MMC 8

Clinical and Translational Report
An Integrated Understanding of the Rapid Metabolic

Benefits of a Carbohydrate-Restricted Diet on
Hepatic Steatosis in Humans
Graphical Abstract Authors
Adil Mardinoglu, Hao Wu,
€ckhed,
Elias Bjornson, ..., Fredrik Ba
Marja-Riitta Taskinen, Jan Borén
Correspondence
fredrik.backhed@wlab.gu.se (F.B.),
jan.boren@wlab.gu.se (J.B.)
In Brief
Mardinoglu et al. use multi-omics to
investigate the effects of a carbohydrate-
restricted diet in obese NAFLD patients.
They show that the diet improves liver fat
metabolism, promotes rapid shifts in the
gut microbiota, increases circulating
folate, and upregulates expression of
genes involved in folate-dependent one-
carbon metabolism in the liver.
Highlights
d A low-carbohydrate diet (LCD) improves liver fat metabolism
in NAFLD patients
d The LCD promotes rapid shifts in the gut microbiota

composition of NAFLD patients
d The LCD-induced microbial changes are associated with

increased circulating folate
d The LCD increases folate-dependent one-carbon metabolism

gene expression in liver
Mardinoglu et al., 2018, Cell Metabolism 27, 559–571

March 6, 2018 ª 2018 Elsevier Inc.
https://doi.org/10.1016/j.cmet.2018.01.005
Cell Metabolism
Clinical and Translational Report
An Integrated Understanding of the Rapid

Metabolic Benefits of a Carbohydrate-Restricted
Diet on Hepatic Steatosis in Humans
Adil Mardinoglu,1,2,8 Hao Wu,3,8 Elias Bjornson,2,3 Cheng Zhang,1 Antti Hakkarainen,4 Sari M. Ra € sa
€nen,5 Sunjae Lee,1
€inen,5,6 Sanni Söderlund,5 Niina Matikainen,5,6
Rosellina M. Mancina,3 Mattias Bergentall,3 Kirsi H. Pietila
Marcus Ståhlman,3 Per-Olof Bergh,3 Martin Adiels,3 Brian D. Piening,7 Marit Granér,5 Nina Lundbom,4 Kevin J. Williams,3
Stefano Romeo,3 Jens Nielsen,2 Michael Snyder,7 Mathias Uhlén,1 Göran Bergström,3 Rosie Perkins,3
Hanns-Ulrich Marschall,3 Fredrik Ba €ckhed,3,* Marja-Riitta Taskinen,5 and Jan Borén3,9,*
1Science for Life Laboratory, KTH Royal Institute of Technology, Stockholm, Sweden
2Department of Biology and Biological Engineering, Chalmers University of Technology, Gothenburg, Sweden
3Department of Molecular and Clinical Medicine, University of Gothenburg, and Sahlgrenska University Hospital, Gothenburg, Sweden
4HUS Medical Imaging Center, Radiology, Helsinki University Hospital, University of Helsinki, Helsinki, Finland
5Research Programs Unit, Diabetes and Obesity, University of Helsinki and Department of Internal Medicine, Helsinki University Hospital,
Helsinki, Finland
6Endocrinology, Abdominal Center, Helsinki University Hospital, Helsinki, Finland
7Department of Genetics, Stanford University, Stanford, CA 94305, USA
8These authors contributed equally
9Lead Contact
*Correspondence: fredrik.backhed@wlab.gu.se (F.B.), jan.boren@wlab.gu.se (J.B.)

https://doi.org/10.1016/j.cmet.2018.01.005
SUMMARY (Chalasani et al., 2012; Rinella and Sanyal, 2016). NAFLD can
progress from simple steatosis to non-alcoholic steatohepatitis
A carbohydrate-restricted diet is a widely recom- (NASH), which is characterized by the additional presence of an
mended intervention for non-alcoholic fatty liver inflammatory infiltrate and hepatocellular injury with or without
disease (NAFLD), but a systematic perspective fibrosis (Chalasani et al., 2012; Rinella and Sanyal, 2016), and
on the multiple benefits of this diet is lacking. may further progress to cirrhosis, liver failure, and hepatocellular
Here, we performed a short-term intervention with carcinoma (Marengo et al., 2016). Increasing evidence also
indicates that NAFLD is a significant independent risk factor for
an isocaloric low-carbohydrate diet with increased
cardiovascular disease and type 2 diabetes (Lonardo et al.,
protein content in obese subjects with NAFLD
2015; Targher et al., 2016), and the dyslipidemia that is present
and characterized the resulting alterations in meta- in many individuals with NAFLD potentially contributes to the
bolism and the gut microbiota using a multi-omics link between these diseases (Gaggini et al., 2013).
approach. We observed rapid and dramatic The pathophysiology of NAFLD has not been resolved, but it
reductions of liver fat and other cardiometabolic develops when the influx of lipids into the liver exceeds hepatic
risk factors paralleled by (1) marked decreases in lipid disposal (by fatty acid oxidation and triglyceride secretion
hepatic de novo lipogenesis; (2) large increases in as lipoprotein particles) (Stefan et al., 2008). Potential sources
serum b-hydroxybutyrate concentrations, reflecting of lipids contributing to fatty liver include fatty acids released
increased mitochondrial b-oxidation; and (3) rapid into the circulation from peripheral adipose tissue, dietary fatty
increases in folate-producing Streptococcus and acids from intestinal chylomicrons, and lipids synthesized
(mostly from carbohydrates) in the liver by de novo lipogenesis
serum folate concentrations. Liver transcriptomic
(DNL) (Donnelly et al., 2005). In hyperinsulinemic subjects with
analysis on biopsy samples from a second cohort
NAFLD, hepatic DNL accounts for approximately 25% of liver
revealed downregulation of the fatty acid synthesis triglyceride content (Diraison et al., 2003; Donnelly et al.,
pathway and upregulation of folate-mediated one- 2005); thus, carbohydrate restriction, combined with exercise
carbon metabolism and fatty acid oxidation path- and regular follow-up, has emerged as an effective dietary
ways. Our results highlight the potential of exploring intervention for obesity (Astrup et al., 2004; Foster et al.,
diet-microbiota interactions for treating NAFLD. 2003) and NAFLD (Rinella and Sanyal, 2016). In addition to their
effects on liver fat, carbohydrate-restricted diets have been
shown to promote marked shifts in the composition of the gut
INTRODUCTION microbiota (David et al., 2014; Duncan et al., 2007). Further-
more, accumulating evidence suggests that microbial changes
In the past 30 years, we have seen a marked increase in non- are implicated in the development and progression of NAFLD
alcoholic fatty liver disease (NAFLD), and it is now the most (Le Roy et al., 2013; Leung et al., 2016; Loomba et al., 2017),
common cause of chronic liver disease in western countries and fecal microbiota transplantation has been shown to be
Cell Metabolism 27, 559–571, March 6, 2018 ª 2018 Elsevier Inc. 559
able to alleviate high-fat-induced steatohepatitis in mice 1986) and a strong correlation between decreases in the concen-
(Zhou et al., 2017). tration of apoC-III and VLDL-triglycerides over the study period
A systematic perspective integrating dietary intervention, (r = 0.91, p = 0.0015; Figure S2). We also analyzed the composi-
microbial profiling, and in-depth metabolic characterizations in tion of VLDL-triglycerides and observed a decreased proportion
humans with NAFLD is lacking. Given the complexity of NAFLD of saturated fatty acids including myristic acid (14:0) and palmitic
pathogenesis, in-depth multi-omics profiling—an approach acid (16:0) and an increased proportion of unsaturated fatty
that has recently been used in studies of both human wellness acids such as oleic acid (18:1) (Table S1). These data are
and disease (Chen et al., 2012; Price et al., 2017; Wu et al., consistent with a significant increase in the number of double
2015)—would likely offer valuable insights into understanding bonds per fatty acid chain observed by MRS of the liver over
how a carbohydrate-restricted diet promotes reduced hepatic the study period (Table S1).
steatosis. Here we performed a 2-week intervention with an To test whether the marked reductions in liver fat were linked
isocaloric carbohydrate-restricted diet in obese subjects with to the diet intervention, we performed a follow-up MRS in seven
NAFLD and used multi-omics profiling to investigate how the of the ten participants 1–3 months after completing the interven-
diet and associated changes in the gut microbiota contribute tion study and returning to their normal diet. We observed that
to improved liver fat metabolism. In addition, we combined their liver fat content returned to a level similar to that measured
plasma metabolomics and liver transcriptomics in a genome- before the diet intervention (11.3% ± 1.6% at follow-up versus
scale metabolic model to further investigate the metabolic 13.8% ± 2.5% at baseline, p = 0.08).
responses to this diet intervention.
Reduced Carbohydrate Consumption Improves Liver Fat
RESULTS AND DISCUSSION Metabolism
To investigate the effects of the diet on processes linked to fat
Reduced Carbohydrate Consumption Has Rapid Effects accumulation in the liver, we measured DNL using stable isotope
on Liver Fat technology and b-hydroxybutyrate (a proxy for b-oxidation). We
Earlier studies designed to reduce hepatic steatosis in subjects observed a rapid and dramatic reduction in absolute DNL (Fig-
with NAFLD have usually combined carbohydrate restriction ure 1H), which was paralleled by a rapid increase in hepatic
with calorie reduction and did not separate out the effect of b-oxidation (Figure 1I). At the end of the 14-day diet intervention,
amplified weight loss (Browning et al., 2006, 2011). To investi- DNL was 79.8% lower and b-hydroxybutyrate was 4.9-fold
gate how liver fat metabolism is affected by reduced carbo- higher than at baseline (Figures 1H and 1I). However, there
hydrate consumption without a concomitant reduction in calorie was no significant change in plasma non-esterified fatty acid
intake, we served a pre-prepared isocaloric low-carbohydrate (NEFA) concentrations over the study period (Table S1). Further-
diet with increased protein content (<30 g of carbohydrates more, there was a strong correlation between the reductions in
and an average of 3,115 kcal per day; Figure 1A; Table S1) DNL and liver fat (r = 0.96, p = 0.0001) and a trend toward a
for 14 days to ten subjects with obesity and high liver fat significant correlation between the reductions in DNL and
(mean ± SEM 16.0% ± 2.3%). The study design is shown in Fig- VLDL-triglycerides (r = 0.67, p = 0.07) (Figure S2). Consistent
ure 1B. To minimize the weight loss that is known to occur on a with improvements in liver function and metabolism, we
short-term isocaloric carbohydrate-restricted diet (Kekwick and observed significant reductions in fasting plasma concentrations
Pawan, 1956), the study subjects were in daily contact with a of alkaline phosphatase and aspartate transaminase, markers of
dietician and were instructed to increase their daily energy intake liver damage, as well as significant decreases in fasting insulin
whenever their weight decreased between two study days by and HOMA-IR (homeostasis model assessment of insulin resis-
more than 0.2 kg. tance) values over the study period (Table S1).
There were no changes in waist circumference during the Studies characterizing the pathogenesis of NAFLD have
study (Table S1). Despite good compliance, we observed a slight focused overwhelmingly on the dysregulation of hepatic DNL
weight loss (1.8% ± 0.2% of their body weight; Figure 1C). Body (Chakravarthy et al., 2005; Fabbrini et al., 2010; Petersen et al.,
composition analysis at baseline and after 14 days on the diet 2007; Wang et al., 2015) and increased delivery of NEFAs from
revealed that decreases in fat mass and water were the major adipose tissue to the liver especially in insulin-resistant subjects
contributors to this minor weight loss (Figure 1D). In contrast to (Lewis et al., 2002); according to one estimation (Donnelly et al.,
the small reduction in weight loss, we observed dramatic reduc- 2005), these pathways account for 25% and 60% of liver fat,
tions in liver fat, as measured by magnetic resonance spectros- respectively. In contrast, the roles of the pathways involved in
copy (MRS), in all the individuals over the 14-day study period hepatic lipid disposal (i.e., mitochondrial b-oxidation and
(mean reduction 43.8%; Figure 1E). Of note, the reduction VLDL-triglyceride export) in NAFLD pathogenesis are less char-
was significant just 1 day after the start of the diet intervention acterized. By restricting carbohydrate intake and thus blunting
(p = 0.027) and was paralleled by a significant decrease in total DNL, we observed a dramatic reduction of liver fat paralleled
liver volume (Figure S1). by a rapid increase of mitochondrial b-oxidation and a marked
We observed marked reductions in very-low-density lipo- reduction of VLDL-triglycerides without any change in plasma
protein (VLDL)-triglycerides (mean reduction 56.7%; Figure 1F) NEFA concentrations. Thus, the reduced hepatic lipid accumula-
and in fasting plasma triglyceride concentrations (mean reduction tion could be explained by both blunted DNL and enhanced
48.4%; Table S1) at the end of the study. In addition, we found a mitochondrial b-oxidation; however, as their regulation is tightly
significant reduction of plasma apolipoprotein C-III (apoC-III) connected (McGarry et al., 1977) it is difficult to estimate the rela-
(Figure 1G), an inhibitor of VLDL clearance (Ginsberg et al., tive contribution of these two sources in our study.
560 Cell Metabolism 27, 559–571, March 6, 2018

