Beyond	  the	  sequence.
“Chromosome-­‐territory–interchroma7n-­‐compartment”	  (CT-­‐IC)	  Model              h"p://www.nature.com/nrg/journal/v2/...
CAGE	  Cap	  analysis	  gene	  expression  “FANTOM	  is	  an	  internaFonal	  research	  consorFum	  established	    by	  ...
RNA-­‐PET
RNA-­‐seq
ChIP-­‐seq
DNase-­‐seq h"p://cshprotocols.cshlp.org/content/2010/2/pdb.prot5384/F1.expansion.html
FAIRE-­‐Seq	  	  Formaldehyde-­‐Assisted	  IsolaFon	  of	  Regulatory	  
DNase-­‐seq	  VS	  FAIRE-­‐seq	  The	  technique	  was	  developed	  in	  the	  laboratory	  of	  Jason	  D.	  Lieb	  at	 ...
•    First,	  FAIRE	  requires	  no	  treatment	  of	  the	  cells	  before	  the	  addi7on	  of	  formaldehyde.	  	      ...
147	  different	  cell	  types	                             h"p://genome.ucsc.edu/ENCODE/cellTypes.html#TOP	  	  Details.	 ...
•  Tier1:	  3	  •  GM12878	  (B-­‐lymphocyte),	  H1-­‐hESC	  (embryonic	  stem	  cells),	  K562	  (leukemia)	  •  Tier2:	 ...
Together:	  Important	  features	  about	  	  the	  organiza7on	  and	  func7on	  	  of	  the	  human	  genomeŒ    The	  ...
Transcribed	  and	  protein-­‐coding	  regions	  •  Covering	  how	  much	  of	  the	  genome?	        •  GENCODE-­‐annota...
RNA	  •  How	  much	  of	  the	  genome	  sequence	  can	  be	  transcribed?	  •  62%	  of	  genomic	  bases	  are	  repro...
Protein	  bound	  regions	  	  “binding	  loca7ons	  of	  119	  different	  DNA-­‐binding	  proteins	  and	  a	  number	  o...
DNase	  I	  hypersensi7ve	  sites	  (DHSs)	  and	  footprints	  	  •    How	  many	  DHSs	  in	  the	  whole	  genome?	  •...
Regions	  of	  histone	  modifica7on	  	  	  “12	  histone	  modificaFons	  and	  variants	  in	  46	  cell	  types,	  inclu...
DNA	  methyla7on	  	  	  •  an	  average	  of	  1.2	  million	  CpGs	  in	  each	  of	  82	  cell	  lines	  and	  Fssues	 ...
Chromosome-­‐interac7ng	  regions	  	  	  •  The	  average	  number	  of	  distal	  ele-­‐	  ments	  interacFng	  with	  a...
Summary	  of	  ENCODE-­‐iden7fied	  elements	  	  	  •  80.4%,	  is	  covered	  by	  at	  least	  one	  ENCODE-­‐idenFfied	 ...
The	  impact	  of	  selec7on	  on	  func7onal	  elements
Promoter-­‐anchored	  integra7on	  	  	  •  two	  relaFvely	  disFnct	  types	  of	  promoter:	  	      •  (1)	  broad,	  ...
Promoter-­‐anchored	  integra7on	  	  	  
Transcrip7on-­‐factor-­‐binding	  site-­‐anchored	  integra7on	  	                                                        ...
Transcrip7on-­‐factor-­‐binding	  site-­‐anchored	  integra7on	  	  	  	  	                                               ...
Transcrip7on	  factor	  co-­‐associa7ons	  	  	                                                       igure	  4	  |	  Co-­...
Transcrip7on	  factor	  co-­‐associa7ons	  	  	                                                       b,	  Three	  classes...
Genome-­‐wide	  integra7on	  	  	         Figure	  5	  |	  IntegraFon	  of	  ENCODE	  data	  by	  genome-­‐wide	  segmenta...
Genome-­‐wide	  integra7on	  	  	                       TranscripFon	  Start	  Site	  (TSS),	  Promoter	  Flanking	  (PF),...
Genome-­‐wide	  integra7on	  	  	                       TranscripFon	  Start	  Site	  (TSS),	  Promoter	  Flanking	  (PF),...
Genome-­‐wide	  integra7on	  	  	                       TranscripFon	  Start	  Site	  (TSS),	  Promoter	  Flanking	  (PF),...
Genome-­‐wide	  integra7on	  	  	  
Genome-­‐wide	  integra7on	  	  	         Figure	  7	  |	  High-­‐resoluFon	  segmentaFon	  of	  ENCODE	  data	  by	  self...
Genome-­‐wide	  integra7on	  	  	  
Genome-­‐wide	  integra7on	  	  	  c,	  The	  associaFon	  of	  Gene	  Ontology	  (GO)	  terms	  on	  the	  same	  represe...
Insights	  into	  human	  genomic	  varia7on	  	  	                                                                       ...
Insights	  into	  human	  genomic	  varia7on	  	  	                                                             the	  corr...
Rare	  variants,	  individual	  genomes	  and	  soma7c	  variants	  	  	  	  	                                            ...
Common	  variants	  associated	  with	  disease	  	  	         Figure	  10	  |	  Comparison	  of	  genome-­‐wide-­‐associa...
Common	  variants	  associated	  with	  disease	  	  	         c,	  Several	  SNPs	  associated	  with	  Crohn’s	  disease...
Limita7ons	  of	  ENCODE	  Annota7ons	  	  •  Cell	  types	  -­‐	  physiologically	  and	  geneFcally	     inhomogeneous.	...
Challenges	  	  •  Adult	  human	  body	  contains	  several	  hundred	  disFnct	     cell	  types	  •  Each	  of	  which	...
Outcome	  	  •  Understanding	  of	  the	  human	  genome	  •  The	  broad	  coverage	  of	  ENCODE	  annotaFons	  enhance...
Future	  goal	  	  •  MechanisFc	  processes	  that	  generate	  these	  elements	       and	  how	  and	  where	  they	  ...
