bioRxiv preprint doi: https://doi.org/10.1101/501940; this version posted December 19, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
1 Drivers of human gut microbial community assembly: Coadaptation,
2 determinism and stochasticity
3 Running title: Drivers of human gut microbiota assembly
4 Kaitlyn Oliphant1, Valeria R. Parreira1, Kyla Cochrane1, and Emma Allen-
5 Vercoe1
6 1Department of Molecular and Cellular Biology, University of Guelph,
7 Guelph, ON, Canada
8 Corresponding author – Kaitlyn Oliphant
9 University of Guelph, 50 Stone Road East, Guelph, ON, Canada N1G 2W1
10 phone: +1 519-824-4120 e-mail: [email protected]
11 Competing interests
12 EA-V is the co-founder and CSO of NuBiyota LLC, a company which is
13 working to commercialize human gut-derived microbial communities for
14 use in medical indications.
15
16
17
18
19
20
21
1 bioRxiv preprint doi: https://doi.org/10.1101/501940; this version posted December 19, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
22 Abstract
23 Microbial community assembly is a complex process shaped by
24 multiple factors, including habitat filtering, species assortment and
25 stochasticity. Understanding the relative importance of these drivers
26 would enable scientists to design strategies initiating a desired
27 reassembly for e.g., remediating low diversity ecosystems. Here, we
28 aimed to examine if a human fecal-derived defined microbial
29 community cultured in bioreactors assembled deterministically or
30 stochastically, by completing replicate experiments under two growth
31 medium conditions characteristic of either high fiber or high protein
32 diets. Then, we recreated this defined microbial community by
33 matching different strains of the same species sourced from distinct
34 human donors, in order to elucidate whether coadaptation of strains
35 within a host influenced community dynamics. Each defined microbial
36 ecosystem was evaluated for composition using marker gene
37 sequencing, and for behaviour using 1H-NMR based metabonomics. We
38 found that stochasticity had the largest influence on the species
39 structure when substrate concentrations varied, whereas habitat
40 filtering greatly impacted the metabonomic output. Evidence of
41 coadaptation was elucidated from comparisons of the two
42 communities; we found that the artificial community tended to exclude
43 saccharolytic Firmicutes species and was enriched for metabolic
44 intermediates, such as Stickland fermentation products, suggesting
2 bioRxiv preprint doi: https://doi.org/10.1101/501940; this version posted December 19, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
45 overall that polysaccharide utilization by Firmicutes is dependent on
46 cooperation.
47 Introduction
48 A critical knowledge gap in the field of microbial ecology is
49 understanding the relative contribution of the forces that drive
50 microbial community assembly. Uncovering this information would
51 facilitate the development of rationally-designed strategies to
52 remediate microbial communities exhibiting undesirable functionality or
53 successive progression after a perturbation. Such forces have been
54 proposed to include environmental selection, historical contingency,
55 dispersal limitation and stochasticity[1]. Environmental selection
56 additionally encompasses several distinctive sub-factors that yield well
57 to manipulation or measurement for predictions, including niche
58 availability (i.e., habitat filtering) and microbial interactivity (i.e., species
59 assortment and coadaptation)[2–4]. Bioreactors present a promising
60 strategy for studying the importance of such drivers, because culture
61 conditions within them can be tightly controlled, and the use of defined
62 microbial consortia can additionally serve to not only deconvolute the
63 system but also to allow for manipulation to address each factor
64 individually[5, 6]. Bioreactors are also currently utilized for both
65 research and industrial processes[7–13], and thus observing and
66 quantifying the contribution of these ecological forces on community
67 assembly is in itself useful information for these applications.
3 bioRxiv preprint doi: https://doi.org/10.1101/501940; this version posted December 19, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
68 The human gut microbial ecosystem (i.e., human gut
69 microbiota) is a suitable testing ground for ecological theory. This
70 ecosystem is known to be critical to health and well-being[14–16], with
71 alterations in both community structure and function reported in
72 several GI disorders[17, 18]. Proper succession of the human gut
73 microbiota during infancy and childhood is also essential to proper
74 development and education of the immune system[19], with such
75 deviations again associated with later onset of autoimmune
76 conditions[20–22]. Thus, therapeutics targeting the human gut
77 microbiota have been trialed in such cases, including probiotics and
78 fecal microbiota transplantation. However, the results of such clinical
79 trials have been mixed[23–25], with several factors having been found
80 to influence outcomes, including dosage, number of strains/donor, and
81 diet. Clearly, the availability of an ecological theoretical framework to
82 contextualize the environmental and microbial constitution would
83 improve the design process of these ecosystem interventions. Further,
84 bioreactor-based models such as the Simulator of Human Intestinal
85 Microbial Ecosystem (SHIME) system[26, 27] and single-vessel units[13],
86 are popular methodological approaches for investigating human gut
87 microbial ecology, due to their replicability, sample yields, cost and lack
88 of ethical constraints.
89 Thus, in our study, we aimed to quantify the relative impact of
90 two of the drivers of microbial community assembly, environmental
91 selection and stochasticity, in terms of both compositional species
4 bioRxiv preprint doi: https://doi.org/10.1101/501940; this version posted December 19, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
92 structure and metabolic behaviour, through use of a defined microbial
93 community derived from a human fecal sample, and single-vessel
94 bioreactor-based models. For the evaluation of environmental selection,
95 we chose to utilize different medium formulations that replicate a high
96 fiber, low protein and a high protein, low fiber diet, as diet has been
97 proposed to be the dominant environmental influencer acting on the
98 human gut microbiota[28]. We additionally scrutinized the two sub-
99 factors of environmental selection, habitat filtering and species
100 assortment, by including a second defined microbial community
101 matching the species constitution of the first, but where each bacterial
102 strain was sourced from a unique human donor. Therefore, the diets
103 would be a representation of habitat filtering, whereas the distinctive
104 communities would model species assortment or coadaptation. To
105 control for co-adaption to the dietary condition that could occur during
106 the initial assembly, we additionally introduced a dietary change after
107 allowing sufficient time for community equilibration to measure the
108 response to a relevant perturbation. Finally, for the evaluation of
109 stochasticity, we assessed the reproducibility of replicates using several
110 multivariate statistical methods to explore steady-state community
111 dynamics.
112 Materials and Methods
113 Creation of defined microbial communities
5 bioRxiv preprint doi: https://doi.org/10.1101/501940; this version posted December 19, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
114 Two defined microbial communities were created to examine
115 the effects of coadaptation on the dynamics of community assembly.
116 The first community served as a control, in which all bacterial strains
117 were derived from the same fecal sample. The second community was
118 constructed to match the species composition of the first (as
119 determined by aligning the 16S rRNA genes from each respective pair),
120 but with each bacterial strain sourced from a unique donor’s fecal
121 sample. The isolation methods and donor description of the control
122 community (CC) is described in Petrof et al.[29]; however, additional
123 species from this isolation round were added to improve the diversity of
124 the formulation (Table S1). The same isolation techniques were utilized
125 to derive the microbial strains for the second, ‘artificial’ community
126 (AC). The donors or international culture collections used to source each
127 species of the AC are indicated in Table S1.
128 Genomic DNA (gDNA) isolated from each strain was individually
129 16S rRNA gene sequenced using an Illumina MiSeq instrument (Illumina
130 Inc., Hayward, CA, USA) in order to use the high read count output as a
131 method to interrogate the purity of each sample. The gDNA was first
132 extracted from each strain following the protocol described in Strauss et
133 al.[30], and the library preparation, sequencing and data processing
134 were then conducted as described in the 16S rRNA based compositional
135 profiling section below. An average read depth in the 103 range was
136 typically achieved when sequencing single strains. The strains were
137 decidedly pure when all amplicon sequence variants (ASVs) that could
6 bioRxiv preprint doi: https://doi.org/10.1101/501940; this version posted December 19, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
138 not be attributed to the target species were of low abundance (< 1%)
139 and could be accounted for as sample cross-contamination through
140 referencing sample blanks. Any strains that were not pure were
141 subjected to serial dilution to extinction, as a method to improve
142 isolation, and were then re-evaluated by the above technique.
