1. Metagenomics: From Bench to Data Analysis 19-23rd September S rRNA-based surveys for Community Analysis: How Quantitative are they?
Dr Mark Alston
Computational Biologist
Organisms and Ecosystems Group

2. Outline Compare sequencing platforms and 16S rRNA regions
Amplicon choice
amplicons vs. full-length rRNA sequencing
Bias and quantification
Comparison to WGS approaches

3. 16S Microbial Community Profiling
16S rRNA gene sequence
conserved (green) and hypervariable (blue) regions
Most common phylogenetic marker
‘gold standard’ in molecular surveys of bacterial and archaeal diversity
Pros
ubiquitous, highly conserved, evolutionarily stable
Cons
often multiple copy, little resolution at/below species level

4. Comparing Different Platforms and Target Regions
‘A comprehensive benchmarking study of protocols and sequencing platforms for 16S rRNA community profiling’
DOI: /s
Compare sequencing platforms
MiSeq (Illumina),
Pacific Biosciences RSII
454 GS-FLX/+ (Roche)
IonTorrent (Life Technologies)
Compare target regions
Assess performance via synthetic microbial communities
mix gDNA from 49 bacterial and 10 archaeal species
even / uneven distribution
Summary of primers and platforms used

5. Ability of Different Platforms and Regions to Reconstruct the Synthetic Community
Even synthetic community
Platform had a significant effect
Species’ frequencies highly unbalanced
Possible causes
primer mismatches
rRNA copy number
amplification bias (associated with target length)
Bacterial Species
Target Region

6. How do Different rRNA Regions reflect Composition?
‘Comparative metagenomic and rRNA microbial diversity characterization using archaeal and bacterial synthetic communities’
DOI: /
Synthetic Bacteria community
Heat map represents accuracy ratio
Perfect agreement has value of 1
underestimated abundance
overestimated abundance

7. How do Different rRNA Regions reflect Composition?
‘Comparative metagenomic and rRNA microbial diversity characterization using archaeal and bacterial synthetic communities’
DOI: /
Synthetic Bacteria community
Heat map represents accuracy ratio
Perfect agreement has value of 1
underestimated abundance
overestimated abundance
Regions suffer from substantial bias

8. Which Region Should I Choose?
16S rRNA gene sequence
conserved (green) and hypervariable (blue) regions
Most common approach
V4, V3–V4 or V4–V5 primers on Illumina platforms
~ 250–430 bp read length
e.g. 16S for V4 on MiSeq

9. Full-length vs. Amplicon 16S Sequencing
Factors affecting taxon abundance estimates and tree-placement
Sequencing platform, primer choice, read length, environmental source, reference database, assignment method [or a combination]
New technologies
short reads sequence ~15-30 % of the full 16S rRNA gene
more quantitative information
reduced taxonomic resolution
species level assignment can be elusive
implications for inferring metabolic traits in various ecosystems

10. Full-length vs. Amplicon 16S Sequencing
Factors affecting taxon abundance estimates and tree-placement
Sequencing platform, primer choice, read length, environmental source, reference database, assignment method [or a combination]
New technologies
short reads sequence ~15-30 % of the full 16S rRNA gene
more quantitative information
reduced taxonomic resolution
species level assignment can be elusive
implications for inferring metabolic traits in various ecosystems
Use full-length 16S rRNA sequencing?

11. Full-length 16S rRNA Sequencing
PacBio
long-read, single-molecule real-time (SMRT) technology
average read lengths > 8 kb at ~ 87% read accuracy
only been used for a few environmental surveys
‘High-resolution phylogenetic microbial community profiling’
DOI: /ismej
MinION™
USB stick-sized device
per-base sequencing accuracy ~85% for 2D reads
additional read length helps resolve 16S rRNA to species level
‘Species level resolution of 16S rRNA gene amplicons sequenced through MinIONTM portable nanopore sequencer’
DOI: /s z

12. Full-length 16S rRNA Sequencing
PacBio
long-read, single-molecule real-time (SMRT) technology
average read lengths > 8 kb at ~ 87% read accuracy
only been used for a few environmental surveys
‘High-resolution phylogenetic microbial community profiling’
DOI: /ismej
MinION™
USB stick-sized device
per-base sequencing accuracy ~85% for 2D reads
additional read length helps resolve 16S rRNA to species level
‘Species level resolution of 16S rRNA gene amplicons sequenced through MinIONTM portable nanopore sequencer’
DOI: /s z

13. Full-length 16S rRNA Sequencing and Gene Variability
non-homogeneous distribution of mutations
varies across different phylogenetic groups
leads to both over- and underestimation of community diversity

14. Full-length 16S rRNA Sequencing and Gene Variability
non-homogeneous distribution of mutations
varies across different phylogenetic groups
leads to both over- and underestimation of community diversity

15. Full-length 16S rRNA Sequencing and Gene Variability
non-homogeneous distribution of mutations
varies across different phylogenetic groups
leads to both over- and underestimation of community diversity
2 Salmonella spp. 97.4% identical across gene
100% identical across V4 region
Underestimate community diversity

16. Full-length 16S rRNA Sequencing and Gene Variability
non-homogeneous distribution of mutations
varies across different phylogenetic groups
leads to both over- and underestimation of community diversity
Mutations accumulated in V4 region
Overestimate community diversity

