Challenges for metagenomic data analysis and lessons from viral metagenomes

[What would you do if sequencing were free?]

Rob Edwards

San Diego State UniversityFellowship for Interpretation of Genomes

The Burnham Institute for Medical Research

Outline

• How and why we sequence environments

• Viral metagenomics– Marine stories– Human stories

• Pyrosequencing– Mine story

• Is there a Future?

Why Metagenomics?

• What is there?

• How many are there?

• What are they doing?

How do you sequence the environment?

• Extract DNA

1.1 g ml-1

1.35 g ml-1

1.5 g ml-1

1.7 g ml-1

CsCl step gradient

CsCl step gradient

How do you sequence the environment?

• Extract DNA

• Create library

Hydroshear

Blunt-ending

Addition of Linkers

Amplification of Fragments

Linker-Amplified Shotgun Libraries (LASLs)

HydroshearBlunt-ending

Addition of LinkersAmplification of Fragments

This method produces high coverage libraries of over 1 million clones from as little as 1 ng DNA

Soil Extraction Kit

David Mead -

Breitbart (2002) PNAS

How do you sequence the environment?

• Extract DNA

• Create library

• Sequence fragments

Outline

• How and why we sequence environments

• Viral metagenomics– Marine stories– Human stories

• Pyrosequencing– Mine story

• Is there a Future?

Why Phages?

• Phages are viruses that infect bacteria– 10:1 ratio of phages:bacteria

– 1031 phages on the planet

• Specific interactions (probably)– one virus : one host

• Small genome size– Higher coverage

• Horizontal gene transfer– 1025-1028 bp DNA per year in the oceans

Uncultured Viruses

200 liters water 5-500 g fresh fecal matter

DNA/RNA LASL

Sequence

Epifluorescent Microscopy

Concentrate and purify viruses

Extract nucleic acids

Bioinformatics

• BLASTagainst NR – blastx, tblastn, tblastx

• BLAST against boutique databases– Complete phage genomes, ACLAME, Other

libraries, 16S

• Parsing to present data in a useful format

BLAST and Parsing

• http://phage.sdsu.edu/blast

• Submit BLAST to local and remote databases– Local (as fast as possible)– NCBI (one search every 3 seconds)

• Many concurrent searches– One search versus 1,000 searches

• Parse data into tables– Access to taxonomy etc

Known22%

Unknown78%

Most Viral Genes are Unknown

Breitbart (2002) PNASRohwer (2003) Cell

TBLAST (E<0.001) 3,093 sequences

60 billion base pairs

60 million sequences

GenBank has more than doubled since 2001 …

GenBank has more than doubled since 2001 …

but the fraction of unknowns remains constant

Edwards (2005) Nature Rev. Microbiol.

All of the new genes in the databases are

coming from environmental sequences

Outline

• How and why we sequence environments

• Viral metagenomics– Marine stories– Human stories

• Pyrosequencing– Mine story

• Is there a Future?

Human-associated viruses

• More bacteria than somatic cells by at least an order of magnitude

• More phages than bacteria by an order of magnitude

• Sample the bacteria in the intestine by sampling their phage

Most Viral DNA Sequences in Adult Human Feces are Unknown Phages

Known40%

Unknown60%

Breitbart (2003) J. Bacteriol.

TBLAST (E<0.001) 532 sequences

Phages94%

Eukaryotic Viruses 6%

No bacteria or viruses in 1st fecal sample

Abundant bacterial and viral communities by 1 week of age

Adults Versus Babies

>108 VLP/g feces

Baby Feces Viruses

• Most sequences are unknown (≈70%)

• Similarities to phages from Lactococcus, Lactobacillus, Listeria, Streptococcus, and other Gram positive hosts

• From microarray studies, sequences are stable in the baby over a 3 month period

• Same types of phage as present in adult feces– one identical sequence to an unrelated adult!

DNA viruses in feces are phages.

Feces ≠ intestines.

RNA viruses?

