These are slides from a talk at the MetaSUB International Consortium Virtual Meeting 2020 on 13th August 2020.

1. COMPUTATIONAL APPROACHES TO
UNDERSTANDING MICROBIAL
INTERACTIONS IN COMMUNITIES
KARTHIK RAMAN
Department of Biotechnology (Bhupat & Jyoti Mehta School of Biosciences)
Initiative for Biological Systems Engineering (IBSE)
Robert Bosch Centre for Data Science and Artificial Intelligence (RBCDSAI)

2. MICROBIAL COMMUNITIES: INTRODUCTION
▪Micro-organisms are ubiquitous, but rarely exist in isolation
▪Often thrive in a multi-species and dynamic environment1
▪Community structure dictated by a complex web of interactions
 Metabolic interactions predominate2!
▪Many microbiomes are also of great interest: soil, gut, skin, …
▪Microbial consortia/co-cultures are useful in many applications, e.g.
metabolic engineering
▪Understanding interactions is crucial
 develop microbial consortia for biotechnological applications3, or
 understand and manipulate the microbiomes e.g. gut
2
1Stolyar et al (2007) Mol Syst Biol 3:92
2Ponomarova O & Patil KR (2015) Curr Opin Microbiol 27:37
3Zhang H et al (2015) PNAS 112:8266

3. THE GUT MICROBIOME
▪A very complex ecosystem
▪Many roles in health and disease
▪Composition is highly variable
 Stage of life
 Diet
 Environmental exposure
 Antibiotic usage
▪Antibiotic usage is a major cause of gut dysbiosis
 Collateral damage to gut microflora
 Induce changes in composition
 Organisms develop resistance
3

4. GUT MICROBIOME AND ANTIBIOTICS
▪Gut microbiome recovers post antibiotic treatment
▪How long does the recovery take?
 Varies from individual to individual
▪Specific groups of organisms accelerate recovery
▪Recovery Associated Bacteria (RABs)1
 21 species identified
 Improved carbohydrate degrading capacity
 Specific synergy between Bacteroides thetaiotaomicron and Bifidobacterium adolescentis
▪Why do these organisms work well together?
4
1Chng et al (2020) Nat Eco Evol doi:10.1038/s41559-020-1236-0

5. HOW CAN WE
UNDERSTAND
MICROBIAL INTERACTIONS
IN COMMUNITIES, SUCH AS THE GUT?
CAN WE USE METABOLIC MODELS?

6. GENOME-SCALE METABOLIC NETWORKS
▪ Reconstructed for many organisms
▪ Present a comprehensive picture of known metabolic
reactions / transports happening in a cell
▪ ‘Draft’ reconstructions are readily obtained from
genome sequence, and databases like ModelSEED
▪ Many methods exist to analyse these networks
6
Image Courtesy of Sigma Aldrich, Order your copy from sigma-aldrich.com/metpath!

7. WHAT CAN GENOME-SCALE METABOLIC MODELS TELL US?
McCloskey D et al (2013) Mol Syst Biol 9:661
▪ Analysis of biological network
properties
▪ Metabolic engineering1
▪ Prediction of cellular phenotypes
▪ Model-driven (biological knowledge)
discovery
▪ Studies of evolutionary processes
▪ Interspecies interactions
1Badri A, Srinivasan A & Raman K (2017)
In silico approaches to metabolic engineering
ISBN 978-0-444-63667-6 pp. 161-200 7

8. Constraint-based Modelling
• Popular for applications such as metabolic engineering
• Demands well-curated models
Network-based (Graph-based) Modelling
• Path-finding in metabolic networks
• Predicting ‘new’ pathways based on atom-atom mapping/
reaction ‘rules’
Need Methods that
• are very scalable and accurate
• can figure all possible routes that exist
HOW TO ANALYSE GENOME-SCALE METABOLIC NETWORKS?
8

9. MODELLING MICROBIAL COMMUNITIES
▪ Again, many methods to model
 Constraint-based
 Graph-based
 Population-based
 Agent-based
▪ More methods being developed
▪ Many challenges1
 Mostly draft reconstructions available
 Difficult to make models talk to one another
▪ Metabolic interactions/exchanges in communities are of
particular interest
▪ Also has applications in understanding other communities,
e.g. gut microbiome
9
1Ravikrishnan A & Raman K (2015) Briefings in Bioinformatics 16:1057–1068
Ravikrishnan & Raman (2018)
ISBN: 978-113859671-9

