1. Sahar Abubucker, Nicola Segata,
Scalable metabolic reconstruction for metagenomic data and the human microbiome
Sahar Abubucker, Nicola Segata,
Johannes Goll, Alyxandria M. Schubert, Jacques Izard, Brandi L. Cantarel, Beltran Rodriguez-Mueller, Jeremy Zucker, Mathangi Thiagarajan, Bernard Henrissat, Owen White, Scott T. Kelley, Barbara Methé, Patrick D. Schloss, Dirk Gevers, Makedonka Mitreva,
Curtis Huttenhower
Harvard School of Public Health
Department of Biostatistics

2. What’s metagenomics?
Total collection of microorganisms within a community
Also microbial community or microbiota
Total genomic potential of a microbial community
Study of uncultured microorganisms from the environment, which can include humans or other living hosts
Total biomolecular repertoire of a microbial community

3. Valm et al, PNAS 2011

4. What to do with your metagenome?
Reservoir of gene and protein functional information
Comprehensive snapshot of microbial ecology and evolution
Who’s there?
What are they doing?
Who’s there varies: your microbiota is plastic and personalized.
What they’re doing is adapting to their environment: you, your body, and your environment.
Public health tool monitoring population health and interactions
Diagnostic or prognostic biomarker for host disease

5. The Human Microbiome Project for a normal population
300 People/ 15(18) Body Sites
Multifaceted data
Multifaceted analyses
>6,000 samples
>50M 16S seqs.
4Tbp unique metagenomic sequence
>1,900 reference genomes
Full clinical metadata
Human population
Microbial population
Novel organisms
Biotypes
Viruses
Metabolism
2 clin. centers, 4 seq. centers, data generation, technology development, computational tools, ethics…

6. Metabolic/Functional Reconstruction: The Goal
Healthy/IBD
BMI
Diet
SNP genotypes
Taxon abundances
Enzyme family abundances
Pathway abundances
Gene expression
LEfSe: LDA Effect Size
Metagenomic biomarker discovery
Nicola Segata

7. HMP: Metabolic reconstruction
Functional seq.
KEGG + MetaCYC
CAZy, TCDB, VFDB, MEROPS…
100 subjects
1-3 visits/subject
~7 body sites/visit
10-200M reads/sample
100bp reads
BLAST
HUMAnN: HMP Unified Metabolic Analysis Network
BLAST → Genes
WGS reads
Genes → Pathways
MinPath (Ye 2009)
Genes (KOs)
Taxonomic limitation
Rem. paths in taxa < ave. ?
Pathways (KEGGs)
Pathways/ modules
Smoothing
Witten-Bell
Xipe
Distinguish zero/low (Rodriguez-Mueller in review)
Gap filling
c(g) = max( c(g), median )

8. HUMAnN: Metabolic reconstruction
← Pathways→
← Samples →
Vaginal
Skin
Nares
Gut
Oral (SupP)
Oral (BM)
Oral (TD)
← Samples →
← Pathways→
Vaginal
Skin
Nares
Gut
Oral (SupP)
Oral (BM)
Oral (TD)
Specifically, HUMAnN produces two outputs for any metagenomic sample. It first makes pathway presence/absence or coverage calls, indicating in a binary way which microbial pathways are present in the community. It then also assigns relative abundances to the pathways that are there, resulting in that matrix I described of communities by pathways, with each cell representing one pathway’s abundance in one community. The pathway coverage matrix looks similar, with each pathway marked as 0 if it’s absent or 1 if it’s present, independent of abundance. What you’re looking at here is an overview of the real HUMAnN results for about 700 HMP samples spanning these seven body sites, but it’s not the most informative view of the data.
Pathway coverage
Pathway abundance

9. HUMAnN: Validating gene and pathway abundances on synthetic data
Validated on individual gene families, module coverage, and abundance
4 synthetic communities: Low (20 org.) and high (100 org.) complexity Even and lognormal abundances
Best-BLAST-hit overshoots false positives, undershoot real pathways as a result
HUMAnN FNs: short genes (<100bp), taxonomically rare pathways
HUMAnN FPs: large and multicopy (not many in bacteria)
Individual gene families
ρ=0.91

