Clostridium difficile (Cd) is an opportunistic diarrheal pathogen and Cd infection
(CDI) represents a major healthcare concern, causing an estimated 15,000 deaths per
year in the United States alone
1
. Several enteric pathogens, including Cd, leverage inflammation and the accompanying
microbial dysbiosis to thrive in the distal gut
2
. Although diet is among the most powerful available tools for affecting the health
of humans and their relationship with their microbiota, investigation into the effects
of diet on CDI has been limited. Here, we show in mice that the consumption of microbiota
accessible carbohydrates (MACs) found in dietary plant polysaccharides has a significant
impact on CDI. Specifically, using a model of antibiotic-induced CDI that typically
resolves within 12 days of infection, we demonstrate that MAC-deficient diets perpetuate
CDI. We show that Cd burdens are suppressed through the addition of either a diet
containing a complex mixture of MACs or a simplified diet containing inulin as the
sole MAC source. We show that switches between these dietary conditions are coincident
with changes to microbiota membership, its metabolic output and Cd-mediated inflammation.
Together, our data demonstrate the outgrowth of MAC-utilizing taxa and the associated
end products of MAC metabolism, namely the short chain fatty acids (SCFAs) acetate,
propionate, and butyrate, are associated with decreased Cd fitness despite increased
Cd toxin expression in the gut. Our findings, when placed into the context of the
known fiber deficiencies of a human Western diet, provide rationale for pursuing MAC-centric
dietary strategies as an alternate line of investigation for mitigating CDI.
CDI is typically associated with antibiotic-mediated dysbiosis, yet 22% of individuals
with community acquired CDI have no recent history of antibiotic use
3
. We and others previously demonstrated that direct microbiota-Cd metabolic interactions
are critical determinants of Cd fitness in the distal gut
4–6
and that the absence of dietary MACs leads to the expression of inflammatory markers
by the host colonic epithelium
7
. Additional in vitro work suggested that MAC-centric metabolic interactions may play
a role in reducing the fitness of Cd in the gut
8–10
, leading us to hypothesize that a MAC-deficient diet reinforces the inflammation
and dysbiosis conducive to CDI.
We used an experimental model of CDI in ex-germ-free Swiss-Webster mice colonized
with the microbiota of a healthy human donor (see Methods). These humanized mice were
fed a diet containing a complex mixture of MACs (MAC+), or two diets that are both
MAC-deficient (MD) (Fig. 1a). Mice fed the MD diets show persistent CDI while mice
fed the MAC+ diet clear the pathogen below detection within 10 days of infection.
After 36 days of persistent infection in mice fed the MD diets, a dietary shift to
the MAC+ diet results in clearance below detection within 9 days (Fig. 1a). This MAC+
diet-mediated CDI suppression is also observed in conventional C57BL/6 and Swiss-Webster
mice and in ex-germ-free Swiss-Webster mice colonized with a conventional Swiss-Webster
mouse microbiota (Fig. S1), demonstrating that MAC-dependent CDI suppression is not
confined to a specific microbiota or host genotype.
To enumerate gut-resident microbes that might suppress Cd, we sequenced 16S rRNA amplicons
from the feces of humanized mice (Fig. 1a). The presence of dietary MACs and treatment
with antibiotics affected both alpha and beta diversity of operational taxonomic units
(OTUs) in the gut microbiota (Figs. 1b–d, S2, S3). Two principle coordinates explain
48% of the variance in weighted UniFrac distances between samples. We traced temporal
changes in the composition of the microbiota through this space. To highlight the
composition of the microbiota in the context of CDI, a log-fold contour plot was drawn
to illustrate burdens of Cd that correspond to these samples (see Fig. S4 for further
explanation of these “contoured PCoA” [cPCoA] plots).
Clindamycin remodels the mouse microbiome to a Cd-permissive state (Figs. 1b–d, dotted
lines; Fig. S3). After inoculation with Cd, the microbiota of MAC+-fed mice changes
significantly, and as Cd burdens decrease, the community returns to resemble the pre-infection
state (Figs. 1b, S3a), illustrating compositional resilience of the microbiota under
the MAC+ dietary condition during CDI. During persistent infection in mice fed the
MD diets, the microbiota composition is similar to the pre-infection MAC-deficient-associated
microbiota. However, after the MAC+ dietary intervention, the microbiota of these
mice transitions to resemble the microbiota of mice prophylactically fed the MAC+
diet as Cd burdens decrease (Figs. 1c, 1d, S3b, S3c). These data suggest that diet
and antibiotic treatment are two major drivers of microbial communities that support
or exclude Cd in our model. Furthermore, the similarities in the microbiota of uninfected
and persistently infected mice fed the MD diets may be due to the metabolic and compositional
constraints imposed by this dietary condition, which we hypothesize is supportive
to Cd during infection.