13	  Threads	  1.  TranscripFon	  factor	  moFfs	  2.  ChromaFn	  pa"erns	  at	  transcripFon	  factor	  binding	  sites	 ...
Spanking	  #ENCODE“We	  note	  that	  ENCODE	  used	  almost	  exclusively	  pluripotent	  stem	  cells	  and	  cancer	  c...
QuesFons:	  How	  much	  of	  the	  genome	  is	  func7onal?	  •  “80	  percent	  is	  the	  figure	  only	  if	  your	  de...
Encode jc 20130412
Encode jc 20130412
Encode jc 20130412
Encode jc 20130412
Encode jc 20130412
Encode jc 20130412
Encode jc 20130412
Encode jc 20130412
Upcoming SlideShare
Loading in...5
×

Encode jc 20130412

246

Published on

An integrated encyclopedia of DNA elements in the human genome
The ENCODE Project Consortium. Nature (6 September 2012)
A journal club presentation

Published in: Health & Medicine
0 Comments
1 Like
Statistics
Notes
  • Be the first to comment

No Downloads
Views
Total Views
246
On Slideshare
0
From Embeds
0
Number of Embeds
0
Actions
Shares
0
Downloads
0
Comments
0
Likes
1
Embeds 0
No embeds

No notes for slide

Encode jc 20130412

  1. 1. Beyond  the  sequence.
  2. 2. “Chromosome-­‐territory–interchroma7n-­‐compartment”  (CT-­‐IC)  Model h"p://www.nature.com/nrg/journal/v2/n4/full/nrg0401_292a.html
  3. 3. CAGE  Cap  analysis  gene  expression “FANTOM  is  an  internaFonal  research  consorFum  established   by  Dr.  Hayashizaki  and  his  colleagues  in  2000  to  assign   funcFonal  annotaFons  to  the  full-­‐length  cDNAs  that  were   collected  during  the  Mouse  Encyclopedia  Project  at  RIKEN.” h"p://www.nature.com/nprot/journal/v7/n3/fig_tab/nprot.2012.005_F1.html
  4. 4. RNA-­‐PET
  5. 5. RNA-­‐seq
  6. 6. ChIP-­‐seq
  7. 7. DNase-­‐seq h"p://cshprotocols.cshlp.org/content/2010/2/pdb.prot5384/F1.expansion.html
  8. 8. FAIRE-­‐Seq    Formaldehyde-­‐Assisted  IsolaFon  of  Regulatory  
  9. 9. DNase-­‐seq  VS  FAIRE-­‐seq  The  technique  was  developed  in  the  laboratory  of  Jason  D.  Lieb  at  the  University  of  North  Carolina,  Chapel  Hill.  In  contrast  to  DNase-­‐Seq,  the  FAIRE-­‐Seq  protocol  doesnt  require  the  permeabiliza7on  of  cells  or  isola7on  of  nuclei,  and  can  analyse  any  cell  types.[1]    DNase-­‐seq  and  FAIRE-­‐seq[2][3]  :  •  produced  strong  cross-­‐validaFon,  with  each  cell  type  having  1-­‐2%  of  the  human   genome  as  open  chromaFn.    •  are  not  fully  overlapping:  FAIRE  being  more  sensiFve  to  find  distal  regulatory   elements  that  are  not  detected  with  DNase-­‐Seq  but  missing  promoter  regions   that  are  detected  with  DNase-­‐Seq    References  ^  Giresi,  PG;  Kim,  J,  McDaniell,  RM,  Iyer,  VR,  Lieb,  JD  (2007  Jun).  "FAIRE  (Formaldehyde-­‐Assisted  IsolaFon  of  Regulatory  Elements)  isolates  acFve  regulatory  elements  from  human  chromaFn.".  Genome  Research  17  (6):  877–85.  doi:10.1101/gr.5533506.  PMC  1891346.  PMID  17179217.  ^  Song,  L;  Zhang,  Z,  Grasfeder,  LL,  Boyle,  AP,  Giresi,  PG,  Lee,  BK,  Sheffield,  NC,  Gräf,  S,  Huss,  M,  Keefe,  D,  Liu,  Z,  London,  D,  McDaniell,  RM,  Chibata,  Y,  Showers,  KA,  Simon,  JM,  Vales,  T,  Wang,  T,  Winter,  D,  Zhang,  Z,  Clarke,  ND,  Birney,  E,  Iyer,  VR,  Crawford,  GE,  Lieb,  JD,  Furey,  TS  (2011-­‐07-­‐12).  "Open  chromaFn  defined  by  DNaseI  and  FAIRE  idenFfies  regulatory  elements  that  shape  cell-­‐type  idenFty.".  Genome  Research  21  (10):  1757–67.  doi:10.1101/gr.121541.111.  PMC  3202292.  PMID  21750106.  ^  Simon,  Jeremy  M;  Giresi,  Paul  G;  Davis,  Ian  J;  Lieb,  Jason  D  (NaN  undefined  NaN).  "Using  formaldehyde-­‐assisted  isolaFon  of  regulatory  elements  (FAIRE)  to  isolate  acFve  regulatory  DNA".  Nature  Protocols  7  (2):  256–267.  doi:10.1038/nprot.2011.444.  
