1. Strain profiling with StrainPhlAn and PanPhlAn
Nicola Segata
Strain profiling with StrainPhlAn and PanPhlAn
Curtis Huttenhower
Galeb Abu-Ali
Ali Rahnavard
STAMPS 2017
Harvard T.H. Chan School of Public Health
Department of Biostatistics

2. Efficient assembly-free meta’omics by leveraging isolatesII
III IV V I II
III IV V II
III II I IV I I II
III II V IV V V
Species pan-genomes
7,677
containing
18.6 million
gene clusters
Core
genes
Marker
genes
NCBI isolate
genomes
Archaea
300
Bacteria
12,926
Viruses
3,565
Eukaryota
112
Open reading frames
49.0 million
total genes
RepoPhlAn
ChocoPhlAn

3. StrainPhlAn: metagenomic strain identification and tracking

4. A tool for strain level population genomics
China
Denmark
Estonia
Finland
Peru’
Hungary
Italy
Norway
France
Spain
Sweden
USA
Germany
P. copri as an example species
Alignment length: 66k nt
Median SNPs: 830 [3.6%]
# pos. samples: 123

5. A tool for strain level population genomics
Alignment length: 62k nt
Median SNPs: 830 [1.3%]
# pos. samples: 123

6. Most bugs (in the gut) are dominated by one stable strain

7. Most bugs (in the gut) are dominated by one stable strain

8. There’s a lot of strain-level variation left to discover
Median divergence from reference markers

9. PanPhlAn: the approach
mapping
Read
Metagenomic sample
Gene coverage
Microbial pangenomes
Cluster to
Gene families
Pan-gene family coverage
Abundance-sorted pan-gene families
Coverage
Multi-copy
genes
Plateau of genes from one metagenome’s strain
Absent genes

10. PanPhlAn for “meta-epidemiology”
Metagenomes from [Loman et al., 2013]

11. Strain-level epidemiology of human-associated E. coli with PanPhlAn
STEC
Scholz et al., Nature Methods, 2016
T2D (China)
German outbreak
Reference genomes
Liver Cirr. (China)
Infants (Italy)
CRC (Europe)
HMP (USA)
Obesity (Europe)
Neilsen (Europe)
T2D (Finland)
Rampelli (Africa)
Liu (Mongolia)
Tito (Peru)
Segre (Skin) B1 B2
~5,000 metagenomes
(and counting)
All continents
Many EU countries A D

12. Multiple options for strain tracking in metagenomes
StrainPhlAn:
Map reads to core markers and call SNPs.
Requires ~10x coverage, ~0.1% error rate.
PanPhlAn:
Map reads to pan-genomes and identify absent genes.
Requires ~1x coverage, ~1% error rate.
Both work uniquely well for meta-analysis.
Not sensitive to typical batch effects.

13. https://bitbucket.org/biobakery/biobakery/wiki/strainphlan
StrainPhlAn tutorial

15. There’s a lot of strain-level variation left to discover
Phylogenetic branch % spanned by reference vs. “wild” bugs

16. Gene-family distribution curves
Select samples with “step” distribution
(colored curves) strain of species present
Base coverage
Reject non-step
(gray) curves
E. coli gene-families

17. Synthetic and semi-synthetic validation
Coverage
Coverage
Coverage
Coverage
Coverage

18. PanPhlAn on Eubacterium rectale
Only one Eubacterium rectale genome used here

19. PanPhlAn on Eubacterium rectale