143 Bioreactor operation
144 A 500 mL Multifors bioreactor system (Infors AG,
145 Bottmingen/Basel, Switzerland) was inoculated with the defined
146 microbial communities and operated as a model of the human distal
147 colon as previously described[31]. The feed medium was designed to
148 replicate two dietary conditions, high fiber, low protein (HF) and high
149 protein, low fiber (HP), based upon the formulation in Marzorati et
150 al.[32] but modified to accommodate a single-vessel system as in
151 McDonald et al.[13] (Table S2). The experimental design is depicted in
152 Figure 1.
153 16S rRNA based compositional profiling
154 The gDNA from bioreactor samples was extracted through use
155 of the QIAamp Fast DNA Stool Mini Kit (Qiagen Inc., Germantown, MD,
156 USA) according to the manufacturer’s directions, but with extra steps
157 included to improve cell lysis. Prior to proceeding with their
158 recommended protocol, cells were first pelleted through centrifugation
159 at 14 000 rpm for 15 min at 4 °C. After resuspension in the lysis buffer,
160 0.2 g of zirconia beads (Biospec Products Inc., Bartlesville, OK, USA)
7 bioRxiv preprint doi: https://doi.org/10.1101/501940; this version posted December 19, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
161 were added, then the samples were bead-beat with a Digital Disruptor
162 Genie (Scientific Industries Inc., New York City, NY, USA) at 3000 rpm for
163 4 min. The samples were subsequently incubated at 90 °C for 15 min,
164 and finally, ultrasonicated at 120 V for 5 min (Branson Ultrasonics,
165 Danbury, CT, USA). Libraries for sequencing were constructed by a one-
166 step PCR amplification with 400 ng of Nextera XT Index v2 sequences
167 (Illumina Inc.) plus standard 16SrRNA v4 region primers[33] and 2 μL of
168 gDNA template in Invitrogen Platinum PCR SuperMix High Fidelity (Life
169 Technologies, Burlington, ON, Canada). Cycler conditions included an
170 initial melting step of 94 °C for 2 min, followed by 50 cycles of 94 °C for
171 30 s, annealing temperature for 30 s and 68 °C for 30 s, with a final
172 extension step of 68 °C for 5 min. The annealing temperature comprised
173 of a 0.5 °C increment touch-down starting at 65 °C for 30 cycles,
174 followed by 20 cycles at 55 °C. The PCR products were subsequently
175 purified using the Invitrogen PureLink PCR Purification Kit (Life
176 Technologies) according to the manufacturer’s directions. Normalization
177 and Illumina MiSeq sequencing was carried out at the Advanced
178 Analysis Center located in the University of Guelph, ON, Canada.
179 The obtained sequencing data was processed using R software
180 version 3.5 with the package DADA2 version 1.8, following their
181 recommended standard protocol[34]. Classification to the genus level
182 was additionally carried out via DADA2 using the SILVA database[35]
183 version 132. Classification to the species level, however, was conducted
184 by uploading the ASVs to NCBI BLAST (https://blast.ncbi.nlm.nih.gov)
8 bioRxiv preprint doi: https://doi.org/10.1101/501940; this version posted December 19, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
185 and selecting the identification with the highest percentage and lowest
186 e-value, while cross-referencing with the known species constitution of
187 the defined microbial communities. The data was then denoised by
188 adding the ASVs that returned identical species classifications together,
189 and after which the ASVs that equated to < 0.01% total abundance
190 across all samples were removed. Finally, the data was normalized by
191 center-log ratio transformation through use of the package ALDeX2
192 version 1.12, taking the median Monte-Carlo instances as the value[36].
193 The statistical analysis of this data is described in Figure 1.
194 1H-NMR based metabonomics
195 Sample preparation, 1H-NMR spectral acquisition and
196 processing, and profiling of metabolites was conducted as previously
197 described[31]. A Bruker Avance III 600 MHz spectrometer with a 5 mm
198 TCI 600 cryoprobe (Bruker, Billerica, MA, USA) at the Advanced Analysis
199 Center located in the University of Guelph, ON, Canada was utilized for
200 spectral acquisition. Spectra were collected at a sample temperature of
201 298 K. The data was analyzed using both an untargeted spectral binning
202 and targeted metabolite profiling approach with the Chenomx NMR
203 suite 8.3 (Chenomx Inc., Edmonton, AB, Canada). For spectral binning,
204 the default parameters of 0.04 ppm sized bins along the 0.04 – 10 ppm
205 region of the spectrum line with omission of water (4.44 – 5.50 ppm)
206 and normalization by standardized area (fraction of the chemical shape
207 indicator, DSS) were implemented. For metabolite profiling, target
208 regions of interest in the spectra were selected by partial least squared
9 bioRxiv preprint doi: https://doi.org/10.1101/501940; this version posted December 19, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
209 discriminant analysis of the spectral binning data (Figure 1). Metabolite
210 identifications were then based on the best fit for the peak regions with
211 the available libraries of compounds. The libraries included both the
212 internal set included with the Chenomx software suite, and the
213 downloaded HMDB[37] set release 2. The statistical analysis of both
214 data sets is described in Figure 1.
215 Results
216 Determination of microbial community ‘steady-state’ stability and
217 replicate reproducibility
218 For this work, it was essential to first determine at which day
219 the bioreactor-grown microbial communities had achieved a stable
220 equilibrium, referred to as ‘steady-state’, in order to make subsequent
221 comparisons. The CC reached steady-state both compositionally and
222 metabolically by day 4 (Table S3). Upon dietary change from HF to HP,
223 the microbial community was compositionally similar from the first time
224 point (day 16) until the end of the run (day 28), however, metabolic
225 stability was not reached until day 18. This latter observation is in line
226 with the calculated amount of time it would take for the bioreactor to
227 shed the excess 2000 mg/mL concentration of fiber from lingering fiber-
228 rich medium following medium change to HP composition, which is 4
229 days post medium change (Figure S1).
230 Results obtained running the AC under the same conditions
231 were much more variable (Table S3). Steady-state was reached for this
10 bioRxiv preprint doi: https://doi.org/10.1101/501940; this version posted December 19, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
232 community both compositionally and metabolically by day 2 in the HF
233 medium, with no significant differences observed at the first time point
234 of day 16 after the change in medium formulation. In the HP medium,
235 however, the AC followed the patterns of the CC more closely than in
236 the HF medium, except for taking longer to reach initial metabolic
237 stability (6 days compared to 4 days for the CC).
238 Next, the reproducibility between replicates of the same
239 condition was evaluated for both the normalized sequence count and
240 1H-NMR spectral binning data. Statistically significant differences
241 between replicates were found within all conditions for both data sets
242 (Table S4). The pairwise Mahalanobis distances ranged in magnitude
243 from 4.3 to 10.6 for the compositional data and 9.5 to 19.1 for the
244 metabolic data by PCA. The untargeted PAM clustering approach failed
245 to recapitulate the expected patterns, i.e., clustering the samples by
246 replicate. The ASWs ranged in magnitude from 0.23 to 0.28 for the
247 compositional data and 0.13 to 0.24 for the metabolic data. Stochastic
248 variation was thus clearly present across each replicate experiment, and
249 thus in order for the changes induced by an introduced environmental
250 pressure to be deemed statistically significant overall, we defined this to
251 mean that it must exceed the within-condition replicate separation
252 (Figure 2). This could be evaluated by overlap of ellipses and magnitude
253 of the Mahalanobis distances for the PCA approach, and recapitulation
254 of the expected clusters by untargeted PAM clustering with improved
255 ASWs for the Euclidean distance approach.
11 bioRxiv preprint doi: https://doi.org/10.1101/501940; this version posted December 19, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
256 Microbial community response to dietary changes
257 With the decided criteria from the above objective, i.e., the
258 between-group Mahalaonbis distance exceeding the found within-group
259 Mahalaonbis distance and untargeted PAM clustering recapitulating the
260 expected pattern of clustering samples by diet, neither community
261 altered its composition in response to dietary change (Table S4; Figure
262 2). However, the dietary change did elicit a clear significant difference in
263 the metabolite profile of both communities (Table S4; Figure 2). The
264 untargeted PAM clustering approach reliably recapitulated two clusters,
265 each containing only the samples of one specific diet, as the best
266 solution. The ASWs also exceeded the values obtained from the
267 clustering solutions within conditions (replicates), at 0.29 and 0.25 for
268 the CC and AC respectively. The Mahalanobis distance derived from PCA
269 was also larger than the maximum value obtained between replicates of
270 the CC, i.e., 19.3 compared to 18.1. For the AC, however, the result was
271 less clear, as the Mahalanobis distance was less than the maximum
272 value obtained between replicates, i.e., 17.6 compared to 19.1. Finally,
273 there was no significant difference between communities that were
274 originally cultured in one medium compared to those eventually
275 cultured in the same medium following a period of culture in a different
276 medium suggesting that within-experiment adaption was not a
277 confounding factor (Table S4).
