ISB nov 2014
- 2. A microbe dominated world
- 3. A microbe dominated world The universal nature of biochemistry. Pace NR. Proc Natl Acad Sci U S A. 2001 Jan 30;98(3):805-8.
- 4. A microbe dominated world Lozupone and Knight PNAS. 2007 Jul 3;104(27):11436-40
- 5. A microbe dominated world Lozupone and Knight PNAS. 2007 Jul 3;104(27):11436-40 Subway image adapted from http://laughingsquid.com/the-gastrointestinal-system-represented-as-a-subway-map/
- 6. A microbe dominated world Ley et al Nat Rev Microbiol. Oct 2008; 6(10): 776–788 Subway image adapted from http://laughingsquid.com/the-gastrointestinal-system-represented-as-a-subway-map/
- 7. How “human” are we, really? Hamady, 2009, PhD thesis
- 8. How “human” are we, really? Hamady, 2009, PhD thesis
- 9. How “human” are we, really? Hamady, 2009, PhD thesis
- 10. Costello et al. Science 326, 1694 (2009)
- 11. Caporaso et al. Genome Biology. 2011;12(5):R50
- 12. Gevers et al. Cell Host and Microbe 2014 Mar 12;15(3)
- 13. Koeing et al. Proc Natl Acad Sci U S A. 2011 Mar 15;108
- 14. Video fun time! Infant time series, https://www.youtube.com/watch?v=Pb272zsixSQ Gut ecosystem restoration https://www.youtube.com/watch?v=-FFDqhM4pks
- 15. Huttenhower et al. 2012 Nature Jun 13:486(7402):207-14
- 16. Yatsunenko et al. Nature. 2012 May 9;486(7402):222-7
- 17. Yatsunenko et al. Nature. 2012 May 9;486(7402):222-7 Human Microbiome Project
- 18. Lozupone et al. Genome Research 2013 Oct;23(10):1704-14
- 19. Blinding, randomization Blinding, randomization Artificial N=2 Sequence generation Data processing Samples: Stool, fresh, N=11 Stool, freeze-dried, N=7 Robogut N=2 Lab 1 Raw Sample Transfer N = 53 Control 96 Tube Sample Set 96 Tube Sample Set Lab 2 Lab 3 . . . Lab 15 HMP DACC FTP Site Site 1 Site 2 Site 3 . . . Site 9 Integrated Data Analysis Slide credit: Dr. Amnon Amir
- 20. Variability in methods Handling DNA extraction kit manufacturer Chemagen 1 MO-BIO 7 Omega BioTek 1 Promega 1 Qiagen 2 Zymo Research Corporation 1 Not reported 3 Homogenizer used? Yes 11 No 3 Not reported 1 PCR primer set 318F/806R 1 EMP V4 515F/806R 11 Schloss 2013 2 Sequencing machine model HiSeq 2500 1 MiSeq 13 Paired end sequencing? Yes 13 No 1 Sequencing read length 150 4 175 1 210 1 250 6 300 2 Fraction of PhiX (range 0-0.3) 0.11 ±0.08 Fraction quality bases (range 0.57-0.96) 0.86 ±0.12 Bioinformatics Primary QC software QIIME 2 Trimmomatic 3 UPARSE 2 Other/custom 2 Primary OTU calling software QIIME 4 UPARSE 5 Unsupervised OTU clustering step? Yes 8 No 2 Taxonomic assignment strategy Classification 3 Clustering 5 Mapping 2 OTU filtering? Yes 3 No 7 Slide credit: Dr. Amnon Amir
- 29. EMPeror http://biocore.github.io/emperor/ Vázquez-Baeza et al. 2013 Gigascience 2(1)
- 31. Genome or microbiome? Yatsunenko et al. Nature. 2012 May 9;486(7402):222-7
- 32. …but with sufficient power Goodrich et al. 2014. Cell 4(6)789-99
- 33. …but with sufficient power Goodrich et al. 2014. Cell 4(6)789-99 Christensenellaceae
- 34. • Analyses are executable, reproducible, and BSD-licensed
- 35. Demo of IPython Notebook
- 36. Where to now?
- 37. Muegge et al. 2011. Science 332(6032):970-974
- 38. Muegge et al. 2011. Science 332(6032):970-974
- 39. Ridaura et al. 2013. Science 341(6150)
- 40. Personal Genome Project Open Humans
- 41. (Some of the) American Gut Collaborators Prof. Rob Knight Co-founder HHMI, CU-Boulder, UCSD Dr. Jeff Leach Co-founder Human Food Project Prof. Cathy Lozupone Creator of UniFrac, focus on ASD Anschutz Medical Center Prof. Marty Blaser ABX impact NYU School of Medicine Prof. Maria Gloria Succession in infants NYU School of Medicine Dominguez Prof. Patrice Cani Type-2 Diabetes Universite Catholique de Louvain Prof. George Church Everything Harvard Medical School, MIT Prof. Pieter Dorrestein Realtime Mass Spec UCSD Prof. Jack Gilbert Hospital microbiome, EMP, ASD U-Chicago, ANL Prof. Ruth Ley Env. relationship with human genetics Cornell University Prof. Owen White Genomics expert U-Maryland School of Medicine; PI of HMP DACC Prof. Phil Hugenholtz Taxonomy, microbial diversity U-Queensland, Australian Centre for Ecogenomics Dr. Tim Spector Genetic epidemiology King’s College
- 42. The American Gut Consortium The Knight Lab Prof. Rob Knight Prof. Robin Dowell Prof. Aaron Clauset Prof. Paul Wischmeyer Dr. Antonio Gonzalez Dr. Gail Ackermann Yoshiki Vázquez-Baeza Justine Debelius Adam Robbins-Pianka Emily TerAvest Alina Prassas Acknowledgements Dr. Embriette Hyde Jeff DeReus Dr. Jenya Kopylova Dr. Eva Nováková Dr. Luke Thompson Justine Debelius Jose Navas
Editor's Notes
- #2: Always have to use this slide…
- #3: Go anywhere in the environment and you can find microbial life drivers of the global ecosystem Estimated at 10^30 cells (whitman)
- #4: …but it isn’t just that they’re everywhere, they are the dominant form of life Billions of years of evolution and incredible genetic diversity Complex microbial relationships But of course, the environment selects (becking) So if you shove a q-tip everywhere in the environment you can imagine, what distinguishes the collection of microbes found the most?
