PREDICTED RELATIVE METABOLOMIC TURNOVER

Predicting Changes in the Environmental Metabolome from the Metagenome

Peter E. Larsen, Frank Collart, Folker Meyer and Jack A. Gilbert

Argonne National Laboratory, Argonne, Illinois, U.S.A.

Keywords: Metagenomics, Metatranscriptmics, Marine, Ecology, Metabolomics, Modeling, Metabolites.

Abstract: Metagenomics, the sequencing and analysis of genomic DNA extracted directly from an environment, can

provide insight into taxonomic and functional diversity, but there are few tools for directly comparing

metabolomes predicted from metagenomic data sets. We present a new method, Predicted Relative

Metabolomic Turnover (PRMT), for comparing the predicted environmental metabolomes encoded in

separate metagenomes and identifying those compounds predicted to be differentially metabolized. The

PRMT method was validated using three separate sets of ocean metagenomic sequence studies, totaling 15

metagenomic samples, over 4.5 million sequence fragments and over 840 million base pairs. These data sets

enable the construction of models representative of the environmental metabolome of the English Channel.

Not only did 88% of the predicted metabolic Predicted Metabolic Relative Turnover shows excellent

correlation with observed oceanographic parameters, but PRMT derived parameters are shown to generate

potentially constructive and testable biological hypotheses that could form the basis for future biological

experiments.

1 INTRODUCTION

Marine biomes dominate the planet’s surface and

single-celled microorganisms are responsible for up

to 98% of the oceans’ primary metabolic

productivity (Jørgensen and Boetius, 2007). These

extremely diverse microbial communities inhabit an

ocean zone containing the largest active pool of

near-surface carbon on the planet (Buesseler et al.,

2007) and are a dominant force in the planet’s

biogeochemical cycles. Understanding the nutrient

and carbon turnoveres of the world’s oceans has key

applications for understanding global ecology. One

now widely used tool for gaining insights into the

components and functionality of this ecosystem is

metagenomic sequencing.

The task of understanding the metabolic

interactions in any microbial community is a

daunting one and undertaking this effort in a

dynamic fluid like the ocean is even more complex.

This problem is compounded by the difficulty

associated with access to water sampling and

obtaining reliable measurements for the metabolites

for which we have reliable analytical methodologies;

let alone the vast number of metabolites and

compounds for which we have no reliable analytical

tools. Metabolomics approaches using techniques

such as NMR or GC-MS ((Bundy et al., 2009),

(Viant, 2007), (Viant, 2008), (Lin et al., 2006))

provide a snap shot of a fraction of the metabolites

present in an ecosystem and enable characterization

of the metabolic fingerprint of a given sample. These

are powerful tools for analyzing the relative

abundance of certain metabolites and can be used in

conjunction with genomic and transcriptomic

techniques to determine the relative importance of

key metabolites in biological processes. This is

important for modeling the metabolomic network

because not all genes in a metagenome are active

when observed, and not all transcripts will form a

functioning protein or enzyme. These constraints

indicate there is a pressing need to develop

techniques that leverage computational and omics

approaches, e.g. linking metagenomics to

metatranscriptomics to metaproteomics to meta-

metabolomics.

Metagenomic perspectives have traditionally

been focused on measuring the differences in

proportions of different genes which are annotated

from shotgun-sequenced datasets (e.g. (Rusch et al.,

2007)); but the real power comes from an ability to

link metabolism to metagenomic data, which could

vastly improve our understanding of the ecosystem

dynamics occurring within an environment. Previous

337

Larsen P., Collart F., Meyer F. and Gilbert J..

PREDICTED RELATIVE METABOLOMIC TURNOVER - Predicting Changes in the Environmental Metabolome from the Metagenome.

DOI: 10.5220/0003314803370345

In Proceedings of the International Conference on Bioinformatics Models, Methods and Algorithms (Meta-2011), pages 337-345

ISBN: 978-989-8425-36-2

 2011 SCITEPRESS (Science and Technology Publications, Lda.)

studies linking metabolomic profiles with nucleic

acid datasets have focused on interpretation of

metatranscriptomes by linking the presence of

specific metabolites to the change in abundance of a

specific protein (Santos et al., 2010). Currently,

methods are being developed to extract a wide

variety of environmental features from metagenomic

sequence data ((Wooley et al., 2010), (Heidelberg et

al, 2010)). Additionally, metagenomics is used to

help determine the taxonomic diversity of a

microbial community, using the taxonomic marker

gene 16S rRNA, and the functional diversity through

annotation of protein coding sequences through

comparison to curated protein databases such as

RefSeq (Pruitt et al., 2007), KEGG (Kanehisa and

Goto, 2000), KEGGnoggs (Muller et al., 2010),

SEED (Overbeek et al., 2005), in cases where the

sequence quality is enabling, more expensive HMM

type searches like PFAM (Finn et al., 2008) or

TIGRfams (Selengut et al., 2007), and linking

environmental conditions with specific biological

processes detected in metagenomic data (Gianoulis

et al., 2009). Finally, metagenomic bio-prospecting

is starting to be used by some groups to find

enzymes with potentially novel activities associated

with bioremediation, pharmaceuticals, and the

search for industrial biocatalysts (e.g.