  10. 10. •  First,  FAIRE  requires  no  treatment  of  the  cells  before  the  addi7on  of  formaldehyde.     •  Formaldehyde  is  applied  directly  to  the  growing  cells  and  enters  quickly  because   of  its  small  size  Therefore,  the  state  of  chromaFn  just  before  the  addiFon  of  the   formaldehyde  is  likely  to  be  captured.     •  Nuclease  sensi7vity  assays  osen  require  that  cells  be  permeabilized,  or  that  nuclei   be  prepared,  both  of  which  allow  Fme  for  arFfacts  based  on  these  preparaFons  to   occur.  •  Second,  each  Fme  a  nuclease-­‐sensiFvity  assay  is  performed,  the  appropriate  enzyme   concentra7on  and  incuba7on  7me  must  be  determined,  because  of  lot-­‐to-­‐lot   variaFons  in  commercial  DNase  acFvity  and  variaFons  in  individual  nuclei  preparaFons.   With  FAIRE,  a  wide  range  of  incuba7on  7mes  (1,  2,  4,  and  7  min)  at  a  single   formaldehyde  concentraFon  (1%)  appears  to  be  equally  effecFve.    •  Third,  in  contrast  with  ChIP,  there  is  no  dependence  on  an7bodies,  FAIRE  can  analyze   any  cells:  wild  type,  mutant,  or  those  that  contain  transgenes  that  would  make  histone   ChIPs  technically  difficult.  •  Another  important  advantage  of  FAIRE  is  that  it  posi7vely  selects  genomic  regions  at   which  nucleosomes  are  disrupted.  These  same  regions  would  be  degraded  in  nuclease   sensiFvity  assays  and  require  idenFficaFon  by  their  absence  or  by  cloning  and   idenFficaFon  of  flanking  DNA.   References   ^  Giresi,  PG;  Kim,  J,  McDaniell,  RM,  Iyer,  VR,  Lieb,  JD  (2007  Jun).   "FAIRE  (Formaldehyde-­‐Assisted  IsolaFon  of  Regulatory  Elements)  isolates  acFve  regulatory  elements  from  human  chromaFn.".  Genome  Research  17  (6):  877–85.   doi:10.1101/gr.5533506.  PMC  1891346.  PMID  17179217.  
  11. 11. 147  different  cell  types   h"p://genome.ucsc.edu/ENCODE/cellTypes.html#TOP    Details.   h"p://www.genome.gov/26524238    RaFonale  for  the  selecFon.   Cell  types  were  selected  largely  for   pracFcal  reasons:   •  wide  availability   •  the  ability  to  grow  them  easily   •  capacity  to  produce  sufficient  numbers   of  cells  for  use  in  all  technologies  being   used  by  ENCODE  invesFgators.       Secondary  consideraFons     •  diversity  in  Fssue  source  of  the  cells   •  germ  layer  lineage  representaFon   •  the  availability  of  exisFng  data   generated  using  the  cell  type,  and   coordinaFon  with  other  ongoing   projects.1640    Data  Sets
  12. 12. •  Tier1:  3  •  GM12878  (B-­‐lymphocyte),  H1-­‐hESC  (embryonic  stem  cells),  K562  (leukemia)  •  Tier2:  15  •  A549,  CD20+,  CD20+_RO01778  CD20+_RO01794,  H1-­‐neurons,  HeLa-­‐S3  HepG2,   HUVEC,  IMR90  LHCN-­‐M2  MCF-­‐7  Monocytes-­‐CD14+,  Monocytes-­‐CD14+_RO01746,   Monocytes-­‐CD14+_RO01826,  SK-­‐N-­‐SH    •  Tier3:    338    a  few  may  be  useful  in  our  studies:  •  8988T,  pancreas  adenocarcinoma;  •  BC_Esophagus_H12817N:esophagus,DNA;  •  Caco-­‐2,  colorectal  adenocarcinoma;  •  HIPEpiC,  iris  pigment  epithelial  cells;  •  HMVEC-­‐dBl-­‐Ad,  adult  lymphaFc  microvascular  endothelial  cells  Note:  •  30  HapMap  Cell  lines,  only  1  Chinese  sample:  Coriell  GM18526  •  Normal  Fssues  and  Fetal  Fssues  •  Stem  Cells  
  13. 13. Together:  Important  features  about    the  organiza7on  and  func7on    of  the  human  genomeŒ  The  vast  majority  (80.4%)  of  the  human  genome  parFcipates  in  at  least  one  biochemical  RNA-­‐   and/or  chromaFn-­‐associated  event  in  at  least  one  cell  type.  95%  of  the  genome  lies  within  8  kb   of  a  DNA–protein  interacFon,  and  99%  is  within  1.7  kb  of  at  least  one  of  the  biochemical   events  measured  by  ENCODE.-­‐Debate!    Primate-­‐specific  elements  as  well  as  elements  without  detectable  mammalian  constraint   show,  in  aggregate,  evidence  of  nega7ve  selec7on.-­‐Func7onal?  Ž  an  iniFal  set  of  399,124  regions  with  enhancer-­‐like  features  and  70,292  regions  with  promoter-­‐ like  features,  as  well  as  hundreds  of  thousands  of  quiescent  regions.      correlate  quan7ta7vely  RNA  sequence  produc7on  and  processing  with  both  chroma7n  marks   and  transcrip7on  factor  binding  at  promoters,  indicaFng  that  promoter  func7onality  can   explain  most  of  the  varia7on  in  RNA  expression.     Many  non-­‐coding  variants  in  individual  genome  sequences  lie  in  ENCODE-­‐annotated   func7onal  regions;  this  number  is  at  least  as  large  as  those  that  lie  in  protein-­‐coding  genes.  ‘   SNPs  associated  with  disease  by  GWAS  are  enriched  within  non-­‐coding  func7onal  elements,   with  a  majority  residing  in  or  near  ENCODE-­‐defined  regions  that  are  out-­‐  side  of  protein-­‐ coding  genes.  In  many  cases,  the  disease  phenotypes  can  be  associated  with  a  specific  cell   type  or  transcrip7on  factor.-­‐right
  14. 14. Transcribed  and  protein-­‐coding  regions  •  Covering  how  much  of  the  genome?   •  GENCODE-­‐annotated  exons  of  protein-­‐  coding  genes  cover  2.94%    (~3%)  of   the  genome     •  or  1.22%  (~1%)  for  protein-­‐coding  exons.    •  Covering  how  much  of  the  gene  models?   •  Protein-­‐coding  genes  span  33.45%  from  the  outermost  start  to  stop  codons   •  or  39.54%  from  promoter  to  poly(A)  site.    •  Can  addi7onal  protein-­‐coding  genes  remain  to  be  found?   •  Analysis  of  mass  spectrometry  data  from  K562  and  GM12878  cell  lines   yielded  57  confidently  idenFfied  unique  pepFde  sequences  in  intergenic   regions  relaFve  to  GENCODE  annotaFon.    •  Other:   •  8,801  automaFcally  derived  small  RNAs  and  9,640  manually  curated  long   non-­‐coding  RNA  (lncRNA)  loci.     •  11,224  pseudogenes,  of  which  863  were  transcribed  and  associated  with   acFve  chromaFn