10. ▪How to convert a metabolic network to a graph?
 Substrate graph
 Nodes: Metabolites
 Edges connect metabolites participating in the same reaction / reactants to products
 Reaction graph
 Nodes: Reactions
 Edges connect reactions sharing metabolites
 Bi-partite graph / Hypergraphs
 Nodes: two sets — metabolites and reactions
 Edges: connect reactants to reaction nodes and reaction nodes to product nodes
 No metabolite–metabolite or reaction–reaction links
▪‘Currency’ metabolites
 Need to be eliminated from substrate graphs!
 Else, we have a two-step glycolysis!
11
GRAPH REPRESENTATIONS OF METABOLIC NETWORKS

11. IDENTIFYING MICROBIAL
INTERACTIONS IN COMMUNITIES

12. OUR ALGORITHM: METQUEST
/RamanLab/MetQuest
Ravikrishnan A, Nasre M & Raman K (2018)
Scientific Reports 8:9932

13. METQUEST: OVERVIEW
▪Novel dynamic-programming based enumeration, which
assembles reactions into pathways of a specified size
producing a given target from a given set of source molecules
▪Employs two phases
 Guided Breadth First Search (BFS)
 Assembly of reactions into pathways
▪Implemented on Python 3.6 & Python 2.7
14

14. METQUEST:
KEY FEATURES
Requires only the topology of reaction
network (rather than stoichiometry and
atom mapping)
Is simple and scalable to larger metabolic
networks
Efficiently handles cyclic and branched
pathways
Examines multiple alternate routes of
conversion
15

15. ▪ Directed bipartite graph G of microbial communities
(consisting of more than one metabolic network) are also
easily constructed
▪ By connecting the graphs of individual organisms
through a common extracellular medium, based on the
overlapping set of exchange reactions
▪ The non-common exchange reactions are connected only
to the extracellular environment
16
HANDLING COMMUNITY METABOLIC NETWORKS

16. METQUEST: INPUTS TO THE ALGORITHM
▪Input to MetQuest
 a directed bipartite graph G derived from a given metabolic network
 a set of seed metabolites, S
 a set of target metabolites, T
 an integer β which bounds the size of any pathway generated
▪Seed Metabolites
 include the source metabolite(s)
 molecules such as co-factors and co-enzymes — commonly present in any cell
 akin to a “medium” for growth
17

17. ▪Input – Directed bipartite graph G derived from metabolic network(s),
seed metabolites
▪Output – Scope of metabolites, Reaction set that can be visited
GUIDED BFS: WALKTHROUGH
18

18. 19
ASSEMBLY OF REACTIONS INTO PATHWAYS

19. ▪Consortium of Clostridium cellulolyticum, Desulfovibrio vulgaris & Geobacter
sulfurreducens1 ⇒ Directed Bipartite graph constructed
▪Size of the network – 14265 nodes, 29073 edges
▪Size of scope – 1135 metabolites
▪Computed pathways of size 20, to all the metabolites within the scope of
cellobiose and other seed metabolites
▪In all the paths, acetate, pyruvate & ethanol were most frequently exchanged as
previously shown1
20
METQUEST ON MICROBIAL COMMUNITIES
1Miller LD et al (2010) BMC Microbiol 10:149

20. QUANTIFYING METABOLIC INTERACTIONS
IN COMMUNITIES
HOW WELL DO ORGANISMS
IN THE GUT SUPPORT ONE ANOTHER?

21. ▪How well does one organism support another?
 In terms of ‘relieving’ blocked reactions
 In terms of improving metabolic capabilities
▪Metabolic Support Index (MSI) measures fraction of
blocked/stuck reactions relieved by the presence of another
organism1
𝑀𝑆𝐼 𝐴; 𝐴 ∪ 𝐵 = 1 −
𝑛 𝑠𝑡𝑢𝑐𝑘, 𝐴;𝐴∪𝐵
𝑛 𝑠𝑡𝑢𝑐𝑘, 𝐴;𝐴
▪If all stuck reactions remain stuck, there is no benefit, i.e. MSI = 0
▪MSI seeks to quantify the extent of benefit in terms of metabolism
enabled
UNDERSTANDING POTENTIAL FOR SYNERGY
METABOLIC SUPPORT INDEX
1Ravikrishnan, Blank, Srivastava & Raman (2020) Comput Struct Biotechnol J 18:1249
22