http://metasystems.riken.jp/metabiome). While

informative, simply describing and counting the

annotations found in a metagenomic dataset will

only help to describe the functional differences

between two or more ecosystems. However,

metagenomic data can be used to make predictions

about the metabolic throughput for an ecosystem and

to generate hypotheses about what chemical

compounds are being actively consumed or

synthesized.

A traditional definition for metabolic turnover is

the rate of turnover of molecules through a

metabolic pathway. Though enzymes work at

reactions rates measured in seconds or minutes,

environmental samples are compared across much

different scales. Therefore, analyses that illustrated

metabolome changes occurring over hours or

months and over distances measured in meters or

miles are more important for understanding the

impact of global changes in the ecosystem. As a

result, we require a new approach, other than

turnover is needed to describe predicted changes in

environmental metabolomes at these large-scale

resolutions.

Here, we propose a methodology for predicting

relative turnover, which we define as the predicted

metabolic consequences of changes in the relative

abundance of genes for specific enzymatic activities

between metagenomic datasets. We call this

technique Predicted Metabolic Relative Turnover

(PRMT). PRMT is not intended as a replacement for

enzyme turnover measurements. Rather the goal of

the PRMT method is to predict the changes in

metabolic capacities of two or more metagenomics

samples and use that information to predict the

effects of those changes on the relative ability to

consume or synthesize specific metabolic

compounds. We will describe our methodology for

predicting metabolic relative turnovers and

demonstrate its validity by comparing our

predictions to measurements of specific

environmental parameters in two datasets from a diel

and seasonal time course of the coastal marine

observatory L4 in the Western English Channel.

2 RESULTS

To interpret metagenomic data in terms of the

synthesis or consumption of metabolites, it is

necessary to define a useful model of an

environmental metabolome and describe that

network of interactions in terms amenable to our

approaches. We must also derive a metric by which

different environmental metabolomes can be

compared. For the purpose of this study, we define a

predicted environmental metabolome as the

complete set of possible enzymatic reactions and the

metabolic compounds implied by those reactions for

the set of enzymes encoded in a set of environmental

metagenomes. A predicted environmental

metabolome can be expressed as a connectivity

matrix, which we term the Environmental

Metabolomic Matrix (EMM). From the annotations

of the predicted protein products in a metagenome,

we can derive a measurement we term the Enzyme

Gene Count (EGC), defined here as the number of

sequences in a metagenomic sample that are

predicted to code for proteins that are annotated with

a specific enzymatic activity. Hence, there is an

EGC value for every enzyme activity in an EMM.

To enable comparisons of individual metagenomes,

this Enzyme Gene Count needs to be normalized

(nEGC) by adjusting for the total EGCs in a

metabolome and to the total EGCs in the set of

metagenomes selected for analysis.

Our proposed method generates an EMM for

every set of metagenomes and a set of PRMT values

BIOINFORMATICS 2011 - International Conference on Bioinformatics Models, Methods and Algorithms

338

Figure 1: Complete predicted environmental metabolome of L4. The complete environmental metabolic capacity network

for the L4 environmental metabolomes, constructed from metagenomic analysis is pictured here. In this figure, nodes

represent metabolic compounds and edges are enzyme-mediated transformations. There are 2692 compounds and 4697

enzymatic transformations in this combined environmental metabolome. This figure was generated in Cytoscape 2.6.1 with

spring embedded layout (Shannon et al., 2003).

for every metabolic compound in an EMM for each

collected metagenome. A PRMT value predict a

change in the rate of turnover of a compound in one

metagenome-predicted metabolome relative to

another. A positive PRMT value indicates that a

reduced turnover is expected and that the compound

is predicted to be either more likely synthesized or

less likely consumed relative to a reference

metagenome. A negative PRMT value indicates

increased compound turnover is expected and that

the compound is less likely synthesized or more

likely consumed relative to a reference metagenome.

