Metformin Is Associated With
Higher Relative Abundance of
Mucin-Degrading
Akkermansia
muciniphila
and Several Short-
Chain Fatty Acid–Producing
Microbiota in the Gut
Diabetes Care 2017;40:54–62 | DOI: 10.2337/dc16-1324
OBJECTIVE
Recent studies suggest the benefici al effects of metform in on glucose metabolism
may be microbially mediated. We examined the association of type 2 diabetes,
metformin, and gut microbiota in community-dwelling Colombian adults. On the
basis of previous research, we hypothesized that metformin is associ ated with
higher levels of short-chain fatty acid (SCFA)–producing and mucin-degrading
microbiota.
RESEARCH DESIGN AND METHODS
Parti cipants were selected from a larger cohor t of 459 partici pants. The present
analyses focus on the 28 participan ts diagnosed with diabetes d14 taking
metformind and t he 8 4 participan ts withou t diab etes who were match ed
(3-t o-1) t o participants w ith diabetes by sex, age, and BMI. We measured de-
mographic information, anthropomet ry, and blood biochemical par ameters and
collected fecal samples from which we performed 16S rRNA gene sequencing to
analyze the composition and structure of the gut microbiota.
RESULTS
We f ound an associat ion between diabetes and gut microbiota that wa s modified
by metformin use. Compared with participants witho ut diabetes, par ticipant s
with diabetes taking metformin had higher relative abundance of Akkermansia
muciniphila, a micr obiota known for mucin degradati on, and several gut micro-
biota known for production of SCFAs, including Butyrivibrio, Bifidobacterium
bifidum, Megasphaera, and an operational taxonomic unit of Prevotella.In
contrast, compared with participants without diabetes, participants with
diabetes not taking metformin had higher relative abundance of Clostridiaceae
02d06 and a distinct operational taxonomic unit of Prevotella and a lower
abundance of Enterococcus casseliflavus.
CONCLUSIONS
Our re sults support the hypothesis that metformin shifts gut microbiota compo-
sition through the enrichment of mucin-degrading A. muciniphila as well as sev-
eral SCFA-producing microbiota. Future studies are needed to determine if these
shifts mediate metformi n’s glycemic and anti-inflammatory properties.
1
VidariumdNutrition, Health and Wellness Re-
search Center, Grupo Empresarial Nutresa, Me-
dellin, Colombia
2
Department of Epidemiology, Johns Hopkins
Bloomberg School of Public Health, Baltimore,
MD
3
Din
´
amica I.P.S.dEspecialista en Ayudas Diag-
n
´
osticas, Medellin, Colombia
4
EPS y Medicina Prepagada Suramericana S.A.,
Medellin, Colombia
Corresponding author: J uan S. Escobar, jsescobar@
serviciosnutresa.com.
Received 20 June 2016 and accepted 27 Septem-
ber 2016.
This article contains Supplementary Data online
at http ://care.diabetesjournals.org/lookup/
suppl/doi:10.2337/dc16-1324/-/DC1.
© 2017 by the American Diabetes Association.
Readers may use this article as long as the work
is properly cited, the use is educational and not
for profit, and the work is not altered. More infor-
mation is available at http://www.diabetesjournals
.org/content/license.
Jacobo de la Cuesta-Zuluaga,
1
Noel T. Mueller,
2
Vanessa Corrales-Agudelo,
1
Elian a P. Vel
´
asquez-Mej
´
ıa,
1
Jenny A. Carmona,
3
Jos
´
e M. Abad,
4
and
Juan S. Escobar
1
54 Diabetes Care Volume 40, Ja nuary 2017
EMERGING TECHNOLOGIES AND THERAPEUTICS
Downloaded from http://diabetesjournals.org/care/article-pdf/40/1/54/519215/dc161324.pdf by guest on 26 August 2022
Microbial communities (microbiota) and
their associated genes (microbiome)
constitute the interface of our environs
and our cells, and the ir composition is
believed to p lay a deterministic role in
human health and disease. In particular,
the development of typ e 2 diab etes, a
disease rising in prevalence around the
globe, has been linked in nonhuman (1)
and human (2–5) studies to imbalances
in microbiota of the intestinal tract
(gut). However, the most recent human
study on this topic found that the asso-
ciation was modulated in a potent-
ially b eneficial manner by metformin
treatment (6).
