Translational Genomics

Ben Hayes – Genomics for livestock breeding

• ref population with known genotypes and phenotypes.
• generate genomic breeding value equation (linear weighted sum of SNPs)
  • selection population from Marker genotypes (with SNPs). Chose which to use for breeding
• SNPs don’t work across population or breeds
  • not dense enough
  • accuracy errodes rapidlyu
  • genome seq data instead of 50K SNP data?
• 27 breeds, 1000 – 2K individual cows sequenced. 35 million SNPs, 2 million INDEL
• BayesRC
  • allows different proportion of variants in each class
  • allows tissue specific differences in expression
  • ‘class’ e.g. synomous vs. non-synonomous variations
• two traits selected
  • lactation genes
  • temperment (farmers don’t like being kicked)
• experiment
  • 20,000 individuals with SNP ChIPs, from Holstein (B&W), Jeresey
  • apply results in a different breed not in the original data (Red cows)
  • BAESRC makes a much clearer distinguishment of SNP in PARE(?) lactase associated gene
  • .1% of genes explain 1% of variance. 95% of genes explain 0%
• temperment
  • TMEM113D – linked in mouse and humans to anxiety and panic behaviors.
• conclusions
  • sequence data + improved method -> more precise mapping
  • need greater information about classes (e.g. ENCODE)

Eli Rodger-Melnick: Open chromatin

• intro to maize: Diploid, 2.3 Gb genome, 10 chromosomes, high diversity (human-to-chimp scale). rapid LD decay
• light and heavy MNase Digest in whole root and whole shoot. Call MNase hyper-sensitive
• gene Tb1 regulates growth of lateral branches.
  • regulated by two transposable elements 65 Kb upstream.
  • MNase hyper sensitive sites ID enhancers and promoters
• compare 2 lines, with 50 SNPs, test algorithm for assigning explanatory variance
• MNase HS sites are a small subset of the low methylation regions
• over half of the HS sites in root are shared in shoot.
• most MNase sites are near genes (though 95% are outside of genes). Minor peaks in frequency around 100 kb out
• GWAS hits are enriched in and around open chromatin.
• MNase HS distribution explains nearly as much of the variance in GWAS as CDS (despite representing less total part of the genome than CDS). Larger erniched if normalized by representation
• shoot specific data explains most traits (but most traits are shoot related).

Questions

• why MNase instead of FAIREseq or DNH

De Groot – Cancer application

• genomic instability, genomic heterogeneity, complex, heterogeneous spatial context
• use tissue engineering and micro-fludics to understand
• why micro-fluidics — more assays on same sample volume, especially
• use passive pumping, — different in pressure, operated with pipettes, no pumps
• exploit low mixing in microfluidic volumes to create diffusion gradients/migration experiments.
• Application to multiple Myeloma
  • a blood cancer that affects the eldery
  • manageable (with difficulty), terminal condition, strong interaction with microenbironment
  • myeloma resides in bone marrow — contributes to drug resistance
• setup:
  • two culture chambers, connected by central well with small channels which allow for difffusion in uniform gradients
• findings
  • monoculture not as predictive (separated but overlappping distributions)
  • in context of whole explaint in outside chambers, response is much better seperated
• new setup: hanging drop culture: two well (allows adding food, treatments etc (and removing?) without disrupting 3D)
  • developed porous polymeric scaffold.
  • compare direct co-culture (physical interaction) and indirect coculture (share extracellular fluid)
  • stromal cells on scaffold added to mylenoma cells in bottom of droplet

Diane Dickel: Large scale in vivo enhancer deletion with CRISPR/Cas9 (Berkeley)