278 Based upon the above results, we determined which features
279 (individual taxa and metabolites) were statistically significantly different
12 bioRxiv preprint doi: https://doi.org/10.1101/501940; this version posted December 19, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
280 between the medium formulations of which the communities were
281 initially grown. The normalized sequencing data was used to evaluate
282 shifts in taxonomic abundance (Table S5), and as expected, no
283 statistically significant differences were found (data not shown). The 1H-
284 NMR metabolite profiles (Table S6), however, revealed > 15 metabolites
285 that exhibited significant changes in concentration in both the CC and
286 AC (Table S7). The concentrations of short-chain fatty acids (SCFAs) and
287 select metabolites of interest that were significantly different between
288 growth medium conditions are depicted in Figure 3 and Figure 4
289 respectively. Most of these alterations were identical between each
290 community, including increased concentrations of several amino acids
291 and their specific fermentation by-products, a lower concentration of
292 methanol and a higher concentration of uracil in the HP medium.
293 Community-specific deviations included increases in concentration of
294 several amino acids and succinate, and a decrease in concentration of
295 valerate for the CC in the HP medium, whereas the AC had a higher
296 concentration of isobutyrate and a lower concentration of glyoxylic acid.
297 Effect of coadaptation on microbial community structure and behaviour
298 Finally, the differences between the CC and AC in both media
299 were evaluated overall and between individual taxa and metabolites as
300 above, to determine if potential coadaptation impacted community
301 composition or behaviour. With the set criteria, there were no
302 significant differences between the overall compositional nor
303 metabolite landscape (Table S4). However, there were several
13 bioRxiv preprint doi: https://doi.org/10.1101/501940; this version posted December 19, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
304 significant differences between individual taxa and metabolites. For the
305 taxa, [Eubacterium] rectale (HF/HP; p-value = 5E-6/4E-4; effect size =
306 74%/62%), Faecalicatena fissicatena (HF/HP; p-value = 2E-5/6E-4; effect
307 size = 60%/54%) and Coprococcus comes (HF only; p-value = 1E-5; effect
308 size = 65%) were significantly altered. Upon examining the raw
309 sequence counts (Table S5), it was observed that both E. rectale and C.
310 comes were virtually undetected in the AC, with the latter likely only
311 reaching statistical significance in the HF condition due to its relatively
312 higher abundance in the CC. On the other hand, F. fissicatena was
313 present in both communities, but at a lower abundance in the CC. For
314 the metabolites, several amino acids, organic acids and uracil had
315 increased concentrations in the CC, whereas branched-chain fatty acids
316 reached higher concentrations in the AC (Table S6; Table S7; Figure 4).
317 Discussion
318 Understanding the drivers of community succession is not only
319 useful for researchers utilizing technologies to simulate microbial
320 ecosystems, such as bioreactors, but also can translate to real world
321 applications. For the human gut microbiota, such an understanding
322 could assist in the design of therapeutic strategies aiming to ameliorate
323 abnormalities exhibited in GI disorders, or the building of predictive
324 tools to project changes over time (for example, during infant
325 development). Here, we examined the relative impact of environmental
326 selection and stochasticity on gut microbial community succession
327 through use of HF and HP medium formulations. Further, we
14 bioRxiv preprint doi: https://doi.org/10.1101/501940; this version posted December 19, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
328 differentiated the effects of habitat and member coadaptation within
329 environmental selection, by creating both a microbial community
330 derived from a single fecal sample and an equivalent community (same
331 or highly similar species) with each member sourced from individual
332 donor fecal samples.
333 First, it was essential to determine the point at which the
334 microbial communities form a stable equilibrium, known as ‘steady-
335 state’. Removing time level variation is important when using
336 bioreactor-based models, as it eliminates technical artifact bias that can
337 confound results. We found that for the CC, steady-state was reached
338 by day 4. This result is much earlier than has been suggested by other
339 studies using SHIME[26, 27] or single-vessel systems[13]. There are
340 several potential reasons for this discrepancy. First, this previous work
341 has used observational techniques, such as moving window analysis or
342 Unifrac clustering, to determine the day at which the microbial
343 community had stabilized. These methods do not statistically validate
344 the variation, resulting in overestimation of the value. Second, steady-
345 state tends to be measured within each replicate, instead of assessed as
346 a dataset. Microbial community dynamics between replicates were not
347 found to be 100% reproducible. Therefore, ‘steady-state’ can ultimately
348 be defined as the point at which time level variation no longer exceeds
349 the extent of replicability, as upon reaching this range statistical tests
350 would no longer be confounded by this technical property. Finally,
351 biological differences could also attribute to this low value, since our
15 bioRxiv preprint doi: https://doi.org/10.1101/501940; this version posted December 19, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
352 experimental communities were of a reduced complexity compared to a
353 typical fecal ecosystem (for example, since there are fewer interaction
354 types and processes in our defined experimental communities
355 compared to fecal communities, equilibrium is achieved more quickly
356 when projected by relevant mathematical models, such as Lotka-
357 Volterra[38]). Of note, we recommend our approach of utilizing whole
358 sequence count and spectral binning data, as opposed to individual taxa
359 and metabolites, to gauge steady-state for a complex microbial
360 community. Not all metabolites stabilized in concentration at the same
361 time, and the rate at which individual metabolites reached stable
362 concentrations was not identical between replicates. Particularly, the
363 metabolites most frequently measured in human fecal-associated
364 communities due to their importance and dominant concentrations, the
365 SCFAs[39, 40], often stabilized faster than other metabolites e.g., amino
366 acid-derived fermentation by-products (Figure 3; Figure 4).
367 The property of significant variation between replicates of the
368 same microbial community under identical conditions is not a new
369 observation for bioreactors[7–11]. It has been termed ‘multistability’,
370 which occurs when numerous entities have non-linear interactions or
371 feedback loops, and thus multiple alternative stable states are able to
372 exist[41]. This phenomenon has been observed both in environmental
373 ecosystems in situ[42, 43], including the human gut microbiota, with the
374 best example of it being reported after antibiotic treatment[41, 44, 45].
375 Therefore, multistability is not an artifact of the modeling technology
16 bioRxiv preprint doi: https://doi.org/10.1101/501940; this version posted December 19, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
376 but rather an accurate representation of this naturally inherent
377 property. Switching between such states can occur either under a
378 gradual external influence or after experiencing a substantial
379 perturbation[41]. Bioreactors thus present a useful tool for quantifying
380 the extent of this stochasticity and determining which factors induce
381 microbial community reassembly through examining within run
382 variability over time. However, it is also important for researchers to
383 ensure that the applied perturbation in their experiment exceeds this
384 inherent variability in order to demonstrate an overall change in
385 microbial community composition or behaviour; simply obtaining a
386 significant p-value is not enough, as such a result can be attained when
387 comparing replicates of identical conditions. We therefore recommend
388 calculating the differences between and within conditions in the
389 Mahalanobis distance of groups after PCA and ASW plus cluster
390 membership after PAM clustering of Euclidean distance matrices (Figure
391 2).