In a previous report, in mice that cleared CDI below the limit of detection, Cd shedding
could be re-initiated upon clindamycin treatment up to approximately 12 weeks after
FMT-mediated suppression
11
, consistent with mice harboring Cd in the absence of disease and not having developed
protective immunity after FMT. In our model, the persistently infected state cannot
be re-established in mice fed a MAC-rich diet that cleared CDI below detection upon
switching them to the MD1 diet (i.e., in the absence of antibiotics). However, persistent
infection can be re-established if mice are first switched to the MD1 diet and treated
with clindamycin (Fig. S5).
Having shown that the MAC+ diet, containing a complex and ill-defined mixture of MACs
(see Methods), is successful in suppressing CDI, we sought to decouple the effects
of MACs from other dietary components (e.g. phytonutrient
12
or protein
13
content). Using a simplified MD1-based diet containing inulin as the sole MAC source,
the findings from our initial dietary intervention experiment in Cd-infected animals
are recapitulated (Figs. 2a, S6). Furthermore, prophylactic inulin feeding (either
10% in the diet as above or 1% in the drinking water of mice fed the MD1 diet), results
in dose-dependent effects on both the maximum Cd burden and on Cd clearance kinetics
(Fig. S7). Taken together, like the complex MAC+ diet, inulin feeding reduces Cd burdens
across experimental paradigms (Figs. 2a, S6–S8).
Notably, 2 of 16 mice given inulin-based dietary interventions failed to clear Cd
below detection (Figs. S6a, S8). However, a critical Cd virulence factor, the glycosylating
toxin TcdB
14
is not detectable in these mice (or the MAC+-fed mice that cleared Cd below detection)
at the endpoint of this experiment (Fig. S6b). In one study that examined toxin concentrations
quantitatively in adults, CDI severity was correlated with fecal toxin level
15
, ranging from asymptomatic carriage to fulminant colitis (up to 15% of healthy adults
are asymptomatically colonized with Cd
16
). However, Cd toxin levels are typically not quantified in current clinical diagnostic
procedures, which involve either qualitative assessment of toxin by immunoassay, in
vitro cytotoxicity assay, or PCR-based detection of toxin genes
17
. This raises the possibility that MACs may play a role in limiting Cd burdens below
a pathogenic threshold in carriers.
Although the complex MAC+ and the inulin-containing diets both negatively impact the
in vivo fitness of Cd, the overall microbiome composition differs between mice fed
these two diets, as illustrated by a two-dimensional cPCoA subspace that explains
62% of the variation in the data and the proportional abundance of taxa (Figs. 2b,
S9, S10). Because increased gut microbiota diversity is associated with resistance
to a number of pathogens and is a hallmark of FMT-mediated suppression of CDI
6,18–20
, one hypothesis is that MACs positively affect CDI outcome by favoring a diverse
microbiota. Though CDI suppression induced by the MAC+ diet is correlated with an
increase in alpha diversity of the gut microbiota, alpha diversity does not increase
when Cd burdens decrease via inulin feeding (Figs. 2c, S11a). Furthermore, community
evenness is lowest in the inulin fed mice (Fig. S11b), consistent with a limited number
of taxa profiting from a single MAC type
21
. Together, these data demonstrate that MACs significantly reduce Cd burdens but that
this reduction is independent of microbiota diversity.
Despite these wholesale differences in community composition, we pursued two of several
possibilities: (1) a common subset of OTUs suppress CDI across dietary interventions
or (2) diet-specific but functionally similar OTUs within dietary conditions suppress
CDI. To identify OTUs that are predictive of Cd presence or that discriminate between
dietary conditions (Table S1, fields ‘Plus_minus_Cd’ and ‘Current_diet,’ respectively),
we performed supervised (random forests) and unsupervised (non-parametric tests for
differential abundance) analyses on OTUs identified in this study. All of the features
identified by supervised analysis as the most discriminating between Cd infection
state or diet for each comparison group (n=15 for each classifier) were also identified
by unsupervised analysis (Bonferroni corrected p<0.05, Table S2). Spearman correlation
analyses were then performed between these features and Cd burdens, refining a list
of high-confidence taxa that are predictive of Cd burdens (Spearman ρ with Bonferroni
corrected p<0.05, Table S3).