• excellent talk, not open

Denis Lo – non-invasive molecular diagnostics

• non-invasive prenatal testing
• Plamsa DNA can be isolated from blood
  • can sequence these and separate
• fetal DNA fragments tend to be small.
  • also see different distribution of peaks
  • fetal DNA depleted in linker after fragmentation (less heterochromatin)
• mtDNA is very short, no peaks (no histones)
• the greater the fraction of the DNA coming from the fetus the shorter the size distribution
• can ID trisomies based on this size diagnostic (not based on counting)
• can combine counting and size for more robust
• can we detect cancer in this way? — cancer has CNV can be detected
• now extending this approach to cohort of 200 patients, 32 healthy, 67 HBV carriers, 90 HCC patients (tumors) 84% detected.
  • this is all traditional sequencing measure
• more tumor DNA the more short sequences detected.
• biggest size difference corresponds with more tumors
• antibodies bind circulating DNA? use anti-DNA antibodies
• patients with active SLE have a larger peak at small levels, related to the amount of anti-DNA in the plasma
  • enriched in hypomethylated DNA.
• increased tumor cell death, plasma DNA goes up more.

Questions

• classify cancer types? – not yet, but maybe in combinations with other data
• histones still associated in plasma DNA? – yes

Boris Rebolledo-Jaramillo on mtDNA

• D-loop, relatively recent part of sequence, also most frequently mutated. involved in replication and transcription (which are interdependent in mtDNA)
• most mutations which cause disease are heteroplasmic
  • severity of disease corresponds with fraction of abberant genomes
  • heteroplasmy goes through stochastic bottleneck
• experiment:
  • blood and cheek swab samples from mother and child. Isolate mtDNA
  • developed robust pipeline to ID artifcats from sample contamination, vial swapping
• Focus on sites with >1000 seq depth and MAF > 1%, -> 172 sites. significant by R. Nileson method
• validate sites using illumina vs Sanger
• ID heteroplasmy only in child or only in mother
  • evidence of bottleneck. Can calculate quantitative effect
• Germ-line mutation rates.
  • D-loop 0.08, full mtDNA 0.013 (lower limit due to filtering for high confidence)
• maternal age dependence on bottleneck frequency?
  • yes, older mothers more heteroplasmies
  • older mothers, more heteroplasmies in the child.
• disease associated alleles – 1 in 8 mothers are carriers (but all are healthy in this study)
  • in 1 child levels of heteroplasmy are comparable to those in patients with the mtDNA disesase.
• checked inference also looking at hair: frequencies are similar.

Siim Sober (U. Tartu, Estonia): RNA-seq of placental transcriptional landscape in normal and complicated pregnacies

• Placenta: an important organ.
• observed in dutch hunger that starvation in mothers during pregancy leads to lasting effects of metabolism of children (e.g. diabetes)
• placenta disorder: preeclampsia – leads to premature delivery, 5% of pregnancies, linked to proteinuria (not enough protein?)
• Experimental design
  • 8 normal births
  • 8 of reach of 4 issues: small gestational age, large gestitational age, preeclampsia, and maternal diabetes.
  • RNA seq, average 17x coverage.
• enriched in placental genes.
• ID genes associated with differential expression as a function of delivery (vaginal vs c-section) fetal gender differences are all sex-chromosome genes.
• birth length and maternal age and number of births did not lead to any differential gene expression in the placenta.
• preclambpsia gene expression is clearly distinct for numerous genes from other pregnancies.
• correlated expression in baby size.

Ana Vinuela

• regulatory landscape of Islet
• 90% of variants associated with traits and diseases by GWAS are non-coding
• effect of non-coding variants largely studied by eQTL — (expected effect on levels of gene expression). But this must be done on the right tissue.
• type 2 diabetes.
  • islet of langerhans produce insulin
• setup
  • difficult to get pancreatic samples
  • combined multiple studies data: 259 samples (whole Islets). 26 samples for Beta cells
  • need approach to remove batch effects by lab
• normalizing
  • PC1 and PC2 cluster the different labs. Correct these effects in the first 10 PCs so the difference between labs disappears.
• now doing eQTL discovery
• ID islet eQTLs (mostly in promoters, as typically observed).
• have an IRX3 eQTL