Most Human RNA Viruses are Known

Known92%

Unknown8%

TBLAST (E<0.001) ≈36,000 sequences

Pepper MildMottle Virus

65%

Other PlantViruses

9%

Other26%

Zhang (2006) PLoS Biology

Pepper Mild Mottle Virus (PMMV)• ssRNA virus; ≈6 kb genome• Related to Tobacco Mosaic Virus• Infects members of Capsicum family• Widely distributed – spread through seeds• Fruits are small, malformed, mottled• Rod-shaped virions

TOBACCO MOSAIC VIRUS http://www.rothamsted.bbsrc.ac.uk/ppi/links/pplinks/virusems/

Viral particles in fecal sample

S1 S2 S3 S4 S5 S6 S7 S8 S9 PMMV

PMMV is common in Human FecesFecal samples

Extract total RNA

RT-PCR for PMMV

San Diego : 78% people are positiveSingapore : 67% people are positive

10-50 fold increase in feces compared to food106-109 PMMV copies per gram dry weight of feces

India

n c

urr

yPork

noodle

red c

hili

Chic

ken r

ice

Chin

ese

food

Hong K

ong c

hili sa

uce

Hong K

ong g

reen c

hili

Vegeta

rian c

hili

Which Foods Contain PMMV?

Chili powder

Chili sauces

NOT FOUND IN FRESH PEPPERS

T

he

su

nm

ac

hin

e.n

et

http://www.sweatnspice.com

Koch’s Postulates

Human microbial metagenome is more

important than human genome

Outline

• How and why we sequence environments

• Viral metagenomics– Marine stories– Human stories

• Pyrosequencing– Mine story

• Is there a Future?

How do you sequence the environment?

• Extract DNA

• Create library

• Sequence fragments

Ever

ythi

ng so

far f

rom

40,0

00 seq

uenc

es

This

is so

200

4

How do you sequence the environment?

• Extract DNA

• Pyrosequence

454 Pyrosequencing

• Emulsion-based PCR

• Luciferase-based sequencing

} SDSU•DNA extraction from environment

•Whole genome amplification

} 454 Inc.

Margulies (2005) Nature

454 Sequence Data(Only from Rohwer Lab)

• 21 libraries– 10 microbial, 11 phage

• 597,340,328 bp total– 20% of the human genome– 50% of all complete and partial microbial

genomes

• 5,769,035 sequences– Average 274,716 per library

• Average read length 103.5 bp– Av. read length has not increased in 7 months

Growth of sequence data

6 million reads600 million bp

Cost of sequencing

• One reaction = $10,000• One reaction = 250,000 reads• 250 reads = $10• 1 read = 4¢• 1 read = 100bp• 1 bp = 0.04¢

($400 per 1x 1,000,000 bp)

• Sanger sequencing ca. $1/rxn, 0.2¢/bp– real cost ca. $5/rxn, 1¢/bp

454 sequencing doescot require cloning, arrayingetc.

Bioinformatics

• 597,340,328 bp total

• 5,769,035 sequences

• 7 months

• Existing tools are not sufficient

Current Pipeline

• Dereplicate

• BLAST against– 16S– Complete phage– nr (SEED)– subsystems

http://phage.sdsu.edu/~rob/Pyrosequencing/

Sequencing is cheap and easy.

Bioinformatics is neither.

Outline

• How and why we sequence environments

• Viral metagenomics– Marine stories– Human stories

• Pyrosequencing– Mine story

• Is there a Future?

The Soudan Mine, Minnesota

Red Stuff OxidizedBlack Stuff Reduced

Red and Black Samples Are Different

Cloned and 454 sequenced16S are indistinguishable

Black stuff

Red

ClonedRed

Annotation of metagenomes by subsystems

A subsystem is a group of genes that work together

– Metabolism– Pathway– Cellular structures– Anything an annotator thinks is

interesting

There are different amounts of metabolism in each environment

There are different amounts ofsubstrates in each environment

BlackStuff

RedStuff

But are the differences significant?