Notably, several taxa are significantly (anti)correlated with Cd burdens regardless
of diet, suggesting that common microbial signatures underlie permissive and non-permissive
states (Table S3). Among these, Parabacteroides, Lachnospiraceae, and Erysipelotrichaceae
are correlated with Cd abundance regardless of diet. The correlation between Parabacteroides
and Cd burden is consistent with previous observations that Parabacteroides are elevated
in Cd “supershedder” mice
22
. In humans, Erysipelotrichaceae and some Lachnospiraceae are enriched in individuals
with CDI compared to nondiarrheal controls
23
. Despite these commonalities, the majority of features identified in Table S3 are
(anti)correlated with Cd in a subset of diets, supporting previous work that distinct
context-dependent communities, rather than core Cd-(un)supportive communities, are
important for determining CDI status
24
. Because others have demonstrated that metabolites, rather than microbes, are able
to differentiate CDI status in humans
25
, we hypothesized that diet creates metabolic landscapes that are either supportive
or unsupportive of Cd.
Therefore, we pursued whether MAC-mediated suppression of CDI could be differentiated
from the permissive condition on a molecular basis relevant to MAC metabolism. We
measured the major metabolic end products of MAC metabolism (the SCFAs acetate, propionate,
and butyrate) in cecal contents of mice (see Fig S6). Acetate and butyrate are elevated
in the ceca of mice fed MAC+ and inulin diets relative to those fed the MD1 diet,
and propionate is elevated in the ceca of MAC+-fed mice relative to those fed the
MD1 or inulin diets (Fig. 3a). Furthermore, acetate, propionate, and butyrate have
concentration-dependent negative effects on Cd growth as measured by differences in
doubling time (Fig. 3b) and concentration-dependent positive effects on expression
of a critical Cd virulence factor, the glycosylating toxin TcdB
14
(Fig. 3c). Our findings using Cd strain 630 are consistent with previous findings
in other Cd strains: (i) SCFAs inhibit the growth of five non-630 strains in a concentration
dependent fashion
10,26
and (ii) butyrate affects toxin expression in Cd strain VPI 10463
27
.
Given these findings, we hypothesize that dietary MACs negatively affect the fitness
of Cd in two interrelated ways. First, MACs drive privileged outgrowth of MAC-utilizing
members of the microbiota (e.g. Bacteroides spp., see Table S2b). Second, the SCFAs
that result from MAC metabolism negatively affect the fitness of Cd, which could be
due to the buildup of endproducts of key metabolic pathways, such as reductive acetogenesis
and butyrogenesis
5,28
. The expression of TcdB (and the co-regulated toxin TcdA) in Cd is controlled by
multiple inputs, such as nutrient availability, quorum sensing, and other environmental
stresses
29
. Therefore, it is possible that SCFAs serve as a signal to Cd of microbiome fermentation,
and this signal of a competitive and inhospitable gut environment leads to an increase
in toxin production.
Since limitation of dietary MACs is known to increase inflammation in the gastrointestinal
tract
7
, we examined colonic histopathology and toxin levels to better understand the inflammation
during MAC-dependent suppression of CDI. Humanized Swiss-Webster mice fed the MD1
diet were infected as in Fig. 1a and were switched to either the MAC+ or inulin diet
at 7 days post infection. At pre- and post-diet shift time points, histopathology
of proximal colon tissue was evaluated. Pathology was significantly increased in all
infected mice relative to uninfected control mice fed the MAC+ or inulin diets (Figs.
4a, S12; Table S4). Notably, inflammation is comparably elevated in both infected
and uninfected mice fed the MD1 diet, consistent with the contribution of the MD diets
to inflammation and Cd persistence.