The ability of the PRMT approach to identify

relative changes in environmental metabolomic

capacities using metagenomic data was validated in

the context of three different sampling distributions:

an environmental metabolome sampled over the

course of a day, seasonal variation in ocean

ecosystems sampled in the course of a year, and with

increasing ocean depth from 10m to 4000m. For

these characterized sample sets, the predicted

changes in metabolism can be correlated with the

relative abundance for biological measurements of

available nutrients and environmental conditions. To

be a useful biological tool, PRMT must be able not

only to replicate prior observations, but it must also

be able to make relevant predictions about a

metabolome. Here, we show that PRMT can yield

testable hypotheses about specific environmental

metabolic interactions, by selecting sub-networks of

the larger predicted metabolome for careful analysis.

Two ocean metagenomic experimental data sets

from the English Channel were used for the

demonstrations of the PRMT approach. Four

metagenomic samples were collected at six hour

intervals at a sampling station in the English

Channel to track the day-night cycle of metabolite

turnover at the surface of the L4 coastal observatory.

Eight metagenomic samples at the same location

were used to investigate seasonal metabolomic

dynamics.

Variability in the population compositions is

observed in each of the metagenomic data sets used

for PRMT validation. The oceanographic parameters

associated with the metagenomic samples also

demonstrate substantial changes in environmental

conditions and nutrient availability in environmental

samples. The change in bacterial community

composition and in environmental parameters

suggests that there is good reason to expect a change

in the overall environmental metabolome across

sampled points in these metagenomic data sets.

Hence, PRMT is a beneficial tool to compare these

environmental samples and predict change in the

environmental metabolomes as a function of time or

depth.

The set of all EC activities for each metagenomic

sampling at L4 used to construct predicted

environmental metabolomes (Figure 1). The L4

metabolome is comprised of 2610 predicted

metabolites and 5067 EC-mediated interactions. The

largest single interconnected network of metabolic

interactions consists of 1551 metabolites and 4030

EC-mediated interactions. The second largest sub-

network has 39 metabolites and 68 EC-mediated

interactions. There are 298 very small sub-networks

comprised of two-metabolites each.

To evaluate the utility of the PRMT approach to

represent the environmental metabolomes, we first

consider how well predicted compound turnoveres

PREDICTED RELATIVE METABOLOMIC TURNOVER - Predicting Changes in the Environmental Metabolome from

the Metagenome

339

correlate with the relative abundance of several

measured oceanographic parameters (chlorophyll A,

total organic nitrogen, total organic carbon, NO

, ammonia, and soluble reactive phosphate) and

consider the results in light of biological

expectations. A positive or negative correlation of

RMF with relative environmental parameter

measurements conveys information about the

predicted network. A positive correlation, increasing

relative abundance of a parameter with increasing

RMF, indicates that when a compound is more

abundant in the environment so is the metabolic

capacity for the synthesis of that compound. A

negative correlation, increasing abundance of a

compound with decreasing RMF, indicates that

when a compound is more abundant in the

environment so is the metabolic capacity for its

consumption.

Diel metabolomic dynamics at the coastal L4

station. A diel cycle model was generated from four

metagenomic data sets derived from biological

samples obtained at the L4 station. In this analysis,

the PRMT model predictions for the relative

synthesis or catabolism of specific metabolites

represents absolute values for entire diel cycle. The

predicted PRMT for specific metabolites had a

significant, positive correlation with the direct

analytical measurements for relative abundance of

total organic nitrogen and total organic carbon

(Table 1). The positive correlation indicates

increased steady state levels (increased synthesis

and/or reduced turnover) of these compounds.

Additionally the PRMT predictions for chlorophyll

A, ammonia, and soluble reactive phosphorus,

demonstrated significant negative correlation with

the direct analytical measurements. However, the

predicted PRMT for nitrate and nitrite showed no

correlation with the observed measurements. The

positive correlation with total organic carbon and

nitrogen and negative correlation with chlorophyll a,

ammonia, and soluble reactive phosphorus indicates

that the PRMT model is potentially suggesting that

there is a synthesis of biomass occurring over the

course of a day, likely drawing on

photosynthetically-derived carbon and energy

inputs, and is consuming dissolved ammonia and

phosphorus in the process. At first look, the

significant negative correlation between PRMT-

values for chlorophyll a and the biological

measurements, indicating potentially increased

catabolic consumption of chlorophyll when it is

most abundant, appears counter-intuitive. It is

important however, to remember that PRMT does

not predict the absolute concentration or changes in

abundance of a compound in an environment, but

instead PRMT calculates the relative rate of turnover

of a metabolite in the pathway. In response to photo-

damage, there is continuous chlorophyll degradation

and repair in photosynthetic organisms and that

chlorophyll hydrolysis need not result in a net

decrease in chlorophyll concentrations((Vavilin and

Vermaas, 2007), (Beisel et al., 2010)). This can be

confirmed by selecting from the complete

metabolome model a sub-network of compounds

within three nearest neighbors of chlorophyll A

(Figure 2). This sub-network identifies a cyclical

metabolic loop by which chlorophyll A is

hydrolyzed and replenished.