17. Compare FL vs. V4 [Sakinaw lake samples]
Community composition profile at genus level
Colour pairs denote samples of the same depth
Bubble sizes indicate read abundance

18. Compare FL vs. V4 [Sakinaw lake samples]
BUT it looks possible to make the same conclusions because there’s a lot of stuff in common!
FL vs. V4 discrepancies highlighted by boxes
e.g. Bacillus greatly underrepresented by V4 c.f. PB [50m samples]
‘High-resolution phylogenetic microbial community profiling’
DOI: /ismej

19. Platforms and Regions Suffer from Substantial Bias
The observed relative frequencies do not reflect the true species frequencies in the community

20. Platforms and Regions Suffer from Substantial Bias
The observed relative frequencies do not reflect the true species frequencies in the community

21. Platforms and Regions Suffer from Substantial Bias
The observed relative frequencies do not reflect the true species frequencies in the community
But, the observed differences between samples could still reflect true differences
Can we have a quantitative method despite the bias?

22. Can 16S rRNA Sequencing be Quantitative?
‘A comprehensive benchmarking study of protocols and sequencing platforms for 16S rRNA community profiling’
DOI: /s
Assembled 2 synthetic communities
one with even distribution, one uneven
Take pairs of samples
Sequence on MiSeq and PacBio platforms

23. Can 16S rRNA Sequencing be Quantitative?
Compare for each species
true ratio of frequencies [known mixtures]
and
observed ratio of frequencies
Highly significant correlation between the two ratios [blue line] and a slope of 1 [red line]
MiSeq
Ratio of Observed Freq.
PacBio
Ratio of True Freq.

24. Can 16S rRNA Sequencing be Quantitative?
Compare for each species
true ratio of frequencies [known mixtures]
and
observed ratio of frequencies
Highly significant correlation between the two ratios [blue line] and a slope of 1 [red line]
Implies 16S rRNA sequencing is strongly quantitative despite being biased
MiSeq more quantitative than PacBio
MiSeq
Ratio of Observed Freq.
PacBio
Ratio of True Freq.

25. MiSeq more quantitative than PacBio
Species responsible for this difference?
Which are more accurately quantified on one platform relative to the other?
MiSeq
Ratio of Observed Freq.
PacBio
Ratio of True Freq.

26. MiSeq vs. PacBio
Species with significantly different quantification accuracies:

27. MiSeq vs. PacBio
Species with significantly different quantification accuracies:
MiSeq the better platform

28. MiSeq vs. PacBio
Species with significantly different quantification accuracies:
MiSeq the better platform
Except for strain resolution
Full-length 16S rRNA sequencing of benefit
Shewanella baltica OS223
Shewanella baltica OS185

29. 16S Microbial Community Profiling
16S rRNA gene sequence
conserved (green) and hypervariable (blue) regions
Most common approach
V4, V3–V4 or V4–V5 primers on Illumina platforms
~ 250–430 bp read length
Economy of scale
single MiSeq run > 10 million reads
High base-calling accuracy
e.g. 16S for V4 on MiSeq

30. Compare Error Rates Across Platforms
Even synthetic community
Platform had a significant effect
MiSeq has the most accurate sequence reads

31. Impact of Overlapping Reads on MiSeq V4 Error Rates
Even synthetic community
Overlapping forward and reverse reads greatly reduces errors
MiSeq Dual Index barcode
Illumina barcodes on both reads
‘stitched’ reads

32. Shotgun Metagenomics vs. Amplicon Sequencing
‘Comparative metagenomic and rRNA microbial diversity characterization using archaeal and bacterial synthetic communities’
DOI: /
Compare amplicon sequencing to Illumina [HiSeq] and 454 metagenomics sequencing

33. Shotgun Metagenomics vs. Amplicon Sequencing
‘Comparative metagenomic and rRNA microbial diversity characterization using archaeal and bacterial synthetic communities’
DOI: /
Compare amplicon sequencing to Illumina [HiSeq] and 454 metagenomics sequencing
Metagenomic data tends to outperform amplicon sequencing

34. Shotgun Metagenomics vs. Amplicon Sequencing
‘A comprehensive benchmarking study of protocols and sequencing platforms for 16S rRNA community profiling’
DOI: /s
MiSeq MG sample
expected
Metagenome sample
benchmark
should be relatively unbiased as fewer PCR amplification steps in library construction
WGS gives the most accurate species estimations

35. Is 16S “Metagenomics” ? Many papers talk about
“metagenomics analysis based on microbial 16S rRNA gene sequencing”
“16S metagenomic studies” etc.
But rRNA surveys focus on a single gene, not genomes
Is this due to a fear of not getting funded if you don’t include a word containing ‘Meta*omics’?
“Referring to 16S surveys as metagenomics is misleading and annoying #badomics #OmicMimicry”

36. In Summary Many sources of bias when we sequence 16S rRNA
e.g. platform, region etc.
Can still be a quantitative
MiSeq V4 a good ‘all round bet’
prior knowledge of taxa may suggest otherwise
combinations of primers?
full-length for strain resolution
Whole genome shotgun
better estimations of species abundances

37. Metagenomics: From Bench to Data Analysis 19-23rd September 2016 Thank You for Listening
Dr Mark Alston
Computational Biologist
Organisms and Ecosystems Group