392 We found that growth medium did significantly alter the
393 microbial community metabolic behaviour but not the composition. The
394 latter result is in line with previous observations that found only minor
395 alterations in microbial community species structure after a dietary
396 change[46–48], and the composition of the human gut microbiota is
397 also reportedly robust in adulthood, with a 60% microbial strain
398 retention rate in a five-year window[49]. Additionally, the metabolic
399 changes we saw were as expected, both proteolysis (higher amino acid
17 bioRxiv preprint doi: https://doi.org/10.1101/501940; this version posted December 19, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
400 concentrations) and amino acid fermentation (higher concentrations of
401 by-products specific to these metabolisms[15, 50]) increased in the HP
402 medium (Figure 4). Our resultant lack of difference between the
403 assembled and switched microbial communities growing in the same
404 medium is supported by Wu et al.[48], as the authors found that dietary
405 changes only influenced the species structure of the human gut
406 microbiota when maintained for a much longer term as opposed to ten
407 days. We would thus conclude that for the typical duration of bioreactor
408 experiments, it is not necessary to consider microbial community
409 adaptation as a confounding factor.
410 The control and artificial communities were similar, but not
411 identical to each other, as indicated by the non-significant differences in
412 overall composition and metabolic behaviour. In terms of species
413 structure, both E. rectale and C. comes failed to integrate into the AC.
414 Both species are saccharolytic and capable of degrading fructans e.g.,
415 inulins; E. rectale can also utilize starch and xylan-derived polymers[51–
416 53]. Intriguingly, Lozupone et al. built a co-occurrence network from
417 fecal metagenomic data collected from 124 unrelated adults and found
418 that C. comes co-occurred with E. rectale[54]. E. rectale in particular is
419 known to require cooperation with other species to utilize resistant
420 starches, as it is incapable of conducting this activity on its own[55, 56].
421 Lozupone et al. found that Bacteroides spp. additionally co-occurred
422 with E. rectale and C. comes, thus the Bacteroides ovatus strains in our
423 communities represent potential collaborators[54]. An elegant study
18 bioRxiv preprint doi: https://doi.org/10.1101/501940; this version posted December 19, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
424 conducted by Rakoff-Nahoum et al. discovered that a strain of B. ovatus
425 produced both membrane-bound and secreted forms of a glycoside
426 hydrolase capable of degrading inulin[57]. When the secreted form of
427 the enzyme was knocked-out, the fitness of B. ovatus was not impacted
428 when grown in monoculture but was significantly diminished when
429 grown in a community setting. Tuncil et al. additionally demonstrated
430 that different Bacteroides spp., namely Bacteroides thetaiotaomicron
431 and B. ovatus, had reciprocal glycan substrate preferences that were
432 maintained from monoculture to coculture[58]. The combination of
433 these two studies would indicate potential mechanisms of coadaptation
434 within a gut microbial ecosystem, such that species would adapt to
435 occupy unique niches or collaborate to exploit the same niche, at least
436 in terms of polysaccharide consumption. Therefore, it is possible that
437 the B. ovatus strain in the AC lacked secretory catabolic enzymes that
438 would have assisted E. rectale and C. comes in utilizing e.g., fructans, or
439 that the glycan substrate preference of the B. ovatus had shifted to
440 consuming the fructans that E. rectale and C. comes could have
441 otherwise degraded themselves, thus effectively outcompeting them.
442 Alternative explanations for our findings, beyond metabolism, might
443 include, for example, a communication mismatch between strains
444 through via quorum sensing[59] or incompatible antimicrobial defense
445 mechanisms[60]. This result thus not only revealed the presence of a
446 possible, cooperative microbial guild within the CC, but also
447 demonstrated that strain level variation is an important property to
19 bioRxiv preprint doi: https://doi.org/10.1101/501940; this version posted December 19, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
448 consider in studies aiming to examine microbial community
449 cooperation.
450 In terms of metabolic difference between the studied
451 communities, an intriguing observation was the elevated amount of
452 amino acid fermentation in the AC compared to the CC, as supported by
453 decreased amino acid concentrations and increased concentrations of
454 their specific fermentation by-products[15, 50]. Further, this heightened
455 activity would explain the lower amount of separation between clusters
456 by medium formulation when evaluating overall statistical significance
457 in the AC comparted to the CC. Particularly, there was evidence of
458 Stickland fermentation occurring (higher concentrations of isobutyrate,
459 isovalerate and valerate), a metabolism specific to the Firmicutes
460 (usually the Clostridia) (Figure 4)[61, 62]. An interesting finding by
461 Shoaie et al. was that when E. rectale was cocultured with B.
462 thetaiotaomicron, it switched its gene expression from fermenting
463 saccharides to amino acids[63]. The E. rectale strain in this case thus
464 likely responded to the introduced competition by occupying a different
465 niche. Together, these results would suggest that polysaccharide
466 utilization by the Firmicutes is dependent on a collaborative effort due
467 to this phylum’s more highly specialized metabolisms[64, 65], and in the
468 absence of cooperation, these species will become putrefactive instead.
469 Many GI disorders are characterized by an inherent low diversity[66],
470 and a loss of bacteria from clostridial cluster XIVa, which includes E.
471 rectale and C. comes[67, 68]. Further, there is often a concurrent lack of
20 bioRxiv preprint doi: https://doi.org/10.1101/501940; this version posted December 19, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
472 butyrate, produced mainly from fermentation by certain Firmicutes
473 members[69], and increased inflammation, which can be promoted by
474 the products of protein fermentation[15, 70, 71]. Therefore, we suggest
475 that an erosion of cooperative interactivity has occurred in these
476 situations, resulting in extinction of particularly dependent species and
477 altered behaviours from those that remain, which exacerbate
478 symptoms.
479 One limitation to our study was the methods utilized to
480 determine microbial strain purity. Despite our best efforts to confirm
481 purity, we found these techniques to be inadequate at detecting and
482 removing all contaminants. Specifically, in the CC, we found that
483 Akkermansia muciniphila bloomed in the bioreactors, which was initially
484 unexpected. Upon completing an Ak. muciniphila specific PCR[72] on
485 the gDNA collected from each strain, we found it was present in the
486 Acidaminococcus intestini stock. This result was unexpected, since Ak.
487 muciniphila was completely undetected by marker gene sequencing,
488 and since an average of >10 000 total reads were attained using this
489 method, its presence was indicated at a rate of less than one in every 10
490 000 cells (Table S8). When we completed a re-extraction and re-
491 sequencing of gDNA obtained from Ac. intestini cultured on FAA
492 supplemented with mucin, the preferred growth substrate of Ak.
493 muciniphila[73], we were then able to detect it; at this point it achieved
494 10% of the total growth (Table S8). To compensate for this unexpected
495 Ak. muciniphila load, we added a strain of Ak. muciniphila to the AC
21 bioRxiv preprint doi: https://doi.org/10.1101/501940; this version posted December 19, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
496 prior to conducting the bioreactor experiments for this community and
497 then scrutinized the sequencing count data for any other outliers once
498 completed (Table S6). We found an unexpected number of reads
499 classified as Phascolarctobacterium, and when Phacolarctobacterium
500 specific PCRs[74, 75] were conducted on the gDNA of the bioreactor
501 samples and strains, we found Phascolarcterobacterium faecium to be
502 present in all six replicates and the E. rectale stock of the AC, but not in
503 the CC. Again, it was present at a rate of 1 per 10 000 cells of E. rectale.
504 P. faecium did not reach high numbers, however, as it was not
505 statistically significantly increased in the AC when compared to the CC.
506 The fact that it had amounted to above 1% total percentage
507 contribution may be attributed to sequencing error or cross-
508 contamination (Table S5). We examined the possibilities as to how its
509 inclusion could confound our experiment. Interestingly, E. rectale did
510 not integrate into the AC, so any interactions resulting from
511 coadaptation between it and P. faecium were absent. Ac. intestini was
512 also able to colonize the CC with a contaminant present at an equivalent
513 amount as E. rectale, so we doubt its inclusion would have negatively
514 impacted the ability of E. rectale to incorporate into the AC either. P.