  15. 15. RNA  •  How  much  of  the  genome  sequence  can  be  transcribed?  •  62%  of  genomic  bases  are  reproducibly  represented  in  sequenced  long   (>200  nucleo7des)  RNA  molecules  or  GENCODE  exons.   •  Of  these  bases,  only  5.5%  are  explained  by  GENCODE  exons.   •  Most  transcribed  bases  are  within  or  overlapping  annotated  gene   boundaries  (that  is,  intronic)   •  Only  31%  of  bases  in  sequenced  transcripts  were  intergenic  •  How  many  transcrip7on  start  sites  (TSSs)    iden7fied?  •  62,403  TSSs  at  high  confidence  (IDR  of  0.01)  in  7er  1  and  2  cell  types.     •  Of  these,  27,362  (44%)  are  within  100  bp  of  the  5’  end  of  a   GENCODE-­‐annotated  transcript  or  previously  reported  full-­‐length   messenger  RNA.     •  the  start  sites  of  novel,  cell-­‐type-­‐specific  transcripts:  The   remaining  regions  predominantly  lie  across  exons  and  3’  UTRs,     cell-­‐type-­‐restricted  expression
  16. 16. Protein  bound  regions    “binding  loca7ons  of  119  different  DNA-­‐binding  proteins  and  a  number  of  RNA  polymerase  components  in  72  cell  types  using  ChIP-­‐seq”  •  How  about  the  sequence  specificity  during  the  binding  process?  •  86%  of  the  DNA  segments  occupied  by  sequence-­‐specific  transcripFon  factors   contained  a  strong  DNA-­‐binding  moFf,  and  in  most  (55%)  cases  the  known  moFf   was  most  enriched    •  How  to  explain  Protein-­‐binding  regions  lacking  high  or  moderate  affinity  cognate   recogni7on  sites?   •  82%  have  high-­‐affinity  recogniFon  sequences  for  other  factors   •  the  median  DNase  I  accessibility  is  twofold  higher  in  the  bo"om  20%  of  peaks   than  in  the  upper  80%
  17. 17. DNase  I  hypersensi7ve  sites  (DHSs)  and  footprints    •  How  many  DHSs  in  the  whole  genome?  •  Dnase-­‐seq:  •  2.89  million  unique,  non-­‐overlapping  DHSs  in  125  cell  types,  most  are  distal  to  TSSs  •  FAIRE-­‐seq:  •  4.8  million  sites  across  25  cell  types  displayed  reduced  nucleosomal,  many  of  which   coincide  with  DHSs.    •  How  many  DHSs  in  one  cell  type?  •  In  Fer  1  and  Fer  2  cell  types,  a  mean  of  205,109  DHSs  per  cell    •  ~  1.0%  of  the  genomic  sequence  in  each  cell  type,  and  3.9%  in  aggregate.     •  On  average,  98.5%  of  the  occupancy  sites  of  transcripFon  factors  mapped  by   ENCODE  ChIP-­‐seq  lie  within  accessible  chromaFn  defined  by  DNase  I  hotspots.  •  Genomic  DNase  I  footprinFng  on  41  cell  types  we  idenFfied  8.4  million  disFnct   DNase  I  footprints.  Our  de  novo  moFf  discovery  on  DNase  I  footprints  recovered  , 90%  of  known  transcripFon  factor  moFfs,  together  with  hundreds  of  novel  evoluFo-­‐ narily  conserved  moFfs,  many  displaying  highly  cell-­‐selecFve  occupancy  pa"erns   similar  to  major  developmental  and  Fssue-­‐specific  regulators.     •     
  18. 18. Regions  of  histone  modifica7on      “12  histone  modificaFons  and  variants  in  46  cell  types,  including  a  complete  matrix  of  eight  modificaFons  across  Fer  1  and  Fer  2.  ”    •  global  pa"erns  of  modificaFon  are  highly  vari-­‐able  across  cell  types,  in  accordance   with  changes  in  transcripFonal  acFvity.    
  19. 19. DNA  methyla7on      •  an  average  of  1.2  million  CpGs  in  each  of  82  cell  lines  and  Fssues    •  96%  of  CpGs  exhibited  differenFal  methylaFon  in  at  least  one  cell  type  or  Fssue  •  The  most  variably  methylated  CpGs  are  found  more  osen  in  gene  bodies  and   intergenic  regions.  •  unexpected  correspondence  between  unmethylated  genic  CpG  islands  and   binding  by  P300,  a  histone  acetyltransferase  linked  to  enhancer  acFvity    •  CpGs  with  allele-­‐specific  methylaFon  consistent  with  genomic  imprinFng,  and   determined  that  these  loci  exhibit  aberrant  methylaFon  in  cancer  cell  lines    •  reproducible  cytosine  methylaFon  outside  CpG  dinucleoFdes  in  adult  Fssues45,   providing  further  support  that  this  non-­‐canonical  methylaFon  event  may  have   important  roles  in  human  biology    
  20. 20. Chromosome-­‐interac7ng  regions      •  The  average  number  of  distal  ele-­‐  ments  interacFng  with  a  TSS  was  3.9,  and  the   average  number  of  TSSs  interacFng  with  a  distal  element  was  2.5,  indicaFng  a   complex  net-­‐  work  of  interconnected  chromaFn.    •  Whereas  promoter  regions  of  2,324  genes  were  involved  in  ‘single-­‐gene’   enhancer–promoter  interacFons,  those  of  19,813  genes  were  involved  in  ‘mulF-­‐ gene’  interacFon  com-­‐  plexes  spanning  up  to  several  megabases,  including   promoter–  promoter  and  enhancer–promoter  interacFons    •  long-­‐range  gene–  element  connecFvity  across  ranges  of  hundreds  of  kilobases  to   several  megabases    •  50–60%  of  long-­‐  range  interacFons  occurred  in  only  one  of  the  four  cell  lines,   indicaFve  of  a  high  degree  of  Fssue  specificity  for  gene–element  connecFvity    
  21. 21. Summary  of  ENCODE-­‐iden7fied  elements      •  80.4%,  is  covered  by  at  least  one  ENCODE-­‐idenFfied  element    •  Order  of  region  classes:  •  1-­‐  different  RNA  types,  covering  62%  of  the  genome  (although  the  majority  is   inside  of  introns  or  near  genes).    •  2-­‐Regions  highly  enriched  for  histone  modificaFons  (56.1%).    •  Excluding  RNA  elements  and  broad  histone  elements,  44.2%  of  the  genome  is   covered   •  open  chromaFn  (15.2%)     •  sites  of  transcripFon  factor  binding  (8.1%)   •  19.4%  covered  by  at  least  one  DHS  or  transcripFon  factor  ChIP-­‐seq  peak   across  all  cell  lines.    •  Using  our  most  conservaFve  assessment,  8.5%  of  bases  are  covered  by  either  a   transcripFon-­‐factor-­‐binding-­‐site  moFf  (4.6%)  or  a  DHS  footprint  (5.7%).  This,   however,  is  sFll  about  4.5-­‐fold  higher  than  the  amount  of  protein-­‐coding  exons,   and  about  twofold  higher  than  the  esFmated  amount  of  pan-­‐mammalian   constraint.    
