The Microbiome and Metagenomics

Report
The Microbiome and
Metagenomics
Catherine Lozupone
CPBS 7711
September 19, 2013
What is the microbiome?
• “The ecological community of commensal,
symbiotic, and pathogenic microorganisms that
share our body space”
• Microbiota: “collection of organisms”
Microbiome: “collection of genes”
• Bacteria, Archaea, microbial eukaryotes (e.g.
fungi or protists) and viruses.
• Body Sites
– Important roles in health and disease: Gut, Mouth,
Vagina, Skin (diverse sites:Nasal epithelial)
– Important roles in disease: Lung, blood, liver, urine
The big tree
• Majority of life’s
diversity is microbial
• Majority of microbial
life cannot be grown
in pure culture
Pace, N.R.,The Universal
Nature of Biochemistry. PNAS
Vol 98(3) pp 805-808.
The Human Gut Microbiota
• 100 trillion microbial cells: outnumber human
cells 10 to 1!
• Most gut microbes are harmless or beneficial.
– Protect against enteropathogens
– Extract dietary calories and vitamins
– Prevent immune disorders
• List of diseases associated with dysbiosis ever
growing
–
–
–
–
Inflammatory Diseases: IBD, IBS
Metabolic Diseases: Obesity, Malnutrition
Neurological Disorders
Cancer
What do we want to understand?
• What does a healthy microbiome look like?
– How diverse is it?
– What types of bacteria are there?
– What is their function?
• How variable is the microbiome?
– Over time within an individual?
– Across individuals?
– Functionally?
• What are driving factors of variability?
– Age, culture, physiological state (pregnancy)
• How do changes affect disease?
– What properties (taxa, amount of diversity) change with disease?
– Cause or affect?
– Functional consequences of dysbiosis
• Host Interactions
– Evolution/adaptation to the host over time.
– Immune system
Culture-independent studies revolutionized
our understanding of gut bacteria
• Culture-based studies over-emphasized
the importance of easily culturable
organisms (e.g. E. coli).
Culture-independent surveys
1.
Extract DNA from
environmental
samples.
2.PCR amplify SSU
rRNA gene (which
species?)
Sequence random
fragments (which
function?)
3. Evaluate
Sequences
Gut microbiota has simple
composition at the phylum level
Data from: Yatsunenko et. al. 2012. Nature.
Different phyla: Animals
and plants
Diversity of Firmicutes in 2 healthy
adults
• Each person
harbors > 1000
species.
• Some species
are unique (red
and blue)
• Some shared
(purple)
• We know very
little about
what most of
these species
do!
Sequencing technology renaissance enabled
more complex study designs
• Sanger Sequencing (thousands)
• Pyrosequencing (millions)
• Illumina (billions!)
Metagenomics
• The study of metagenomes, genetic material
recovered directly from environmental
samples.
• Marker gene
– PCR amplify a gene of interest
– Tells you what types of organisms are there
– Bacteria/Archaea (16S rRNA), Microbial Euks (18S
rRNA), Fungi (ITS), Virus (no good marker)
• Shotgun
– Fragment DNA and sequence randomly.
– Tells you what kind of functions are there.
Small Subunit Ribosomal RNA
• Present in all known life
forms
• Highly conserved
• Resistant to horizontal
transfer events
16S rRNA secondary structure
Other ‘Omics
• MetaTranscriptomics (sequence version of
microarray)
–
–
–
–
Isolate all RNA
Deplete rRNA
Sequence all transcripts
Sometimes phenotype only seen in activity of the
microbiota
• Metabolomics
– What metabolites does a community produce?
– E.g. in feces or urine
• MetaProteomics
– What proteins does a community produce?
Integrating Data Types
• 16S rRNA -> shotgun metagenomics
– What gene differences cannot be explained by
16S?
– Selection by HGT
• 16S/ genomics -> transcriptomics->
metabolomics
– What species or genes (or combination of species
or genes), when expressed, are responsible for
producing a given metabolite?
