OPEN
ORIGINAL ARTICLE
Defining the functional traits that drive bacterial
decomposer community productivity
Rachael Evans
1
, Anna M Alessi
1
, Susannah Bird
1
, Simon J McQueen-Mason
1
, Neil C Bruce
1
and Michael A Brockhurst
1,2
1
Department of Biology, University of York, York, UK and
2
Department of Animal and Plant Sciences,
University of Sheffield, Sheffield, UK
Microbial communities are essential to a wide range of ecologically and industrially important
processes. To control or predict how these communities function, we require a better understanding
of the factors which influence microbial community productivity. Here, we combine functional
resource use assays with a biodiversity–ecosystem functioning (BEF) experiment to determine
whether the functional traits of constituent species can be used to predict community productivity.
We quantified the abilities of 12 bacterial species to metabolise components of lignocellulose and
then assembled these species into communities of varying diversity and composition to measure
their productivity growing on lignocellulose, a complex natural substrate. A positive relationship
between diversity and community productivity was caused by a selection effect whereby more
diverse communities were more likely to contain two species that significantly improved community
productivity. Analysis of functional traits revealed that the observed selection effect was primarily
driven by the abilities of these species to degrade β-glucan. Our results indicate that by identifying
the key functional traits underlying microbial community productivity we could improve industrial
bioprocessing of complex natural substrates.
The ISME Journal (2017) 11, 1680–1687; doi:10.1038/ismej.2017.22; published online 21 March 2017
Introduction
Microbial communities underpin the functioning of
natural ecosystems (Soliveres et al., 2016) and the
efficiency of a wide range of industrial bioprocesses
(for example, waste bioreactors) (Cydzik-
Kwiatkowska and Zielińska, 2016; Widder et al.,
2016). The form of the biodiversity–ecosystem
functioning (BEF) relation ship is therefore an impor-
tant property of microbial communities both in
nature and the simpler communities used in a range
of industrial bioprocesses. Several studies have
identified positive BEF relationships for microbial
community productivity (Bell et al., 2005; Gravel
et al., 2011), stability (Awasthi et al., 2014), micro-
pollutant degradation (Johnson et al., 2015) and
resistance to invasion (Elsas et al., 2012), suggesting
that for a range of functions microbial community
performance improves with increasing species rich-
ness. Positive BEF relationships can arise via the
complementarity effect, whereby diverse commu-
nities use more of the available resource space
through niche differentiation or facil itation (Salles
et al., 2009; Singh et al., 2015), or the selection effect
(also termed the sampling effect ), whereby diverse
communities are more likely to contain species
which have a large impact on community function-
ing (Hooper et al., 2005; Langenheder et al., 2012,
2010; Awasthi et al., 2014). Both complementarity
and selection effects depend on the functional traits
of constituent species and several studies have now
shown functional diversity to be a better predictor of
community function than phylogenetic diversity
(Mokany et al., 2008; Salles et al., 2009; Krause
et al., 2014). However, for many ecologically and
biotechnologically important microbial communities
it is still unclear how the functional traits of
individual species scale-up to determine the perfor-
mance of a diverse community.
One of the most important ecosystem functions
microbial communities perform is the decomposi-
tion of plant material and subsequent nutrient
cycling (Van Der Heijden et al., 2008; McGuire and
Treseder, 2010). Understanding the decomposition
of plant material also has important industrial
relevance. Plant biomass (collectively referred to as
lignocellulose) is the most abundant raw mater ial on
Earth (Pauly and Keegstra, 2008). It is typically
composed of approximately 40–50% cellulose, 20–
40% hemicellulose and 20–35% lignin which
together form a complex, recalcitrant structure
(Himmel et al., 2007; Liao et al., 2016). The high
sugar content and abundance of lignocellulose make
Correspondence: R Evans, Department of Biology, University of
York, Wentworth Way, York YO10 5DD, UK.