22. 𝑛 𝑠𝑡𝑢𝑐𝑘,𝐴|𝐴 = 2
𝑛 𝑠𝑡𝑢𝑐𝑘,𝐵|𝐵 = 5
𝑛 𝑠𝑡𝑢𝑐𝑘,𝐴|𝐴∪𝐵 = 0
𝑛 𝑠𝑡𝑢𝑐𝑘,𝐵|𝐴∪𝐵 = 3
𝑀𝑆𝐼 𝐴 𝐴 ∪ 𝐵 = 1
𝑀𝑆𝐼 𝐵 𝐴 ∪ 𝐵 = 0.4
COMPUTING MSI
Individual metabolic networks
X
X
XX
X
X
X
Community metabolic network
✓ ✓ ✓
X
X
X
X
X
X
X
𝑀𝑆𝐼 𝐴; 𝐴 ∪ 𝐵 = 1 −
𝑛 𝑠𝑡𝑢𝑐𝑘, 𝐴;𝐴∪𝐵
𝑛 𝑠𝑡𝑢𝑐𝑘, 𝐴;𝐴 23

23. METABOLIC SUPPORT: KNOWN CO-CULTURES
WHICH ORGANISM BENEFITS MORE?
Co-culture organisms MSIA MSIB Experimental observations
Ketogulonicigenium vulgare
(A) & Bacillus megaterium (B)
0.224 0.016
B. megaterium is a helper strain for K. vulgare1
Yarrowia lipolytica &
Cellulomonas fimi
0.120 0.018
C. fimi provides additional metabolites to Y. lipolytica in a co-culture setup2
Desulfovibrio vulgaris &
Methanococcus maripaludis
0.179 0.036
D. vulgaris benefits the interaction with M. maripaludis3
Clostridium cellulolyticum &
C. acetobutylicum
0.052 0.003
C. acetobutylicum helps C. cellulolyticum to grow and metabolise cellulose
under non-favourable conditions through metabolic exchanges4
Pichia stipitis &
Saccharomyces cerevisiae
0.032 0.016
P. stipitis benefits from the interactions with S. cerevisiae: ethanol & acetalde-
hyde exchanged; involved in pathways producing 10+ amino acids5
In all cases, organism with higher MSI exhibited higher biomass in co-culture!
1Zhang J et al (2010) Process Biochem 45:602–606
2Stolyar S et al (2007) Mol Syst Biol 3:92
3Lubbe A et al (2017) Metabolites 7:4
4Salimi F & Mahadevan R (2013) BMC Biotechnol 13:95
5Ravikrishnan et al (2020) Comput Struct Biotechnol J 18:1249
24

24. GUT MICROBIOME: METABOLIC INTERACTIONS
▪21 microbial species — antibiotic recovery associated bacteria (RABs)1
▪Joint metabolic networks (of 21C2 =210 combinations of RAB organisms)
constructed
▪Microbial association network constructed based on MSI values
▪Also possible to analyse for specific metabolites exchanges, such as amino acids
1Chng et al (2020) Nat Eco Evol doi:10.1038/s41559-020-1236-0
25

25. MSI FOR GUIDING EXPERIMENTS
26
▪ RABs with high metabolic support to
other RABs: top 10% values shown
▪ Node sizes reflect the number of
incoming edges
▪ B. thetaiotaomicron and
B. adolescentis were evaluated
further in an in vivo model for
microbiome recovery
▪ Bt + Ba combination → recovery
of higher microbial diversity
Chng et al (2020) Nat Eco Evol doi:10.1038/s41559-020-1236-0

26. LIMITATIONS
▪ Only a static snapshot of interactions happening in the gut
▪ Nevertheless, graph-based approaches are very useful and complement constraint-
based models
▪ Some predictions agree with previous experiments, but many remain to be
validated
▪ MetQuest algorithm
 Predictions are obviously heavily contingent on the quality of the input network
 No weights or ranking attached to the metabolites/paths
 Difficult to identify very long pathways — but they may not be very interesting!
▪ Also, we view cellular interactions only through a metabolic lens — lots more
happening in reality!
27

27. SUMMARY

28. SUMMARY
▪MetQuest – a tool to identify metabolic exchanges and biosynthetic pathways
 Particularly interesting to identify metabolic cross-talks happening in a community
▪MSI – a metric to quantify support an organism receives from another in a
community
▪Generic framework to
 Understand microbial interactions
 Identify metabolic exchanges between organisms in a community
▪Paves way for
 Rational design of microbial consortia
 Obtaining mechanistic insights into microbial communities such as gut, skin microbiome etc.
 Useful to study the kinds of microbial communities unravelled by MetaSUB!
29

29. ACKNOWLEDGEMENTS
▪ Dr. Aarthi Ravikrishnan
 DST-INSPIRE
 DAAD-UGC Exchange Programme
▪ Dr. Meghana Nasre (CSE, IITM)
▪ Dr. Niranjan Nagarajan (A*STAR, Singapore)
30

30. Contact me at
kraman@iitm.ac.in
home.iitm.ac.in/kraman/lab
MORE INFO
/RamanLab/MetQuest

31. THANK YOU!