515 faecium is asaccharolytic and incapable of Stickland fermentation, and
516 instead consumes succinate as a substrate to produce propionate[74,
517 76]. Therefore, the inclusion of P. faecium would not have altered any of
518 our discussion points regarding polysaccharide utilization networks or
519 increased putrefactive activity. The only influence it thus potentially had
22 bioRxiv preprint doi: https://doi.org/10.1101/501940; this version posted December 19, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
520 was the significant decrease in the concentration of succinate and
521 increase in the concentration of propionate in the AC (Figure 3). We
522 thus determined that our experiment remained of sufficient validity to
523 answer our hypothesis, and decidedly continued due to a lack of
524 available and validated alternatives for detecting contaminants.
525 Hopefully, new technology that can improve sequencing accuracy,
526 reduce cross-contamination and enhance the obtained number of reads
527 will address this issue in the future; for a review of current next-
528 generation sequencing developments, please refer to Goodwin et
529 al.[77]. However, these drawbacks should presently be carefully
530 considered by other researchers.
531 We have concluded that stochasticity is a property inherent to
532 human gut microbial ecosystems but is exceeded by forces of
533 environmental selection, at least in terms of driving microbial
534 community behaviour. Substrate availability also seems to dictate
535 functionality over cooperative interactivity, but that does not preclude
536 the existence of coadaptation. Our work fits into the observations of
537 previous studies, that indicate microbial community assembly in the
538 human gut is a deterministic process[78], habitat filtering predominates
539 species assortment[2], competitive interactions are more numerous
540 than cooperative ones[79], but microbial guilds covering metabolic
541 modules exist[3, 4]. We have also suggested a methodology to elucidate
542 steady-state, that additional testing is required to determine overall
543 statistically significant differences of a treatment, and that microbial
23 bioRxiv preprint doi: https://doi.org/10.1101/501940; this version posted December 19, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
544 community alterations exhibited in GI disorders could result from a
545 break-down of cooperation. Future work studying microbial interactions
546 should consider strain level variation, and could, for example, compare
547 interactions between strains derived from ‘healthy’ versus low diversity
548 ecosystems.
549 Acknowledgements
550 We would like to acknowledge the Natural Sciences and
551 Engineering Research Council of Canada and Ontario Ministry of
552 Training, Colleges and Universities scholarships to KO for providing
553 funding. Funding was gratefully received from an NSERC Discovery
554 Grant and a National Institutes of Health R33 to EA-V. We are most
555 thankful to our collaborators, Dr. Mike Surette, McMaster University,
556 Canada and Dr. Sydney Finegold, VA Wadsworth Laboratory, USA for the
557 gift of suitable matching strains for our AC community, as well as
558 members of the EA-V lab for providing other needed isolates – Dr. Mike
559 Toh, Michelle Daigneault, Erin Bolte and Dr. Rafael Peixoto.
560 Competing Interests
561 EA-V is the co-founder and CSO of NuBiyota LLC, a company
562 which is working to commercialize human gut-derived microbial
563 communities for use in medical indications.
564 References
24 bioRxiv preprint doi: https://doi.org/10.1101/501940; this version posted December 19, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
565 1. Costello EK, Stagaman K, Dethlefsen L, Bohannan BJM, Relman DA.
566 The application of ecological theory towards an understanding of
567 the human microbiome. Science 2012; 336: 1255–1262.
568 2. Levy R, Borenstein E. Metabolic modeling of species interaction in
569 the human microbiome elucidates community-level assembly rules.
570 Proc Natl Acad Sci U S A 2013; 110: 12804–12809.
571 3. Larsen OFA, Claassen E. The mechanistic link between health and
572 gut microbiota diversity. Sci Rep 2018; 8: 2183.
573 4. Zhang C, Yin A, Li H, Wang R, Wu G, Shen J, et al. Dietary
574 Modulation of Gut Microbiota Contributes to Alleviation of Both
575 Genetic and Simple Obesity in Children. EBioMedicine 2015; 2: 968–
576 984.
577 5. Großkopf T, Soyer OS. Synthetic microbial communities. Curr Opin
578 Microbiol 2014; 18: 72–77.
579 6. Zhou J, Ning D. Stochastic Community Assembly: Does It Matter in
580 Microbial Ecology? Microbiol Mol Biol Rev 2017; 81: e00002-17.
581 7. Pagaling E, Strathdee F, Spears BM, Cates ME, Allen RJ, Free A.
582 Community history affects the predictability of microbial ecosystem
583 development. ISME J 2014; 8: 19–30.
584 8. Graham DW, Knapp CW, Van Vleck ES, Bloor K, Lane TB, Graham CE.
585 Experimental demonstration of chaotic instability in biological
586 nitrification. ISME J 2007; 1: 385–393.
25 bioRxiv preprint doi: https://doi.org/10.1101/501940; this version posted December 19, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
587 9. Zhou J, Liu W, Deng Y, Jiang Y-H, Xue K, He Z, et al. Stochastic
588 Assembly Leads to Alternative Communities with Distinct Functions
589 in a Bioreactor Microbial Community. mBio 2013; 4: e00584-12.
590 10. Kohrs F, Heyer R, Bissinger T, Kottler R, Schallert K, Püttker S, et al.
591 Proteotyping of laboratory-scale biogas plants reveals multiple
592 steady-states in community composition. Anaerobe 2017; 46: 56–
593 68.
594 11. Gast CJVD, Ager D, Lilley AK. Temporal scaling of bacterial taxa is
595 influenced by both stochastic and deterministic ecological factors.
596 Environ Microbiol 2008; 10: 1411–1418.
597 12. Van de Wiele T, Van den Abbeele P, Ossieur W, Possemiers S,
598 Marzorati M. The Simulator of the Human Intestinal Microbial
599 Ecosystem (SHIME®). In: Verhoeckx K, Cotter P, López-Expósito I,
600 Kleiveland C, Lea T, Mackie A, et al. (eds). The Impact of Food
601 Bioactives on Health: in vitro and ex vivo models. 2015. Springer,
602 Cham (CH).
603 13. McDonald JAK, Schroeter K, Fuentes S, Heikamp-Dejong I,
604 Khursigara CM, de Vos WM, et al. Evaluation of microbial
605 community reproducibility, stability and composition in a human
606 distal gut chemostat model. J Microbiol Methods 2013; 95: 167–
607 174.
608 14. Flint HJ, Scott KP, Duncan SH, Louis P, Forano E. Microbial
609 degradation of complex carbohydrates in the gut. Gut Microbes
610 2012; 3: 289–306.
26 bioRxiv preprint doi: https://doi.org/10.1101/501940; this version posted December 19, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
611 15. Portune KJ, Beaumont M, Davila A-M, Tomé D, Blachier F, Sanz Y.
612 Gut microbiota role in dietary protein metabolism and health-
613 related outcomes: The two sides of the coin. Trends Food Sci
614 Technol 2016; 57: 213–232.
615 16. Round JL, Mazmanian SK. The gut microbiota shapes intestinal
616 immune responses during health and disease. Nat Rev Immunol
617 2009; 9: 313–323.
618 17. Cho JA, Chinnapen DJF. Targeting friend and foe: Emerging
619 therapeutics in the age of gut microbiome and disease. J Microbiol
620 Seoul Korea 2018; 56: 183–188.
621 18. Daliri EB-M, Tango CN, Lee BH, Oh D-H. Human microbiome
622 restoration and safety. Int J Med Microbiol IJMM 2018.
623 19. Renz H, Holt PG, Inouye M, Logan AC, Prescott SL, Sly PD. An
624 exposome perspective: Early-life events and immune development
625 in a changing world. J Allergy Clin Immunol 2017; 140: 24–40.
626 20. Fujimura KE, Sitarik AR, Havstad S, Lin DL, Levan S, Fadrosh D, et al.
627 Neonatal gut microbiota associates with childhood multisensitized
628 atopy and T cell differentiation. Nat Med 2016; 22: 1187–1191.
629 21. Arrieta M-C, Stiemsma LT, Dimitriu PA, Thorson L, Russell S, Yurist-
630 Doutsch S, et al. Early infancy microbial and metabolic alterations
631 affect risk of childhood asthma. Sci Transl Med 2015; 7: 307ra152.