• Sample 10,000 proteins from site 1• Count frequency of each subsystem• Repeat 20,000 times

• Repeat for sample 2

• Combine both samples• Sample 10,000 proteins 20,000 times• Build 95% CI

• Compare medians from sites 1 and 2 with 95% CI

Rodriguez-Brito (2006). In Review

Examples of significantly different subsystems

Red Stuff

Arg, Trp, His UbiquinoneFA oxidationChemotaxis, FlagellaMethylglyoxal

metabolism

Black Stuff

Ile, Leu, ValSiderophoresGlycerolipidsNiFe hydrogenasePhenylpropionate

degradation

Subsystem differences & metabolism

Iron acquisitionBlack Stuff

Siderophore enterobactin biosynthesisferric enterobactin transportABC transporter ferrichromeABC transporter heme

Black stuff: ferrous iron (Fe2+, ferroan [(Mg,Fe)6(Si,Al)4O10(OH)8])

Red stuff: ferric iron (goethite [FeO(OH)])

Nitrification differentiates the samples

Edwards (2006)In review

Not all biochemistry happens in a single organism

Anaerobic methane oxidationBoetius et al. Nature, 2000.

CH4 + SO42- -> HCO3

- + HS- + H2S

ArchaeaCH4 + H2O ->

HCO3- + OH + H2 -> CO2 +

H2OBacteriaSO4

2- + H2O ->

HS- + OH + 2O2

The challenge is explaining the differences between samples

Red Sample

Arg, Trp, His UbiquinoneFA oxidationChemotaxis, FlagellaMethylglyoxal

metabolism

Black Sample

Ile, Leu, ValSiderophoresGlycerolipidsNiFe hydrogenasePhenylpropionate

degradation

We are moving away from one organism one reaction andtowards studying the biochemistry of whole environments

Bacteria don’t live alone

Summary

From 454 sequence:– Identify microbial composition– Identify metabolic function– Identify statistically significant

differences in metabolism

– Who, what, why of microbial ecology

Marine Near-shore water (~100 samples) Off-shore water (~50 samples) Near- and off-shore sediments

Metazoanassociated Corals Fish Human blood Human stool

Sampling Sites

Terrestrial/Soil Amazon rainforest Konza prairie Joshua Tree desert Singapore Air

Freshwater Aquifer Glacial lake

ExtremeHot springs (84oC; 78oC)Soda lake (pH 13)Solar saltern (>35% salt)

SDSUForest RohwerMya BreitbartBeltran Rodriguez-Brito

Rohwer Lab:Linda WegleyFlorent AnglyMatt Haynes

Also at SDSUAnca SegallWillow SegallStanley Maloy

Math Guys@SDSU Peter Salamon Joe Mahaffy James Nulton Ben Felts David Bangor Steve Rayhawk Jennifer Mueller

MIT: Ed DeLong

NSF - Biotic Surveys and Inventories - Biological Oceanography

- Biocomplexity

FIG Veronika Vonstein Ross Overbeek Annotators

Genome Institute of Singapore: Zhang Tao Charlie Lee Chia Lin Wei Yijun Ruan

Viral Community Structure

• Contigs assembled from fragments with >= 98% identity over 20 bp are a resampling of a single phage genome

• Contig specturm is the number of contigs that have one sequence, the number that have two sequences, and so on

• Use both analytical and Monte-Carlo simulations to predict community structure from contig spectrum

The Math Guys (2006) In preparation

0.00

0.20

0.40

0.60

0.80

1.00

1.20

1.40

1.60

1.80

2.00

0 10 20 30 40 50

Species Rank

Abundance of the species (%)

Determine the actual contig spectrum of the sample

Predict a contig spectrum using a species abundance modelCompute the error between the actual and predicted

Adjust the parameters in the species abundance model to minimize errors

Continue this procedure until we obtain the smallest error

Find the smallest error, a global minimum

Model parameters

Error

Fecal

Seawater

MarineSediments

Viral Communities are Extremely Diverse

Lots of rare viral genotypes

0

1

2

3

4

5

6

7

8

9

10

Cropland Earthworms

Fossil CoralsForest Amphibians

River Bacteria

Sediment Viruses

Seawater VirusesSeawater Viruses

Fecal Viruses

Bacteria on CoralsAgriculture Soil Bacteria

Soil Nematodes

Rainforest SpidersAmazon Fish

Rainforest BirdsForest Mammals

Temperate Forest Beetles

Shannon-WienerIndex