Metformin (1,1-dimethylbiguanide
hydrochloride) is the most frequent
medication used to treat type 2 diabet es
(7), and findings from recent studies
suggest it may also prevent cancer (7)
and cardiovascular events (8). Metfor-
min h as pleiotropic effects, yet the
majority of mechanistic studies have
focused on changes in liver function
(7,9,10). Although metformin certainly
alters hepatic glucose production via
effects on AMPK, there is growing evi-
dence that the ge nesis of its action is in
the gut (11–15).
Metformin is ;50% bioavailable , al-
lowing for near-equal intestinal and
plasma exposure, but intestinal accumu-
lation of metformin is 300 times that of
the plasma (16), making the gut the pri-
mary reservoir for metformin in humans.
Unlike oral administration, intravenous
administration of metformin in humans
does not improve glycemia (14). More-
over, in mice, oral admi nistration of a
broad-spectrum antibiotic cocktail with
oral metformin ab rogates metformin’s
glucose- lowering effect (12). Providing
yet further evidence that the glucose-
lowering effect of metformin may originate
in the lower bowel, a delayed-release oral
metformin, which targets the ileum,
had a similar or greater glucose-lowering
effect than immediate-release or extended-
release metformin, despite the delayed-
release metformin having lower systemic
exposure than the other metformin for-
mulations (15).
Recent studies in animal s (11,12,13)
and hu mans (6,17) provide evidence
that metformin may partially restore
gut dysbiosis associated with type 2 di-
abetes. In mice fed a high-fa t diet, met-
formin treatment increased the relative
abundance of Akkermansia muciniphila
(11–13), a m ucin-degrading bac teria
that has been shown to rev erse meta -
bolic disorders (1,12). In humans, partic-
ipants with diabetes taki ng metformin
had similar abundance of Subdoligranu-
lum and, to some extent, Akkermansia
compared with control subjects without
diabetes, suggesting that metformin
may help ameliorate a type 2 diabetes–
associated gut microbiome (6). It has also
been shown that people with diabetes
taking metformin had a higher relative
abundance of Adlercreutzia (17), and
metagenomic functiona l analyses dem-
onstrated significantly enhanced butyrate
and propionate production in people with
diabetes using metformin (6). In contrast,
people with diabetes who were not treat-
ed with metformin had a higher abun-
dance of Eubacterium and Clostridiaceae
SMB53 (17) and lower levels of short-
chain fatty acid (SCFA) producers, such
as Roseburia, Subdoligranulum,anda
cluster of butyrate-producing Clostri-
diales ( 6). These findings provide evi-
dence that gut microbes may contribute
to the antidiabetes effects of metformin
through pathways that include mucin
degradation and SCFA production.
In this study we aimed to test the gen-
eralizability of previous observations con-
cerning the influence of metformin on the
association of type 2 diabetes and gut
dysbiosis in a Colombian adult popula-
tion. Given the considerable variation in
the microbiota associated with type 2 di-
abetes and that the gut microbiota of Co-
lombians is different to t hat of other
populations (18), we hypothesized that
the microbial taxa involved in the type 2
diabetes dysbiosis of Colombians are dif-
ferent to those observed in Chinese and
European populations (2,3) but that the
effect of metformin is similar, i.e., through
enrichment of mucin-degrading and SCFA-
producing microbiota.
RESEARCH DESIGN AND METHODS
Study Design
Between July and November 2014, we en-
rolled 459 men and women 18–62 years
old, with BMI $18.5 kg/m
2
, living in
the Colombian cities of Medellin, Bogota,
Barranquilla, Bucaramanga, and Cali. All
participants enrolled in the study were in-
sured by the health insurance provider
EPS y Medicina Prepagada Suramericana
S.A. (EPS SURA). We excluded pregnant
women, individuals who consumed anti-
biotics or antiparasitics ,3 months prior
to enrollment, and individuals diagnosed
with Alzhe imer disease, Parkinson dis-
ease, or any other neurodegenerative dis-
eases; current or recent cancer (,1year);
and gastrointestinal diseases (Crohn dis-
ease, ulcerative colitis, short bowel syn-
drome, diverticulosis, or celiac disease).