A B
C D E
p p
F G H I
p p p
Figure 1. Reduced Carbohydrate Consumption Improves Liver Lipid Metabolism and Reduces Inflammation in Obese Subjects with NAFLD
(A) Percentage of energy from carbohydrate, protein, and fat at baseline (inner circle) and during the 14-day dietary intervention (outer circle).
(B) Study design indicating time points used in the multi-omics analysis. D, day.
(C–E) Boxplots (with median) and individual data showing (C) weight changes across the study period, (D) changes in body composition between day 0 and day
14, and (E) liver fat changes across the study period.
(F–I) Boxplots (with median) showing (F) plasma concentrations of VLDL-triglycerides (TG) (n = 10), (G) plasma concentrations of apoC-III (n = 10), (H) DNL (n = 9),
and (I) plasma concentrations of b-hydroxybutyrate (n = 10) at days 0, 3, and 14.
(J) Heatmap showing statistically significant reductions in inflammatory markers and FGF-21 over the study period (FDR < 0.05; n = 10). Post hoc group
comparisons (D0 versus D3 and D0 versus D14) were performed by paired t test with Bonferroni correction. *p < 0.05; +p < 0.01; #p < 0.001. p or FDR values
across time were obtained by one-way ANOVA with repeated measurements.
See also Figures S1 and S2 and Table S1.
Cell Metabolism 27, 559–571, March 6, 2018 561

Reduced Carbohydrate Consumption Decreases cantly altered strains confirmed the dramatic compositional
Inflammatory Markers and FGF21 shifts of the gut microbiome after 1 day (Adonis test; p = 0.049;
NAFLD has been linked to a low-grade systemic inflammation Figure 2B). These observations are consistent with earlier
(Asrih and Jornayvaz, 2013; Braunersreuther et al., 2012), and studies that reported rapid alterations in both microbial compo-
high carbohydrate intake has been shown to associate with sition (David et al., 2014; Wu et al., 2011) and bacterial
higher inflammation (Solga et al., 2004). We therefore investigated growth rate (Korem et al., 2015) in response to extreme dietary
the effect of the dietary intervention on a range of inflammation- interventions. We observed further compositional shifts after
related plasma proteins. We observed significant reductions in 7 days, but not between 7 and 14 days (Figure 2B), suggesting
many of the inflammatory markers over the study period (Fig- that the compositional shifts stabilized after 1 week on the low-
ure 1J). In particular, we showed that plasma concentrations of carbohydrate diet with increased protein content. Among the
interleukin-6 (IL-6) and tumor necrosis factor alpha (TNF-a) top ten most abundant of 25 significantly altered genera, only
were significantly reduced after 14 days of the carbohydrate- Streptococcus, Lactococcus, and Eggerthella were increased
restricted diet (Figure 1J). IL-6 and TNF-a have attracted partic- over the study period (Figure 2C; Table S2). The carbohydrate-
ular attention because the severity of NAFLD and progression degrading bacteria Ruminoccocus, Eubacterium, Clostridium,
to NASH are associated with higher systemic levels of these and Bifidobacterium were all decreased (Figure 2C; Table S2).
cytokines (Paredes-Turrubiarte et al., 2016; Wieckowska et al., In parallel, we observed decreased carbohydrate fermentation
2008). However, we did not observe significant reductions in as shown by reductions in fecal concentrations of short-chain
these cytokines after 3 days on the diet, indicating that they are fatty acids (SCFAs) (Figure 2D), in agreement with previous
not the main drivers of liver fat. studies (David et al., 2014; Duncan et al., 2007).
Several studies have shown that concentrations of the peptide
hormone fibroblast growth factor 21 (FGF21) are elevated in Diet-Induced Functional Shifts in the Gut Microbiota
subjects with NAFLD and correlate with hepatic fat content Increase Folate Production
(Dushay et al., 2010; Li et al., 2010), and plasma FGF21 has To further investigate functional changes in the gut microbiome
been suggested as a potential diagnostic marker of NAFLD in response to the dietary intervention, we annotated genes to
(Li et al., 2013). Consistent with these findings, we showed that KEGG orthology (KO) (Kanehisa and Goto, 2000). In total, 598
plasma concentrations of FGF21 decreased rapidly in response and 580 KOs were significantly increased and decreased,
to the carbohydrate-restricted diet, with significant reductions respectively, after 14 days on the diet (Figure S3; Table S2).
after both 3 and 14 days on the diet (Figure 1J). Paradoxically, Pathway enrichment analysis demonstrated that these 1,178
plasma FGF21 concentrations in rodents increase in response KOs were involved in 18 KEGG pathways (Figure 3A). Of these,
to low-carbohydrate ketogenic diets (Badman et al., 2007, the most upregulated pathways were those involved in folate
2009; Laeger et al., 2014). One potential explanation for this dif- biosynthesis and the two-component system (which controls
ference is that FGF21 expression in rodents has been shown to gene expression in response to changes in environmental condi-
be sensitive to the protein content of the diet (Laeger et al., 2014). tions in many bacterial species), whereas the pathways involved
Ketogenic diets given to rodents typically contain reduced in starch and sucrose metabolism and biosynthesis of amino
protein levels compared with control diet, in contrast to our acids including branched-chain amino acids (BCAAs; valine,
study diet with increased protein content. An earlier study also leucine, and isoleucine) were significantly decreased (Figure 3A).
reported diet-induced reductions in FGF21 concentrations in Notably, the folate biosynthesis and starch and sucrose meta-
humans upon increased protein intake (Christodoulides et al., bolism pathways were significantly altered after only 1 day of di-
2009). Moreover, plasma concentrations of FGF21 are increased etary intervention (Figure 3A). In-depth analysis of bacterial folate
in individuals who have undergone bariatric surgery (Lips et al., biosynthesis pathways showed that seven of 12 related KOs
2014), a treatment that is commonly associated with protein were significantly increased over the study period (Figure 3B).
malnutrition (Bal et al., 2012). We therefore speculate that the Of note, folate is a common food fermentation product during
protein content in the diet plays important roles in regulating the growth of Streptococcus and Lactococcus (LeBlanc et al.,
human FGF21 expression. 2011), and both of these lactic acid bacteria increased in abun-
dance over the study period (Figure 2C).
Reduced Carbohydrate Consumption Induces Rapid Both the gut microbiota and the diet are sources of serum
Shifts in the Gut Microbiota Composition folate. Although folate is known to be present at higher levels
Because of the increasing evidence supporting a link between in diets rich in carbohydrates (e.g., from vegetables and cereals)
diet, the gut microbiota, and NAFLD (Leung et al., 2016), we (Witthöft et al., 1999), we did not observe any significant
investigated how the gut microbiome was affected by the dietary changes in dietary folate when the participants changed to the
intervention. We performed whole-genome shotgun sequencing carbohydrate-restricted diet (Table S1). Furthermore, in contrast
of fecal samples obtained from the participants at five time to previous studies that reported reduced or no significant
points across the study (0, 1, 3, 7, and 14 days). On average, changes in serum folate levels in response to carbohydrate-
we obtained 17.5 million paired-end reads for each sample restricted diets (Keogh et al., 2008; Noakes et al., 2005; Nuttall
(ranging from 14.2 million to 21.3 million; Table S2). Ninety-four and Gannon, 2006; Wycherley et al., 2010), we observed a signif-
bacterial strains were significantly altered upon dietary interven- icant increase in serum folate concentrations after just 1 day of
tion (Figure 2A; Table S2), and major shifts occurred after only the diet (Figure 3C; p = 0.0039), in parallel with the rapid effect
1 day of the dietary intervention (Figure 2A). Principal coordinate of the diet on liver fat (Figure 1E). This discrepancy between
analysis (PCoA) of the relative abundance of all these signifi- our study and earlier studies may be explained by differences

A
B C D
p
Figure 2. Reduced Carbohydrate Consumption Rapidly Alters Gut Microbial Composition

(A) Heatmap showing significant changes in the abundance of bacterial strains over the 14-day study period. D, day. FDR < 0.05; likelihood ratio test.
(B) PCoA plot based on the relative abundance of all bacterial strains that are significantly altered over the study period at the indicated time points
(mean ± SEM; p D1 versus D0 = 0.049; Adonis test based on 5,000 permutations).
(C) Boxplots (with median) showing the relative abundance of the ten most abundant genera that were significantly altered by the diet over the study period
(n = 10). FDR < 0.05; likelihood ratio test.
(D) Boxplots (with median) showing fecal concentrations of SCFAs at days 0, 1, 3, 7, and 14 (n = 10). p values across time were obtained by one-way ANOVA with
repeated measurements.
See also Table S2.
in the macronutrient composition. Our diet had both increased with increased protein content (Figure S4), suggesting that
protein content compared with the baseline diet and very low bacterially produced folate could contribute to the diet-induced
carbohydrate content (<10% energy from carbohydrates). It is increase in serum folate concentrations.
thus likely that the high dietary protein content combined with Our findings thus indicate that the increased circulating folate
the reduction in carbohydrate-dependent bacteria would drive could be a consequence of the altered microbial composition.
the growth of the folate-producing bacteria Streptococcus and In support of this, previous studies have shown that the gut
Lactococcus, resulting in increased bacterial folate production. microbiota can produce large quantities of folate (Kim et al.,
We also observed a trend toward an increase in fecal folate 2004; Rossi et al., 2011; Strozzi and Mogna, 2008; Sugahara
when the participants changed to the low-carbohydrate diet et al., 2015) and that folate is absorbable across the colon in