We also measured burdens of Cd and levels of TcdB in feces during the shift from the
MD1 diet to either the MAC+ or inulin diets. TcdB is detected during persistent infection
but its expression is further elevated at 2 and 4 days after the shift to the MAC+
or inulin diet (Fig. 4b). Though cfu-normalized TcdB abundance is elevated after the
diet shift, the overall abundance of TcdB decreases from day 2 to day 4 post diet
shift (Fig. S13). This shows that toxin expression is elevated on a per-cell basis
in response to MACs but that overall TcdB abundance scales with decreasing burdens
of Cd. Together, these data support a model where MAC deficient diets facilitate a
level of inflammation supportive of Cd survival in the gut, enabling Cd persistence
despite lower levels of toxin expression. Cd responds to the non-permissive MAC-driven
environmental change by elevating toxin expression. However, Cd is unable to maintain
or regain its niche in the gastrointestinal tract upon the sustained consumption of
MACs by the host.
In the context of CDI, the relative contribution of inflammation to the exclusion
of inflammation-sensitive competitors versus the creation of privileged nutrients,
analogous to strategies delineated for the enteric pathogen Salmonella Typhimurium,
remains to be determined
2
. Furthermore, though SCFAs can directly inhibit bacterial growth and affect Cd toxin
production, additional work is needed to investigate the potential roles of other
relevant factors that could be impacting infection dynamics. For example, SCFA can
signal through host pathways (e.g. improved barrier function via hypoxia-inducible
factor
30
) which may influence inflammation and microbiome composition independently of Cd
burdens and toxin production; other metabolites such as bile acids can influence CDI
6
and may interact with dietary effects.
Despite the individuality in the gut microbiota of patients with CDI, there is a consistent
metabolic response that underlies CDI across individuals
25
. In light of observations that MACs profoundly alter the composition and function
of the microbiota and host physiology
31
, our findings raise the possibility that SCFAs are a critical part of the metabolic
landscapes tied to CDI status across individuals. To further address our hypothesis
that SCFAs affect Cd growth and toxin production in the gut, we conducted SCFA and
tributyrin feeding experiments in mice that were fed the MD1 diet. These experimental
approaches are both intended to increase the SCFAs (acetate, propionate, and butyrate,
or only butyrate, respectively) experimentally in the MD1 dietary condition, in the
absence of MAC introduction and accompanying microbiota metabolic activity. Specifically,
starting at 1 day pre-infection, mice were fed a cocktail of acetate, propionate,
and butyrate in their drinking water ad libitum
32
or were gavaged daily with 200 µL tributyrin
30
(see Methods and Fig. S14). There was neither a significant difference in the levels
of Cd recovered from the feces of these mice nor a significant difference in the levels
of TcdB detected in the feces of these mice. The levels of SCFAs in cecal contents
at D7 post infection also was not different between treatment groups.
SCFAs are metabolites that play regulatory and metabolic roles in the host and the
microbiota, and the pleiotropic roles that they play in mediating interspecies interactions
are just beginning to be appreciated. Currently, several technical hurdles exist to
experimentally manipulating SCFAs within the colon, presenting a tremendous opportunity
to develop methods that enable exploring this important class of fermentation end-products
and their effects on gut ecology. For example, the negative results reported in Fig.
S14 highlight the need to develop precise methods to reproducibly alter SCFA concentrations
in the distal gut. The cocktail of SCFAs that we used was previously shown to affect
SCFA levels in the cecal contents and feces of germ free mice
32,33
. Perhaps mice need to be provided with this cocktail for a longer period of time
to see an effect (at least 21 days in the above studies) or the presence of a resident
microbiome affects host absorption or microbial metabolism of SCFAs. We also gavaged
mice with tributyrin in an attempt to elevate levels of butyrate (which differentiates
the Cd permissive and Cd non-permissive dietary conditions, see Fig. 3a) in a more
targeted fashion in the distal gut
30,34
in our infected animals. In our work, we reasoned that the levels of butyrate would
remain elevated for the duration of the experiment if mice were gavaged daily with
tributyrin, especially given that a previous study reported elevated butyrate in the
ceca of mice at 24 hours post tributyrin gavage
30
. Notably, a difference between our work and the previous studies that used tributyrin
to alter levels of butyrate in the gut was timing of measurement of butyrate in the
ceca relative to antibiotic treatment. In both studies, butyrate was measured soon
after antibiotic treatment (within 35 hours post antibiotic treatment
34
). In our work, we measured SCFAs 8 days after antibiotic treatment. So it is possible
that differences in the microbiota at the time of analysis (e.g., antibiotics depleting
microbes that encode tributyrin-active lipases
35
in the small intestine, enabling delivery of higher concentrations to the cecum) are
the basis of the observed differences. Other factors that differentiate the experimental
set-up, such as the use of a MAC-deficient diet in our experiments, may also be contributing.