E-mail: re638@york.ac.uk
Received 13 September 2016; revised 21 December 2016; accepted
22 January 2017; published online 21 March 2017
The ISME Journal (2017) 11, 1680–1687
www.nature.com/ismej
it a promising substrate for biofuel production (Naik
et al., 2010). However, lignin is highly recalcitrant to
enzymatic attack causing a bottleneck in the efficient
conversion of lignocellulose to biofuels reducing
cost-effectiveness (Jorgensen et al., 2007; Naik et al.,
2010). Understanding how natural microbial co m-
munities (for example, in soils (Lynd et al., 2002),
compost (Lopez-Gonzalez et al., 2014) or termite guts
(Brune, 2014)) achieve efficient lignocellulose degra-
dation could inform both the prediction of nutrient
cycling in natural systems and the design of efficient
microbial communities for industrial processes (Wei
et al. , 2012). Both biodiversity and the presence of
certain species have been shown to influence the rate
of decomposition by bacterial communities (Bell
et al., 2005; Bonkowski and Roy, 2005; Langenheder
et al., 2012) but the mechanisms which determine
community decomposition performance remain
poorly understood (McGuire and Treseder, 2010).
A key question therefore is to what extent commu-
nity functioning is predictable from the combined
functional traits of constituent species?
Using culturable bacterial strains isolated from
compost we performed a random partition design
BEF experiment (Bell et al., 2009) to test the
contributions of species richness and composition
to productivity of communities when grown on
wheat straw. Although using only the culturable
fraction of the community is likely to overlook so me
functionally important species in the natural com-
munity, culturability is a key feature of microbes that
could feasibly be used in industrial bioprocessing.
Next we tested how the functional traits of indivi-
dual species shaped the productivity of these
communities to determine the extent to which
community productivity was predictable from the
functional traits of the constituent species and to
determine the contributi on of each func tional trait to
overall productivity. We quantified the functional
resource use traits of each species by their ability to
utilise a range of known components of lignocellu-
lose (that is, cellulose, hemicellulose, pectin and
lignin).
Materials and methods
Bacterial isolates
Bacterial strains used in this study were isolated
from wheat straw compost enrichment cultures
(700 ml M9 minimal media (22 mM KH
2
PO
4
,42mM
Na
2
HPO
4
,19mM NH
4
Cl, 1 mM MgSO
4
, 0.09 mM
CaCl
2
,9mM NaCl), 1% (w/v) wheat straw compost,
5% (w/v) milled wheat straw). Cultures were grown
on an orbital shaker (150 r.p.m.) at 30 °C for 8 weeks.
The enrichment culture process will have favoured
those species able to grow at 30 °C in a well aerated
environment which are required characteristics for
further experiments. As a result the isolated bacteria
used in this study do not provide a full representa-
tion of the complex microbial community present in
compost, but do represent a diverse collection of
naturally coexisting isolates that could potentially be
used in industrial biop rocessing. Each week serial
dilutions were prepared and spread onto nutrient
agar, potato dextrose agar and M9 minimal media
containing 1.5% (w/v) agar and 1% (w/v) milled
wheat straw. Single colonies that appeared morpho-
logically distinct on agar plates (Supplementary
Table 1) were assayed for activity against carbox-
ymethylcellulose (CMC) and xylan (both from
Sigma-Aldrich, Dorset, UK) using Congo red staining
assays (Teather and Wood, 1982). Species with
activity against CMC and/or xylan were identified
by 16S rRNA gene sequencing (16S sequences were
deposited in GenBank under the accession numbers
KX527645-KX527656). The twelve species included
in this study were chosen as they represent phylo-
genetic or functional diversity based on 16S rRNA
sequences and CMC and xylan assays (Supple-
mentary Figure 1; Supp lementary Table 1).
Biodiversity–ecosystem functioning experiment
Communities for the BEF experime nt were designed
using the random partition design described by Bell
et al. (2009). Species were randomly divided into
communities with species richness levels of 1, 2, 3,
4, 6 and 12 species with each isolate represented an
equal number of times at each richness level. This
process was repeated to produce 12 monocultures,
66 two-isolate communities, 60 three-isolate com-
munities, 63 four-isolate communities, 66 six-isolate
communities and one 12-isolate community. Each
community was replicated five times to give a total
of 1340 communities. The twelve species were
grown for two days in 5 ml nutrient broth on an
orbital shaker (150 r.p.m.) at 30 °C. Cultures were
harvested by centrifugation, washed and suspended
in M9 minimal media and left at room temperature
for 2 h to metabolise remaining nutr ients before
OD
600
was standardised to 0.1 to ensure similar
starting densities. Deep well plates containing 380 μl
M9 minimal media with 1% (w/v) milled wheat
straw per well were inoculated with a total of 120 μl
cultures, for example, monocultures were inoculated
with 120 μl single species culture whereas the 12
species community was inoculated with 10 μlof
each culture. The MicroResp system was used to
measure respiration of cultures (Campbell et al.,
2003). Briefly, each well in the deep well plate is
sealed to a microplate well containing indicator dye
which changes colour in response to CO
2
concentra-
tion. Microplates containing indicator gel were
replaced every 24 h to prevent cultures becoming
anaerobic. Community productivity was estimated
as cumulati ve respiration (Tiunov and Scheu, 2005;
Armitage, 2016). Specifically, cultures were grown
for 7 days at 30 °C and productivity was measured as
the cumulative change in absorbance (λ = 595 nm) of
the indicator gel immediately before and after being
sealed to deep well cultures plates. The change in
Functional traits predict community productivity
R Evans et al
1681
The ISME Journal
OD of the indicator gel from control wells containing
no inoculum was used to account for atmospheric
CO
2
concentration. Note that due to the presence of
particles of wheat straw in the growth medium it was
not possible to measure change in microbial biomass
by absorbance.