A B
C
p
Figure 3. Reduced Carbohydrate Consumption Promotes Microbial Shifts toward Folate Production
(A) KEGG pathway analysis showing pathways that were significantly altered at the indicated times over the 14-day study period (n = 10). D, day. Pathways that
are reduced or increased are shown in blue and brown, respectively. FDR < 0.1; hypergeometric test. The size of the bubble is proportional to the enrichment
score for each KEGG pathway term.
(B) Bacterial folate biosynthesis pathway showing KOs that significantly increased over the study period (brown boxes).
(C) Boxplots (with median) showing changes in serum concentrations of folate changes over the study period (n = 10). p values across time were obtained by
one-way ANOVA with repeated measurements. S, screen.
humans (Aufreiter et al., 2009). We thus provide the first analysis between these two variables for each individual in the
evidence indicating that a low-carbohydrate diet with increased first cohort using data collected at seven time points across
protein content can shift the gut microbiota toward increased the study (0, 1, 3, 4, 7, 10, and 14 days). Despite the limited
production of folate that is potentially utilized by humans. How- number of sampling points, we observed significant correla-
ever, we cannot exclude the possibility that the carbohydrate- tions in five of ten of these individuals and a trend toward signif-
restricted diet promoted enhanced dietary folate absorption in icance in a further two (Figure 4A). No correlation between liver
the small intestine, resulting in increased circulating concentra- fat and serum folate was observed in three individuals (Fig-
tions of folate and favoring the growth of bacteria that produce ure 4A), two of whom had the lowest liver fat in the cohort at
folate. baseline (<10%); a lack of correlation in these two is likely
explained by the fact that their liver fat had reduced to normal
Folate and Associated Metabolites Are Linked to levels (<5%) by day 7 and thus plateaued (Figure 1E). However,
Improved Liver Fat Metabolism by performing a partial regression analysis between liver fat
Animal studies have revealed that folate deficiency increases and serum folate based on a linear mixed-effect model, we
expression of pro-inflammatory cytokines such as IL-6 and observed an overall significant association between liver fat
TNF-a (Kolb and Petrie, 2013) and has detrimental effects on and serum folate (p = 0.0001); according to this model, serum
hepatic lipid metabolism (Akesson et al., 1982), leading to folate could explain 35.4% and 19.5% of the variation in liver
accumulation of liver fat (Kelley et al., 1950) and steatosis fat before and after adjusting for BMI, respectively (Figure 4B).
(Christensen et al., 2010; Halsted et al., 2002). Growing To investigate whether there are other metabolites that could
evidence also indicates a role for folate deficiency in the patho- potentially play a role in improved liver fat metabolism in our
genesis of NAFLD in humans (Hirsch et al., 2005; Xia et al., study, we also performed untargeted metabolomics on plasma
2017). We therefore investigated whether serum folate corre- samples from the participants at baseline and after 3 and
lated with liver fat in our study. We first performed a correlation 14 days on the dietary intervention. We observed significant

A B C
Figure 4. Serum Folate Is Associated with Improved Liver Lipid Metabolism

(A) Pearson correlations between liver fat and serum folate in each individual in the first cohort. D, day.
(B) Linear mixed-effect model (with 95% confidence bands highlighted) between liver fat and serum folate after adjusting for contributions from BMI to both
variables; samples from the same individual are colored as in (A) (n = 10).
(C) Integration of liver fat, folate-producing bacteria, serum folate, and metabolomics data from the first cohort (n = 10) using a multivariate method (mixDIABLO
with repeated measurements) with a correlation cutoff of 0.8. Components are linear combinations of variables from each omics dataset that are maximally
correlated using a full-design matrix.
changes in the relative signal intensity of 202 of 648 detected Liver Transcriptional Changes Reflect Improved Lipid
metabolites over the study period, most of which are involved Metabolism
in lipid or amino acid metabolism (Figure S5; Table S3). We also served the identical isocaloric low-carbohydrate diet with
In particular, we observed significant decreases in metabolites increased protein content to a second cohort of seven subjects
involved in fatty acid synthesis (including those involved in phos- for 7 days. In this second cohort of similar age (p = 0.379) and
pholipid, lysolipid, and diacylglycerol metabolism). Circulating BMI (p = 0.230), we confirmed that the diet reduced DNL, fasting
diacylglycerol concentrations have been repeatedly linked with plasma triglycerides, plasma concentrations of alkaline phos-
insulin resistance (Petersen et al., 2017); thus, reductions in diac- phatase and insulin, and HOMA-IR values (Table S4). Analysis
ylglycerol are consistent with the improved insulin sensitivity of liver biopsies taken from these subjects 7 days before the
observed in this study (Table S1). We also observed significant study start and after 7 days of dietary intervention showed that
increases in metabolites involved in b-oxidation pathways and the diet significantly reduced the hepatic triglycerides by a
plasmalogens, which are known antioxidants (Wallner and mean of 43.9% (Table S4). Notably, we also observed a significant
Schmitz, 2011) (Figure S5; Table S3). Increased oxidative stress increase in serum folate concentrations in this cohort (Table S4).
is considered to contribute to the pathogenesis of NAFLD (Haas To investigate whether alterations in hepatic gene expression
et al., 2016), and antioxidants help to maintain the cellular redox reflected the diet-induced improvements in lipid metabolism, we
balance by neutralizing the oxidative stress produced by fatty also performed a global transcriptomic analysis on the liver sam-
acid oxidation (Gambino et al., 2011), thereby alleviating NAFLD. ples collected from this second cohort. We observed significant
Nicotinamide and its methylation product 1-methylnicotinamide, decreases in 9.1% of genes, whereas only 3.8% of genes
which has been shown to suppress liver triglyceride levels in increased after 7 days on the diet (false discovery rate [FDR]
mice (Hong et al., 2015), were also increased during the dietary adjusted p value < 0.1; Table S5). KEGG pathway enrichment
intervention (Table S3). analysis revealed that 34 pathways were upregulated, including
By using multivariate dimension reduction and discriminant fatty acid degradation, amino acid metabolism, and peroxisome
analysis, we revealed that 48 of the 202 significantly altered proliferator-activated receptor (PPAR) signaling pathways,
metabolites strongly correlated with serum folate (and thus liver whereas only seven pathways were downregulated, such as
fat) as well as with the folate-producing bacteria Streptococcus ribosome and oxidative phosphorylation (Table S5).
and Lactococcus (correlation coefficient cutoff 0.8; Figure 4C; Consistent with the diet-induced reductions in plasma
Table S3). Of note, b-hydroxybutyrate and most of the identified apoC-III concentrations and DNL noted earlier, we observed
plasmalogens associated with folate (Figure 4C; Table S3), decreased hepatic expression of APOC3 and FASN (although
suggesting that folate may improve liver fat metabolism by pro- no significant changes in SREBP-1c) in the second cohort after
moting an increase in mitochondrial b-oxidation and a reduction 7 days on the diet (Table S5). We also observed diet-induced in-
of oxidative stress. In agreement, previous studies have demon- creases in the hepatic expression of PPAR-a and its downstream
strated that folate can confer protective effects against oxidative genes (Figure 5), most of which are known to encode proteins
stress (Sid et al., 2017). However, further efforts are needed involved in the fatty acid oxidation including mitochondrial and
to explore the molecular and cellular mechanisms behind the peroxisomal b-oxidation as well as microsomal u-oxidation (Fig-
multiple potential benefits provided by folate. ure 5) (Kersten, 2014; Pawlak et al., 2015). Intriguingly, both the

Figure 5. Liver Transcriptome Changes Reflect Improved Hepatic Lipid Metabolism
Gene regulatory network for PPARa showing significant increases (salmon) and decreases (light blue) in genes in liver from individuals in the second cohort
after 7 days on a low-carbohydrate diet with increased protein content (n = 7). FDR < 0.1; Wald test with paired design.
See also Table S5.
gut microbiota (Crawford et al., 2009) and folate (Tryndyak et al., MTRR, SHMT1, SHMT2, MTHFD1, and ALDH1L1 (Figure 6A),
2012) have been shown to be able to regulate hepatic lipid meta- and the proton-coupled folate transporter gene SLC46A1 (Table
bolism via a PPAR-a-regulated lipid catabolic pathway. S5). Serine hydroxymethyltransferase (SHMT) catalyzes the
Expression of CAT, the gene encoding catalase, was also up- interconversion of glycine and serine, and serine is required for
regulated in response to the diet (Table S5). Hepatic expression the generation of glutathione (GSH), an antioxidant that also
of the fatty acid transporter gene CD36 was not significantly plays a role in maintaining b-oxidation (Mardinoglu et al., 2017;
changed (Figure 5), in line with the lack of effect of the diet on Nguyen et al., 2013). We have previously shown that individuals
plasma NEFA concentrations in both cohorts (Tables S1 and with NAFLD have serine deficiency (Mardinoglu et al., 2014,
S4). Notably, consistent with the diet-induced increase in serum 2017) and that de novo GSH synthesis is altered in humans
folate concentrations in both cohorts (Figure 3C; Table S4), we with hepatic steatosis (Mardinoglu et al., 2017).
found diet-induced increases in the hepatic expression of genes Finally, we integrated the liver transcriptomics and plasma
involved in folate-mediated one-carbon metabolism, including metabolomics data using iHepatocytes2322, a functional

glycine, serine, and methionine (Ducker and Rabinowitz,
A
2017; Sid et al., 2017).
Study Limitations
There are several strengths with this study, but there are also
limitations. For example, the study lacks a control group,
in which a normal carbohydrate diet was provided to subjects.
In addition, the types of carbohydrate, protein, and fat provided
during the diet intervention were not matched with the diet the
subjects consumed before the study. Therefore, it is possible
that changes in the types of macronutrients (e.g., refined sugars
versus complex carbohydrates, saturated versus unsaturated
fats, plant versus animal proteins) could have affected the
outcome measures.
B
Conclusions
In summary, by using a multi-omics approach, we showed that
short-term intervention with an isocaloric low-carbohydrate
diet with increased protein content promotes multiple
metabolic benefits in obese humans with NAFLD. In particular,
we observed a dramatic reduction of liver fat resulting from a
marked decrease in DNL and increase in b-oxidation. Of note,
our data indicate that rapid microbial shifts toward folate produc-
tion paralleled by increased circulating folate concentrations
may be partially behind the improved lipid metabolism, balanced
oxidative stress, and reduced inflammation. Taken together,
these findings are of important clinical interest in understanding
the pathogenesis and prevention of NAFLD. However, although
studies in mice have shown that folate supplementation protects
against high-fat-diet-induced liver steatosis (Sid et al., 2015,
2017), care should be taken before implementing folate supple-
mentation into clinical practice since long-term effects remain to
Figure 6. Integration of Liver Transcriptomics and Plasma Metabo-
be elucidated.
lomics Data
(A) Reactions involved in folate-mediated one-carbon metabolism showing STAR+METHODS
significant increases (red) and decreases (light blue) in genes in liver from
individuals after 7 days on a low-carbohydrate diet with increased protein Detailed methods are provided in the online version of this paper
content (n = 7). FDR < 0.1; Wald test with paired design.
and include the following:
(B) A genome-wide metabolic model for the liver integrating both serum
metabolomics data and liver transcriptomics data. Red lines indicate
d KEY RESOURCES TABLE
increased fluxes, and blue lines indicate decreased fluxes.
See also Tables S5 and S6.
d CONTACT FOR REAGENT AND RESOURCE SHARING
d EXPERIMENTAL MODEL AND SUBJECT DETAILS
B Clinical Study Design
genome-scale metabolic model of hepatocytes (Mardinoglu B Isocaloric Carbohydrate-Restricted Diet
et al., 2014). As expected, given the diet-induced reduction in B Influence of Gender
DNL, the model predicted a significant reduction of fluxes car- d METHOD DETAILS
ried by the reactions associated with glycolysis, the pentose B Genotyping of PNPLA3 and TM6SF2 Variants
phosphate pathway, and the tricarboxylic acid (TCA) cycle (Fig- B Body Composition, Liver Fat and Liver Volume
ure 6B; Table S6). The decreased fluxes in the reactions B Measurement of Hepatic DNL
involved in glycolysis and the TCA cycle were partly compen- B Lipoproteins and Biochemical Analyses
sated by increased fluxes in b-oxidation and BCAAs, required B Fecal Genomic DNA and Genome Sequencing
to provide sufficient energy as well as substrates for the B Metagenomics Analyses
biosynthesis of essential metabolites. Of note, we observed B Measurement of Fecal SCFAs and Folate
that the fluxes carried by the reactions involved in folate and B Untargeted Metabolomics Analyses
methionine metabolism, serine uptake, as well as in the de B Liver Lipid Analysis
novo synthesis of GSH were significantly increased in response B Transcript Profiling of Liver Biopsies
to the diet. Our data are thus consistent with a central role of B Model Simulation
folate in mediating one-carbon metabolism and its involvement d QUANTIFICATION AND STATISTICAL ANALYSIS
in a number of biosynthetic processes including synthesis of B Statistical Analysis