Previously, it was shown that MACs influence lipolysis of tributyrin in vitro
36
, raising the possibility that a MAC-rich diet might allow for higher levels of butyrate
to be delivered more distally in mice fed MAC-rich diets.
Current microbiota-centric therapies for CDI, such as fecal microbiota transplant
and probiotic administration, focus on the introduction of exogenous organisms. Our
work in mice shows that dietary intervention supports microbial communities that exclude
Cd without the requirement for microbe introduction. Regardless of the inter-experiment
and inter-animal variations in clearance kinetics (Fig. S8), the effect is highly
reproducible. Clearance kinetics may be further affected by host genetics, initial
microbiota composition, or overall dietary MAC concentration/composition.
Notably, two independent human trials have shown cooked green bananas (rich in MACs
as evident by elevated SCFAs in the stool of treated patients) aid host recovery from
another enteric pathogen, Shigella
37,38
. More recently, it was shown that a MAC-deficient diet leads to microbiota-dependent
mucus degradation and attachment-dependent lethal colitis by the murine pathogen,
Citrobacter rodentium
39
. Taken together, our work is part of a growing body of literature providing evidence
that dietary manipulation of the metabolic networks of the intestinal tract is a powerful
and underexplored way to influence gastrointestinal pathogens.
Methods
Media and bacterial growth conditions
Frozen stocks of Clostridium difficile (Cd) strain 630
40
were maintained under anaerobic conditions in septum-topped vials. Cd 630 was routinely
cultured on CDMN agar, composed of Cd agar base (Oxoid) supplemented with 7% defibrinated
horse blood (Lampire Biological Laboratories), 32 mg/L moxalactam (Santa Cruz Biotechnology),
and 12 mg/L norfloxacin (Sigma-Aldrich) in an anaerobic chamber at 37° (Coy). After
16–24 hours of growth, a single colony was picked into 5 mL of pre-reduced Reinforced
Clostridial medium (RCM, Oxoid) and grown for 16 hours. This 16-hour culture was used
to inoculate mice, below.
For in vitro experiments, Cd 630 was cultured on CDMN as above. Single colonies were
picked into pre-reduced Cd minimal medium (CDMM) without glucose, as described previously
41
. After 16 hours of growth, subcultures were prepared at a 1:200 dilution in pre-reduced
CDMM supplemented with 0, 10, or 30 mM of sodium acetate (Fisher), sodium propionate
(Sigma Aldrich), sodium butyrate (Sigma Aldrich), or sodium chloride (EMD Millipore)
in sterile polystyrene 96 well tissue culture plates with low evaporation lids (Falcon).
To further minimize evaporation of culture media during growth, the 36 wells along
the perimeter of the 96 well plates were filled with water rather than culture. Cultures
were grown anaerobically as above in a BioTek Powerwave plate reader. At 15-minute
intervals, the plate was shaken on the ‘slow’ setting for 1 minute and the optical
density (OD600) of the cultures was recorded using Gen5 software (version 1.11.5).
After 24 hours of growth, culture supernatants were collected after centrifugation
(5 minutes at 2,500 × g) and stored at −20°C for quantification of TcdB (see Quantification
of C. difficile toxin TcdB, below).
Murine model CDI
All animal studies were conducted in strict accordance with the Stanford University
Institutional Animal Care and Use Committee (IACUC) guidelines. Murine model CDI was
performed on age- and sex-matched mice between 8 and 17 weeks of age, possessing one
of three gut microbiota colonization states: (1) Humanized Swiss-Webster mice: female
germ free mice (SWGF, Taconic; bred in house) were inoculated with a fecal sample
obtained from a healthy anonymous donor, which was used in previous studies from our
laboratory
42,43
, (2) male conventionally-reared Swiss-Webster mice (SWRF, Taconic; bred in house);
or male C57BL/6 mice (B6EF, Taconic; experiments conducted on animals acquired directly
from vendor), and (3) Conventionalized mice: male germ free Swiss-Webster mice were
inoculated with a fecal sample obtained from SWRF mice. The gut microbiomes of the
humanized and conventionalized mice were allowed to engraft for at least 4 weeks
44
. The fecal sample used for humanizations was de-identified and obtained with informed
consent of the donor.