Functional trait assays
To quantify the fundamental niche of each species,
growth assays were performed on several polysac-
charides present in lignocellulose. Hemicellulose
substrates included xylan (Sigma-Aldr ich), arabinox-
ylan (P-WAXYL, Megazyme, Bray, Ireland) and
galactomannan (P-GALML, Megazyme); cellulose
substrates included β-glucan (P-BGBL, Megazyme)
and Whatman filter paper; additional substrates
included pectin (Sigma-Aldrich) and Kraft lignin
(Sigma-Aldrich). Cultures were prepared as
described for the BEF experiment. These cultures
(5 μl) were used to inoculate 495 μl of M9 minimal
media with 0.2% (w/v) of each carbon source or one
6 mm steri le filter paper disc in 96-well deep well
plates. Cultures were replicated six times and several
blank wells containing no inoculum were included
as negative controls. Cultures were grown for 7 days
at 30 °C and the MicroResp system was used to
measure culture respiration as described above.
Statistical analysis
The biodiversity and ecosystem functioning relation-
ship was analysed usi ng the linear model method
described by Bell et al. (2009). The species coeffi-
cients provided by this method give a measure of the
effect of each species on community productivity
relative to an average species: values of 41 indicate
an above average contribution while values of o1
indicate a below average contribution to community
productivity. To assess the effect of Paenibacillus sp.
A8 and Cellulomonas flavigena D13 on community
productivity, communities containing both species,
Paenibacillus sp. A8 only, C. flavigena D13 only or
neither of these species were compared using
analysis of variance (ANOVA) followed by post hoc
Tukey tests. Linear models were used to compare the
ability of species richness and the presence or
absence of Paenibacillus sp. A8 and C. flavigena
D13 to predict commun ity productivity.
To standardise measures of functional traits across
diverse substrates, performance on each substrate
was normalised by dividing by the maxim um
observed respiration on that substrate. For each
bacterial isolate we can then calculate its funda-
mental niche (along the carbon degradation axis) by
summing performance on all substrates. To estimate
the niche space of each community we used the
community niche (CN) metric described by Salles
et al. (2009), which sum s the maximal performance
on each substrate: CN ¼
P
7
i¼1
max
n
j¼1
P
ij
, where P
ij
is the performance of species j on carbon source i and
n is the number of species in each community.
The ability of each functional trait to predict
community productivity was analysed by summing
performance of all species in a community on each
carbon source to give a measure of the total
fundamental niche space of that community. To
approximate the realised niche space of commu-
nities we also a ssessed the ability of the maximum
performance on each carbon source in a community
to predict community productivity; this metric
assumes that the species best able to grow on a
given carbon source in a community dom inates
consumption of that carbon source providing a
conservative estimate of realised niche. Linear
regressions were used to analyse how well CN and
functional trait performa nce predicted community
productivity. It is important to note that be cause all
species can grow on several carbon sources, sum-
ming functional trait use may act as a proxy of
species richness. To control for this effect we
analysed whether summed community functional
traits remained significant when fitted to the resi-
duals of the species richness model (that is, commu-
nity productivity predicted by species richness).
Competing models were compared using the Akaike
information criterion.
Results
Biodiversity–ecosystem function relationship
We observed a positive relationship between species
richness and community productivity (F
1, 264
= 60.1;
Po0.001, Figure 1) with species richness explaining
19% of variation in productivity. As highlighted by
the variance in productivity within species richness
levels, species identity also had a significant effect
on community productivity (F
12, 254
= 45.3, Po0.001).