This study was conducted in accor-
dance with the principles of the Decla-
ration of Helsinki, as revised in 2 008, and
had minimal risk according to the
Colombian Ministry of Health (Resolution
008430 of 1993). All of the partici-
pants were thoroughly informed about
the study and p rocedures. Participants
were assured of anonymity and confi-
dentiality. Written informed consent was
obtained from all the participants before
beginning the study. The Bioethics
Committee of Sede de Investigaci
´
on
UniversitariadUniversity of Antioquia re-
viewed the protocol and the consent
forms and approved the procedures de-
scribed here (approbation act 14–24–588
dated 28 May 2014).
Anthropometric, Clinical, and Dietary
Evaluations
We calculated BMI as weight (kg)/height
squared (m
2
) to clas sify participants as
lean (18.5 # BMI , 25.0 kg/m
2
), over-
weight (25.0 # BMI , 30.0 kg/m
2
), or
obese (BMI $30 kg/m
2
). In addition, val-
ues of HDL, LDL, VLDL, total cholester ol,
triglycerides, apolipoprotein B, fasting
glucose, glycated hemoglobin (HbA
1c
),
fasting insulin, adiponectin, an d hs-CRP
were obtained (collection and measure-
ment explained in Supplementary Data).
Dietary intake was evaluated through
24-h dietary recalls (see Supplementary
Data).
DNA Extraction and Sequence
Analysis
Each participant collect ed their own fe-
cal sample in a hermetically sealed, ster-
ile receptacle provided by t he research
team. Samples were immediately refrig-
erated in household freezers and brought
to an EPS SURA facility in each city within
12 h; receipt of samples occurred exclu-
sively in the morning (6
A.M.–12 P.M.). As
such, stools were collected between the
evening of the day before and the morn-
ing of the day of sample receipt. Fecal
sampleswerestoredondryiceand
sent to a central laboratory via next-day
delivery. Before DNA e xtraction, stool
consistency was evaluated by trained lab-
oratory technicians.
care.d iabetesjournals.org de la Cuesta-Zuluaga and Associates 55
Downloaded from http://diabetesjournals.org/care/article-pdf/40/1/54/519215/dc161324.pdf by guest on 26 August 2022
Total microbial DNA was extracted
using the QIAamp DNA Stool Mini Kit
(Qiagen, Hilden, Germany) following
the manuf acturer’s instructions, with a
slight modification consisting in a bead-
beating step with the lysis buffer (20 s at
15 Hz using a stainless steel bead with a
5-mm diameter). After e xtraction, we
quantified DNA concentration using a
NanoDrop spectrophotometer (Nyxor
Biotech, Paris, France) and sent it to the
Microbial Systems Laboratory, University
of Michigan Medical School (Ann Arbor,
MI). The V4 hypervariable region of the
16S rRNA gene was amplified using the
F515 (59-CA CGGTCGKCGGCGCCATT-39)
and R806 (59-GGACTACHVGGGTWTC
TAAT-39) primers and sequenced usi ng
the Illumina MiSeq sequencing platform
with V2 chemistry and the dual-index
sequencing strategy (19). In addition to
DNA from fecal samples, we sequenced
negative controls (ultrapure water and
the QIAamp DNA Stool Mini Kit’sEBelu-
ti o n buffer), a DNA extraction blank,
and a mock community (HM-782D, BEI
Resources, Manassas, VA) in each instru-
ment’s run. Sequences were curated
following the MiSeq standard operating
procedure implemented by Mo thur
v.1.36 (20) (see Supplementary Data). Raw
sequences were deposited at the NCBI and
can be accessed through the BioProject
(accession number PRJNA325931).
Definition of Type 2 Diabetes and
Selection of Control Subjects
We identified 28 participants in our study
that had type 2 diabetes; 26 self-reported
physician-diagnosed diabetes prior to
the beginning of the study and 2 were
diagnosed through laboratory testing
(fasting blood glucose $126 mg/dL
and HbA
1c
$6.5%). Of the 28 participants
with type 2 diabetes, 14 were under met-
formin treatment, 14 were not (1 was
treated with insulin alone, 2 with gliben-
clamide, and 11 were under no pharmaco-
logical treatment for type 2 diabetes,
including the 2 participants unaware of
their diabetes status) (Suppleme ntary
Table 1).