632 22. Kostic AD, Gevers D, Siljander H, Vatanen T, Hyötyläinen T,
633 Hämäläinen A-M, et al. The dynamics of the human infant gut
27 bioRxiv preprint doi: https://doi.org/10.1101/501940; this version posted December 19, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
634 microbiome in development and in progression toward type 1
635 diabetes. Cell Host Microbe 2015; 17: 260–273.
636 23. Quraishi MN, Widlak M, Bhala N, Moore D, Price M, Sharma N, et al.
637 Systematic review with meta-analysis: the efficacy of faecal
638 microbiota transplantation for the treatment of recurrent and
639 refractory Clostridium difficile infection. Aliment Pharmacol Ther
640 2017; 46: 479–493.
641 24. Cao Y, Zhang B, Wu Y, Wang Q, Wang J, Shen F. The Value of Fecal
642 Microbiota Transplantation in the Treatment of Ulcerative Colitis
643 Patients: A Systematic Review and Meta-Analysis. Gastroenterol Res
644 Pract 2018; 2018.
645 25. Sun J, Marwah G, Westgarth M, Buys N, Ellwood D, Gray PH. Effects
646 of Probiotics on Necrotizing Enterocolitis, Sepsis, Intraventricular
647 Hemorrhage, Mortality, Length of Hospital Stay, and Weight Gain in
648 Very Preterm Infants: A Meta-Analysis. Adv Nutr 2017; 8: 749–763.
649 26. Liu L, Firrman J, Tanes C, Bittinger K, Thomas-Gahring A, Wu GD, et
650 al. Establishing a mucosal gut microbial community in vitro using an
651 artificial simulator. PLOS ONE 2018; 13: e0197692.
652 27. Van den Abbeele P, Grootaert C, Marzorati M, Possemiers S,
653 Verstraete W, Gérard P, et al. Microbial community development in
654 a dynamic gut model is reproducible, colon region specific, and
655 selective for Bacteroidetes and Clostridium cluster IX. Appl Environ
656 Microbiol 2010; 76: 5237–5246.
28 bioRxiv preprint doi: https://doi.org/10.1101/501940; this version posted December 19, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
657 28. Xu Z, Knight R. Dietary effects on human gut microbiome diversity.
658 Br J Nutr 2015; 113: S1–S5.
659 29. Petrof EO, Gloor GB, Vanner SJ, Weese SJ, Carter D, Daigneault MC,
660 et al. Stool substitute transplant therapy for the eradication of
661 Clostridium difficile infection: ‘RePOOPulating’ the gut. Microbiome
662 2013; 1: 3.
663 30. Strauss J, White A, Ambrose C, McDonald J, Allen-Vercoe E.
664 Phenotypic and genotypic analyses of clinical Fusobacterium
665 nucleatum and Fusobacterium periodonticum isolates from the
666 human gut. Anaerobe 2008; 14: 301–309.
667 31. Yen S, McDonald JAK, Schroeter K, Oliphant K, Sokolenko S,
668 Blondeel EJM, et al. Metabolomic analysis of human fecal
669 microbiota: a comparison of feces-derived communities and
670 defined mixed communities. J Proteome Res 2015; 14: 1472–1482.
671 32. Marzorati M, Vilchez-Vargas R, Bussche JV, Truchado P, Jauregui R,
672 El Hage RA, et al. High-fiber and high-protein diets shape different
673 gut microbial communities, which ecologically behave similarly
674 under stress conditions, as shown in a gastrointestinal simulator.
675 Mol Nutr Food Res 2017; 61.
676 33. Caporaso JG, Lauber CL, Walters WA, Berg-Lyons D, Lozupone CA,
677 Turnbaugh PJ, et al. Global patterns of 16S rRNA diversity at a depth
678 of millions of sequences per sample. Proc Natl Acad Sci U S A 2011;
679 108 Suppl 1: 4516–4522.
29 bioRxiv preprint doi: https://doi.org/10.1101/501940; this version posted December 19, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
680 34. Callahan BJ, McMurdie PJ, Rosen MJ, Han AW, Johnson AJA, Holmes
681 SP. DADA2: High-resolution sample inference from Illumina
682 amplicon data. Nat Methods 2016; 13: 581–583.
683 35. Quast C, Pruesse E, Yilmaz P, Gerken J, Schweer T, Yarza P, et al. The
684 SILVA ribosomal RNA gene database project: improved data
685 processing and web-based tools. Nucleic Acids Res 2013; 41: D590–
686 D596.
687 36. Fernandes AD, Reid JN, Macklaim JM, McMurrough TA, Edgell DR,
688 Gloor GB. Unifying the analysis of high-throughput sequencing
689 datasets: characterizing RNA-seq, 16S rRNA gene sequencing and
690 selective growth experiments by compositional data analysis.
691 Microbiome 2014; 2: 15.
692 37. Wishart DS, Tzur D, Knox C, Eisner R, Guo AC, Young N, et al. HMDB:
693 the Human Metabolome Database. Nucleic Acids Res 2007; 35:
694 D521-526.
695 38. Gavina MKA, Tahara T, Tainaka K, Ito H, Morita S, Ichinose G, et al.
696 Multi-species coexistence in Lotka-Volterra competitive systems
697 with crowding effects. Sci Rep 2018; 8: 1198.
698 39. Koh A, De Vadder F, Kovatcheva-Datchary P, Bäckhed F. From
699 Dietary Fiber to Host Physiology: Short-Chain Fatty Acids as Key
700 Bacterial Metabolites. Cell 2016; 165: 1332–1345.
701 40. Morrison DJ, Preston T. Formation of short chain fatty acids by the
702 gut microbiota and their impact on human metabolism. Gut
703 Microbes 2016; 7: 189–200.
30 bioRxiv preprint doi: https://doi.org/10.1101/501940; this version posted December 19, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
704 41. Pepper JW, Rosenfeld S. The emerging medical ecology of the
705 human gut microbiome. Trends Ecol Evol 2012; 27: 381–384.
706 42. Caruso T, Chan Y, Lacap DC, Lau MCY, McKay CP, Pointing SB.
707 Stochastic and deterministic processes interact in the assembly of
708 desert microbial communities on a global scale. ISME J 2011; 5:
709 1406–1413.
710 43. Dumbrell AJ, Nelson M, Helgason T, Dytham C, Fitter AH. Relative
711 roles of niche and neutral processes in structuring a soil microbial
712 community. ISME J 2010; 4: 337–345.
713 44. Dethlefsen L, Relman DA. Incomplete recovery and individualized
714 responses of the human distal gut microbiota to repeated antibiotic
715 perturbation. Proc Natl Acad Sci U S A 2011; 108 Suppl 1: 4554–
716 4561.
717 45. Gonze D, Lahti L, Raes J, Faust K. Multi-stability and the origin of
718 microbial community types. ISME J 2017; 11: 2159–2166.
719 46. Duncan SH, Belenguer A, Holtrop G, Johnstone AM, Flint HJ, Lobley
720 GE. Reduced Dietary Intake of Carbohydrates by Obese Subjects
721 Results in Decreased Concentrations of Butyrate and Butyrate-
722 Producing Bacteria in Feces. Appl Environ Microbiol 2007; 73: 1073–
723 1078.
724 47. Walker AW, Ince J, Duncan SH, Webster LM, Holtrop G, Ze X, et al.
725 Dominant and diet-responsive groups of bacteria within the human
726 colonic microbiota. ISME J 2011; 5: 220–230.
31 bioRxiv preprint doi: https://doi.org/10.1101/501940; this version posted December 19, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
727 48. Wu GD, Chen J, Hoffmann C, Bittinger K, Chen Y-Y, Keilbaugh SA, et
728 al. Linking Long-Term Dietary Patterns with Gut Microbial
729 Enterotypes. Science 2011; 334: 105–108.
730 49. Faith JJ, Guruge JL, Charbonneau M, Subramanian S, Seedorf H,
731 Goodman AL, et al. The long-term stability of the human gut
732 microbiota. Science 2013; 341: 1237439.