  22. 22. The  impact  of  selec7on  on  func7onal  elements
  23. 23. Promoter-­‐anchored  integra7on      •  two  relaFvely  disFnct  types  of  promoter:     •  (1)  broad,  mainly  (C+G)-­‐rich,  TATA-­‐less  promoters;     •  (2)  narrow,  TATA-­‐box-­‐containing  promoters.    •  a  limited  set  of  chromaFn  marks  are  sufficient  to  ‘explain’  transcripFon  and  that  a   variety  of  transcripFon  factors  might  have  broad  roles  in  general  transcripFon   levels  across  many  genes  •  there  is  enough  informaFon  present  at  the  promoter  regions  of  genes  to  explain   most  of  the  variaFon  in  RNA  expression.    
  24. 24. Promoter-­‐anchored  integra7on      
  25. 25. Transcrip7on-­‐factor-­‐binding  site-­‐anchored  integra7on     Figure  3  |  Pa"erns  and   asymmetry  of  chromaFn   modificaFon  at  transcripFon-­‐ factor-­‐binding  sites.  a,  Results   of  clustered  aggregaFon  of   H3K27me3  modificaFon  signal   around  CTCF-­‐binding  sites  (a   mulFfuncFonal  protein   involved  with  chromaFn   structure).  The  first  three  plots   (les  column)  show  the  signal   behaviour  of  the  histone   modificaFon  over  all  sites   (top)  and  then  split  into  the   high  and  low  signal   components.  The  solid  lines   show  the  mean  signal   distribuFon  by  relaFve   posiFon  with  the  blue  shaded   area  delimiFng  the  tenth  and   nineFeth  percenFle  range.   The  high  signal  component  is   then  decomposed  further  into   six  different  shape  classes  on   the  right  (see  ref.  30  for   details).  The  shape   decomposiFon  process  is   strand  aware.  
  26. 26. Transcrip7on-­‐factor-­‐binding  site-­‐anchored  integra7on           Figure  3  |  Pa"erns  and  asymmetry  of   chromaFn  modificaFon  at  transcripFon-­‐ factor-­‐binding  sites.  b,  Summary  of   shape  asymmetry  for  DNase  I,   nucleosome  and  histone  modificaFon   signals  by  plo|ng  an  asymmetry  raFo   for  each  signal  over  all  transcripFon-­‐ factor-­‐  binding  sites.  All  histone   modificaFons  measured  in  this  study   show  predominantly  asymmetric   pa"erns  at  transcripFon-­‐factor-­‐binding   sites.  An  interacFve  version  of  this   figure  is  available  in  the  online  version   of  the  paper.
  27. 27. Transcrip7on  factor  co-­‐associa7ons       igure  4  |  Co-­‐associaFon   between  transcripFon   factors.  a,  Significant  co-­‐   associaFons  of  transcripFon   factor  pairs  using  the  GSC   staFsFc  across  the  enFre   genome  in  K562  cells.  The   colour  strength  represents   the  extent  of  associaFon   (from  red  (strongest),  orange,   to  yellow  (weakest)),  whereas   the  depth  of  colour   represents  the  fit  to  the   GSC20  model  (where  white   indicates  that  the  staFsFcal   model  is  not  appropriate)  as   indicated  by  the  key.  Most   transcripFon  factors  have  a   nonrandom  associaFon  to   other  transcripFon  factors,   and  these  associaFons  are   dependent  on  the  genomic   context,  meaning  that  once   the  genome  is  separated  into   promoter  proximal  and  distal   regions,  the  overall  levels  of   co-­‐associaFon  decrease,  but   more  specific  relaFonships   are  uncovered.  
  28. 28. Transcrip7on  factor  co-­‐associa7ons       b,  Three  classes  of  behaviour   are  shown.  The  first  column   shows  a  set  of  associaFons  for   which  strength  is  independent   of  locaFon  in  promoter  and   distal  regions,  whereas  the   second  column  shows  a  set  of   transcripFon  factors  that  have   stronger  associaFons  in   promoter-­‐proximal  regions.   Both  of  these  examples  are   from  data  in  K562  cells  and  are   highlighted  on  the  genome-­‐ wide  co-­‐associaFon  matrix  (a)   by  the  labelled  boxes  A  and  B,   respecFvely.  The  third  column   shows  a  set  of  transcripFon   factors  that  show  stronger   associaFon  in  distal  regions  (in   the  H1  hESC  line).  