To initiate CDI, mice were given a single dose of clindamycin by oral gavage (1 mg;
200 µL of a 5 mg/mL solution) and were infected 24 hours later with 200 µL of overnight
culture grown in RCM (approximately 1.5×107 cfu/mL) or mock infected with 200 µL filter
sterilized PBS. To reactivate CDI in mice that had cleared the infection below detection,
mice were given a single dose of clindamycin as above.
Feces were collected from mice directly into microcentrifuge tubes and placed on ice.
To monitor Cd burdens in feces, 1 µL of each fecal sample was resuspended in PBS to
a final volume of 200 µL, 10-fold serial dilutions of fecal slurries (through 10−3-fold)
were prepared in sterile polystyrene 96 well tissue culture plates (Falcon). For each
sample, duplicate 10 µL aliquots of each dilution (technical replicates) were spread
onto CDMN agar. After 16–24 hours of anaerobic growth at 37°C, colonies were enumerated
and technical replicates were averaged to give cfu values (limit of detection = 2×104
cfu/mL feces). Cd was undetectable in all mice prior to inoculation with Cd (Figs.
1A, S1, S6–S8) and in all mice that were mock infected with PBS (Fig. S8), supporting
that the animals used in this work were not pre-colonized with Cd (e.g. Cd LEM1, as
seen by Etienne-Mesmin and colleagues
45
). After serial dilution of fecal samples, the remaining amounts of fecal samples
were immediately frozen at -80°C until needed for 16S rRNA analysis and TcdB ELISAs,
below. It was not possible to blind researchers to infection or dietary status of
the animals.
Cd burdens were chosen as the benchmark for significance between persistent infection
and MAC-mediated CDI suppression. The number of mice required per experimental group
was chosen based on a power calculation for continuous variables for a power of 80%
and a significance of 95% according to the equation n=1+2C(s/d)2, where n = individuals
required per experimental group, C= a constant based on significance and power (C=7.85),
s = estimated population standard deviation, and d = difference to be detected
46
. Based on the burdens of Cd in persistently infected mice fed the MD1 diet (Fig.
1a, 1–36 days post infection, 60 independent measurements), the mean log-transformed
Cd burden in feces is 7.58 log cfu/mL feces with a standard deviation of 0.59 log
cfu/mL feces. Assuming a 95% confidence interval and 80% experimental power, 2.37
mice (3 mice, rounded up) are needed to accurately determine at least a 2-log reduction
in Cd burdens from the persistently infected state. Therefore, our choice of 3–5 mice
per group provides adequate power to discern when the burdens of Cd have decreased
below 4.3 log cfu/mL feces.
Mouse diets
Mice were fed one of four diets in this study ad libitum: (1) a diet containing a
complex mixture of MACs (MAC+, Purina LabDiet 5010); (2) a custom MAC-deficient diet
47
[MD1, 68% glucose (w/v), 18% protein (w/v), and 7% fat (w/v) (Bio-Serv); (3) a commercially
available MAC-deficient diet [MD2, 34% sucrose (w/v), 17% protein (w/v), 21% fat (w/v);
Harlan TD.88137], or (4) a custom diet containing inulin as the sole MAC source
47
, which is based on the MD1 diet [58% glucose (w/v); 10% inulin (w/v) [Beneo-Orafti
group; OraftiHP]; 18% protein (w/v), and 7% fat (w/v) (Bio-Serv)]. Where applicable
(see Fig. S7), the drinking water of mice fed the MD1 diet was supplemented with 1%
inulin (w/v) [Beneo-Orafti group; OraftiHP]. Because mice consume approximately 5
grams of food per day and 5 mL of water per day
48
, water with 1% inulin gives an approximate 10-fold reduction in inulin consumed relative
to the 10% inulin diet. Where applicable (see Fig. S14), mice fed the MD1 diet were
gavaged daily with 200 µL tributyrin (Sigma Aldrich) or were given a cocktail of sodium
acetate (67.5 mM, Fisher), sodium propionate (25 mM, Sigma Aldrich), and sodium butyrate
(40 mM, Sigma Aldrich) in their drinking water
32
starting 1 day pre-infection. Groups of mice were randomly assigned to dietary conditions.
Histology and histopathological scoring
Proximal colon sections harvested for histopathologic analyses were fixed in 10% buffered
neutral formalin and routinely processed for paraffin embedding, sectioned at 4 microns,
mounted on glass slides and stained with hematoxylin and eosin (Histo-tec Laboratory,
Hayward, CA). Analyses were performed by a board certified veterinary pathologist,
using a semiquantitative scoring system
49
that evaluated distribution and severity of cellular infiltrates (inflammation), gland
hyperplasia, edema, and epithelial disruption using a severity score of 0 to 5 (0
= no significant lesion, 1 = minimal, 2 = mild, 3 = moderate, 4 = marked, 5 = severe,
see Table S4).