733 50. Kanehisa M, Goto S. KEGG: Kyoto Encyclopedia of Genes and
734 Genomes. Nucleic Acids Res 2000; 28: 27–30.
735 51. Scott KP, Martin JC, Duncan SH, Flint HJ. Prebiotic stimulation of
736 human colonic butyrate-producing bacteria and bifidobacteria, in
737 vitro. FEMS Microbiol Ecol 2014; 87: 30–40.
738 52. Rosero JA, Killer J, Sechovcová H, Mrázek J, Benada O, Fliegerová K,
739 et al. Reclassification of Eubacterium rectale (Hauduroy et al. 1937)
740 Prévot 1938 in a new genus Agathobacter gen. nov. as
741 Agathobacter rectalis comb. nov., and description of Agathobacter
742 ruminis sp. nov., isolated from the rumen contents of sheep and
743 cows. Int J Syst Evol Microbiol 2016; 66: 768–773.
744 53. HOLDEMAN LV, MOORE WEC. New Genus, Coprococcus, Twelve
745 New Species, and Emended Descriptions of Four Previously
746 Described Species of Bacteria from Human Feces. Int J Syst Evol
747 Microbiol 1974; 24: 260–277.
748 54. Lozupone C, Faust K, Raes J, Faith JJ, Frank DN, Zaneveld J, et al.
749 Identifying genomic and metabolic features that can underlie early
32 bioRxiv preprint doi: https://doi.org/10.1101/501940; this version posted December 19, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
750 successional and opportunistic lifestyles of human gut symbionts.
751 Genome Res 2012; 22: 1974–1984.
752 55. Cockburn DW, Orlovsky NI, Foley MH, Kwiatkowski KJ, Bahr CM,
753 Maynard M, et al. Molecular details of a starch utilization pathway
754 in the human gut symbiont Eubacterium rectale. Mol Microbiol
755 2015; 95: 209–230.
756 56. Cockburn DW, Suh C, Medina KP, Duvall RM, Wawrzak Z, Henrissat
757 B, et al. Novel carbohydrate binding modules in the surface
758 anchored α-amylase of Eubacterium rectale provide a molecular
759 rationale for the range of starches used by this organism in the
760 human gut. Mol Microbiol 2018; 107: 249–264.
761 57. Rakoff-Nahoum S, Foster KR, Comstock LE. The evolution of
762 cooperation within the gut microbiota. Nature 2016; 533: 255–259.
763 58. Tuncil YE, Xiao Y, Porter NT, Reuhs BL, Martens EC, Hamaker BR.
764 Reciprocal Prioritization to Dietary Glycans by Gut Bacteria in a
765 Competitive Environment Promotes Stable Coexistence. mBio 2017;
766 8.
767 59. Kerényi Á, Bihary D, Venturi V, Pongor S. Stability of Multispecies
768 Bacterial Communities: Signaling Networks May Stabilize
769 Microbiomes. PLOS ONE 2013; 8: e57947.
770 60. Cordero OX, Wildschutte H, Kirkup B, Proehl S, Ngo L, Hussain F, et
771 al. Ecological populations of bacteria act as socially cohesive units of
772 antibiotic production and resistance. Science 2012; 337: 1228–1231.
33 bioRxiv preprint doi: https://doi.org/10.1101/501940; this version posted December 19, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
773 61. Fischbach MA, Sonnenburg JL. Eating For Two: How Metabolism
774 Establishes Interspecies Interactions in the Gut. Cell Host Microbe
775 2011; 10: 336–347.
776 62. de Vladar HP. Amino acid fermentation at the origin of the genetic
777 code. Biol Direct 2012; 7: 6.
778 63. Shoaie S, Karlsson F, Mardinoglu A, Nookaew I, Bordel S, Nielsen J.
779 Understanding the interactions between bacteria in the human gut
780 through metabolic modeling. Sci Rep 2013; 3: 2532.
781 64. Rogowski A, Briggs JA, Mortimer JC, Tryfona T, Terrapon N, Lowe
782 EC, et al. Glycan complexity dictates microbial resource allocation in
783 the large intestine. Nat Commun 2015; 6.
784 65. Flint HJ, Scott KP, Duncan SH, Louis P, Forano E. Microbial
785 degradation of complex carbohydrates in the gut. Gut Microbes
786 2012; 3: 289–306.
787 66. Cho I, Blaser MJ. The Human Microbiome: at the interface of health
788 and disease. Nat Rev Genet 2012; 13: 260–270.
789 67. Lane ER, Zisman TL, Suskind DL. The microbiota in inflammatory
790 bowel disease: current and therapeutic insights. J Inflamm Res
791 2017; 10: 63–73.
792 68. Vital M, Karch A, Pieper DH. Colonic Butyrate-Producing
793 Communities in Humans: an Overview Using Omics Data. mSystems
794 2017; 2.
34 bioRxiv preprint doi: https://doi.org/10.1101/501940; this version posted December 19, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
795 69. Louis P, Flint HJ. Diversity, metabolism and microbial ecology of
796 butyrate-producing bacteria from the human large intestine. FEMS
797 Microbiol Lett 2009; 294: 1–8.
798 70. Fan P, Li L, Rezaei A, Eslamfam S, Che D, Ma X. Metabolites of
799 Dietary Protein and Peptides by Intestinal Microbes and their
800 Impacts on Gut. Curr Protein Pept Sci 2015; 16: 646–654.
801 71. Yao CK, Muir JG, Gibson PR. Review article: insights into colonic
802 protein fermentation, its modulation and potential health
803 implications. Aliment Pharmacol Ther 2016; 43: 181–196.
804 72. Collado MC, Derrien M, Isolauri E, de Vos WM, Salminen S.
805 Intestinal Integrity and Akkermansia muciniphila, a Mucin-
806 Degrading Member of the Intestinal Microbiota Present in Infants,
807 Adults, and the Elderly. Appl Environ Microbiol 2007; 73: 7767–
808 7770.
809 73. Derrien M, Vaughan EE, Plugge CM, de Vos WM. Akkermansia
810 muciniphila gen. nov., sp. nov., a human intestinal mucin-degrading
811 bacterium. Int J Syst Evol Microbiol 2004; 54: 1469–1476.
812 74. Wu F, Guo X, Zhang J, Zhang M, Ou Z, Peng Y.
813 Phascolarctobacterium faecium abundant colonization in human
814 gastrointestinal tract. Exp Ther Med 2017; 14: 3122–3126.
815 75. Watanabe Y, Nagai F, Morotomi M. Characterization of
816 Phascolarctobacterium succinatutens sp. nov., an Asaccharolytic,
817 Succinate-Utilizing Bacterium Isolated from Human Feces. Appl Env
818 Microbiol 2012; 78: 511–518.
35 bioRxiv preprint doi: https://doi.org/10.1101/501940; this version posted December 19, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
819 76. Del Dot T, Osawa R, Stackebrandt E. Phascolarctobacterium faecium
820 gen. nov, spec. nov., a Novel Taxon of the Sporomusa Group of
821 Bacteria. Syst Appl Microbiol 1993; 16: 380–384.
822 77. Goodwin S, McPherson JD, McCombie WR. Coming of age: ten years
823 of next-generation sequencing technologies. Nat Rev Genet 2016;
824 17: 333–351.
825 78. Jeraldo P, Sipos M, Chia N, Brulc JM, Dhillon AS, Konkel ME, et al.
826 Quantification of the relative roles of niche and neutral processes in
827 structuring gastrointestinal microbiomes. Proc Natl Acad Sci U S A
828 2012; 109: 9692–9698.
829 79. Coyte KZ, Schluter J, Foster KR. The ecology of the microbiome:
830 Networks, competition, and stability. Science 2015; 350: 663–666.
831 80. Szymańska E, Saccenti E, Smilde AK, Westerhuis JA. Double-check:
832 validation of diagnostic statistics for PLS-DA models in
833 metabolomics studies. Metabolomics Off J Metabolomic Soc 2012;
834 8: 3–16.
835 81. Anderson MJ. A new method for non-parametric multivariate
836 analysis of variance. Austral Ecol 2001; 26: 32–46.
837 82. Efron B, Tibshirani RJ. An Introduction to Bootstrap. 1994. Chapman
838 and Hall/CRC, London, UK.