Seasonal metabolomic dynamics at the coastal L4

station. The predicted PRMT values for the relative

abundance of measured metabolites between

January, April and August 2008 at the L4 coastal

observatory all demonstrate significant negative

correlations except for nitrate concentrations (Table

2). This significant negative correlation for the

PRMT indicates that between January and August

there is a relative increase in the consumption of all

metabolites, hence when a nutrient is available the

enzymatic capacity for the metabolism of that

Figure 2: Identified metabolism pathway for chlorophyll a

turnover in L4 diel cycle data. A sub-network was

generated from the complete L4 EMM by selecting all

compounds within three nearest neighbors of chlorophyll

a. In the network, chlorophyll a is highlighted in green and

the cycle of metabolic interactions predicted to be

responsible for chlorophyll a turnover are highlighted in

light green.

nutrient is also present. Obviously, the extended

time-frame for analysis permits greater variability in

measured parameters and hence represents periods

of tremendous turnover. However, this suggests that

increase in biological productivity between January

BIOINFORMATICS 2011 - International Conference on Bioinformatics Models, Methods and Algorithms

340

and August causes an overall decrease in the

availability of nutrients, as would be expected.

Table 1: Diel Cycle Correlation of Calculated PRMT and

Observed Parameters at the L4 Station.

Measured

Oceanographc

Parameter

KEGG

Metabolites

Correlation

with all

other

metabolites

Correlation

Chlorophyll A Chlorophyll a 0.00 (0.58)

-0.86

Total Organic

Nitrogen L-Amino acid 0.00 (0.57)

0.71

Total Organic

Carbon Starch 0.00 (0.72)

0.93

NO2 + NO3 Nitrite 0.01 (0.65) 0.24

Ammonia NH3 0.00 (0.51)

-0.76

SRP

Orthophosphat

e 0.00 (0.57)

-0.91

Table 2: Seasonal Correlation of Calculated PRMT and

Observed Parameters at the L4 Station.

Measured

Oceanographc

Parameter

KEGG

Metabolites

Correlation

with all other

metabolites

Correlation

Chlorophyll A Chlorophyll a 0.03 (0.65)

-0.76

Total Organic

Nitrogen L-Amino acid 0.03 (0.76)

-0.88

Total Organic

Carbon Starch -0.03 (0.78)

-1.00

NO2 + NO3 Nitrite -0.02 (0.63)

-0.69

Ammonia NH3 0.00 (0.64)

-0.96

SRP

Orthophosphat

e -0.02 (0.61)

-0.68

3 DISCUSSION

We have presented PRMT, a technique for

producing a prediction of the metabolic relative

turnover of specific metabolites between two or

more predicted environmental metabolomes inferred

from metagenomic datasets. This represents an

initial step on the way to true multi’omic

comparative analysis by creating predictive or

modeled meta-metabolomic data from metagenomic

data. PRMT can generate hypotheses about

environmental metabolism that immediately propose

experiments in environmental monitoring. PRMT

was applied to generate biological hypotheses, as for

identification of accelerated rates of chlorophyll

turnover in response to photo damage from sunlight,

and used to deduce the metabolism of biologically or

chemically difficult to measure environmental

parameters.

One of the most significant problems facing

ecological systems science is parameterization of the

ecosystem dynamics. Long term monitoring of any

site requires considerable investment and

infrastructure to provide the capacity for collecting

samples from the site and have co-acquisition of

environmental information, e.g. salinity, pressure,

temperature, porosity, humidity, light availability.

The problems regarding the measurement of these

physical parameters pales by comparison when we

consider the measurement of nutrients and other

biological metabolites, e.g. nitrate, nitrite, ammonia,

methane, sulfate, phosphate, etc. Measuring and

storing such a wide array of environmental

parameters requires not only extensive expertise but

also financial support ,which limited the extent of

many observatories. Ongoing and future projects

such as the Global Ocean Survey, TARA Oceans,

Hawaiian Ocean Time Series, Western Channel

Observatory, Bermudan Ocean Time Series, Long

Term Ecological Research sites, NEON,

Terragenome, etc., are going to great lengths to

include as many measurements as possible of

environmental parameters. These initiatives will

help to characterize the ecology of these

communities, however, they are still only broad

stroke analyses, there are potentially millions of

metabolites associated with the microbial

communities of any environment and to perform true

metabolomic interpretation of community dynamics

it would be necessary to examine the concentrations

of all of these and how they change with time or

space.