Sequencing Technologies
• Sanger -> 454 Pyrosequencing -> Illumina
Short reads (pyrosequencing)
can recapture the result.
• UW UniFrac
clustering with Arb
parsimony insertion
of 100 bp reads
extending from
primer R357.
• Assignment of
short reads to an
existing phylogeny
(e.g. greengenes
coreset) allows for
the analysis of very
large datasets.
Liu Z, Lozupone C, Hamady M, Bushman FD & Knight R (2007) Short pyrosequencing
reads suffice for accurate microbial community analysis. Nucleic Acids Res 35: e120.
Preprocessing pyrosequencing datasets
• Quality filtering: Discard sequences that:
–
–
–
–
Are too short and too long (200-1000 range)
With low quality scores
With long homopolymers
Can trim poor quality regions from the ends
• PyroNoise and Chimeras
– Can greatly inflate OTU counts
– Pyronoise algorithm uses SFF files to fix noisy
sequences
• Use barcodes to assign sequences to
samples
Defining species: OTU picking
• Cluster sequences based on % identity
– 97% id typical for species
– CD-HIT, UCLUST
• For Phylogenetic diversity measures need
to make a tree
– Align sequences: NAST, PyNAST
– Denovo tree building: FastTree
– Assign reads to sequences in a pre-defined
reference tree
Comparing Diversity
• Overview of methods for evaluating/comparing microbial
diversity across samples using 16S rRNA
  diversity: Measures how much is there?
  diversity: How much is shared?
• Phylogenetic verses taxon based diversity.
• Quantitative verses Qualitative diversity.
• What types of taxa are driving the patterns? Which
species are associated with measured properties?
• Tools: UniFrac/QIIME/Topiary Explorer
• Lozupone, C.A. and R. Knight (2008) Species divergence and the
measurement of microbial diversity. FEMS Microbiol Rev. 1-22.
How do we describe and compare
diversity?
  Diversity:
A
– “How many species are in a sample?”
• (e.g. 6 colors in A and 6 in B)
– e.g.: Are polluted environments less diverse than pristine?
  Diversity:
– “How many species are shared between samples?”
• (e.g. 2 shared colors between A and B)
– e.g.: Does the microbiota differ with different disease
states?
B
Quantitative versus Qualitative measures
• Qualitative: Considers presence absence only
A
– : How many species are in a sample?
• e.g.: 6 colors in both A and B.
– How many species are shared between
samples?
• e.g.: A and B are identical because the same colors
are present in both.
• Quantitative: Also considers relative abundance.
– : Accounts for “evenness”:
• e.g. B, where the population is evenly distributed
across the 6 species, is more diverse than A, where
all species are present but red dominates.
– Samples will be considered more similar if the
same species are numerically dominant versus
rare.
• e.g. B and A no longer look identical because of
differences in abundance.
B
What is a phylogenetic diversity
measure?
A
  Diversity:
– Taxon: “How many species are in a sample?”
– Phylogenetic: “How much phylogenetic divergence is in a
sample?”
• (e.g. B more individually diverse than A - more
divergent colors)
  Diversity:
– Taxon: “How many species are shared between samples?”
– Phylogenetic: “How much phylogenetic distance is shared
between samples?”
• (only related colors from B are in A)
B
Advantages of phylogenetic techniques.
• Phylogenetically related organisms are more likely to have similar
roles in a community.
• Taxon-based methods assume a “star phylogeny,” where all
relationships between taxa are ignored.
• Phylogeny and Taxon-based methods can be complementary.
Diversity Measures
• Diversity
– Phylogenetic Diversity: PD
– Taxon-based:
• observed # species (richness)
• Correct for undersampling (Chao1, Ace)
• Richness + evenness (Shannon-Weaver index)
•  Diversity
– Test if samples have significantly different membership.
• UniFrac Significance, P test, Libshuff (Phylogenetic)
– Identify environmental variables associated with differences
between many samples.