10. A portrait of the healthy human microbiome: Who’s there vs
A portrait of the healthy human microbiome: Who’s there vs. what they’re doing
← Relative abundance →
← Phylotypes →
← Relative abundance →
Nares
Oral (BM)
Vaginal
Skin
Gut
Oral (SupP)
Oral (TD)
← Relative abundance →
← Pathways →
Instead, consider this view that Dirk showed earlier of the taxa in a collection of the HMP samples, specifically the gut communities. Each color here represents a taxon, and as he described, there’s huge variation among individuals, with Bacteroides ranging from tremendously dominant to a minority organism. We can use this same view for the metabolic reconstruction data, where now each color represents a specific metabolic pathway. There’s much less variation in community function among individuals than there is in community composition – roughly the same pathways are present at the same abundances in each individual’s gut, regardless of variation in membership. This is true not only in the gut, but in all of the HMP’s body sites. The body sites themselves differ somewhat in metabolism, as I’ll discuss later, but again not nearly as dramatically as they differ in composition. This is now an overview of most of the HMP’s 16S data and all 700 of our metabolic reconstructions, and it captures the important functional patterns in the human microbiome perhaps better than the previous slide. From this high-level perspective, though, I’d like to zoom in to a specific pathway or set of enzymes, though, and talk about…
← Relative abundance →

11. Niche specialization in human microbiome function
Metabolic modules in the KEGG functional catalog enriched at one or more body habitats
Linking this back to the HMP’s metabolic reconstruction data, we used LEfSe to summarize pathways with significant differences in abundance between body sites, resulting in this set of pretty colors plotted on top of the KEGG functional hierarchy. So here, you’re looking at a tree of pathways and processes drawn from KEGG, in which each outermost leaf node represents an individual small metabolic module. Portions of the hierarchy are colored based on the body site where they’re most significantly enriched, so for example this yellow dot corresponds to an overabundance of spermidine biosynthesis in the GI tract, which is also fairly abundant throughout the oral cavity. You’ll notice that there’s a tremendous amount of niche specialization in the form of over- and under-abundant processes, which is one way of seeing metabolic function in the human microbiome adapting specifically to each niche throughout the body. This regulation of metabolic abundances is somewhat in contrast to the complete absence of pathways, however, in that relatively few processes tend to be completely absent from any body site. This outer ring, for example, shows just 24 pathways that show such a presence/absence pattern, such as type six secretion in the oral cavity, which is completely absent from other body sites. Such absences seem to be rare, though, with most functionality following an everything-is-everywhere pattern, and the environment selects for how abundant particular metabolic processes are based on their utility to the organisms in each habitat.
Nicola Segata

12. Proteoglycan degradation by the gut microbiota
Glycosaminoglycans (Polysaccharide chains)
AA core

13. Proteoglycan degradation: From pathways to enzymes
10-3
10-8
Enzyme relative abundance
Heparan sulfate degradation missing due to the absence of heparanase, a eukaryotic enzyme
Other pathways not bottlenecked by individual genes
HUMAnN links microbiome-wide pathway reconstructions → site-specific pathways → individual gene families

14. Patterns of variation in human microbiome function by niche
Three main axes of variation
Eukaryotic + oxidative metabolism on exterior surfaces (skin/airways)
Low-complexity vaginal community
Unique carbohydrate metabolism in the gut
Oral habitats covary consistently, with significant differences on the non-mucosal tooth surface
These are only broad patterns: as seen above, every human-associated microbial habitat is functionally distinct!

15. Patterns of variation in human microbiome function by niche
Three main axes of variation
Eukaryotic exterior
Low-diversity vaginal
Gut metabolism
Oral vs. tooth hard surface
Only broad patterns: every human-associated habitat is functionally distinct!
Three main axes of variation
Eukaryotic + oxidative metabolism on exterior surfaces (skin/airways)
Low-complexity vaginal community
Unique carbohydrate metabolism in the gut
Oral habitats covary consistently, with significant differences on the non-mucosal tooth surface
These are only broad patterns: as seen above, every human-associated microbial habitat is functionally distinct!