We matched each participant with di-
abetes with three participants without
diabetes in our study based on sex, age
(to the cl osest possible age; maximum
difference between case and control
subjects 6 years; mean 1.5 years; median
1 year), and BMI category (lean, over-
weight, or obese). This left us with a total
analytic sample of 112 study participants
comprising 14 with type 2 diabetes using
metformin (T2D-met
+
), 14 with type 2 di-
abetes not using metformin (T2D-met
2
),
and 84 without diabetes (ND).
Statistical Analyses
Anthropometric and clinical variables
were compared across study groups us-
ing ANOVA and t tests after checking for
homoscedasticity and normal distrib u-
tion of residuals (using Fligner-Killeen
tests of homogeneity of variances and
Shapiro-Wilk normality tests). When
necessary, variables were appropriately
transformed ( natural log for uncon-
strained variables or arcsin square root
for proportions). Sex ratio and s tool con-
sistency were compared using x
2
tests.
Statistical analyses were performed
with R v.3.2.2 (21).
Curated DNA sequences ranged f rom
69 to 102,660 sequences per sample
(median 28,699). To limit the effects of
uneven sampling, we rarefied the data
set to 4,091 sequences per sample, result-
ing in the exclusion of one T2D-met
2
par-
ticipant with 69 reads. Although rarefaction
may lead to missing low-abundance data, it
is a powerful way to reduce the likelihood
of detecting false positives, especially
among those operational taxonomic units
(OTUs) with very low abundance.
The gut microbiota structure and com-
position was assessed by quantifying and
interpreting similarities based on intra-
and intergroup diversity analyses (a and
b diversity, respectively). For a diversity,
we calculated Good’s coverage and the
number of OTUs of each sample using
Mothur and constructed rarefaction
curves. We compared these indices among
groups of participants using analysis of
similarity (ANOSIM) with 1,000 permuta-
tions using the Vegan package of R (22).
b Diversity was assessed using
phylogeny-based generalized UniFrac dis-
tances (with the a parameter controlling
weight on abundant lineages = 0.5) calcu-
lated with the GUniFrac package of R (23).
For this, we first reduced the alignment
and the OTU table to one representative
sequence per OTU, then obtained a dis-
tance matrix from uncorrected pairwise
distances between aligned sequences,
and finally constructed a relaxed neighbor-
joining phylogenetic tree using Mothur
and Clearcut. Comparisons among
groups of participants were performed
using the adonis function (ANOVA using
distance matrices) of the permutational
multivariate ANOVA (PERMANOVA) im-
plemented in the Vegan package of R (22).
We next used linear discriminant
analysis (LDA) e ffect size (LEfSe) (24) t o
agnostically identify microbial bio-
markers. LEfSe uses the nonparametric
factorial Krus kal-Wallis sum rank test to
detect individual OTUs with significant
differential abundance among groups
of participants, then performs a set of
pairwise tests among groups of partici-
pants using the unpaired Wilcoxon rank
sum test, and finally uses LDA to esti-
mate the effect siz e of each differen-
tially abundant OTU (24). The strength
of LEfSe c ompared with standard stat is-
tical approaches is that, in addition to
prov iding P values, it provides an estima-
tion of the magnitude of the association
between each OTU and the grouping cat-
egories (e.g., metformin, type 2 diabetes)
through the LDA score. For stringency,
microbial biomarkers in our study were
retained if they had a P , 0.05 and a
(log10) LDA score $3, i.e., one order of
magnitude greater than LEfSe’sdefault.
Finally, across groups, we tested for dif-
ferences in relative abundance of the
mucin-degrading A. muciniphila and
major butyrate-producing microbial
genera, including Butyrivibrio, Ros eburia,
Subdoligranulum,andFaecalibacterium.
For th is analysis, we pooled all OTUs
classified in each of these phylotypes
(4 for A. muciniphila,11forButyrivibrio,
4forRoseburia,10forSubdoligranulum,
and 5 for Faecalibacterium) and tested for
differences using ANOVA and t tests on
arcsin square root transformed relative
abundances. T his pooling served to
examine whether differences in relative
abundance of these groups of bacteria
occurred across all OTUs or only in
specific OTUs.