Although calculated PRMT values tended to

correlate well with relative abundance of measured

ocean parameters, there were a few cases where

good correlation was not observed. Measured ocean

parameters might not necessarily match intracellular

concentrations where metabolite concentrations are

being affected by the metabolism of more complex

molecules not represented in the KEGG enzymatic

reactions. There may also be instances for which the

number of times an enzymatic activity is detected in

the metagenome does not directly correlate with the

relative rate of enzyme activity. Even where there

was good correlation, the data presented in Tables 1-

3 indicate that the best correlations of predicted

metabolite’s PRMT value was not always with the

metabolite’s corresponding environmental

parameter. One reason for this is the high degree of

interdependence in the EMM. A metabolite’s PRMT

is not independent, but is closely linked to the

predicted metabolites in its immediate

neighborhood. Also, in some cases an exact

correspondence between measured environmental

parameter and KEGG metabolite. While, for

example, relative abundance of TOC and PRMT

values for the KEGG metabolite starch may

PREDICTED RELATIVE METABOLOMIC TURNOVER - Predicting Changes in the Environmental Metabolome from

the Metagenome

341

correlate well, there is no reason to think that there

might not be other carbon-related metabolites that

correlate as well or better. Finally, analysis was

partially hampered by the small number of samples

relative to the large number of calculated PRMT

values and nEGCs. With biological replication of

metagenomic data, the PRMT method will be able to

interpret results in the context of statistical

significance.

A question that remains is how much

metagenomic sequencing data is ‘enough’ to capture

a useful picture of an environmental metabolome.

This study analyzed approximately less than a 1000

of one percent of the genetic material in the sample

volume analyzed, and of that only less that 50% of

predicted proteins could be annotated. In seeking

better annotation of metagenomic sequence data, it

is possible that PRMT analysis will also have an

application. Looking for inconsistencies in the

predicted metabolic relative turnover when

compared to environmental parameters and then

using massively parallel computing to determine the

combination of sequences, both annotated and

orphaned, which would help to improve the

relationship between observed and predicted

turnover, may help to narrow down our

investigations regarding unknown or orphaned

metagenomic sequences.

Much enzymatic functional diversity in the ocean

has yet to be discovered and the KEGG metabolic

pathways do not contain all possible enzymatic

transformations in even the ocean environmental

metabolome. Although PRMT predicted the realized

turnovers in environmental metabolomes very well

in this study, with increased understanding of

environmental metabolomes, richer sets of known

metabolic interactions, and better annotations of

metagenomic data, PRMT will perform even better,

providing a powerful tool for the analysis of

metagenomic data. While we have chosen KEGG

databases and EC activities for this study, PRMT

could also accommodate definitions of enzyme

activity by methods other than EC annotation and

can utilize other databases of metabolic interactions

such as the curated databases of metabolic networks

Metacyc (Caspi et al., 2010) or BRENDA

(Schomburg et al., 2002). The ability to predict

relative metabolic consequences of changes in

environmental metabolomes encoded by sequenced

metagenomes is expected to have applications in

carbon management, bioremediation, and annotation

of previously unknown function in metagenomic

sequence data. As additional volumes of

metagenomic data are collected, as each sequenced

metagenome contains more sequence data, and as

the accuracy and completeness of metagenomic

sequence annotations continues to improve, the

PRMT method will be able to incorporate those

advances and expand its ability to serve as an

important tool for metagenomic analysis.

4 METHODS

The principle steps of the PRMT method are the

acquisitions of annotated metagenomic data sets, the

generation of EMM and nEGC from annotated

metagenomes, and calculating PRMT values using

EMM and nEGC. Generate Environmental

Metabolome Matrix (EMM) For the purpose of

this study, an Environmental Metabolome is the set

of all detected enzyme activities encoded in a

metagenome and all of the metabolite compounds

implied by those activities. One EMM is generated

for each set of metagenomic data collected from a

given environment. This network of predicted

metabolic reactions represents the theoretical

metabolic capability of an environmental

metabolome, not necessarily the actual reactions and

compounds associated with an environment or the

reactions taking place in an ecosystem. For this

study, the set of Kyoto Encyclopedia of Genes and

Genomes (KEGG) metabolic reactions were used

the represent the set of all possible metabolic

reactions in an environmental metabolome and

Enzyme commission number (EC) annotations of

activity are used to assign function to predicted

proteins encoded in metagenomes. KEGG contains a

large collection of manually curated metabolic

pathways taken from literature references and

published materials. EC numbers are a classification

scheme for enzyme activities based on the chemical

reactions they catalyze. KEGG reactions contain

identities for the reactant compound, product

compounds, and mediating EC activity. The KEGG

compounds water, di- and tri-phospho nucleotides,

NADP+, NADPH, and CoA were excluded from the

list of possible reactants as being non-specific to

particular reactions and metabolic processes. The set

of predicted interactions in the environmental

metabolome is converted into a connectivity matrix,

referred to as the Environmental Metabolome Matrix

(EMM) by the method. If an EC-mediated reaction

is identified as reversible in the KEGG database,

then both forward and reverse reactions are included

in the EMM. If a particular transformation is

attributed to more than one EC activity, then each

activity is incorporated into the EMM. Essentially,

BIOINFORMATICS 2011 - International Conference on Bioinformatics Models, Methods and Algorithms