16. How do microbes and function vary within each body site across the population?

17. How do body sites compare between individuals across the population?

18. HMP: Prevalence of species (OTUs) across the population
Cumulative prevalence

19. HMP: Prevalence of pathways across the population
16 (of 251) modules strongly “core” at 90%+ coverage in 90%+ individuals at 7 body sites
24 modules at 33%+ coverage
71 modules (28%) weakly “core” at 33%+ coverage in 66%+ individuals at 6+ body sites
Contrast zero phylotypes or OTUs meeting this threshold!
Only 24 modules (<10%) differentially covered by body site
Compare with 168 modules (>66%) differentially abundant by body site
Cumulative prevalence

20. Linking function to community composition
Plus ubiquitous pathways: transcription, translation, cell wall, portions of central carbon metabolism…
← 52 posterior fornix microbiomes →
Lactobacillus crispatus
Phosphate and peptide transport
Lactobacillus jensenii
Sugar transport
Embden-Meyerhof glycolysis, phosphotransferases
Lactobacillus gasseri
Lactobacillus iners
F-type ATPase, THF
← Taxa and correlated metabolic pathways →
As Dirk mentioned, the vaginal community is particularly low-diversity, and it’s been observed by Jacques Ravel and Larry Forney in their HMP Demonstration Project to assemble into one of about five different states within each individual. Each state is dominated by just one or two specific organisms, and you can see that here in a view of these Lactobacilli and other organisms in the HMP’s 52 posterior fornix metagenomes. However, we’ve co-clustered these organisms’ abundances with that of microbial metabolism that’s also differentially abundant and specific to these community types – the vaginal community tends to adopt one of just a few possible states within each individual, dominated by one or two organisms that bring a specific set of unique metabolism to the community. These examples are, of course, complemented by processes that are present in every community type, including basic housekeeping and metabolism.
Gardnerella/Atopobium
AA and small molecule biosynthesis
Candida/Bifidobacterium
Eukaryotic pathways

21. Linking communities to host phenotype
Body Mass Index
Top correlates with BMI in stool
Vaginal pH (posterior fornix)
Normalized relative abundance
Finally, it’s important to emphasize that in this case, community structure and function also correlate with host phenotype in the form of vaginal pH. If you compare host vaginal pH in these 50 subjects to the abundance of specific organisms and metabolism, there’s a relatively large set of processes that are depleted in high-pH communities, plus the organism Lactobacillus crispatus. These are instead replaced by a collection of other organisms and processes that come to dominate high-pH communities. This is one of the strongest examples we’ve observed in the HMP of a direct correlation all the way from community membership to function to host phenotype, and I should emphasize the strength of this effect in comparison to other phenotypes of interest like BMI in the gut, where this is the same view of the best correlations we observe there.
Vaginal pH (posterior fornix)
Vaginal pH, community metabolism, and community composition represent a strong, direct link between phenotype and function in these data.

22. Who’s there varies even in health
Microbial biomolecular function and metabolism in the human microbiome: the story so far?
HUMAnN
Accurate metagenomic metabolic reconstruction
Sequences → genes → pathways → phenotypes
Validated on 4x synthetic communities
Who’s there varies even in health
What they’re doing doesn’t (as much)
There are patterns in this variation
Communities in related environments adapt using related functions
Function correlates with membership and phenotype
~1/3 to 2/3 of human metagenome characterized
Job security!

23. Ask both what you can do for your microbiome and what your microbiome can do for you
This, along with an egregious misquote of JFK, is the message I’d like to conclude with today. It’s important to remember that you and your microbiome interact constantly, and that its biomolecular functionality is an important complement to your own. The human genome is a fixed structure – we all carry the same genes for our entire lives. To draw a parallel with Waddington’s energy landscape, though, our microbiome and its functionality is fluid – it’s plastic, constantly responding to selection from its environment. That selection includes you, the landscape of microbial environments throughout the human body, and each of our unique diets and environments that shape and personalize the functionality of our microbiomes over the course of our lives.
Waddington

24. Interested? We’re recruiting students and postdocs!
Thanks!
Human Microbiome Project
Dirk Gevers
George Weinstock
Owen White
Rob Knight
Makedonka Mitreva
Erica Sodergren
Mihai Pop
Vivien Bonazzi
Jane Peterson
Lita Proctor
Nicola Segata
Levi Waldron
Fah Sathira
Johannes Goll
Yuzhen Ye
Beltran Rodriguez-Mueller
Jeremy Zucker
Mathangi Thiagarajan
Brandi Cantarel
Qiandong Zeng
Maria Rivera
Barbara Methe
Bill Klimke
Daniel Haft
Sahar Abubucker
Jacques Izard
HMP Metabolic Reconstruction
Bruce Birren Ramnik Xavier
Doyle Ward Eric Alm
Ashlee Earl Lisa Cosimi
Alyx Schubert
Pat Schloss
Ben Ganzfried
Interested? We’re recruiting students and postdocs!
Vagheesh Narasimhan
Larisa Miropolsky

26. HMP: Prevalence of genera (phylotypes) across the population
Cumulative prevalence