Quantification of C. difficile toxin TcdB
Levels of TcdB in culture supernatants and feces were quantified relative to a standard
curve of purified TcdB using the “Separate detection of C. difficile toxins A and
B” kit (TGC Biomics) according to the instructions that are readily available on the
manufacturer’s website. For culture supernatants which were harvested at 24 hours
post-inoculation (see Media and bacterial growth conditions, above), toxin abundance
was normalized by the final OD600 of the culture and “Normalized TcdB Abundance in
Culture Supernatant” = [(x ng toxin) / (y OD600)] and is reported in Fig. 3c. For
each fecal sample, toxin abundance was normalized by the number colony forming units,
as determined by selective culture on CDMN agar (above), for each sample; “Normalized
TcdB Abundance in Feces” = [(x ng toxin gram−1 feces) / (y cfu C. difficile mL−1 feces)]
and is reported in Fig. 4b. Non cfu-normalized TcdB abundance (ng toxin gram−1 feces)
in mouse feces is reported in Fig. S13.
16S rRNA amplicon sequencing and OTU picking methods
Total DNA was extracted from frozen fecal material using the PowerSoil DNA Isolation
Kit (MoBio) or the Powersoil-htp 96 well DNA isolation kit (MoBio). Barcoded primers
were used to amplify the V3–V4 region of the 16S rRNA gene from extracted bacterial
DNA using primers 515f and 806rB via PCR
50
. Amplicon cleanup was performed using the UltraClean PCR Clean-Up Kit (MoBio) and
quantification was performed using the high sensitivity Quant-iT dsDNA Assay Kit (Thermo
Fisher). Amplicons were pooled to an equimolar ratio. Amplicons from 3 mouse experiments
were sequenced in 3 different paired-end Illumina MiSeq runs, with each experiment
occurring on a separate run. The sample split/run corresponds to the field ‘Experiment’
in Table S1.
For commands executed for the 16S rRNA-based bioinformatics analysis, please see Code
S1, an ipython notebook. Runs were demultiplexed independently due to some non-unique
barcodes, and then concatenated prior to OTU picking using ‘split_libraries_fastq.py’
with default quality parameters in QIIME 1.9.1
51
. Open reference OTU picking was conducted with default parameters using the QIIME
script ‘pick_open_reference_otus.py’ (with default clustering algorithm UCLUST
52
) on the 24,582,127 reads that passed quality filtering. OTUs whose representative
sequence failed to align to the Greengenes reference alignment with at 85% identity
using PyNAST were discarded
53,54
.
We removed OTUs occurring in at least 10 samples, and/or having less than 26 counts
in the entire dataset. This filtering reduced the number of OTUs by 95.04% (211,884
to 10,504) but removed 5.2% of the feature-mass (23,293,178 to 22,078,743). This type
of filtering removes a vast number of features that are likely artefacts, boosts power
by reducing false discovery penalties, and concentrates analysis on biologically meaningful
features. We rarefied our data to correct for differences sequencing depth. To ensure
our results were not artefacts of rarefaction depth we conducted analyses at multiple
rarefaction levels and our conclusions were not changed. We use OTU tables rarefied
to 7,000 in this study, facilitating inter-run comparisons.
Supervised learning
Using the ‘supervised_learning.py’ script from QIIME 1.9.1, the random forests classification
method (with 10-fold cross validation error estimation) was trained using an OTU table
as prepared above in “16S sequencing and OTU picking methods.” Presence or absence
of Cd in a fecal sample (as determined by selective culture) or current diet were
used as the class label category, corresponding to the field ‘Plus_minus_Cd’ and ‘Current_diet’
of Table S1. The OTU table was modified for this analysis by querying each of the
11 Cd 630 rRNA sequences against a BLAST database built from the representative set
of OTUs created during OTU picking (see “16S sequencing and OTU picking methods”),
after which the OTUs that matched Cd 630 rRNA sequences (cutoff 97% identity) were
collapsed into a single Cd OTU, “k__Bacteria;p__Frimicutes;c__Clostridia;o__Clostridiales;f__Peptostreptococcaceae;g__Clostridioides;s__putative_difficile.”