  29. 29. Genome-­‐wide  integra7on       Figure  5  |  IntegraFon  of  ENCODE  data  by  genome-­‐wide  segmentaFon.   a,  IllustraFve  region  with  the  two  segmentaFon  methods  (ChromHMM  and  Segway)  in  a  dense  view  and  the  combined  segmentaFon   expanded  to  show  each  state  in  GM12878  cells,  beneath  a  compressed  view  of  the  GENCODE  gene  annotaFons.  Note  that  at  this  level  of   zoom  and  genome  browser  resoluFon,  some  segments  appear  to  overlap  although  they  do  not.  SegmentaFon  classes  are  named  and   coloured  according  to  the  scheme  in  Table  3.  Beneath  the  segmentaFons  are  shown  each  of  the  normalized  signals  that  were  used  as  the   input  data  for  the  segmentaFons.  Open  chromaFn  signals  from  DNase-­‐seq  from  the  University  of  Washington  group  (UW  DNase)  or  the   ENCODE  open  chromaFn  group  (Openchrom  DNase)  and  FAIRE  assays  are  shown  in  blue;  signal  from  histone  modificaFon  ChIP-­‐seq  in  red;   and  transcripFon  factor  ChIP-­‐seq  signal  for  Pol  II  and  CTCF  in  green.  The  mauve  ChIP-­‐seq  control  signal  (input  control)  at  the  bo"om  was  also   included  as  an  input  to  the  segmentaFon.  
  30. 30. Genome-­‐wide  integra7on       TranscripFon  Start  Site  (TSS),  Promoter  Flanking  (PF),  Enhancer  (E),  Weak  Enhancer  (WE),  CTCF   binding  (CTCF),  Transcribed  Region  (T)  and  Repressed  or  InacFve  Region  (R) Figure  5  |  IntegraFon  of  ENCODE  data  by  genome-­‐wide  segmentaFon.   b,  AssociaFon  of  selected  transcripFon  factor  (les)  and  RNA  (right)  elements  in  the  combined   segmentaFon  states  (x  axis)  expressed  as  an  observed/expected  raFo  (obs./exp.)  for  each   combinaFon  of  transcripFon  factor  or  RNA  element  and  segmentaFon  class  using  the  heat-­‐   map  scale  shown  in  the  key  besides  each  heat  map.    
  31. 31. Genome-­‐wide  integra7on       TranscripFon  Start  Site  (TSS),  Promoter  Flanking  (PF),  Enhancer  (E),  Weak  Enhancer  (WE),  CTCF   binding  (CTCF),  Transcribed  Region  (T)  and  Repressed  or  InacFve  Region  (R) Figure  5  |  IntegraFon  of  ENCODE  data  by  genome-­‐wide  segmentaFon.   c,  Variability  of  states  between  cell  lines,  showing  the  distribuFon  of  occurrences  of  the  state   in  the  six  cell  lines  at  specific  genome  locaFons:  from  unique  to  one  cell  line  to  ubiquitous  in   all  six  cell  lines  for  five  states  (CTCF,  E,  T,  TSS  and  R).    
  32. 32. Genome-­‐wide  integra7on       TranscripFon  Start  Site  (TSS),  Promoter  Flanking  (PF),  Enhancer  (E),  Weak  Enhancer  (WE),  CTCF   binding  (CTCF),  Transcribed  Region  (T)  and  Repressed  or  InacFve  Region  (R) Figure  5  |  IntegraFon  of  ENCODE  data  by  genome-­‐wide  segmentaFon.   d,  DistribuFon  of  methylaFon  level  at  individual  sites  from  RRBS  analysis  in  GM12878  cells   across  the  different  states,  showing  the  expected  hypomethylaFon  at  TSSs  and   hypermethylaFon  of  genes  bodies  (T  state)  and  repressed  (R)  regions.    
  33. 33. Genome-­‐wide  integra7on      
  34. 34. Genome-­‐wide  integra7on       Figure  7  |  High-­‐resoluFon  segmentaFon  of  ENCODE  data  by  self-­‐organizing  maps  (SOM).  a–c,  The  training  of  the  SOM  (a)  and   analysis  of  the  results  (b,  c)  are  shown.  IniFally  we  arbitrarily  placed  genomic  segments  from  the  ChromHMM  segmentaFon   on  to  the  toroidal  map  surface,  although  the  SOM  does  not  use  the  ChromHMM  state  assignments  (a).  We  then  trained  the   map  using  the  signal  of  the  12  different  ChIP-­‐seq  and  DNase-­‐seq  assays  in  the  six  cell  types  analysed.  Each  unit  of  the  SOM  is   represented  here  by  a  hexagonal  cell  in  a  planar  two-­‐dimensional  view  of  the  toroidal  map.  Curved  arrows  indicate  that   traversing  the  edges  of  two  dimensional  view  leads  back  to  the  opposite  edge.  The  resulFng  map  can  be  overlaid  with  any   class  of  ENCODE  or  other  data  to  view  the  distribuFon  of  that  data  within  this  high-­‐resoluFon  segmentaFon.    
  35. 35. Genome-­‐wide  integra7on      
  36. 36. Genome-­‐wide  integra7on      c,  The  associaFon  of  Gene  Ontology  (GO)  terms  on  the  same  representaFon  of  the  same  trained  SOM.  We  assigned  genes  that  are  within  20  kb  of  a  genomic  segment  in  a  SOM  unit  to  that  unit,  and  then  associated  this  set  of  genes  with  GO  terms  using  a  hypergeometric  distribuFon  aser  correcFng  for  mulFple  tesFng.  Map  units  that  are  significantly  associated  to  GO  terms  are  coloured  green,  with  increasing  strength  of  colour  reflecFng  increasing  numbers  of  genes  significantly  associated  with  the  GO  terms  for  either  immune  response  (les)  or  sequence-­‐specific  transcripFon  factor  acFvity  (centre).  In  each  case,  specific  SOM  units  show  associaFon  with  these  terms.  The  right-­‐hand  panel  shows  the  distribuFon  on  the  same  SOM  of  all  significantly  associated  GO  terms,  now  colouring  by  GO  term  count  per  SOM  unit.  For  sequence-­‐specific  transcripFon  factor  acFvity,  two  example  genomic  regions  are  extracted  at  the  bo"om  of  panel  c  from  neighbouring  SOM  units.  These  are  regions  around  the  DBX1  (from  SOM  unit  26,31,  les  panel)  and  IRX6  (SOM  unit  27,30,  right  panel)  genes,  respecFvely,  along  with  their  H3K27me3  ChIP-­‐seq  signal  for  each  of  the  Fer  1  and  2  cell  types.  For  DBX1,  representaFve  of  a  set  of  primarily  neuronal  transcripFon  factors  associated  with  unit  26,31,  there  is  a  repressive  H3K27me3  signal  in  both  H1  hESCs  and  HUVECs;  for  IRX6,  representaFve  of  a  set  of  body  pa"erning  transcripFon  factors  associated  with  SOM  unit  27,30,  the  repressive  mark  is  restricted  largely  to  the  embryonic  stem  (ES)  cell.  An  interacFve  version  of  this  figure  is  available  in  the  online  version  of  the  paper.