• Phylogenetic
– Unweighted and Weighted UniFrac
– DPCoA
• Taxon-based: Jaccard/Sorenson indices
Phylogenetic Diversity (PD)
• Sum of branches leading to sequences in a sample.
• Sample with taxa spanning the most branch length in this tree
represents the most phylogenetically and perhaps functionally
divergent community.
Faith, D.P. (1992) Conservation evaluation and phylogenetic diversity.
Biological Conservation 61, 1-10.
PD Rarefaction
• Plot the amount of branch length against the # of observations.
• Shape of curve allows for estimating how far we are from sampling all of
the phylogenetic diversity.
• Allows for comparison of phylogenetic diversity between samples.
Eckburg, P.B., et al. (2005) Diversity of the human intestinal microbial flora. Science 308,
1635-1638.
Phylogenetic and OTU based techniques can
be complementary
• Results of analyzing the
same data with Chao1
and PD.
• Samples from stool,
mouth, lung, plasma,
and negative controls.
• Differentiation between
the stool/mouth and
negative controls greater
with Chao1 than with PD
• The negative controls
have few OTUs but they
are phylogenetically
diverse
• Chao1 estimates go up
with sampling effort.
Phylogenetic  diversity: How is
diversity partitioned across
samples?
• Do two samples contain significantly
different microbial populations?
• Can we see broad trends that relate
many samples and explain them in
terms of environmental factors?
Unique Fraction (UniFrac) metric
•
•
Qualitative phylogenetic  diversity.
Distance = fraction of the total branch length that is unique to any particular
environment.
Lozupone and Knight, 2005, Appl Environ Microbiol 71:8228
Clustering with the UniFrac Algorithm
Can we see broad trends that relate many samples and explain them in terms of
environmental factors?
What types of environments have similar
phylogenetic diversity?
pH
Temperature
0-100°C
Pressure
1-12
Nutrient
Availability
Oligotrophic
Eutrophic
1-200 atm
Lozupone CA & Knight R (2007) Global patterns in bacterial diversity. Proc
Natl Acad Sci U S A 104: 11436-11440.
Salinity is the most important factor
PCoA of
UniFrac
Distance
Matrix
Hierarchical
clustering
(UPGMA)
of the same
UniFrac distance
matrix
Qualitative vs Quantitative measures of
Phylogenetic  Diversity
• Qualitative:
– Unweighted UniFrac
– Detects factors restrictive for microbial growth.
– High temperature, low pH, founder effects.
• Quantitative:
– Weighted UniFrac, DPCoA.
– Detects transient changes.
– Seasonal changes, nutrient availability, response to
pollution.
• Yield different, complementary results and applying
both to same data can provide insight into nature of
community changes.
Weighted UniFrac
Qualitative
Quantitative
Lozupone et al., 2007. Appl Environ Microbiol 73:1576
Obesity and Gut Microbiota
• Mice heterozygous
for mutation in
Leptin gene
interbreed.
• 16S gene
sequenced for
bacteria in gut of
mothers and
offspring.
Ley et al., (2005)Obesity Alters Gut Microbiota, PNAS Vol 102: pp 11070-11075
So how about the obese mice?
Mice cluster
perfectly by
mother
Ley et al., (2005)Obesity Alters Gut Microbiota, PNAS Vol 102: pp 11070-11075
Stronger clustering with obesity with
Weighted UniFrac
Unweighted UniFrac
Weighted UniFrac
Comparison of human
stool and mucosal
microbes
• Unweighted: all
samples cluster by
individual.
• Weighted: stool looks
different.
Eckburg, P.B., et al. (2005) Diversity of the human intestinal microbial flora. Science 308, 16351638.
Measures in the same class cluster
the data similarly
• Double principal coordinates
analysis (DPCoA)
– Another quantitative  diversity
measure.
– A matrix of species distances is first
used to ordinate the species using
PCoA.
– The position of the communities in
coordinate space is the average
position of the species that they
contain, weighted by relative
abundances.