d DATA AND SOFTWARE AVAILABILITY Aufreiter, S., Gregory, J.F., 3rd, Pfeiffer, C.M., Fazili, Z., Kim, Y.I., Marcon, N.,
B Datasets Generated and Deposited Kamalaporn, P., Pencharz, P.B., and O’Connor, D.L. (2009). Folate is
d ADDITIONAL RESOURCES absorbed across the colon of adults: evidence from cecal infusion of (13)C-
labeled [6S]-5-formyltetrahydrofolic acid. Am. J. Clin. Nutr. 90, 116–123.
SUPPLEMENTAL INFORMATION Badman, M.K., Kennedy, A.R., Adams, A.C., Pissios, P., and Maratos-Flier, E.
(2009). A very low carbohydrate ketogenic diet improves glucose tolerance in
Supplemental Information includes seven figures and six tables and can be ob/ob mice independently of weight loss. Am. J. Physiol. Endocrinol. Metab.
found with this article online at https://doi.org/10.1016/j.cmet.2018.01.005. 297, E1197–E1204.
Badman, M.K., Pissios, P., Kennedy, A.R., Koukos, G., Flier, J.S., and
ACKNOWLEDGMENTS Maratos-Flier, E. (2007). Hepatic fibroblast growth factor 21 is regulated by
PPARalpha and is a key mediator of hepatic lipid metabolism in ketotic states.
The authors thank Ara Koh for helpful discussions, Anna Hallén for preparation Cell Metab. 5, 426–437.
of the graphical abstract, all staff for excellent laboratory work and patient Bal, B.S., Finelli, F.C., Shope, T.R., and Koch, T.R. (2012). Nutritional
care, and the clinical biomarker facility at SciLifeLab, Sweden. Whole-genome deficiencies after bariatric surgery. Nat. Rev. Endocrinol. 8, 544–556.
shotgun sequencing was performed at the Genomics Core Facility at the Sahl- , K. (2016). MuMIn: multi-model inference. R package, version 1.15.6.
Barton
grenska Academy, University of Gothenburg. The computations for metage-
https://CRAN.R-project.org/package=MuMIn.
nomics analyses were performed on resources provided by the Swedish Na-
€chler, M., Bolker, B., and Walker, S. (2015). Fitting linear mixed-
Bates, D., Ma
tional Infrastructure for Computing (SNIC) through Uppsala Multidisciplinary
Center for Advanced Computational Science (UPPMAX). This work was sup- effects models using lme4. J. Stat. Software 67, https://doi.org/10.18637/jss.
ported by grants from the Swedish Research Council, Swedish Heart-Lung v067.i01.
Foundation, Swedish Diabetes Foundation, Sahlgrenska University Hospital Benjamini, Y., and Hochberg, Y. (1995). Controlling the false discovery rate: a
ALF funds, Sigrid Jusélius Foundation, Academy of Finland, Finnish Diabetes practical and powerful approach to multiple testing. J. Roy. Stat. Soc. B 57,
Research Foundation, and EU Innovative Medicines Initiative EMIF, and by the 289–300.
FP7-HEALTH-2012-INNOVATION-1 program RESOLVE no. 305707. F.B. is a Bian, H., Hakkarainen, A., Lundbom, N., and Yki-Jarvinen, H. (2014). Effects of
recipient of a European Research Council consolidator grant 615362 - META- dietary interventions on liver volume in humans. Obesity 22, 989–995.
BASE and Torsten Söderberg Professor in Medicine.
Braunersreuther, V., Viviani, G.L., Mach, F., and Montecucco, F. (2012). Role of
cytokines and chemokines in non-alcoholic fatty liver disease. World J.
AUTHOR CONTRIBUTIONS
Gastroenterol. 18, 727–735.
M.-R.T., H.-U.M., K.H.P., F.B., K.J.W., N.M., and S.S. were involved in study Bray, N.L., Pimentel, H., Melsted, P., and Pachter, L. (2016). Near-optimal
design. S.M.R., S.S., N.M., and M.G. were involved in clinical studies in the first probabilistic RNA-seq quantification. Nat. Biotechnol. 34, 525–527.
cohort. A.H. and N.L. collected and analyzed the magnetic resonance data of Breheny, P., and Burchett, W. (2016). Visualization of regression models using
the first cohort. H.-U.M. and G.B. were involved in clinical studies in the second visreg. Version 2.3.0. https://CRAN.R-project.org/package=visreg.
cohort. A.M., C.Z., and S.L. performed the transcriptome pre-processing and
Browning, J.D., Baker, J.A., Rogers, T., Davis, J., Satapati, S., and Burgess,
integration with metabolomics data using GEM. E.B. performed the pre-
S.C. (2011). Short-term weight loss and hepatic triglyceride reduction:
processing of the inflammation markers. R.M.M. and S.R. performed genetic
evidence of a metabolic advantage with dietary carbohydrate restriction.
screening and helped with the metadata preparation. B.D.P., J.N., M. Snyder,
Am. J. Clin. Nutr. 93, 1048–1052.
M.A., and M.U. were involved in data analyses. H.W. conducted computational
and statistical analyses as well as data visualization. M.B. assisted in metage- Browning, J.D., Davis, J., Saboorian, M.H., and Burgess, S.C. (2006). A low-
nomic analysis. M. Ståhlman and P.-O.B. measured the fecal SCFA and folate carbohydrate diet rapidly and dramatically reduces intrahepatic triglyceride
and performed lipidomics analyses. H.W., R.P., and J.B. wrote the paper. content. Hepatology 44, 487–488.
Corresponding authors: F.B. is responsible for analyses related to the gut Brugger, B., Erben, G., Sandhoff, R., Wieland, F.T., and Lehmann, W.D. (1997).
microbiome, and J.B. is responsible for all other parts of the study. All authors Quantitative analysis of biological membrane lipids at the low picomole level
commented on the manuscript. by nano-electrospray ionization tandem mass spectrometry. Proc. Natl.
Acad. Sci. USA 94, 2339–2344.
DECLARATION OF INTERESTS Chakravarthy, M.V., Pan, Z., Zhu, Y., Tordjman, K., Schneider, J.G., Coleman,
The authors declare no competing interests. T., Turk, J., and Semenkovich, C.F. (2005). "New" hepatic fat activates
PPARalpha to maintain glucose, lipid, and cholesterol homeostasis. Cell
Metab. 1, 309–322.
Received: September 11, 2017
Revised: December 6, 2017 Chalasani, N., Younossi, Z., Lavine, J.E., Diehl, A.M., Brunt, E.M., Cusi, K.,
Accepted: January 10, 2018 Charlton, M., and Sanyal, A.J. (2012). The diagnosis and management of
Published: February 15, 2018 non-alcoholic fatty liver disease: practice guideline by the American associa-
tion for the study of liver diseases, American college of gastroenterology,
REFERENCES and the American gastroenterological association. Am. J. Gastroenterol.
107, 811–826.
Akesson, B., Fehling, C., Jagerstad, M., and Stenram, U. (1982). Effect of Chen, R., Mias, G.I., Li-Pook-Than, J., Jiang, L., Lam, H.Y., Chen, R., Miriami,
experimental folate deficiency on lipid metabolism in liver and brain. Br. J. E., Karczewski, K.J., Hariharan, M., Dewey, F.E., et al. (2012). Personal omics
Nutr. 47, 505–520. profiling reveals dynamic molecular and medical phenotypes. Cell 148,
Anders, S., and Huber, W. (2010). Differential expression analysis for sequence 1293–1307.
count data. Genome Biol. 11, R106. Christensen, K.E., Wu, Q., Wang, X., Deng, L., Caudill, M.A., and Rozen, R.
Asrih, M., and Jornayvaz, F.R. (2013). Inflammation as a potential link between (2010). Steatosis in mice is associated with gender, folate intake, and expres-
nonalcoholic fatty liver disease and insulin resistance. J. Endocrinol. 218, sion of genes of one-carbon metabolism. J. Nutr. 140, 1736–1741.
R25–R36. Christodoulides, C., Dyson, P., Sprecher, D., Tsintzas, K., and Karpe, F. (2009).
Astrup, A., Meinert Larsen, T., and Harper, A. (2004). Atkins and other low- Circulating fibroblast growth factor 21 is induced by peroxisome proliferator-
carbohydrate diets: hoax or an effective tool for weight loss? Lancet 364, activated receptor agonists but not ketosis in man. J. Clin. Endocrinol.
897–899. Metab. 94, 3594–3601.