342

metagenomic data is mapped to specific enzymes

that enable inference of metabolic compounds

associated with these activities. The sum of these

activities and compounds as organized into a cellular

network represents the predicted environmental

metabolome. The connectivity of this network can

be represented as a matrix and normalized by

restricting the sum of all inputs or outputs to 1 or -1,

respectively. Calculate Normalized Enzyme Gene

Counts (nEGC) for Each Metagenome For each

enzyme activity in the EMM, the Enzyme Gene

Count (EGC) is determined. The EGC is the number

of sequence reads in a metagenome that are

predicted to code for proteins that are annotated to

enzymes with a specific EC enzyme activity. For

each enzyme activity, the normalized EGC (nEGC)

is calculated using the following formula:

log

EC EC

All

Sample sample

Sample

Ave

nEGC EGC

Ave

⎛⎞

⎜⎟

⎝⎠

(1)

Where

ample

nEGC

is the normalized enzyme gene

count for activity ‘EC’ in a specific metagenome.

Sample

EGC

is the number of times enzyme activity

‘EC’ was assigned to a sequence in metagenome ‘x’.

Ave

Sample

is the average of all EGC in a metagenome.

Ave

All

is the average of all EGC in the set of all

analyzed metagenomes.

Calculation of Predicted Metabolic Relative

Turnover (PRMT) It is currently impractical to

model the metabolic turnover and metabolite

concentrations for every nutrient potentially present

in the ocean environment. However, we can model

the differences in two or more environmental

metabolic networks and predict the relative synthesis

or metabolism of specific metabolites for different

environmental metabolomes and their respective

nEGCs. Here we present a novel method, Predicted

Metabolic Relative Turnover (PRMT) Analysis, for

the analysis of metagenomic data and the prediction

of differential metabolic capacities between

metagenomes. For calculating PRMT, grant certain

assumptions about the predicted environmental

metabolic network. First, we assume that all EC-

activity mediated reactions can be modeled as zero-

order for the range of conditions encountered in the

environment. Second, we assume all reactions are at

steady state equilibrium. We also assume that the

expressed enzyme concentration is proportionate to

the frequency at which the gene for that enzyme is

found in the metagenome. Additionally, we assume

that the reaction rate of an enzyme is proportionate

to enzyme concentration for the conditions

encountered in the environment.

PRMT is calculated:

(

)

xyXyx

ggMc

−

(2)

Where

is the vector of PRMT for all compounds

in EMM in metagenome ‘x’ relative to metagenome

‘y’. M

is the connectivity matrix for predicted

Environmental Metabolism Matrix (EMM) for the

set of all metagenomes X.

and

are vectors of

normalized enzyme gene counts (nEGC) for all

enzyme activities in metagenomes for metagenome

‘x’ and metagenome ‘y’, in the set of metagenomes

X. An example of an application of PRMT using a

simple network can be found in supplemental file

Text S1.

Calculations of PRMT returns unit-less values,

analogous to a fold change, for the predicted relative

metabolic activity in an experimental metabolome

relative to a reference metabolome for each

compound in the EMM. A positive PRMT indicates

that a compound is either more likely synthesized or

less likely consumed in the experimental

metagenome. A negative PRMT indicates that a

compound is more likely consumed or less likely

synthesized in the reference metagenome. Like a

fold change, a PRMT does not contain information

regarding concentration. A positive or negative

PRMT does not suggest an absolute tendency

towards synthesis or consumption of a compound

respectively, only its predicted relative rate of

metabolic turnover compared to another

metagenome. PRMT Correlation with Biological

Observations For the metagenomes analyzed here

by PRMT, measurements of oceanographic

parameters were also collected. To validate the

PRMT method, we considered how well PRMT

predicted changes in environmental metabolomes

compared with the actual measured changes in

oceanographic parameters. Although there are few

direct overlaps between measured parameters and

available KEGG compounds that were used in

EMM, in most cases a suitable compound useful for

comparison could be selected, such as starch for

total organic carbon or orthophosphate for soluble

reactive phosphate. For each data set, for each type

of metadata collected for an environmental sample is

expressed as relative abundance, calculated as the

base 2 log of the measured parameter value divided

by the average of parameter measurements across all

data points. Pearson’s Correlation Coefficient (PCC)