The linear model coefficient for each species
provides the estim ated contribution of that species
to community productivity relative to an average
species (Bell et al., 2009). Two species, Paenibacillus
sp. A8 and C. flavigena D13, made significantly
greater contributions to community function relative
to an average species (F
1, 254
= 73.1, Po0.001 and
F
1, 254
= 256.3, Po0.001 respectively, Supplementary
Figure 2). Of the remaining species, the contribution
of Rheinheimera sp. D14A and Stenotrophomonas
sp. D12, did not significantly differ from the average
species while the remaining eight species made
significantly below average contributions to commu-
nity functioning (Supplementary Figure 2).
To further investigate the effects of Paenibacillus
sp. A8 and C. flavigena D13, the productivity of
communities containing either one, both or neither
of these species was compared. Communities that
contained both Paenibacillus sp. A8 and C. flavigena
D13 were significantly more productive than com-
munities containing either one or neither of these
species (post hoc Tukey tests, Po0.001, Figure 1).
Functional traits predict community productivity
R Evans et al
1682
The ISME Journal
The productivity of communities containing both
Paenibacillus sp. A8 and C. flavigena D13 did not
significantly differ across species richness levels
suggesting additional species within these commu -
nities are not cont ributing to community productiv-
ity (F
1, 28
= 0.42, P40.05, green line Figure 1).
Communities containing only C. flavigena D13 were
more productive than those containing onl y Paeni-
bacillus sp. A8 (post hoc Tukey test, Po0.001), while
communities which did not contain these species
were significantly less productive than communities
containing either one of these species (post hoc
Tukey test, Po0.001). These results indicate that the
positive BEF relationship is predomin antly driven
by the selection effect, that is, more diverse commu-
nities are more likely to contain the highly perform-
ing species Paenibacillus sp. A8 and C. flavigena
D13 and are therefore more productive.
Quantification of functional traits
To determine if differences in productivity could be
explained by the function al traits of species we
assayed the ability of species to utilise various
components of lignocellulose. All species were able
to grow to varying degrees on the labile substrates,
hemicellulose (xylan, arabinoxylan and galactoman-
nan) and pectin, whereas growth on recalcitrant
substrates (β-glucan, filter paper and lignin) was less
universal (Figure 2). This pattern is consistent with
the hypothesis that functional groups that degrade
recalcitrant substrates are not as common as those
that degrade labile substrates (Schimel and Gulledge,
1998; Waldrop and Firestone, 2004). A linear model
revealed significant main effects of both species
(F
11,336
=30.3, Po0.001) and carbon source (F
6,336
=105.8,
Po0.001) on productivity and a significant interac-
tion between these factors (F
66, 336
= 6.8, Po0.001),
suggesting niche differentiation in resource use
among the species. It is notable that some species,
in particular Rhodococcus sp. E31, displayed gen-
eralist resource use, being able to grow on recalci-
trant substrates like lignin as well as on the more
labile substrates.
Community productivity and functional traits
To determine if the functional niche of communities
could be used to predict productivity we calculated
community niche as described by Salles et al. (2009).
This index sums the maximum growth achieved by a
constituent species on each substrate. We found a
significant positive relationship between community
niche and community product ivity (F
1, 264
= 73.31,
Po0.001, Figure 3a). Similar to the results of Salles
et al. (2009), community niche explained more
variation in community productivity than species
richness (22% and 19%, respectively).
When calculating community niche, each func-
tional trait is weighted equally despite differences in
the abundances of substrates in wheat straw ligno-
cellulose, for example, cellulose constitutes 40–50%
whereas pectin only constitutes 1–2%. To determine
which functional traits were important for predicting
community productivity we summed the growth of
constituent species on each carbon source used in
functional trait assays to calculate the total funda-
mental niche of that community. The summed
activity on β-glucan had a significant positive
relationship with productivity (F
1, 264
= 182.7,
Po0.001) and was the best predictor of community
productivity, explaining 41% of variation
(Figure 3b). The ability to utilise arabinoxylan and
xylan also had significant positive relationships with
productivity (F
1, 264
=105.8,Po0.001 and F
1, 264
= 98.6,
Po0.001, respectively), explaining 29% and 27% of
variation, respectively. There were significant posi-
tive relationships between the remaining carbon
sources and community productivity though these
explained less variation than community richness
and were not significant when species richness was
included in models. The fundamental niche space of
community is unlikely to be achieved due to
interactions between species such as competition
for resources. Therefore to approximate the realised
niche space of each community we also analysed the
maximum performance per carbon source in a
community. Consistent with the analysis of summed
performance, maximum performance on β-glucan,
arabinoxylan and xylan had significant positive
relationships with productivity (F
1, 264
= 134.8,
Po0.001, F
1, 264
= 76.2, Po0.001 and F
1, 264
= 74.5,
Po0.001, respectively) explaining 34%, 23% and
Figure 1 Relationship between community productivity and
species richness. Black line shows linear regression for all data
points (F
1, 264
= 60.1, R
2
= 0.19, Po0.001). Each point is the mean
productivity of five replicate communities. Points are coloured by
the presence or absence of C. flavigena D13 and Paenibacillus sp.