Crawford, P.A., Crowley, J.R., Sambandam, N., Muegge, B.D., Costello, E.K., Kanehisa, M., and Goto, S. (2000). KEGG: Kyoto encyclopedia of genes and
Hamady, M., Knight, R., and Gordon, J.I. (2009). Regulation of myocardial genomes. Nucleic Acids Res. 28, 27–30.
ketone body metabolism by the gut microbiota during nutrient deprivation. Karlsson, F.H., Nookaew, I., and Nielsen, J. (2014). Metagenomic data utiliza-
Proc. Natl. Acad. Sci. USA 106, 11276–11281. tion and analysis (MEDUSA) and construction of a global gut microbial gene
David, L.A., Maurice, C.F., Carmody, R.N., Gootenberg, D.B., Button, J.E., catalogue. PLoS Comput. Biol. 10, e1003706.
Wolfe, B.E., Ling, A.V., Devlin, A.S., Varma, Y., Fischbach, M.A., et al. Kekwick, A., and Pawan, G.L. (1956). Calorie intake in relation to body-weight
(2014). Diet rapidly and reproducibly alters the human gut microbiome. changes in the obese. Lancet 271, 155–161.
Nature 505, 559–563.
Kelley, B., Totter, J.R., and Day, P.L. (1950). The lipotropic effect of folic acid
Diraison, F., Moulin, P., and Beylot, M. (2003). Contribution of hepatic de novo on rats receiving various purified diets. J. Biol. Chem. 187, 529–535.
lipogenesis and reesterification of plasma non esterified fatty acids to plasma
Keogh, J.B., Brinkworth, G.D., Noakes, M., Belobrajdic, D.P., Buckley, J.D.,
triglyceride synthesis during non-alcoholic fatty liver disease. Diabetes Metab.
and Clifton, P.M. (2008). Effects of weight loss from a very-low-carbohydrate
29, 478–485.
diet on endothelial function and markers of cardiovascular disease risk in
Diraison, F., Pachiaudi, C., and Beylot, M. (1996). In vivo measurement of subjects with abdominal obesity. Am. J. Clin. Nutr. 87, 567–576.
plasma cholesterol and fatty acid synthesis with deuterated water: determina-
Kersten, S. (2014). Integrated physiology and systems biology of PPARalpha.
tion of the average number of deuterium atoms incorporated. Metabolism 45,
Mol. Metab. 3, 354–371.
817–821.
Kim, T.H., Yang, J., Darling, P.B., and O’Connor, D.L. (2004). A large pool of
Donnelly, K.L., Smith, C.I., Schwarzenberg, S.J., Jessurun, J., Boldt, M.D., and available folate exists in the large intestine of human infants and piglets.
Parks, E.J. (2005). Sources of fatty acids stored in liver and secreted via lipo- J. Nutr. 134, 1389–1394.
proteins in patients with nonalcoholic fatty liver disease. J. Clin. Invest. 115,
Kindt, R., and Coe, R. (2005). Tree Diversity Analysis. A Manual and Software
1343–1351.
for Common Statistical Methods for Ecological and Biodiversity Studies
Ducker, G.S., and Rabinowitz, J.D. (2017). One-carbon metabolism in health (World Agroforestry Centre).
and disease. Cell Metab. 25, 27–42.
Kolb, A.F., and Petrie, L. (2013). Folate deficiency enhances the inflammatory
Duncan, S.H., Belenguer, A., Holtrop, G., Johnstone, A.M., Flint, H.J., and response of macrophages. Mol. Immunol. 54, 164–172.
Lobley, G.E. (2007). Reduced dietary intake of carbohydrates by obese sub-
Korem, T., Zeevi, D., Suez, J., Weinberger, A., Avnit-Sagi, T., Pompan-Lotan,
jects results in decreased concentrations of butyrate and butyrate-producing
M., Matot, E., Jona, G., Harmelin, A., Cohen, N., et al. (2015). Growth dynamics
bacteria in feces. Appl. Environ. Microbiol. 73, 1073–1078.
of gut microbiota in health and disease inferred from single metagenomic
Dushay, J., Chui, P.C., Gopalakrishnan, G.S., Varela-Rey, M., Crawley, M., samples. Science 349, 1101–1106.
Fisher, F.M., Badman, M.K., Martinez-Chantar, M.L., and Maratos-Flier, E.
Kotronen, A., Peltonen, M., Hakkarainen, A., Sevastianova, K., Bergholm, R.,
(2010). Increased fibroblast growth factor 21 in obesity and nonalcoholic fatty
Johansson, L.M., Lundbom, N., Rissanen, A., Ridderstrale, M., Groop, L.,
liver disease. Gastroenterology 139, 456–463.
et al. (2009). Prediction of non-alcoholic fatty liver disease and liver fat using
Fabbrini, E., Sullivan, S., and Klein, S. (2010). Obesity and nonalcoholic fatty metabolic and genetic factors. Gastroenterology 137, 865–872.
liver disease: biochemical, metabolic, and clinical implications. Hepatology
Laeger, T., Henagan, T.M., Albarado, D.C., Redman, L.M., Bray, G.A., Noland,
51, 679–689.
R.C., Munzberg, H., Hutson, S.M., Gettys, T.W., Schwartz, M.W., et al. (2014).
Foster, G.D., Wyatt, H.R., Hill, J.O., McGuckin, B.G., Brill, C., Mohammed, FGF21 is an endocrine signal of protein restriction. J. Clin. Invest. 124,
B.S., Szapary, P.O., Rader, D.J., Edman, J.S., and Klein, S. (2003). A random- 3913–3922.
ized trial of a low-carbohydrate diet for obesity. N. Engl. J. Med. 348, Le Roy, T., Llopis, M., Lepage, P., Bruneau, A., Rabot, S., Bevilacqua, C.,
2082–2090. Martin, P., Philippe, C., Walker, F., Bado, A., et al. (2013). Intestinal microbiota
Gaggini, M., Morelli, M., Buzzigoli, E., DeFronzo, R.A., Bugianesi, E., and determines development of non-alcoholic fatty liver disease in mice. Gut 62,
Gastaldelli, A. (2013). Non-alcoholic fatty liver disease (NAFLD) and its 1787–1794.
connection with insulin resistance, dyslipidemia, atherosclerosis and coronary LeBlanc, J.G., Laino, J.E., del Valle, M.J., Vannini, V., van Sinderen, D.,
heart disease. Nutrients 5, 1544–1560. Taranto, M.P., de Valdez, G.F., de Giori, G.S., and Sesma, F. (2011). B-group
Gambino, R., Musso, G., and Cassader, M. (2011). Redox balance in the vitamin production by lactic acid bacteria–current knowledge and potential
pathogenesis of nonalcoholic fatty liver disease: mechanisms and therapeutic applications. J. Appl. Microbiol. 111, 1297–1309.
opportunities. Antioxid. Redox Signal. 15, 1325–1365. Lee, S., Zhang, C., Kilicarslan, M., Piening, B.D., Bjornson, E., Hallstrom, B.M.,
Ginsberg, H.N., Le, N.A., Goldberg, I.J., Gibson, J.C., Rubinstein, A., Wang- Groen, A.K., Ferrannini, E., Laakso, M., Snyder, M., et al. (2016). Integrated
Iverson, P., Norum, R., and Brown, W.V. (1986). Apolipoprotein B metabolism network analysis reveals an association between plasma mannose levels
in subjects with deficiency of apolipoproteins CIII and AI. Evidence that and insulin resistance. Cell Metab. 24, 172–184.
apolipoprotein CIII inhibits catabolism of triglyceride-rich lipoproteins by Leung, C., Rivera, L., Furness, J.B., and Angus, P.W. (2016). The role of the gut
lipoprotein lipase in vivo. J. Clin. Invest. 78, 1287–1295. microbiota in NAFLD. Nat. Rev. Gastroenterol. Hepatol. 13, 412–425.
Haas, J.T., Francque, S., and Staels, B. (2016). Pathophysiology and mecha- Lewis, G.F., Carpentier, A., Adeli, K., and Giacca, A. (2002). Disordered fat
nisms of nonalcoholic fatty liver disease. Annu. Rev. Physiol. 78, 181–205. storage and mobilization in the pathogenesis of insulin resistance and type 2
Halsted, C.H., Villanueva, J.A., Devlin, A.M., Niemela, O., Parkkila, S., Garrow, diabetes. Endocr. Rev. 23, 201–229.
T.A., Wallock, L.M., Shigenaga, M.K., Melnyk, S., and James, S.J. (2002). Li, H., Fang, Q., Gao, F., Fan, J., Zhou, J., Wang, X., Zhang, H., Pan, X., Bao, Y.,
Folate deficiency disturbs hepatic methionine metabolism and promotes Xiang, K., et al. (2010). Fibroblast growth factor 21 levels are increased in
liver injury in the ethanol-fed micropig. Proc. Natl. Acad. Sci. USA 99, nonalcoholic fatty liver disease patients and are correlated with hepatic
10072–10077. triglyceride. J. Hepatol. 53, 934–940.
Hirsch, S., Poniachick, J., Avendano, M., Csendes, A., Burdiles, P., Smok, G., Li, H., Zhang, J., and Jia, W. (2013). Fibroblast growth factor 21: a novel meta-
Diaz, J.C., and de la Maza, M.P. (2005). Serum folate and homocysteine levels bolic regulator from pharmacology to physiology. Front. Med. 7, 25–30.
in obese females with non-alcoholic fatty liver. Nutrition 21, 137–141. Lips, M.A., de Groot, G.H., Berends, F.J., Wiezer, R., van Wagensveld, B.A.,
Hong, S., Moreno-Navarrete, J.M., Wei, X., Kikukawa, Y., Tzameli, I., Prasad, Swank, D.J., Luijten, A., van Dijk, K.W., Pijl, H., Jansen, P.L., et al. (2014).
D., Lee, Y., Asara, J.M., Fernandez-Real, J.M., Maratos-Flier, E., et al. (2015). Calorie restriction and Roux-en-Y gastric bypass have opposing effects on
Nicotinamide N-methyltransferase regulates hepatic nutrient metabolism circulating FGF21 in morbidly obese subjects. Clin. Endocrinol. (Oxf.) 81,
through Sirt1 protein stabilization. Nat. Med. 21, 887–894. 862–870.

