sterol and stanol concentrations and serum lipid levels between the three and six-month time points (p > 0.05, Wilcoxon ranked-sum test), attesting to the temporal robustness of the reported findings.Moreover, the detection of shared features between the studied cohorts, such as conserved fractions of cholesterol high/low-converters, shared taxonomic microbiota correlations and dietinduced alterations shifts in cholesterol conversion, demonstrates the strength of the identified associations.It is possible, however, that additional, subgroup-specific associations of the cholesterol converter type with fecal microbiota or serum lipid profiles could not Significance determined by GLMM and post-hoc Tukey's test (Benjamini-Hochberg-corrected): q > 0.1 ns, q < 0.1 *, q < 0.05 **, q < 0.01 ***, q < 0.001 ****.Pooled data are represented as mean G SD.

Figure 1 .Figure 2 .
Figure 1.Equal distributions of cholesterol high and low-converter types among humans with and without obesity (A) Sterol and stanol concentrations as determined by LC-MS/HRMS in 28 fecal samples from the KETO study participants before dietary intervention.(B) Negative correlation (Spearman's rank) between fecal coprostanol and cholesterol concentrations and bimodal distribution of cholesterol high (N = 17), intermediate (N = 4) and low-converter (N = 7) types, as classified based on the fecal coprostanol/cholesterol ratio.(C and D) Comparable negative correlations (Spearman's rank) between the fecal concentrations of the phytosterols sitosterol and campesterol and the corresponding stanol conversion products stigmastanol (n = 26), and 5b-campestanol (n = 23).(E) Similar fecal sterol and stanol concentration profiles in individuals with obesity from the CARBFUNC study before dietary intervention (n = 145 samples), compared to lean KETO study participants (N = 89/26/30 for cholesterol high/intermediate/low-converters).(F)Negative correlation (Spearman's rank) of fecal coprostanol and cholesterol concentrations in CARBFUNC study participants and bimodal distribution into high and low-converter types.Spearman's rank correlation, Benjamini-Hochberg (BH) corrected: q > 0.05 ns, q < 0.01 **, q < 0.001 ***.Pooled data are represented as mean G SD.

Figure 4 .
Figure 4. Circulating blood lipids in cholesterol high and low-converters (A) Comparable total cholesterol, but increased serum TAG and decreased HDL-C levels in lean cholesterol low (N = 7) compared to high-converters (N = 17) from the KETO study.(B) No significant difference in blood lipid levels between cholesterol high (N = 89) and low-converters (N = 30) with obesity from the CARBFUNC study.Wilcoxon rank-sum, p > 0.05 ns, p < 0.05 *.Pooled data are represented as mean G SD.

Figure 5 .
Figure 5. Continued fold-change of > 0.25 in estimated marginal means (EMM) are marked with red dots and labels, unless they were detected by the GLMM as cohort and/or sexassociated (black dots).(C) Distribution of cholesterol high and low-converters among all KETO and CARBFUNC study participants before (N = 143) and after (N = 54) LCHF dietary intervention (gray lines connecting pre and post-intervention samples).(D) Increased cholesterol-to-coprostanol conversion in low-converters from both cohorts on the LCHF diets (N PRE = 37, N LCHF = 11), as evidenced by reduced fecal cholesterol and increased fecal coprostanol levels and increased coprostanol/cholesterol ratios.(E) The increased cholesterol-to-coprostanol conversion in low-converters on LCHF diets was accompanied by an increased fecal relative abundance of E. coprostanoligenes.group(N PRE, LOW = 37, N LCHF, LOW = 11), resulting in similar relative abundances in high and low-converters on the LCHF diets (N PRE, HIGH = 106, N LCHF, HIGH = 43).(F) Decreased cholesterol-to-coprostanol conversion in high-converters from both cohorts after LCHF diet intervention (N PRE = 106, N LCHF = 43), at least based on increased fecal cholesterol levels and an increased coprostanol/cholesterol ratio.Individuals were classified as cholesterol high/low-converters based on preintervention time points, with symbol colors indicating the classification during the LCHF diet.Significance determined by GLMM and post-hoc Tukey's test (Benjamini-Hochberg-corrected): q > 0.1 ns, q < 0.1 *, q < 0.05 **, q < 0.01 ***, q < 0.001 ****.Pooled data are represented as mean G SD.

Table 1 .
Semi-quantitative food questionnaire-based dietary habits of lean KETO study participants