See Code S1 for the code used for this analysis.
Quantification of short chain fatty acids (SCFAs)
Immediately following euthanasia at 32 days post infection, cecal contents were removed
from mice described in Fig. S6, weighed, and flash frozen in liquid nitrogen. Cecal
contents (70–150 mg) were suspended in a final volume of 600 µl in ice-cold ultra
pure water and blended with a pellet pestle (Kimble Chase) on ice. The slurry was
centrifuged at 2,350 × g for 30 seconds at 4°C and 250 µL of the supernatant was removed
to a septum-topped glass vial and acidified with 20µL HPLC grade 37% HCl (Sigma Aldrich).
Diethyl ether (500 µL) was added to the acidified cecal supernatant to extract SCFAs.
Samples were then vortexed at 4°C for 20 minutes on ‘high’ and then were centrifuged
at 1,000 × g for 3 minutes. The organic phase was removed into a fresh septum-topped
vial and placed on ice. Then, a second extraction was performed with diethyl ether
as above. The first and second extractions were combined for each sample and 250 µL
of this combined solution was added to a 300 µL glass insert in a fresh glass septum-topped
vial containing and the SCFAs were derivitized using 25 µL N-tert-butyldimethylsilyl-N-methyltrifluoroacetamide
(MTBSTFA; Sigma Aldrich) at 60°C for 30 minutes.
Analyses were carried out using an Agilent 7890/5975 single quadrupole GC/MS. Using
a 7683B autosampler, 1 µL split injections (1:100) were made onto a DB-5MSUI capillary
column (30 m length, 0.25 mm ID, 0.25 µm film thickness; Agilent) using helium as
the carrier gas (1 mL/minute, constant flow mode). Inlet temperature was 200°C and
transfer line temperature was 300°C. GC temperature was held at 60°C for 2 minutes,
ramped at 40°C/min to 160°C, then ramped at 80°/min to 320°C and held for 2 minutes;
total run time was 8.5 minutes. The mass spectrometer used electron ionization (70eV)
and scan range was m/z 50–400, with a 3.75-minute solvent delay. Acetate, propionate,
and butyrate standards (20 mM, 2 mM, 0.2 mM, 0.02 mM, 0 mM) were acidified, extracted,
and derivatized as above, were included in each run, and were used to generate standard
curves to enable SCFA quantification.
Measurement of doubling time for in vitro growth experiments
Raw OD600 measurements of cultures grown in CDMM (see “Media and bacterial growth
conditions,” above) were exported from Gen5 to MATLAB and analyzed using the growth_curve_analysis_v2_SCFA.m
script and analyze_growth_curve_SCFA.m function (Code S2 and Code S3, respectively).
Growth rates were determined for each culture by calculating the derivative of natural
log-transformed OD600 measurements over time. Growth rate values at each time point
were then smoothed using a moving average over 75-minute intervals to minimize artefacts
due to noise in OD measurement data. To mitigate any remaining issues with noise in
growth rate values, all growth rate curves were also inspected manually. Specifically,
in cases where the analyze_growth_curve_SCFA function selected an artefactual maximum
growth rate, the largest local maximum that did not correspond to noise was manually
assigned as the maximum growth rate. Doubling time was then computed by dividing the
natural log of 2 by maximum growth rate. The investigator that conducted growth curve
analysis was blinded to the experimental conditions in which growth curve data were
obtained.
Statistical methods
Alpha and beta diversity, correlations, and random forests were computed using QIIME
1.9.1 (‘alpha_diversity_through_plots.py’, ‘beta_diversity_through_plots.py’, ‘observation_metadata_correlations.py’,
‘supervised_learning.py’). Kruskal-Wallis, Mann-Whitey, Student’s T, ANOVA, and D’Agostino-Pearson
tests were performed using standard statistical analyses embedded in the Prism 7 software
package (GraphPad Software Inc.). Spearman correlations were calculated in Python
using Code S1 under the heading ‘feature correlations by diet’. Specific statistical
tests are noted in figure legends or tables as applicable.
Data availability
The data that support the findings of this study are available from the corresponding
author upon request. The 16S sequence data is uploaded to Qiita (https://qiita.ucsd.edu/study/description/11349).
Code availability
For custom code used in this study, see Code S1–S3.
Supplementary Material
1
2
3
Code 1
Code 2
Code 3