Lofgren, L., Forsberg, G.B., and Stahlman, M. (2016). The BUME method: a Petersen, M.C., Vatner, D.F., and Shulman, G.I. (2017). Regulation of hepatic
new rapid and simple chloroform-free method for total lipid extraction of glucose metabolism in health and disease. Nat. Rev. Endocrinol. 13, 572–587.
animal tissue. Sci. Rep. 6, 27688. Pramfalk, C., Pavlides, M., Banerjee, R., McNeil, C.A., Neubauer, S., Karpe, F.,
Lonardo, A., Ballestri, S., Marchesini, G., Angulo, P., and Loria, P. (2015). and Hodson, L. (2015). Sex-specific differences in hepatic fat oxidation and
Nonalcoholic fatty liver disease: a precursor of the metabolic syndrome. Dig. synthesis may explain the higher propensity for NAFLD in men. J. Clin.
Liver Dis. 47, 181–190. Endocrinol. Metab. 100, 4425–4433.
Loomba, R., Seguritan, V., Li, W., Long, T., Klitgord, N., Bhatt, A., Dulai, P.S., Price, N.D., Magis, A.T., Earls, J.C., Glusman, G., Levy, R., Lausted, C.,
Caussy, C., Bettencourt, R., Highlander, S.K., et al. (2017). Gut microbiome- McDonald, D.T., Kusebauch, U., Moss, C.L., Zhou, Y., et al. (2017). A wellness
based metagenomic signature for non-invasive detection of advanced fibrosis study of 108 individuals using personal, dense, dynamic data clouds. Nat.
in human nonalcoholic fatty liver disease. Cell Metab. 25, 1054–1062.e5. Biotechnol. 35, 747–756.
Love, M.I., Huber, W., and Anders, S. (2014). Moderated estimation of fold Rinella, M.E., and Sanyal, A.J. (2016). Management of NAFLD: a stage-based
change and dispersion for RNA-seq data with DESeq2. Genome Biol. 15, 550. approach. Nat. Rev. Gastroenterol. Hepatol. 13, 196–205.
Lundbom, J., Hakkarainen, A., Soderlund, S., Westerbacka, J., Lundbom, N., Rossi, M., Amaretti, A., and Raimondi, S. (2011). Folate production by probiotic
and Taskinen, M.R. (2011). Long-TE 1H MRS suggests that liver fat is more bacteria. Nutrients 3, 118–134.
saturated than subcutaneous and visceral fat. NMR Biomed. 24, 238–245.
Roza, A.M., and Shizgal, H.M. (1984). The Harris Benedict equation reeval-
Mardinoglu, A., Agren, R., Kampf, C., Asplund, A., Nookaew, I., Jacobson, P.,
uated: resting energy requirements and the body cell mass. Am. J. Clin.
Walley, A.J., Froguel, P., Carlsson, L.M., Uhlen, M., et al. (2013). Integration of
Nutr. 40, 168–182.
clinical data with a genome-scale metabolic model of the human adipocyte.
Mol. Syst. Biol. 9, 649. Sid, V., Siow, Y.L., and O, K. (2017). Role of folate in nonalcoholic fatty liver
disease. Can. J. Physiol. Pharmacol. 95, 1141–1148.
Mardinoglu, A., Agren, R., Kampf, C., Asplund, A., Uhlen, M., and Nielsen, J.
(2014). Genome-scale metabolic modelling of hepatocytes reveals serine defi- Sid, V., Wu, N., Sarna, L.K., Siow, Y.L., House, J.D., and O, K. (2015). Folic acid
ciency in patients with non-alcoholic fatty liver disease. Nat. Commun. 5, 3083. supplementation during high-fat diet feeding restores AMPK activation via an
AMP-LKB1-dependent mechanism. Am. J. Physiol. Regul. Integr. Comp.
Mardinoglu, A., Bjornson, E., Zhang, C., Klevstig, M., Soderlund, S., Stahlman,
Physiol. 309, R1215–R1225.
M., Adiels, M., Hakkarainen, A., Lundbom, N., Kilicarslan, M., et al. (2017).
Personal model-assisted identification of NAD+ and glutathione metabolism Singh, A., Gautier, B., Shannon, C.P., Vacher, M., Rohart, F., Tebutt, S.J., and
as intervention target in NAFLD. Mol. Syst. Biol. 13, 916. Le Cao, K.-A. (2016). DIABLO - an integrative, multi-omics, multivariate
method for multi-group classification. bioRxiv. https://doi.org/10.1101/
Mardinoglu, A., Shoaie, S., Bergentall, M., Ghaffari, P., Zhang, C., Larsson, E.,
067611.
Backhed, F., and Nielsen, J. (2015). The gut microbiota modulates host amino
acid and glutathione metabolism in mice. Mol. Syst. Biol. 11, 834. Solga, S., Alkhuraishe, A.R., Clark, J.M., Torbenson, M., Greenwald, A., Diehl,
Marengo, A., Rosso, C., and Bugianesi, E. (2016). Liver cancer: connections A.M., and Magnuson, T. (2004). Dietary composition and nonalcoholic fatty
with obesity, fatty liver, and cirrhosis. Annu. Rev. Med. 67, 103–117. liver disease. Dig. Dis. Sci. 49, 1578–1583.
McGarry, J.D., Mannaerts, G.P., and Foster, D.W. (1977). A possible role for Ståhlman, M., Davidsson, P., Kanmert, I., Rosengren, B., Borén, J., Fagerberg,
malonyl-CoA in the regulation of hepatic fatty acid oxidation and ketogenesis. B., and Camejo, G. (2008). Proteomics and lipids of lipoproteins isolated at low
J. Clin. Invest. 60, 265–270. salt concentrations in D2O/sucrose or in KBr. J. Lipid Res. 49, 481–490.
McMurdie, P.J., and Holmes, S. (2013). phyloseq: an R package for reproduc- Stefan, N., Kantartzis, K., and Haring, H.U. (2008). Causes and metabolic
ible interactive analysis and graphics of microbiome census data. PLoS One 8, consequences of fatty liver. Endocr. Rev. 29, 939–960.
e61217. Strozzi, G.P., and Mogna, L. (2008). Quantification of folic acid in human feces
Murphy, R.C., James, P.F., McAnoy, A.M., Krank, J., Duchoslav, E., and after administration of Bifidobacterium probiotic strains. J. Clin. Gastroenterol.
Barkley, R.M. (2007). Detection of the abundance of diacylglycerol and triacyl- 42 (Suppl. 3 Pt 2 ), S179–S184.
glycerol molecular species in cells using neutral loss mass spectrometry. Anal. Sugahara, H., Odamaki, T., Hashikura, N., Abe, F., and Xiao, J.Z. (2015).
Biochem. 366, 59–70. Differences in folate production by bifidobacteria of different origins. Biosci.
Nguyen, D., Samson, S.L., Reddy, V.T., Gonzalez, E.V., and Sekhar, R.V. Microbiota Food Health 34, 87–93.
(2013). Impaired mitochondrial fatty acid oxidation and insulin resistance in Targher, G., Byrne, C.D., Lonardo, A., Zoppini, G., and Barbui, C. (2016). Non-
aging: novel protective role of glutathione. Aging Cell 12, 415–425. alcoholic fatty liver disease and risk of incident cardiovascular disease: a
Noakes, M., Keogh, J.B., Foster, P.R., and Clifton, P.M. (2005). Effect of an meta-analysis. J. Hepatol. 65, 589–600.
energy-restricted, high-protein, low-fat diet relative to a conventional high-
Taskinen, M.R., Kuusi, T., Helve, E., Nikkila, E.A., and Yki-Jarvinen, H. (1988).
carbohydrate, low-fat diet on weight loss, body composition, nutritional status,
Insulin therapy induces antiatherogenic changes of serum lipoproteins in
and markers of cardiovascular health in obese women. Am. J. Clin. Nutr. 81,
noninsulin-dependent diabetes. Arteriosclerosis 8, 168–177.
1298–1306.
R Core Team. (2015). R: A Language and Environment for Statistical
Nuttall, F.Q., and Gannon, M.C. (2006). The metabolic response to a high-
Computing (R Foundation for Statistical Computing).
protein, low-carbohydrate diet in men with type 2 diabetes mellitus.
Metabolism 55, 243–251. Tryndyak, V., de Conti, A., Kobets, T., Kutanzi, K., Koturbash, I., Han, T.,
Fuscoe, J.C., Latendresse, J.R., Melnyk, S., Shymonyak, S., et al. (2012).
Paredes-Turrubiarte, G., Gonzalez-Chavez, A., Perez-Tamayo, R., Salazar-
Interstrain differences in the severity of liver injury induced by a choline- and
Vazquez, B.Y., Hernandez, V.S., Garibay-Nieto, N., Fragoso, J.M., and
folate-deficient diet in mice are associated with dysregulation of genes
Escobedo, G. (2016). Severity of non-alcoholic fatty liver disease is associated
involved in lipid metabolism. FASEB J. 26, 4592–4602.
with high systemic levels of tumor necrosis factor alpha and low serum
interleukin 10 in morbidly obese patients. Clin. Exp. Med. 16, 193–202. Wallner, S., and Schmitz, G. (2011). Plasmalogens the neglected regulatory
and scavenging lipid species. Chem. Phys. Lipids 164, 573–589.
Pawlak, M., Lefebvre, P., and Staels, B. (2015). Molecular mechanism of
PPARalpha action and its impact on lipid metabolism, inflammation and Wang, Y., Viscarra, J., Kim, S.J., and Sul, H.S. (2015). Transcriptional regula-
fibrosis in non-alcoholic fatty liver disease. J. Hepatol. 62, 720–733. tion of hepatic lipogenesis. Nat. Rev. Mol. Cell Biol. 16, 678–689.
Petersen, K.F., Dufour, S., Savage, D.B., Bilz, S., Solomon, G., Yonemitsu, S., Wichmann, A., Allahyar, A., Greiner, T.U., Plovier, H., Lundén, G.O., Larsson,
Cline, G.W., Befroy, D., Zemany, L., Kahn, B.B., et al. (2007). The role of T., Drucker, D.J., Delzenne, N.M., Cani, P.D., and Backhed, F. (2013).
skeletal muscle insulin resistance in the pathogenesis of the metabolic Microbial modulation of energy availability in the colon regulates intestinal
syndrome. Proc. Natl. Acad. Sci. USA 104, 12587–12594. transit. Cell Host Microbe 14, 582–590.

Wieckowska, A., Papouchado, B.G., Li, Z., Lopez, R., Zein, N.N., and Wu, H., Tremaroli, V., and Backhed, F. (2015). Linking microbiota to human
Feldstein, A.E. (2008). Increased hepatic and circulating interleukin-6 diseases: a systems biology perspective. Trends Endocrinol. Metab. 26,
levels in human nonalcoholic steatohepatitis. Am. J. Gastroenterol. 103, 758–770.
1372–1379. Wycherley, T.P., Brinkworth, G.D., Keogh, J.B., Noakes, M., Buckley, J.D., and
Clifton, P.M. (2010). Long-term effects of weight loss with a very low carbohy-
€gerstad, M. (1999).
Witthöft, C.M., Forssén, K., Johannesson, L., and Ja drate and low fat diet on vascular function in overweight and obese patients.
Folates - food sources, analyses, retention and bioavailability. Scand. J. J. Intern. Med. 267, 452–461.
Nutr. 43, 138–146.
Xia, M.F., Bian, H., Zhu, X.P., Yan, H.M., Chang, X.X., Zhang, L.S., Lin, H.D.,
Hu, X.Q., and Gao, X. (2017). Serum folic acid levels are associated with the
Wu, G.D., Chen, J., Hoffmann, C., Bittinger, K., Chen, Y.Y., Keilbaugh,
presence and severity of liver steatosis in Chinese adults. Clin. Nutr. https://
S.A., Bewtra, M., Knights, D., Walters, W.A., Knight, R., et al. (2011).
doi.org/10.1016/j.clnu.2017.06.021.
Linking long-term dietary patterns with gut microbial enterotypes.
Science 334, 105–108. Young, M.D., Wakefield, M.J., Smyth, G.K., and Oshlack, A. (2010). Gene
ontology analysis for RNA-seq: accounting for selection bias. Genome Biol.
Wu, H., Esteve, E., Tremaroli, V., Khan, M.T., Caesar, R., Manneras-Holm, L., 11, R14.
Stahlman, M., Olsson, L.M., Serino, M., Planas-Felix, M., et al. (2017). Zhou, D., Pan, Q., Shen, F., Cao, H.X., Ding, W.J., Chen, Y.W., and Fan, J.G.
Metformin alters the gut microbiome of individuals with treatment-naive type (2017). Total fecal microbiota transplantation alleviates high-fat diet-induced
2 diabetes, contributing to the therapeutic effects of the drug. Nat. Med. 23, steatohepatitis in mice via beneficial regulation of gut microbiota. Sci. Rep.
850–858. 7, 1529.

STAR+METHODS
KEY RESOURCES TABLE
REAGENT or RESOURCE SOURCE IDENTIFIER

Antibodies
Proseek Multiplex Inflammation I96x96 array Olink Bioscience article 94300
Chemicals, Peptides, and Recombinant Proteins
DNeasy Blood and Tissue kit QIAGEN Cat# 69506
Critical Commercial Assays
FGF-21 R&D Systems Cat# DF2100
IL-6 R&D Systems Cat# D6050
TNF-a R&D Systems Cat# DTA00C
Triglycerides Thermo Fisher Scientific TR22421
Cholesterol Thermo Fisher Scientific TR13421
Insulin Thermo Fisher Scientific KAQ1251
NEFA-HR(2) Assay Wako 434-91795
b-hydroxybutyrate DiaSys Diagnostic Systems 1 3701 99 10 930
Folate Roche Diagnostics USA A98032
Genotyping PNPLA3 rs738409 TaqMan Applied Biosystems ID: C_7241_10
Genotyping TM6SF2 rs58542926 TaqMan Applied Biosystems ID: C_89463510_10
Deposited Data
RNAseq liver data NCBI Gene Expression Omnibus GEO: GSE107650
Metagenomics data Sequence Read Archive SRA: PRJNA420817
Software and Algorithms
jMRUI v3.0 software jMRUI http://www.jmrui.eu/license-and-download/jmrui-license/
LCModel v6.3g software LCModel http://www.lcmodel.com/lcmodel.shtml
DESeq2 bioconductor http://bioconductor.org/packages/release/bioc/
html/DESeq2.html
Phyloseq (1.19.1) bioconductor https://bioconductor.org/packages/release/bioc/
html/phyloseq.html
Goseq bioconductor https://bioconductor.org/packages/release/bioc/
html/goseq.html
visreg R https://cran.r-project.org/web/packages/visreg/index.html
lme4 package R https://cran.r-project.org/web/packages/lme4/index.html
GRCh38 Genome Reference Consortium https://www.ncbi.nlm.nih.gov/grc/human
mixDIABLO mixOmics http://mixomics.org/mixDIABLO/
CONTACT FOR REAGENT AND RESOURCE SHARING
Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Jan Borén
(jan.boren@wlab.gu.se).
EXPERIMENTAL MODEL AND SUBJECT DETAILS
Clinical Study Design

Ten middle-aged (54±4 years) subjects (2 women, 8 men) with high liver fat were recruited to the first cohort in this study. All subjects
had stable (±5 kg) weight in the three months preceding enrollment.
The inclusion criteria were high liver fat (>5.5% measured by MRS), body mass index 27–40 kg/m2, negative viral hepatitis
serology, transferrin saturation <40%, normal a1-antitrypsin and ceruloplasmin concentrations, negative autoimmune liver disease
markers, negative pregnancy test for pre-menopausal women on enrollment, and no history of medication use associated with
hepatic steatosis. The exclusion criteria were liver cirrhosis, portal hypertension, chronic liver disease other than NAFLD, including
e1 Cell Metabolism 27, 559–571.e1–e5, March 6, 2018

alcoholic liver disease as defined by daily alcohol intake of >40 g (men) and >20 g (women), respectively, significant endocrine
disease including diabetes, any medication acting on nuclear hormone receptors or inducing CYP P450, self-administration of
supplements other than calcium, any significant cardiovascular co-morbidity, history of non-compliance, and genetic variations
in the two main genetic determinants of liver fat accumulation: PNPLA3 (homozygous for I148M) or TM6SF2 (homozygous or
heterozygous for E167K).
Participants in the first cohort followed an isocaloric low-carbohydrate diet with increased protein content for 14 days and came to
the study center on days 0, 1, 3, 4, 7, 10 and 14 of the diet after an overnight fast. At each of these visits, subjects were weighed, blood
samples were taken and liver fat was measured (by MRS). Fecal samples were collected on days 0, 1, 3, 7 and 14. The study design
with time-points for the different analyses performed is shown in Figure 1B.
To obtain liver biopsies from obese individuals with high liver fat before and after the low-carbohydrate diet with increased protein
content, we recruited a second cohort (5 men, 2 women; mean age 47.3±14.6 years). These subjects were mildly obese
(BMI 32.1±3.8 kg/m2) with ultrasound-verified fatty liver (but without diabetes or any medication) who underwent a Menghini liver
biopsy for the diagnosis of NASH. Those individuals from whom we obtained biopsies >4 cm (of which 2 cm was used for histological
confirmation of NAFLD/NASH with no advanced fibrosis; see Table S4) were asked to agree to a second biopsy within two weeks and
to follow the identical low-carbohydrate diet with increased protein content as the first cohort for 7 days until the morning before the
second biopsy. Blood samples were taken from the participants on the same day as their biopsies.
The study was performed in accordance with the Declaration of Helsinki and it was approved by the Helsinki University Central
Hospital Ethics Committee (for the first cohort) and the Ethics Committee at the University of Gothenburg (liver biopsy cohort).
Each subject gave written informed consent before participation in the study. A CONSORT flow diagram for the study is shown in
Figure S7.
Isocaloric Carbohydrate-Restricted Diet