839 83. Konietschke F, Bathke AC, Harrar SW, Pauly M. Parametric and
840 nonparametric bootstrap methods for general MANOVA. J Multivar
841 Anal 2015; 140: 291–301.
36 bioRxiv preprint doi: https://doi.org/10.1101/501940; this version posted December 19, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
842 84. Friedrich S, Pauly M. MATS: Inference for potentially singular and
843 heteroscedastic MANOVA. J Multivar Anal 2018; 165: 166–179.
844 85. Goodpaster AM, Kennedy MA. Quantification and statistical
845 significance analysis of group separation in NMR-based
846 metabonomics studies. Chemom Intell Lab Syst Int J Spons Chemom
847 Soc 2011; 109: 162–170.
848 86. Mahalanobis PC. On the generalised distance in statistics. Proc Natl
849 Inst Sci India 1936; 2: 49–55.
850 87. Hennig C, Liao TF. How to find an appropriate clustering for mixed-
851 type variables with application to socio-economic stratification. J R
852 Stat Soc Ser C Appl Stat 2013; 62: 309–369.
853 88. Sokolenko S, Blondeel EJM, Azlah N, George B, Schulze S, Chang D,
854 et al. Profiling Convoluted Single-Dimension Proton NMR Spectra: A
855 Plackett–Burman Approach for Assessing Quantification Error of
856 Metabolites in Complex Mixtures with Application to Cell Culture.
857 Anal Chem 2014; 86: 3330–3337.
858 89. Natividad JM, Pinto-Sanchez MI, Galipeau HJ, Jury J, Jordana M,
859 Reinisch W, et al. Ecobiotherapy Rich in Firmicutes Decreases
860 Susceptibility to Colitis in a Humanized Gnotobiotic Mouse Model.
861 Inflamm Bowel Dis 2015; 21: 1883–1893.
862 90. Human Microbiome Jumpstart Reference Strains Consortium,
863 Nelson KE, Weinstock GM, Highlander SK, Worley KC, Creasy HH, et
864 al. A catalog of reference genomes from the human microbiome.
865 Science 2010; 328: 994–999.
37 bioRxiv preprint doi: https://doi.org/10.1101/501940; this version posted December 19, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
866 Figure legends
867 Figure 1. Experimental design of bioreactor runs and statistical testing.
868 Bioreactors were either inoculated with the control microbial
869 community (CC) or the artificial microbial community (AC). Bioreactors
870 were either fed a high fiber, low protein medium or a high protein, low
871 fiber medium, which are distinguished by colour in the figure. After
872 allowing a 14-day equilibration, the medium formulations were
873 changed, and then the bioreactors were run for an additional 14 days.
874 Sampling occurred every 2 days, in which 2 x 2 mL were taken and
875 stored at -80 °C for 16S rRNA sequencing and 1H-NMR based
876 metabonomics, respectively. The box with the dotted line showcases
877 this sampling schematic. Both the composition, i.e., species structure,
878 and metabolic signature, i.e., binning data, were utilized for all
879 comparisons. Comparison 1 (*) was done to resolve steady-state;
880 samples were first divided into two groups, ‘early’ and ‘late’, and
881 moving window analysis was done using a valid orthogonal partial-least
882 squared discriminant analysis (O-PLS-DA) model, as determined by
883 permutation testing described in Szymańska et al.[80] using the R
884 software (version 3.5) package ropls 1.12. As well, the p-value
885 computed from a PERMANOVA of the Euclidean distance matrix as
886 described in Anderson[81] using the R package vegan version 2.5.2, with
887 permutations restricted in a repeated measures design [81, 82]. Only
888 steady-state time points were used for subsequent comparisons. The
889 remaining comparisons (**) conducted were within replicates of the
38 bioRxiv preprint doi: https://doi.org/10.1101/501940; this version posted December 19, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
890 same condition, between media (controlling for microbial community),
891 between microbial communities (controlling for medium), and between
892 the treatment versus starting diet. These tests included dimension
893 reduction by principal component analysis (PCA) then the Wald-type
894 statistic (WTS) for multivariate data as according to Konietschke et
895 al.[83] and the modified ANOVA-type statistic (MATS) as according to
896 Fredrich and Pauly[84]. Resampling was achieved with a parametric
897 bootstrap approach via the R packages ropls and MANOVA.RM version
898 0.3.1. Additionally, the pairwise Mahalanobis distances[85, 86] were
899 calculated using R package HDMD version 1.2. PERMANOVA of the
900 Euclidean distance matrix was also computed as described in
901 Anderson[81] using the R package vegan. Additionally, PAM clustering
902 was conducted, selecting the best solution from the average silhouette
903 widths (ASWs)[87] using the package fpc version 2.1.11. Metabolites of
904 interest were selected from partial least squared discriminant analysis
905 (PLS-DA, ***) of all samples grouped by the microbial community and
906 medium combination via variable importance of projection (VIP) scores
907 (> 1) using R package ropls. Statistically significant differences between
908 individual taxa and metabolites (****) were then determined between
909 media and between microbial communities grown within each medium
910 by a series of Kruskal-Wallis (KW) tests with Benjamini-Hochberg (BH)
911 correction and Dunn’s post-hoc analysis via the package dunn.test
912 version 1.3.5, in addition to the calculation of effect size (η2). Only
39 bioRxiv preprint doi: https://doi.org/10.1101/501940; this version posted December 19, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
913 features that remained significant after post-hoc analysis (q-value <
914 0.05) and had an effect size above 50% were considered significant.
915 Figure 2. Analysis of overall statistically significant differences in the 1H-
916 NMR spectral binning data obtained for the control community. Panel A
917 depicts samples from days 2 – 14 of the three bioreactors fed the high
918 fiber medium formulation, grouped by replicate (1 – 3). Panel B depicts
919 samples from days 2 – 14 of all six bioreactors grouped by medium
920 formulation (high fiber and high protein). The left panel is the result of
921 principal component analysis (PCA) conducted in R software version 3.5
922 by the package ropls version 1.12, with plots, including ellipses
923 assuming the multivariate t distribution, drawn by the package ggplot2
924 version 2.2.1. The right panel is the result of non-multidimensional
925 scaling (NMDS) of Euclidean distance matrices conducted by the
926 package vegan version 2.5.2, with plots, including the best solution of
927 untargeted partitioning around medoids clustering determined by
928 average silhouette width using the package fpc version 2.1.11, drawn by
929 the package ggplot2. Panel B presents a superior separation than panel
930 A, indicating that the effect of diet exceeds stochasticity.
931 Figure 3. Concentrations of short-chain fatty acids determined by 1H-
932 NMR metabolite profiling in bioreactor samples over time. Replicates 1
933 – 3 are the communities that were initially grown in the high fiber
934 medium formulation, whereas replicates 4 – 6 are the communities that
935 were initially grown in the high protein medium formulation. Panel A
936 depicts the control community, and panel B depicts the artificial
40 bioRxiv preprint doi: https://doi.org/10.1101/501940; this version posted December 19, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
937 community. Plots were drawn in R software version 3.5 by the package
938 ggplot2 version 2.2.1, with 10% error bars representing the expected
939 amount of technical measurement inaccuracy[88].
940 Figure 4. Concentrations of select metabolites that were statistically
941 significantly different between medium formulations or communities as
942 determined by 1H-NMR metabolite profiling in bioreactor samples over
943 time. Replicates 1 – 3 are the communities that were initially grown in
944 the high fiber medium formulation, whereas replicates 4 – 6 are the
945 communities that were initially grown in the high protein medium
946 formulation. Panel A depicts the control community, and panel B
947 depicts the artificial community. Plots were drawn in R software version
948 3.5 by the package ggplot2 version 2.2.1, with 10% error bars
949 representing the expected amount of technical measurement
950 inaccuracy[88].
41 bioRxiv preprint doi: https://doi.org/10.1101/501940; this version posted December 19, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. bioRxiv preprint doi: https://doi.org/10.1101/501940; this version posted December 19, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. bioRxiv preprint doi: https://doi.org/10.1101/501940; this version posted December 19, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. bioRxiv preprint doi: https://doi.org/10.1101/501940; this version posted December 19, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.