  37. 37. Insights  into  human  genomic  varia7on       Preferen7al   binding  towards   each  parental   allele Figure  8  |  Allele-­‐specific  ENCODE  elements.  a,  RepresentaFve  allele-­‐specific  informaFon  from  GM12878  cells  for  selected  assays  around  the   first  exon  of  the  NACC2  gene  (genomic  region  Chr9:  138950000–138995000,  GRCh37).  TranscripFon  signal  is  shown  in  green,  and  the  three   secFons  show  allele-­‐  specific  data  for  three  data  sets  (POLR2A,  H3K79me2  and  H3K27me3  ChIP-­‐  seq).  In  each  case  the  purple  signal  is  the   processed  signal  for  all  sequence  reads  for  the  assay,  whereas  the  blue  and  red  signals  show  sequence  reads  specifically  assigned  to  either   the  paternal  or  maternal  copies  of  the  genome,  respecFvely.  The  set  of  common  SNPs  from  dbSNP,  including  the  phased,  heterozygous  SNPs   used  to  provide  the  assignment,  are  shown  at  the  bo"om  of  the  panel.  NACC2  has  a  staFsFcally  significant  paternal  bias  for  POLR2A  and  the   transcripFon-­‐associated  mark  H3K79me2,  and  has  a  significant  maternal  bias  for  the  repressive  mark  H3K27me3.
  38. 38. Insights  into  human  genomic  varia7on       the  correla7on  of  selected  allele-­‐specific  signals   across  the  whole  genome.  For  instance,  we   found  a  strong  allelic  correla7on  between   POL2RA  and  BCLAF1  binding,  as  well  as  nega-­‐   7ve  correla7on  between  H3K79me2  and   H3K27me3,  both  at  genes  (Fig.  8b,  below  the   diagonal,  bokom  lel)  and  chromosomal   segments  (top  right).  Overall,  we  found  that   posi7ve  allelic  correla7ons  among  the  193   ENCODE  assays  are  stronger  and  more  frequent   than  nega  7ve  correla7ons.   Figure  8  |  Allele-­‐specific  ENCODE  elements.     b,  Pair-­‐wise  correlaFons  of  allele-­‐specific  signal  within   single  genes  (below  the  diagonal)  or  within  individual   ChromHMM  segments  across  the  whole  genome  for   selected  DNase-­‐seq  and  histone  modificaFon  and   transcripFon  factor  ChIP-­‐seq  assays.  The  extent  of   correlaFon  is  coloured  according  to  the  heat-­‐map  scale   indicated  from  posiFve  correlaFon  (red)  through  to   anF-­‐correlaFon  (blue).  An  interacFve  version  of  this   figure  is  available  in  the  online  version  of  the  paper.    
  39. 39. Rare  variants,  individual  genomes  and  soma7c  variants           A:  variants  annota7on   B:  1%  of  transcripFon-­‐factor-­‐ binding  sites  in  GM12878  cells   are  detected  in  a  haplotype-­‐ specific  fashion.  (Fig.  9b  -­‐a   CTCF-­‐binding  site)   C:Overall,  somaFc  variaFon  is   relaFvely  depleted  from   ENCODE  annotated  regions,   parFcularly  for  elements   specific  to  a  cell  type  matching   the  putaFve  tumour  source    
  40. 40. Common  variants  associated  with  disease       Figure  10  |  Comparison  of  genome-­‐wide-­‐associaFon-­‐study-­‐idenFfied  loci  with  ENCODE  data.  a,  Overlap  of  lead  SNPs  in  the  NHGRI  GWAS   SNP  catalogue  (June  2011)  with  DHSs  (les)  or  transcripFon-­‐factor-­‐binding  sites  (right)  as  red  bars  compared  with  various  control  SNP  sets  in   blue.  The  control  SNP  sets  are  (from  les  to  right):  SNPs  on  the  Illumina  2.5M  chip  as  an  example  of  a  widely  used  GWAS  SNP  typing  panel;   SNPs  from  the  1000  Genomes  project;  SNPs  extracted  from  24  personal  genomes  (see  personal  genome  variants  track  at  h"p:// main.genome-­‐browser.bx.psu.edu  (ref.  80)),  all  shown  as  blue  bars.  In  addiFon,  a  further  control  used  1,000  randomizaFons  from  the   genotyping  SNP  panel,  matching  the  SNPs  with  each  NHGRI  catalogue  SNP  for  allele  frequency  and  distance  to  the  nearest  TSS  (light  blue   bars  with  bounds  at  1.5  Fmes  the  interquarFle  range).  For  both  DHSs  and  transcripFon-­‐factor-­‐  binding  regions,  a  larger  proporFon  of   overlaps  with  GWAS-­‐implicated  SNPs  is  found  compared  to  any  of  the  controls  sets.  b,  Aggregate  overlap  of  phenotypes  to  selected   transcripFon-­‐factor-­‐binding  sites  (les  matrix)  or  DHSs  in  selected  cell  lines  (right  matrix),  with  a  count  of  overlaps  between  the  phenotype   and  the  cell  line/factor.  Values  in  blue  squares  pass  an  empirical  P-­‐value  threshold  #0.01  (based  on  the  same  analysis  of  overlaps  between   randomly  chosen,  GWAS-­‐matched  SNPs  and  these  epigeneFc  features)  and  have  at  least  a  count  of  three  overlaps.  The  P  value  for  the  total   number  of  phenotype–transcripFon  factor  associaFons  is  ,0.001.    