was calculated for measured values for an

oceanographic parameter and the corresponding

PREDICTED RELATIVE METABOLOMIC TURNOVER - Predicting Changes in the Environmental Metabolome from

the Metagenome

343

PRMT value for a KEGG compound in the EMM. A

correlation close to one indicates that when a

compound is present in the environment, PRMT

predicts that the enzymes involved in its synthesis

are present. A correlation close to -1 indicates that

when a compound is present in the environment,

PRMT predicts that the enzymatic pathways relevant

for its catabolism are present. A PCC close to zero

suggests that there is no relationship between the

environmental presence of a compound and its

calculated PRMT. To estimate how well paired

parameter measurements and PRMT values

correlated was generated, metabolome PCC scores

between a measured parameter and every compound

PRMT values in an EMM was calculated. The

average and standard deviation for all metabolome

PCC scores was calculated. A correlation between a

measured oceanographic parameter and its paired is

considered successful if the PCC score was less than

the average minus one standard deviation for

metabolome PCC scores, or if the PCC score was

greater than the average plus the standard deviation

of metabolome PCC scores.

REFERENCES

Jørgensen B. B, Boetius A. (2007) Feast and famine--

microbial life in the deep-sea bed. Nat Rev Microbiol

10:770-81.

Buesseler K. O, Lamborg C. H., Boyd P. W., Lam P. J.,

Trull T. W., Bidigare R. R, Bishop J. K., Casciotti K.

L., Dehairs F., Elskens M., Honda M., Karl D. M.,

Siegel D. A., Silver M. W., Steinberg D. K., Valdes J.,

Van Mooy B., Wilson S. (2007) Revisiting carbon

turnover through the ocean's twilight zone. Science

316(5824):567-70.

Bundy J. G., M. P. Davey, M. R. Viant (2009)

Environmental metabolomics: A critical review and

future perspectives. Metabolomics 5, 3-21.

Viant M. R. (2007), Metabolomics of aquatic organisms:

the new omics on the block. Mar. Ecol. Prog. Series

332, 301-306.

Viant M. R. (2008), Recent developments in

environmental metabolomics. Molecular Biosystems 4,

980-986.

Lin C-Y, M. R. Viant and R. S. Tjeerdema (2006)

Metabolomics: Methodologies and Applications in the

Environmental Sciences. J. Pestic. Sci. 31, 245-251.

Rusch D. B., Halpern A. L., Sutton G, Heidelberg K. B.,

Williamson S., et al. (2007) The Sorcerer II Global

Ocean Sampling Expedition: Northwest Atlantic

through Eastern Tropical Pacific. PLoS Biol 5(3): e77.

doi:10.1371/journal.pbio.0050077.

Santos E. M., J. S. Ball, T. D. Williams, H. Wu, F. Ortega,

R. van Aerle, I. Katsiadaki, F. Falciani, M. R. Viant, J.

K. Chipman, C. R. Tyler (2010) Identifying health

impacts of exposure to copper using transcriptomics

and metabolomics in a fish model. Environ. Sci.

Technol. 44, 820-826.

Wooley J. C., Godzik A., Friedberg I. (2010) A Primer on

Metagenomics. PLoS Comput Biol 6(2): e1000667.

doi:10.1371/journal.pcbi.1000667.

Heidelberg K. B., Gilbert J. A. and Joint I. (2010).

Review: The revolution in genomic approaches to

describe environmental microbial diversity. Microbial

Biotechnology. In press.

Pruitt K. D., Tatusova, T., Maglott D. R. (2007) NCBI

Reference Sequence (RefSeq): a curated non-

redundant sequence database of genomes,

transcripts and proteins. Nucleic Acids Res

35(Database issue):D61-5.

Kanehisa, M. and Goto, S. (200) KEGG: Kyoto

Encyclopedia of Genes and Genomes. Nucleic Acids

Res. 28, 27-30.

Muller J, Szklarczyk D, Julien P, Letunic I., Roth A.,

Kuhn M., Powell S., von Mering C., Doerks T., Jensen

L. J., Bork P. (2010) eggNOG v2.0: extending the

evolutionary genealogy of genes with enhanced non-

supervised orthologous groups, species and functional

annotations. Nucleic Acids Res. (Database

issue):D190-5.

Overbeek R., Begley T., Butler R. M., Choudhuri J. V.,

Chuang H. Y., et al. (2005) The subsystems approach

to genome annotation and its use in the project to

annotate 1000 genomes. Nucleic Acids Res. 17:5691-

702.