- #5: And it turns out, whether samples are from saline or non-saline envs This is a principal coordinates plot, will be showing a few, and will go into detail later on how these plots are created. For now, I want you to interpret as “each point is a sample, points closer are more similar, and PC1 is more meaningful than PC1” Hotsprings/hydrothermal vents are not outliers Estuarys in the center How do animals fit in?
- #6: Rather, what about the organisms that have specialized to gastrointestinal tracts?
- #7: Incredibly, notice signal is washed out, primary difference is now env or vertebrate Notice salne/nonsaline split is still present 3lbs of organisms in and on you
- #10: - Specialization of function - Extends what our bodies can do - vitamin synthesis - scfa production - break down of complex plant fibers - protect us, even educate immune system Microbiome implicated in: autism, obesity, wasting diseases, tylenol toxicity What do our communities look like from all over?
- #11: Three primary clusters of samples Not explained by person, day or sex
- #13: Not just PCoAs, though they do make data exploration like a video game Enrichment in Crohn’s disease for nasty bugs, highlight fuseobacteria and colorectal cancer relationship Depletion of organisms that are tied to butyrate and other SCFA production, note that butyrate feeds IECs which coordinate with the innate immune system ABX usage in these patients appears to amplify the loss of beneficial microbes and gain of detrimental ones
- #14: SWITCH TO VIDEO!
- #15: Infant time series video, using the short one that rob uses for presentations
- #16: INTENTIONALLY CUT OFF What I think is the coolest picture in the hmp, but this wasn’t new Lots of taxonomic variation, but what they defined as healthy were individuals aged around 20-40, and not that many individuals at that But I think the study design issue is highlighted better by Yatsunenko et al
- #17: Age gradient in global gut, and primary clusters explained by population And the diversity represented by the HMP is…
- #18: - This little region
- #19: Population for the reference is one issue, but technical artifacts abound Highlight primer choices (eg hmp v13 and v35) Differing protocols used Reinforces why it is hard to compare the results of study foo with bar if the protocols differ Should mention study x used instrument y too… Practically: Protocols, primers, voodoo, and sequencing technologies can matter - Design for reuse and integration - Sample sizes
- #20: Process thousands of samples, entirely blinded for protocols and data processing Assess variation, what matters what doesn’t’ Includes clinical samples Vital for medical applications
- #21: Manuscript in prep
- #23: Add in pop comparisons against HMP
- #25: Which is further emphasized with respect to the AG (which is primarily US individuals currently) For this figure, randomly subsample N samples I times and collect the mean and std of the min observed distance, do this over an increasing number of samples Note, this is r1-14 and sortmerna, filtered for blooms, and all ag fecal samples @ 1k rarefaction
- #26: ICU microbiome data in isolation Background: two time points, near to admission, at day 10 or dischange 5 sites around the world No restriction on diagnosis We see clustering by site, but we don’t know if they are systematically different from the rest of the population
- #27: - Combined with full ag (and filtered for blooms) ICU are large dots, small dots are AG Trending toward saliva, afaik not observed before Have not eval’d specific differences yet
- #29: - Sleep duration more interesting, exercise freq. and location was weak (p==0.02 and 0.06 respectively, and uncorrected)
- #31: NOTE ASD COHORT British Gut, just off the ground, in conjunction with Dr. Tim Spector Reduced overhead for participants, can ship domestically instead of internationally EMP is the project that contains the AG Aiming to collect 200,000 samples from everywhere, everything (10s of thousands of samples sequenced now) Over 160 indep. Research projects, and > 200 collaborators Strict sequencing protocols to minimize technical variability, and maximize reuse of data Adherence to MIMARKS, rigorous metadata standards Open access, open source Works close with GSC
- #32: Explains more of the variance than genome
- #33: Samples from nearly 1000 individuals, 171 MZ pairs and 245 DZ pairs, all female
- #34: Identified christensenellaceae through this as significantly heritable In mouse transplant experiments, adding c. minuta to innoculum that lacked it resulted in reduced adiposity (donor source was obese human)
- #36: Haven’t decided specific one yet, but probably shared OTUs as I know that one the best
- #37: Predominantly 16S (but predicting other data works _really_ well!) Need to get more samples! Can begin to use AG to provide prelim. evidence for more focused studies Systems approach: integrating composition with other data types like metabolomics
- #38: - Can we apply linear transformations to PCs as to best get the points to match up?
- #39: For instance, we can ask how well composition and metagenomic potential are correlated And it turns out, that it is correlated pretty well. In this data set, fecal samples were collected from a range of mammals Primary clustering is by diet type, and paired metagenomic-16S samples are highly correlated
- #40: Going a step further, in a study of twins discordant for obesity, we can see that transcriptomic and metabolomic data correlate well too
- #41: - In conjunction with Jason Bobe, George Church, and the Personal Genome Project, we’ve been collecting microbiome samples from PGP participants AG is the first outside project on Open Humans Aiming to be the go to place for members of the general public that want to participate in research Link studies Provide mechanisms to link individuals between studies while protecting privacy
- #42: …just some of the collaborators