  41. 41. Common  variants  associated  with  disease       c,  Several  SNPs  associated  with  Crohn’s  disease  and  other  inflammatory  diseases  that  reside  in  a  large   gene  desert  on  chromosome  5,  along  with  some  epigeneFc  features  indicaFve  of  funcFon.  The  SNP   (rs11742570)  strongly  associated  to  Crohn’s  disease  overlaps  a  GATA2  transcripFon-­‐factor-­‐binding  signal   determined  in  HUVECs.  This  region  is  also  DNase  I  hypersensiFve  in  HUVECs  and  T-­‐helper  TH1  and  TH2   cells.  An  interacFve  version  of  this  figure  is  available  in  the  online  version  of  the  paper.
  42. 42. Limita7ons  of  ENCODE  Annota7ons    •  Cell  types  -­‐  physiologically  and  geneFcally   inhomogeneous.  •  Local  microenvironments  in  culture  may  also  vary  •  Use  of  DNA  sequencing  to  annotate  funcFonal  genomic   features  is  also  constrained.  •  Considerable  quanFtaFve  variaFon  in  the  signal  strength   along  the  genome      (The  ENCODE  Project  ConsorFum,  2011)
  43. 43. Challenges    •  Adult  human  body  contains  several  hundred  disFnct   cell  types  •  Each  of  which  expresses  a  unique  subset  of  the  1,800   TFs  encoded  in  the  human  genome  •  Brain  alone  contains  thousands  of  types  of  neurons   that  are  likely  to  express  not  only  different  sets  of  TFs   but  also  a  larger  variety  of  non-­‐coding  RNAs  •  A  truly  comprehensive  atlas  of  human  funcFonal   elements  is  not  pracFcal  with  current  technologies    (The  ENCODE  Project  Consor7um,  2011)
  44. 44. Outcome    •  Understanding  of  the  human  genome  •  The  broad  coverage  of  ENCODE  annotaFons  enhances   our  understanding  of  common  diseases  with  a  geneFc   component,  rare  geneFc  diseases  •  119  of  1,800  known  transcripFon  factors  and  13  of  more   than  60  currently  known  histone  or  DNA  modificaFons   across  147  cell  types  •  Overall  these  data  reflect  a  minor  fracFon  of  the   potenFal  funcFonal  informaFon  encoded  in  the  human   genome       (The  ENCODE  Project  ConsorFum,  2012)
  45. 45. Future  goal    •  MechanisFc  processes  that  generate  these  elements   and  how  and  where  they  funcFon  •  Enlarge  the  data  set  to  addiFonal  factors,   modificaFons  and  cell  types,  complemenFng  the  other   related  projects  •  ConsFtute  foundaFonal  resources  for  human   genomics,  allowing  a  deeper  interpretaFon  of  the   organizaFon  of  gene  and  regulatory  informaFon  and   the  mechanisms  of  regulaFon,  and  thereby  provide   important  insights  into  human  health  and  disease       (The  ENCODE  Project  ConsorFum,  2012)
  46. 46. 13  Threads  1.  TranscripFon  factor  moFfs  2.  ChromaFn  pa"erns  at  transcripFon  factor  binding  sites  3.  CharacterizaFon  of  intergenic  regions  and  gene  definiFon  4.  RNA  and  chromaFn  modificaFon  pa"erns  around  promoters  5.  EpigeneFc  regulaFon  of  RNA  processing  6.  Non-­‐coding  RNA  characterizaFon  7.  DNA  methylaFon  8.  Enhancer  discovery  and  characterizaFon  9.  Three-­‐dimensional  connecFons  across  the  genome  10. CharacterizaFon  of  network  topology  11. Machine  learning  approaches  to  genomics  12. Impact  of  funcFonal  informaFon  on  understanding  variaFon  13. Impact  of  evoluFonary  selecFon  on  funcFonal  regions
  47. 47. Spanking  #ENCODE“We  note  that  ENCODE  used  almost  exclusively  pluripotent  stem  cells  and  cancer  cells,  which  are  known  as  transcrip7onally  permissive  environments.”    Another  cri7cism  in  the  paper  is  the  sensi7vity  vs  specificity  choice  for  repor7ng  on  the  data.    Unfortunately,  the  ENCODE  data  are  neither  easily  accessible  nor  very  useful—without  ENCODE,  researchers  would  have  had  to  examine  3.5  billion  nucleo7des  in  search  of  func7on,  with  ENCODE,  they  would  have  to  siQ  through  2.7  billion  nucleo7des.-­‐-­‐“Big  Science,”  “small  science,”  and  “ENCODE”.
  48. 48. QuesFons:  How  much  of  the  genome  is  func7onal?  •  “80  percent  is  the  figure  only  if  your  definiFon  is  so  loose  as  to  be  all  but   meaningless.”  •  “FuncFonal"  simply  means  a  li"le  bit  of  DNA  thats  been  idenFfied  in  an  assay  of   some  sort  or  another.  That’s  a  remarkably  silly  definiFon  of  funcFon  and  if   youre  using  it  to  discount  junk  DNA  its  downright  disingenuous.”  •  “The  upshot  is  that  you’d  expect  many  of  the  elements  that  ENCODE  idenFfied   if  you  just  wrote  out  a  random  string  of  As,  Gs,  Cs,  and  Ts.”  •  “does  an  onion  have  around  five  Fmes  as  much  non-­‐coding  DNA  as  we  do?  Or   why  pufferfishes  can  get  by  with  just  a  tenth  as  much?  “  •  Junk  Vs  Garbage  

×