• Produces same results as weighted
UniFrac.
Fast UniFrac
• Computation enhancements create order of magnitude increases in
speed and reduced memory requirements.
Hamady, Lozupone and Knight, The ISME Journal. 2009. Epub ahead of print.
Avoiding bias
• Pyrosequencing often produces high variability in the number of
sequences per sample.
• This can introduce bias because undersampling creates inflated
beta diversity values
•
•
•
Lozupone et al. 2011. ISME. 5:169-72
Randomly
resampled a
dataset at different
depths and
calculated the
average UniFrac
distance.
Samples with fewer
sequences look
artificially different.
Rarefaction:
randomly select an
even amount of
sequences
Web interfaces have >2200 registered users.
Unifrac papers have collectively 1250 citations.
461 citations
www.microbio.me/qiime
Study effects drive clustering of
Western adults
Lozupone et al. Genome Research. 2013
Age and culture drive differences
Supervised Learning, classical
statistics, taxonomic classification,
and phylogenetic trees; How can we
use these tools to understand which
microbial taxa change across
treatments?
Identifying compositional changes that
drive diversity patterns
• Histograms
Histograms and trees can pain a different picture
•
Peterson 2008 Cell Host Microbe: 3:417-27
Cluster XIVa ~43% of the total bacteria in the stool
of healthy individuals (Maukonen 2006. J Med
Microbiol. 55:625-33.)
16S rRNA gene tree of OTUs
prevalent in 2 studies of diet/obesity
–
–
•
Turnbaugh 2009 Sci Transl Med. 1:6ra14
Ley 2006. Nature. 444:1022-3
Clostridia clusters XIVa and IV are
the most abundant in the healthy gut.
Identifying taxonomic determinants
• Which taxa are significantly different between
health and disease?
– Using OTUs versus classifier derived taxa.
• PCoA Biplots:Which taxa are correlated with
overall clustering patterns?
• Finding discriminatory OTUs with Supervised
Learning.
• Applying classical statistical tests with
out_category_significance.py
• Exploring relationships in trees.
Defining Taxa
• 2 methods
– OTUs
– Classifiers (e.g. the RDP classifier)
• For both methods phylogenetic depth of the taxa can be
varied.
– OTUs – different %IDs (97%, 95%, 90%)
– Classifiers – different levels (species, genus, family)
• Advantage of using OTUs
– Can evaluate phylotypes not related to known species or in
taxonomic groups with poorly defined systematics.
– Each OTU represents an equal amount of phylogenetic
divergence.
• Advantage of using Classifiers
– Can more easily relate results to other published results.
– Fewer taxa than OTUs.
At what level should I classify?
• Shallow
– 97% ID OTU or species-level taxonomy assignments
– Advantage
• Biological properties of taxa have the potential to be more
strictly defined
– Disadvantage
• Can loose power to find associations in broader lineages in
which a trait is conserved
• Broad
– 90% ID OTUs or family-level taxonomic assignments
– Advantage
• More powerful for conserved traits
– Disadvantage
• Association in a broader group is often driven by only a subset
of its members (i.e. if you detect that Gamma Proteobacteria
go up you cannot say that E. coli did it!)
When ill-defined systematics can cause
Clostridium cluster XIVa
trouble
Lachnospiraceae
Clostridium
Lozupone et al 2012
Genome Research
Ruminococcus
Ruminococcus
Blautia
Ruminococcus
Ruminococcus
Blautia
Clostridium
Eubacterium
Clostridium
Eubacterium
Clostridium
Eubacterium
Clostridium
PCoA Bi-plots
• Allows visualization of taxa and samples in the
same PCoA space
Finding discriminative OTUs
• 2 methods
– Supervised learning
– Classical statistics
• Supervised learning
– Evaluates how well OTUs/taxa can be used to classify
by treatment.