Subjects in the first cohort were instructed to follow an isocaloric low-carbohydrate diet with increased protein content (providing a
mean of 3114.6 kcal energy per day) for 14 days. Each subject’s energy requirement was estimated (using the Harris Benedict
equation) from the basal metabolic rate (Roza and Shizgal, 1984), which was multiplied by 1.5 corresponding to the subject’s activity
level. The study diet provided approximately 4% energy from carbohydrate, 72% energy from fat and 24% energy from
protein whereas the corresponding macronutrient composition in the baseline diet was 40%, 42% and 18%, respectively (Figure 1A).
The carbohydrate content of the diet was restricted to 23-30 g per day.
To ensure compliance with the diet, subjects underwent a teaching session and had daily contact with an experienced dietitian.
Study subjects were asked to keep dietary records for 3 days beforehand to record the macronutrient content of the baseline diet.
All meals and snacks were pre-prepared in a metabolic kitchen and weighed and delivered to the subjects twice a week during the
14-day study period. The meals were frozen and the snacks and vegetables were fresh. The subjects were permitted to consume
calorie-free beverages and sweeteners. The dietician also instructed the participants to increase their daily energy intake by
200 kcal from macadamia nuts whenever their weight decreased between two study days by more than 0.2 kg. During the study,
all study subjects filled in a checklist of the meal consumed to keep track of meals eaten. The actual macronutrient intake
during the low-carbohydrate diet with increased protein content was calculated based on the checklists recorded by the subjects.
The dietitian contacted the participants daily (by phone or email) to verify adherence to the meal plan and check weight stability
(measured by home scales).
Influence of Gender
The first cohort consisted of ten obese (BMI 34.1±1.2 kg/m2) middle-aged (54±4 years) subjects (2 women, 8 men) with high liver fat.
The second cohort consisted of seven obese (BMI 32.1±3.8 kg/m2) middle-aged (47±15 years) subjects (2 women, 5 men) with high
liver fat. All subjects were analyzed as their own control, during and after the diet intervention. Subgroup analysis did not reveal
differences between women and men in their responses of liver fat content during the diet intervention.
METHOD DETAILS
Genotyping of PNPLA3 and TM6SF2 Variants

DNA was extracted from whole blood using DNeasy Blood and Tissue kit according to the manufacturer’s instructions. PNPLA3
rs738409 and TM6SF2 rs58542926 variants were genotyped by 5’-nuclease Taqman Assays. Primers and probes (FAM and VIC
labelled) were directly supplied by ThermoFisher Scientific (TaqMan Applied Biosystems, Assay ID: C_7241_10 and
C_89463510_10 for PNPLA3 rs738409 and TM6SF2 rs58542926, respectively). Post-PCR allelic discrimination was performed on
a CFX384 Real-Time System (Bio-Rad) by measuring allele-specific fluorescence. For both variants, genotyping was performed in
triplicate with concordance between replicates and success rate of 100%.
Body Composition, Liver Fat and Liver Volume

Body composition of the first cohort was determined using a Tanita MC-980 MA body composition analyzer at baseline and on day
14. Liver fat and liver volume were assessed by MRS and magnetic resonance imaging (Bian et al., 2014) using a clinical 1.5 imager
(Avantofit, Siemens, Erlangen, Germany). For determination of liver fat content, PRESS localization technique was used to collect
Cell Metabolism 27, 559–571.e1–e5, March 6, 2018 e2

25325325 mm3 spectra with five different echo times (TE): 30, 50, 80, 135 and 200 ms with 4, 4, 4, 8 and 64 number of averages,
respectively. Acquisition was triggered to end-expiration (TR > 3000 ms). Intensities of water and methylene resonances were
assessed using LCModel v6.3g software. The apparent T2 relaxation times of methylene and water resonances did not change during
the study (Table S1). For each individual experiment, the apparent T2 of water and methylene was determined by monoexponential fit
to five TEs. T2 relaxation effects on intensities were corrected, and liver fat content was calculated as the ratio methylene:
(methylene+water) and further converted to mass fractions as described (Kotronen et al., 2009). Unsaturation indices were
determined from 200 ms spectra using jMRUI v3.0 software (Lundbom et al., 2011).
Measurement of Hepatic DNL

DNL was analyzed as described previously (Diraison et al., 1996). Absolute DNL [i.e., the contribution of DNL to triglycerides within
plasma very low-density lipoproteins (VLDL-TG)] was calculated by multiplying the relative contribution of newly synthesized
palmitate from DNL to VLDL-TG by the concentration of triglycerides in VLDL (Pramfalk et al., 2015). One subject was excluded
from the analysis of DNL for technical reasons.
Lipoproteins and Biochemical Analyses

In the first cohort, lipoprotein fractions were separated as described previously (Taskinen et al., 1988). Serum VLDL from the second
cohort was isolated by ultracentrifugation using deuterium oxide as described previously (Ståhlman et al., 2008). Plasma concentra-
tions of triglyceride, cholesterol, NEFA, b-hydroxybutyrate, glucose and serum insulin were analyzed using a Konelab 60i analyzer
(Thermo Fisher Scientific). Plasma levels of FGF21, IL-6, TNF-a and apoC-III were measured by ELISA (R&D Systems, Minneapolis,
USA). Other inflammatory protein biomarkers were analyzed using the Proseek Multiplex Inflammation I96x96 array (Olink Bioscience,
Uppsala, Sweden). Serum folate was measured using a chemiluminescent immunoassay (Roche Diagnostics USA).
Fecal Genomic DNA and Genome Sequencing

Fecal genomic DNA was extracted from 100 mg of frozen stools from the first cohort using the QIAamp DNA mini stool kit
(Qiagen, Courtaboeuf, France) following repeated bead-beating (6500 r.p.m., 3 x 30 s). The DNA was extracted from 48 fecal
samples, obtained from the 10 participants at 5 different time points during the study (2 fecal samples from day 1 were not obtained).
DNA fragments of approximately 300 bp were sequenced on an Illumina NextSeq 500 instrument (150 bp; paired-end) at the
Genomics Core Facility, Sahlgrenska Academy, University of Gothenburg.
Metagenomics Analyses
We obtained a total of 151 Gb of raw paired-end reads. The taxonomic and KO composition was obtained by using an updated
version of the MEDUSA pipeline (Karlsson et al., 2014), as described before (Wu et al., 2017). The mean mapping rates for the genome
and gene catalogues were 43.1% and 72.6%, respectively (Table S2). The obtained taxonomic composition and KO profile
matrix were further analyzed by DESeq2 package (Love et al., 2014). Pathway enrichment analyses are based on KEGG annotation
(Kanehisa and Goto, 2000) and hypergeometric test using goseq (Young et al., 2010). The beta diversity and PCoA analysis were
calculated based on the relative abundance of all significant strains (further transformed by square root to reduce the influence of
dominant KOs as suggested previously (Kindt and Coe, 2005) using phyloseq (version 1.19.1) (McMurdie and Holmes, 2013).
Measurement of Fecal SCFAs and Folate

Fecal SCFAs were measured using gas chromatography coupled to mass spectrometry detection (GC-MS) as described previously
(Wichmann et al., 2013). In brief, approximately 50-250 mg of feces was mixed with internal standards, added to glass vials and
freeze dried. All samples were then acidified with HCl, and SCFAs were extracted with two rounds of diethyl ether extraction. The
organic supernatant was collected, the derivatization agent N-tert-butyldimethylsilyl- N-methyltrifluoroacetamide (Sigma-Aldrich,
Stockholm, Sweden) was added, and samples were incubated overnight. SCFAs were quantified with a 7090A gas chromatograph
coupled to a 5975C mass spectrometer (Agilent Technologies 5975C, Santa Clara, CA). SCFA standards were obtained from Sigma-
Aldrich (Stockholm, Sweden).
For folate measurements, approximately 100 mg of fecal sample was transferred to pre-weighed Sarstedt tubes along with six
ceramic beads. 500 mL methanol:water [80:20] containing 20 mM ammonium acetate, 0.1% ascorbic acid and 100 nM of the internal
standard 5-methyl-THF-13C1,d3 (Toronto Research Chemicals, Downsview, ON, Canada) was added and the samples were
homogenized for 10 min at 25 Hz using a TissueLyser instrument (Quiagen Nordic, Copenhagen, Denmark). The samples were
then centrifuged at 20,000g for 10 min and 300 mL of supernatant was transferred to glass vials and evaporated under a stream
of nitrogen. The dried samples were reconstituted in 150 mL of injection solvent (methanol:water [10:90] + 0.1% HCl) and vortexed
for 5 min at 1500 rpm. The sample were added to an OSTRO 96-well plate (Waters, Milford, MA) and eluted under positive pressure
into the 96-well collection plate.
The samples were then injected into a Waters Acquity UPLC system equipped with a Waters BEH Phenyl column (2.1 x 100 mm;
1.7 mm particle size). Column temperature was maintained at 30 C and the flow rate was 400 mL/min. The temperature of the auto
sampler was set at 12 C. The mobile phases consisted of water with 0.1% formic acid (A) and acetonitrile with 0.1% formic acid
(B). Gradient elution was performed using a linear gradient from 1-15% B over 3 min. The gradient was then increased linearly to
99% B over 4 min. After 0.5 min of isocratic elution the gradient returned to 1% B and held for 2.5 min for a total runtime of 10 min.