A8 and linear regressions between community productivity and
species richness are shown for each of these groups: green points
represent communities containing both these species (F
1, 28
= 0.42,
P40.05); red points represent communities containing C. flavi-
gena D13 (F
1, 50
= 4.43, Po0.05); blue points represent commu-
nities containing Paenibacillus sp. A8 (F
1, 50
= 1.01, P40.05); grey
points represent communities containing neither of these species
(F
1, 129
= 60.1, Po0.001). Productivity is measured as the cumula-
tive change in OD
595
of MicroResp indicator plate after 7 days
growth.
Functional traits predict community productivity
R Evans et al
1683
The ISME Journal
22% of variation, respectively. There were signifi-
cant positive relationships between the maximum
performance on lignin (F
1, 264
= 7.4, Po0.01), pectin
(F
1, 264
= 20.8, Po0.001) and galactomannan (F
1,
264
= 47.1, Po0.001) and community productivity
though these explained less variation than commu-
nity richness. Ther e was no significant relationship
between the maximum ability to degrade filter paper
and community productivity (Supplementary Table
2). This suggests that identifying and measuring key
functional traits could be a better predictor of
community product ivity than either species richness
or community niche.
Discussion
Understanding the factors that influence micr obial
community productivity has potentially important
ecological and industrial applications (Widder et al.,
2016). The ability of community niche to predict
functioning in well-defined media has been demon-
strated previously (Salles et al., 2009). Here, we
define for communities growing in complex unde-
fined media, the key functional resource use traits
that predict decomposer community productivity.
Crucially, functional resource use traits explained
more variation in productivity than either species
richness or measures of community niche. Indeed, a
single function, the ability to degrade β-glucan,
explained a larger proportion of variation than
community niche. This key functional trait was
shared by two dominant strains which were
shown to significantly increase the productivity of
communities.
As with several previous B EF studies (Bell et al.,
2009; Gravel et al., 2011; Awasthi et al., 2014), we
identified a posi tive relationship between species
richness and community productivity. By analys-
ing the effect of community composition we found
that the presence of two highly functioning species,
Paenibacillus sp. A8 and C. flavigena D13, sig-
nificantly increased community productivity sug-
gesting this positive BEF relationship is driven by
the selection effect. To determine i f t he dominance
of these two species could be explained by their
functional traits, we compared the ability of these
species to utilise the various carbon sources used in
functional trait assays to the other species. With the
exception of Rhodococcus sp. E31, Paenibacillus
sp. A8 and C. flavigena D13 were the highest
performing species on β-glucan (Figure 2). The
ability to utilise β-glucan may suggest these species
areabletometabolisethecelluloseportionof
wheat s traw in addition to the more labi le hemi-
cellulose and pectin fractions. Interestingly, when
the productivity of communities containing either
one, both or neither of these species is compared
across each day of the experiment (Supplementary
Figure 3), it is noticeable that communities contain-
ing neither of these s pecies have very low produc-
tivity during the lat er days o f t he experiment.
A possible explanation is that easily accessible
labile substrates are being used within the first two
days of growth after which only recalcitrant and
inaccessible substrates r emain. The ability to
degrade cellulose would allow Paenibacillus sp.
A8 and C. flavigena D13 to maintain higher levels
of growth when labile substrates become depleted.
Figure 2 Productivity of species grown on each carbon source. Filter paper and β-glucan represent cellulose like substrates (red); xylan,
arabinoxylan and galactomannan represent hemicelluloses (blue). Productivity is measured as the cumulative change in OD of MicroResp
indicator plates over 7 days.
Functional traits predict community productivity
R Evans et al
1684
The ISME Journal