– Discriminative OTUs are those for which classification
power is reduced when they are removed from the set
– Advantage:
• evaluates OTUs contextually rather than independently
– Disadvantage:
• only works with Discrete sample groupings (i.e. will not
handle correlations with disease severity or changes over
time)
Feature importance scores
• All OTUs with scores
> 0.001 considered
‘important’
– Yatsunenko et al
Nature 2012
• Problem: We do not
know the direction
of change.
• With only two
categories –
compare the means.
Classical Statistics Tests in QIIME
• otu_category_significance.py
–
–
–
–
i: otu table
m: category mapping
c: category (e.g. health status)
s: statistical test
•
•
•
•
ANOVA
Pearson correlation
Paired T test
G-test of independence
– f: minimum number of samples found in to be considered
– Removes OTUs that don’t pass the filter, performs a
statistical test on each OTU, corrects for multiple
comparisons with FDR and Bonferroni correction.
– Can also be run on Taxa Summary tables files if in BIOM
format.
Assign statistical significance values to bar charts
ANOVA output
• I use these means and their significance to
assess direction of change in Supervised
learning results.
Are discriminatory OTUs related to
each other and to type strains?
• Relate them in a tree.
• ARB to make the tree using parsimony
insertion.
– http://www.mpi-bremen.de/ARBSILVA.html
• Topiary explorer to visualize/color the tree and
make publication quality graphics
– http://topiaryexplorer.sourceforge.net
Sometimes associations are
phylogenetically shallow
Erysipelotrichales with HIV infection
Genomics
• Genomics : Thousands of
complete and draft genome
sequences for human
commensals publicly available
– Promise: translate 16S into
functional predictions (PiCRUST)
– Challenges: no genomes for
Distribution
(16S rRNA)
unculturable microbes
– Genes with high HGT
Comparative
genomics
(complete
genomes)
Experimental
Confirmation
(anaerobic
culture)
Annotating genes to functions
• Based on similarity to genes of known function.
NCBI genomes
have functions
listed for
predicted
proteins
Databases for functional assignments
• COGs (Clusters of Orthologous Groups;
http://www.ncbi.nlm.nih.gov/COG/)
• KEGG (Kyoto Encyclopedia of Genes
and Genomes;
http://www.genome.jp/kegg/)
• CAZy (Carbohydrate Active Enzymes
database; http://www.cazy.org/)
• pFAM (protein family database;
http://www.sanger.ac.uk/resources/da
tabases/pfam.html)
COG database
• Orthologous groups
– A group of proteins that are
expected to perform the same
function in the different
organisms in which they are
found.
– Function is inferred for the whole
group based on experimental
work with one of its members.
– COGs are grouped into larger
functional groups.
KEGG database
• Orthologous groups
(assigned KO
numbers)
• Metabolic
pathways.
– Boxes contain
enzyme
commission
database (EC)
numbers.
• Each EC is
associated with KO
numbers (a protein
family that is known
to perform that
reaction).
Shotgun metagenomics
KEGG pathway Ontology
Glycoside Hydrolases (GH)
Degradation: hydrolyze glycosidic bonds between two carbs
or between a carb and a non-carb.
Important for degradation of plant polysaccharides.
GlycosylTransferases (GT)
Biosynthesis: catalyze the transfer of sugar moeties.
Important for communication with host immune system.
• Database
describing
protein
families
predicted to
be
carbohydrat
e active
based on
homology
• Uses HMMs
• Exact
reaction
performed
does not
need to be
known.
• Similar to CAZy but with a broader scope.
• Hidden Markov Models that describe
sequence motifs of a known function
Annotating genes to taxonomic groups
• Based on similarity to genes in a
database of reference genomes.
– http://www.genomesonline.org/cgibin/GOLD/index.cgi
• Mg-RAST uses best BLAST hit:
M5N4
Annotating metagenomes
• MgRAST
http://metagenomics.anl.gov/metagenomics.c
gi?page=Analysis
• Produces Table mapping samples to
annotations that can be further processed in
QIIME

similar documents