Finn R. D., Tate J., Mistry J., Coggill P. C., Sammut S. J.,

Hotz H. R., Ceric G., Forslund K., Eddy S. R.,

Sonnhammer E. L., Bateman A. (2008) The Pfam

protein families database. Nucleic Acids Res. 2008

Jan;36(Database issue):D281-8. Epub 2007 Nov 26.

Selengut J. D., Haft D. H., Davidsen T., Ganapathy A.,

Gwinn-Giglio M., Nelson W. C., Richter A. R., White

O. (2007) TIGRFAMs and Genome Properties: tools

for the assignment of molecular function and

biological process in prokaryotic genomes. Nucleic

Acids Res. 35(Database issue):D260-4.

Gianoulis T. A., Raes J., Patel P. V., Bjornson R., Korbel

J. O., Letunic I., Yamada T., Paccanaro A., Jensen L.

J., Snyder M., Bork P., Gerstein M. B. Quantifying

environmental adaptation of metabolic pathways in

metagenomics. Proc Natl Acad Sci U S A. 2009 Feb

3;106(5):1374-9. Epub 2009 Jan 22.

Vavilin D., W. Vermaas (2007) Continuous chlorophyll

degradation accompanied by chlorophyllide and

phytol reutilization for chlorophyll synthesis in

Synechocystis sp. PCC 6803. Biochim Biophys Acta.

1767(7):920-9.

Beisel KG, Jahnke S, Hofmann D, Köppchen S, Schurr U,

Matsubara S (2010) Continuous turnover of carotenes

and chlorophyll a in mature leaves of Arabidopsis

revealed by 14CO2 pulse-chase labeling. Plant

Physiol 152(4):2188-99.

Gilbert J. A., Thomas S., Cooley N. A., Kulakova A.,

Field D., Booth T., McGrath J. W., Quinn J. P., Joint I.

(2009) Potential for phosphonoacetate utilization by

BIOINFORMATICS 2011 - International Conference on Bioinformatics Models, Methods and Algorithms

344

marine bacteria in temperate coastal waters. Environ

Microbiol 1:111-25.

Quinn J. P., Anna N. Kulakova, Natalie A. Cooley and

John W. McGrath (2007) New ways to break an old

bond: the bacterial carbon–phosphorus hydrolases and

their role in biogeochemical phosphorus cycling.

Environmental Microbiology 9(10), 2392–2400.

Panas, P., Ternan, N. G., Dooley, J. S., and McMullan, G.

(2006) Detection of phosphonoacetate degradation and

phnA genes in soil bacteria from distinct geographical

origins suggest its possible biogenic origin. Environ

Microbiol 8: 939–945.

Caspi R., Altman T., Dale J. M., Dreher K., Fulcher C. A.,

Gilham F., Kaipa P., Karthikeyan A. S., Kothari A.,

Krummenacker M., et al (2010) The MetaCyc

database of metabolic pathways and enzymes and the

BioCyc collection of pathway/genome databases.

Nucleic Acids Res 38: D473–D479

Schomburg I., Chang A., Schomburg D. BRENDA,

enzyme data and metabolic information. Nucleic

Acids Res. 2002 Jan 1;30(1):47-9.

Meyer F., Paarmann D., D'Souza M., Olson R., Glass E.

M. et al. (2008) The metagenomics RAST server - a

public resource for the automatic phylogenetic and

functional analysis of metagenomes. BMC

bioinformatics 9, 386.

Gilbert J. A., Field D., Huang Y., Edwards R., Li W.,

Gilna P., Joint I. (2008) Detection of large numbers of

novel sequences in the metatranscriptomes of complex

marine microbial communities. PLoS One 3(8):e3042.

DeLong E. F., Preston C. M., Mincer T., Rich V., Hallam

S. J., Frigaard N. U., Martinez A., Sullivan M. B.,

Edwards R., Brito B. R., Chisholm S. W., Karl D. M.

(2006). Community genomics among stratified

microbial assemblages in the ocean's interior. Science

311(5760):496-503.

Karl, D. M. and R. Lukas (1996) The Hawaii Ocean Time-

series (HOT) Program: Background, rationale and

field implementation. Deep-Sea Res. II, 43, 129-156.

Shannon P., Markiel A., Ozier O., Baliga N. S., Wang J.

T., Ramage D., Amin N., Schwikowski B., Ideker T.

(2003) Cytoscape: a software environment for

integrated models of biomolecular interaction

networks. Genome Res. 13(11):2498-504.

PREDICTED RELATIVE METABOLOMIC TURNOVER - Predicting Changes in the Environmental Metabolome from

the Metagenome

345