The folates were detected using a Xevo TQ-XS (Waters, Milford, MA) using positive electrospray. Fine tune of the mass spectrom-
eter for each metabolite was performed. After optimization the ion source parameters were: Capillary: 1.00 kV, Desolvation temper-
ature 500 C, Desolvation gas flow 1000 L/h, Cone gas flow 150 L/h, Nebulizer gas flow 7.0 L/h, Cone voltage 20 V, Collision energy 20
eV. For each analyte the most intense precursor/product ion transition was selected. Consequently, the MRM transitions used were:
Folic acid m/z 442.2 / 295.2, 5-methyl-THF m/z 460.2 / 313.2, 5-methyl-THF-13C1,d3 m/z 464.2 / 317.2. Dwell time was 44 ms
for all analytes.
Untargeted Metabolomics Analyses

Untargeted metabolomic profiling of plasma was performed by Metabolon (Durham, NC), as previously described (Lee et al., 2016).
Samples were prepared using the automated MicroLab STAR system from Hamilton Company. A recovery standard was added
before the first step in the extraction process for quality control purposes. Sample preparation was conducted using aqueous meth-
anol extraction process to remove the protein fraction while allowing maximum recovery of small molecules. The resulting extract
was divided into four fractions: one for analysis by UPLC/MS/MS (positive mode), one for UPLC/MS/MS (negative mode), one for
GC/MS, and one for backup. Samples were placed briefly on a TurboVap (Zymark) to remove the organic solvent. Each sample
was then frozen and dried under vacuum. Samples were then prepared for the appropriate instrument (either UPLC/MS/MS or
GC/MS).
Liver Lipid Analysis

Liver biopsies from the second cohort were collected in tubes with RNAlater (Qiagen, Hilden, Germany) and stored at -80 C. Lipids
from liver biopsies were extracted as previously described (Lofgren et al., 2016). Lipid extracts were diluted in chloroform:methanol
[1:2] containing 5 mM ammonium acetate and infused into a QTRAP 5500 mass spectrometer (Sciex, Concord, Canada) using a
NanoMate Triversa (Advion Biosciences, Ithaca, NY). Phosphatidylcholines and triglycerides were detected in positive mode using
precursor ion scanning and neutral loss scanning as previously described (Brugger et al., 1997; Murphy et al., 2007). Quantification
was made against internal standards added during the extraction.
Transcript Profiling of Liver Biopsies

Total RNA was extracted from the liver biopsies using the RNeasy Mini Kit (Qiagen, Hilden, Germany) according to the manufacturer’s
instructions, and used in the subsequent mRNA sample preparation for sequencing using Illumina HiSeq 2500 systems with the stan-
dard Illumina RNA-seq protocol. The RNAseq was performed at GATC Biotech AG [project number NG-10969] (Konstanz, Germany).
Gene abundance (in both raw counts and transcripts per million) was quantified using the kallisto pipeline (Bray et al., 2016) based
on human genome assembly version GRCh38. The DESeq2 package (Anders and Huber, 2010) was used to identify differentially
expressed genes.
Model Simulation
A previously established genome-scale metabolic model for liver, iHepatocyte2322 (Mardinoglu et al., 2014), was used to investigate
the metabolic shifts in response to the diet. The substrate uptake of the model is constrained partly based on a previous study
(Mardinoglu et al., 2017) and the fluxes are provided in Table S6. To investigate the flux change in response to the diet, both tran-
scriptomic and metabolomic changes were integrated as constraints into the model using a previously developed method, RMetD
(Mardinoglu et al., 2015). The flux distribution was calculated based on the average of 10,000 random sampled fluxes as previously
described (Mardinoglu et al., 2013).
QUANTIFICATION AND STATISTICAL ANALYSIS
Statistical Analysis
All statistical analyses were performed in the R environment (R Core Team, 2015). The Wald test or the Likelihood Ratio Test was used
to analyze differential abundances for all count data (for metagenomics and liver transcriptomics) depending on whether the dataset
contains samples collected at two time points or at more than two time points from each individual. One-way ANOVA with repeated
measurements was used for all other longitudinal datasets (for phenomics and metabolomics). Otherwise, two-tailed Wilcoxon
rank–sum tests or Wilcoxon signed–rank tests were used throughout the study, depending on whether the samples were paired.
The linear mixed effect model was built using lme4 package (Bates et al., 2015) in which liver fat adjusted for BMI was entered as
a response variable whereas serum folate adjusted for BMI was entered for fixed effect and individual IDs as random effect. The
model was visualized by visreg (Breheny and Burchett, 2016) and conditional and marginal coefficients of the model were determined
using MuMIn (Barton , 2016). The residual plot and quantile-quantile plot are shown in Figure S6. The multivariate analysis and
integration of phenomics, metagenomics, and metabolomics are based on mixDIABLO (Singh et al., 2016). The Spearman’s
rank–order correlation was used to determine the strength and direction of the monotonic relationships between two variables unless
strong collinearity was observed, in which case the Pearson product-moment correlation was calculated. Raw P values were
adjusted by the Benjamini–Hochberg method (Benjamini and Hochberg, 1995) with a false discovery rate of 5%, unless indicated
otherwise. Data are shown as mean±s.e.m. unless otherwise indicated.
Cell Metabolism 27, 559–571.e1–e5, March 6, 2018 e4

DATA AND SOFTWARE AVAILABILITY
Datasets Generated and Deposited

Metagenomics data from the first cohort were deposited at Sequence Read Archive with accession number SRA: PRJNA420817.
RNA-seq raw and processed files from double liver needle biopsies taken before and after one-week diet intervention from the
second cohort were deposited at NCBI Gene Expression Omnibus with accession number GEO: GSE107650.
ADDITIONAL RESOURCES
The clinical trial was registered at ClinicalTrials.gov with identifier: NCT02558530; https://clinicaltrials.gov/ct2/show/NCT02558530

Cell Metabolism, Volume 27
Supplemental Information
An Integrated Understanding of the Rapid

Metabolic Beneﬁts of a Carbohydrate-Restricted
Diet on Hepatic Steatosis in Humans
Adil Mardinoglu, Hao Wu, Elias Bjornson, Cheng Zhang, Antti Hakkarainen, Sari M.
Räsänen, Sunjae Lee, Rosellina M. Mancina, Mattias Bergentall, Kirsi H. Pietiläinen, Sanni
Söderlund, Niina Matikainen, Marcus Ståhlman, Per-Olof Bergh, Martin Adiels, Brian D.
Piening, Marit Granér, Nina Lundbom, Kevin J. Williams, Stefano Romeo, Jens
Nielsen, Michael Snyder, Mathias Uhlén, Göran Bergström, Rosie Perkins, Hanns-Ulrich
Marschall, Fredrik Bäckhed, Marja-Riitta Taskinen, and Jan Borén
Figure S1. Box plots (with median) showing liver volume at baseline (D0) and after
14 days of the carbohydrate-restricted diet (D14) in the first cohort (n = 10). Related
to Figure 1 and Table S1. **P < 0.01 (Wilcox signed-rank test).
Figure S2. Pairwise Pearson correlations between delta changes (day 14 versus day 0
of the carbohydrate-restricted diet) for liver fat, VLDL-triglycerides (TG), apoCIII,
DNL, and β-hydroxybutyrate in the first cohort (n = 8). Related to Figure 1 and Table
S1. Boxes in the lower triangle (which show correlation plots) are mirrored by boxes in the
upper triangles (which present the corresponding correlation coefficient and P value). Note
the variables are color coded (e.g. yellow liver fat, blue border for VLDL-TG). One
individual with an extreme DNL value (red circle) and one individual without a DNL
measurement were not included in this analysis. Arrows indicate how to read the
correlation matrix (in this case the correlation between liver fat and β-hydroxybutyrate).
Solid arrows show the plot, and dashed lines the correlation coefficient and P value.
Figure S3. MA plot for all KOs annotated from metagenomics data in the first cohort
at day 14 versus day 0 of the carbohydrate-restricted diet (n = 10). Related to Figure 3
and Table S2. KOs that are significantly upregulated and downregulated at day 14 are
shown in red and blue, respectively. FDR < 0.05 (likelihood ratio test).
Figure S4. Box plots (with median) showing concentrations of fecal folate at baseline
(D0) and after 3 days of the diet (D3) in the first cohort (n = 7). Related to Figure 3.
Note that only fecal samples obtained at these time points were used here because many
fecal samples were missing due to earlier analyses (metagenomics and SCFAs
measurements). One-tailed Wilcox signed-rank test.
Figure S5. Heatmap plot showing metabolites that significantly changed across the
study period in serum from the first cohort (n = 10). Related to Figure 4 and Table S3.
D, day. Pathways highlighted in pink background are upregulated whereas the ones with
green background are downregulated. FDR < 0.05 (one-way ANOVA with repeated
measurements).
Figure S6. Residual plot and quantile-quantile (Q-Q) plot for evaluation of the linear
mixed-effect model. Related to Figure 4B.
First cohort Second cohort
Screening of obese study subjects (n=13) Patients with ultrasound-verified fatty liver
with high liver fat (>5.5% measured by MRS) referred to specialist for clinical evaluation
including a Menghini liver biopsy for the
diagnosis of NASH (n=9)
Excluded (n=3)
• not meeting inclusion criteria (n=3)
Patients invited to participate in study (n=9)
Enrolment in study (n=10): dietary records for

3 days followed by teaching session with a Excluded (n=2)
dietitian • not meeting inclusion criteria (n=1)
• declined to participate (n=1)
Diet intervention during 2 weeks (n=10).

Daily contact with dietitian Diet intervention during 1 week (n=7).
Daily contact with dietitian
During 2 weeks diet intervention (n=10).

analyses according to study protocol During 1 week diet intervention (n=7).
Analyses according to study protocol
Excluded (n=3)
• unable to participate (n=1)
After 1 week diet intervention: Menghini liver
• not returning to normal diet (n=2)
biopsy (n=7)
Follow-up MRS 1-3 months after

completing the intervention study and
returning to their normal diet (n=7)
Figure S7. CONSORT flow diagram for the study. Related to Figure 1B.

MMC 8

Uploaded by

Copyright:

Available Formats

MMC 8

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

MMC 8

Uploaded by

Copyright:

Available Formats

Clinical and Translational Report

An Integrated Understanding of the Rapid Metabolic

d The LCD promotes rapid shifts in the gut microbiota

d The LCD-induced microbial changes are associated with

d The LCD increases folate-dependent one-carbon metabolism

Mardinoglu et al., 2018, Cell Metabolism 27, 559–571

Clinical and Translational Report

An Integrated Understanding of the Rapid

*Correspondence: fredrik.backhed@wlab.gu.se (F.B.), jan.boren@wlab.gu.se (J.B.)

560 Cell Metabolism 27, 559–571, March 6, 2018

Cell Metabolism 27, 559–571, March 6, 2018 561

562 Cell Metabolism 27, 559–571, March 6, 2018

Figure 2. Reduced Carbohydrate Consumption Rapidly Alters Gut Microbial Composition

Cell Metabolism 27, 559–571, March 6, 2018 563

564 Cell Metabolism 27, 559–571, March 6, 2018

Figure 4. Serum Folate Is Associated with Improved Liver Lipid Metabolism

Cell Metabolism 27, 559–571, March 6, 2018 565

566 Cell Metabolism 27, 559–571, March 6, 2018

Cell Metabolism 27, 559–571, March 6, 2018 567

568 Cell Metabolism 27, 559–571, March 6, 2018

Cell Metabolism 27, 559–571, March 6, 2018 569

570 Cell Metabolism 27, 559–571, March 6, 2018

Cell Metabolism 27, 559–571, March 6, 2018 571

KEY RESOURCES TABLE

REAGENT or RESOURCE SOURCE IDENTIFIER

CONTACT FOR REAGENT AND RESOURCE SHARING

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Clinical Study Design

e1 Cell Metabolism 27, 559–571.e1–e5, March 6, 2018

Isocaloric Carbohydrate-Restricted Diet

Genotyping of PNPLA3 and TM6SF2 Variants

Body Composition, Liver Fat and Liver Volume

Cell Metabolism 27, 559–571.e1–e5, March 6, 2018 e2

Measurement of Hepatic DNL

Lipoproteins and Biochemical Analyses

Fecal Genomic DNA and Genome Sequencing

Measurement of Fecal SCFAs and Folate

e3 Cell Metabolism 27, 559–571.e1–e5, March 6, 2018

Untargeted Metabolomics Analyses

Liver Lipid Analysis

Transcript Profiling of Liver Biopsies

QUANTIFICATION AND STATISTICAL ANALYSIS

Cell Metabolism 27, 559–571.e1–e5, March 6, 2018 e4

Datasets Generated and Deposited

e5 Cell Metabolism 27, 559–571.e1–e5, March 6, 2018

An Integrated Understanding of the Rapid

Enrolment in study (n=10): dietary records for

Diet intervention during 2 weeks (n=10).

During 2 weeks diet intervention (n=10).

Follow-up MRS 1-3 months after

You might also like

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.