Results from LEfSe and from the pooled
analysis of phylotypes were corrected for
multiple testing using the Bayesian ap-
proach implemented in the qvalue pack-
age of R (25). Tests were considered
significant if they had a P value # 0.05
andaqvalue#0.1.
RESULTS
In Table 1 we present the characteristics
of T2D-met
+
, T2D-met
2
, and ND partic-
ipants. There were no statis tically signif-
icant differences (all P values . 0.10) in
demographic, anthropometric, or clini-
cal parameters between T2D-met
+
and
56 Type 2 Diabetes, Metformin, and Gut Microbiota Diabetes Care Volume 40, Ja nuary 2017
Downloaded from http://diabetesjournals.org/care/article-pdf/40/1/54/519215/dc161324.pdf by guest on 26 August 2022
Table 1—General characteristics of T2D-met
+
, T2D-met
2
, and ND participants among community-dwelling Colombian adult s
Group P*
T2D-met
+
T2D-met
2
ND T2D-met
+
vs. T2D-met
2
T2D-met
+
vs. ND T2D-met
2
vs. ND
n 14 14 84 ddd
Age (years) 50 6 10 44 6 9476 9 0.11 0.33 0.24
Sex (F/M) 0.36 0.50 0.43 0.70 0.83 0.84
Anthropometry
BMI (kg/m
2
) 31.88 6 4.63 32.15 6 6.36 31.11 6 4.53 0.98 0.56 0.62
Body fat (%) 0.41 6 0.05 0.41 6 0.03 0.40 6 0.04 0.96 0.68 0.58
Waist circumference (cm) 104.6 6 9.8 102.8 6 14.2 102.0 6 11.3 0.70 0.39 0.85
Clinical parameters
Total cholesterol (mg/dL) 178 6 51 208 6 44 187 6 30 0.11 0.54 0.10
HDL (mg/dL) 40 6 11 38 6 6446 12 0.77 0.21 0.0136
LDL (mg/dL) 105 6 35 128 6 39 115 6 28 0.11 0.31 0.26
VLDL (mg/dL) 38 6 40 52 6 52 31 6 15 0.19 0.64 0.0471
Apolipoprotein B (mg/dL) 102 6 30 108 6 30 97 6 26 0.56 0.55 0.18
Triglycerides (mg/dL) 176 6 202 244 6 253 154 6 75 0.15 0.97 0.07
Fasting glucose (mg/dL) 127 6 47 145 6 76 90 6 10 0.57 0.0047 0.0060
HbA
1c
[% (mmol/mol)] 6.9 6 1.4 (52.0 6 15.3) 7.1 6 1.7 (54.0 6 18.6) 5.6 6 0.3 (38.0 6 3.3) 0.78 0.0026 0.0047
Fasting insulin (mU/mL) 22.24 6 12.58 24.24 6 12.86 15.20 6 8.79 0.49 0.11 0.0029
Insulin re sistance index 2.9 6 1.5 3.7 6 3.0 1.9 6 1.1 0.39 0.0420 0.0018
Leptin (ng/mL) 7.39 6 6.09 8.34 6 5.40 7.89 6 6.14 0.38 0.91 0.25
Adiponectin (mg/mL) 4.59 6 1.97 4.96 6 3.22 6.79 6 3.90 0.91 0.0195 0.07
hs-CRP (mg/L) 2.70 6 2.5 3 3.35 6 2.76 4.04 6 5.02 0.61 0.21 0.62
Dietary intake
Energy intake (calories) 1,843 6 268 1,865 6 584 1, 945 6 572 0.75 0.61 0.57
Carbohydrate (g) 250 6 37 260 6 86 267 6 88 0.92 0.52 0.68
Protein (g) 74.7 6 9.1 69.7 6 13.1 74.1 6 13.8 0.26 0.85 0.26
Fat (g) 60.7 6 13.0 60.2 6 21.3 62.9 6 17.1 0.64 0.76 0.51
Cholesterol (mg) 336 6 31 330 6 41 347 6 38 0.67 0.28 0.18
Dietary fiber (g) 17.9 6 4.8 18.6 6 6. 4 17.5 6 4.8 0.88 0.73 0.64
Stool consistency [n (%)] 0.38 0.58 0.50
Hard 4 (28) 2 (14) 13 (15) ddd
Normal 8 (57) 9 (64) 54 (64) ddd
Mushy 2 (14) 1 (7) 13 (15) ddd
Diarrheic 0 (0) 2 (14) 4 (5) ddd
Values presented as mean 6 SD. *All P values from t tests except in sex and stool consistency (x
2
tests).
care.d iabetesjournals.org de la Cuesta-Zuluaga and Associates 57
Downloaded from http://diabetesjournals.org/care/article-pdf/40/1/54/519215/dc161324.pdf by guest on 26 August 2022
T2D-met
2
participants. Compared with
ND participants, T2D -met
+
participants
had higher fasting glucose, HbA
1c
,andin-
sulin resistance than ND participants and
lower levels of the insulin-sensitizing hor-
mone adiponectin (P , 0.05). No other
demographic, anthropometric, or clinical
parameters were statistically different.
16S rRNA Gene Sequencing
Gut microbiota communities were spe-
cific to each participant with marked in-
tersubject differences (Fig. 1A) (overall
interindividual generalized UniFrac dis-
tance = 0.720 6 0.009). We found high
coverage across all groups of participants
(mean Good’s coverage 6 SD = 0.990 6
0.001); 99% of OTUs were detected at
least by two DNA reads, demonstrating
thorough sampling of the gut micro-
biota. We next tested for differences
in the number of observed OTUs across
the groups of participants. We found no
differences between participants with
diabetes and N D participants (ANOSIM
statistic R = 0.005, P value=0.425)or
between T2D-met
+
and T2D-met
2
partic-
ipants (ANOSIM statistic R = 20.018, P =
0.557). The number of observed OTUs
tended to be more similar between T2D-
met
+
and ND than between T2D-met
2
and ND participants (Supplementary Fig.
1); however, these differences were not
statistically significant (T2D-met
+
vs. ND:
ANOSIM statistic R = 0.018, P = 0.348;
T2D-met
2
vs. ND: ANOSIM statistic R =
0.009, P =0.409).
We observed no significant differences
in b diversity estimates among the three
groups of participants (PERMANOVA: R
2
=
0.019, P = 0.335) (Fig. 1A) or between
participants with diabetes and ND partic-
ipants (R
2
= 0.009, P = 0.416) (Fig. 1B).
However, the comparison between met-
formin and nonmetformin users reached
significance (R
2
=0.013,P =0.036)(Fig.
1C), demonstrating differences in the
bacterial community structure associated
with metformin use. The difference was
also s ignificant when comparing T2D-
met
+
and ND participants (R
2
=0.015,
P = 0.036) but not when comparing
T2D-met
2
and ND participants (R
2
=
0.008, P = 0.943). These results suggested
the microbial communities of T2D-met
+
versus T2D-met
2
were modestly phylo-
genetically dissimilar.
We next used LEfSe to examine differ-
ences in the relative abundance of gut
microbiota at the OTU level. Note that
we were only interested in OTUs dis-
playing str ong a sso ciatio ns in the LDA
(represented by OTUs with [log10] LDA
scores $3); such st ringency resulted in
fewer retained, but more likely biologically
relevant, OTUs (19 displayed in Fig. 2 and
PCo2 (6.03%)
PCo1 (10.34%)
ND
R
2
= 0.019
P = 0.34
T2D-met
+
T2D-met
-
PCo2 (6.03%)
R
2
= 0.009
P = 0.42
ND
T2D
PCo1 (10.34%)
PCo2 (6.03%)
met-
R
2
= 0.013
P = 0.036
T2D-met
+
PCo1 (10.34%)
A
B
C
Figure 1—Principal coordinates analysis based on generalized UniFrac. A: Comp aris on among the
three groups of participants. B: Comparison between participants with diabetes and ND part ici-
pants. C: Comparison between T2D-met
+
and participants not taking metformin (T2D-met
2
and
ND). Ellipses encompass 75% of data distribution in each group of parti cipants. R
2
and P values from
PERMANOVA.
58 Type 2 Diabetes, Metformin, and Gut Microbiota Diabetes Care Volume 40, Ja nuary 2017
Downloaded from http://diabetesjournals.org/care/article-pdf/40/1/54/519215/dc161324.pdf by guest on 26 August 2022