Lecture 22: More protein-DNA interactions

BMMB 554: Lecture 22

Transcript

1. L22: DIGGING DEEPER INTO PROTEIN-DNA INTERACTIONS Foundations in Data Driven Life Sciences BMMB 554
2. Learning objectives • How can we determine the precise locations of protein-DNA interaction from ChIP-seq data? • How can we find differences in regulatory signals across conditions? • Can we determine whether proteins are bound directly or indirectly to DNA? • Can we assay three-dimensional interactions in chromatin?
3. PINPOINTING PROTEIN-DNA BINDING EVENTS IN CHIP-SEQ DATA
4. Identifying precise TF binding locations • ChIP-seq reads are distributed bimodally around binding sites. • Regions of ChIP-enrichment are known as “peaks”. Valouev, et al. Nature Methods (2008)
5. Identifying precise TF binding locations
6. Computational challenge: resolve the structure of TF binding events from ChIP-seq data How many binding events are here? How close to the actual bound bases are event predictions? + -
7. Mixture models of ChIP-seq binding events Lhx3 ChIP-seq
8. Mixture models of ChIP-seq binding events Lhx3 ChIP-seq
9. Mixture models of ChIP-seq binding events Lhx3 ChIP-seq
10. Mixture models of ChIP-seq binding events Lhx3 ChIP-seq Predicted binding events: Motif instances: 140bp 160bp
11. GPS/GEM/ChExMix allow more accurate spatial resolution of binding events GEM: http://groups.csail.mit.edu/cgs/gem/
12. DETECTING DIFFERENTIAL PROTEIN- DNA INTERACTIONS
13. Next steps in regulatory analysis • We’ve found: – TF ChIP-seq peaks, or – Histone modification domains – ATAC-seq domains, or… etc. in different cell types / conditions. • Statistical significance ≠ biological significance. • How do we find loci with differential signals?
14. Detecting differential binding: Can’t we just use a Venn diagram? Peaks called in condition A Peaks called in condition B Peaks in condition A that are within 200bp of peaks in condition B
15. Beware of comparisons between thresholded quantitative variable Fake 2-condition ChIP-seq data NO DIFFERENCES (simulated replicates)
16. Beware of comparisons between thresholded quantitative variable Peak in B
17. Beware of comparisons between thresholded quantitative variable A > B B > A A = B Non-peak A B
18. Beware of comparisons between thresholded quantitative variable Fake 2-condition ChIP-seq data 40% of datapoints are 4-fold different across conditions
19. Beware of comparisons between thresholded quantitative variable A > B B > A A = B Non-peak A B
20. 0% 20% 40% 60% 80% 100% 1 2 3 4 5 % Differential Events Sensitivity Mean Read Count Quintile (1=lowest, 5=highest) Venn diagrams are bad at detecting differential binding events Sn. Sp. 0% 20% 40% 60% 80% 100% 1 2 3 4 5 % Differential Events Specificity Mean Read Count Quintile (1=lowest, 5=highest) MACS & list comparison
21. What should we do instead? • Treat ChIP/ATAC/DNase-seq data like any other quantitative assay in biology! • Simple procedure: – Merge peaks across conditions (bedtools merge) – Quantify per-replicate read counts (featureCounts) – Perform differential enrichment analysis (DEseq2) • More sophisticated tools: – DBChIP – DiffBind – MultiGPS
22. A principled approach to characterizing differential binding activity 1. Perform ChIP-seq experiments in multiple conditions 2. Detect & quantify consistent binding events across experiments 3. Perform differential count analysis Condition 1 Condition 2 +8 0 -8 Scaled mean read count log fold difference
23. MultiGPS aligns binding events across experiments ChrnX cluster ~30Kbp Plotted window = 1,650bp Isl1 (spinal motor neurons) Isl1 (cranial motor neurons) Isl1 (ES cells)
24. Isl1 (sMN) Isl1 (cMN) Isl1 (ES) Isl1 (sMN) Isl1 (cMN) Isl1 (ES)
25. motif instances Isl1 (sMN) Isl1 (cMN) Isl1 (ES) Isl1 (sMN) Isl1 (cMN) Isl1 (ES)
26. MultiGPS accurately detects differential binding events 0% 20% 40% 60% 80% 100% 1 2 3 4 5 % Differential Events Specificity Mean Read Count Quintile (1=lowest, 5=highest) MultiGPS (inter-expt prior, no motif prior) MACS & edgeR MACS & DBChIP SISSRs & edgeR 0% 20% 40% 60% 80% 100% 1 2 3 4 5 % Differential Events Sensitivity Mean Read Count Quintile (1=lowest, 5=highest) Sn. Sp. MultiGPS: Mahony, et al., PLoS Computational Biology, 2014
27. What if the concentration of signal changes globally? • Example scenarios: – TF that become more highly expressed during time-course. – Histone marks after (de)methylase knock-down. • Many normalization strategies assume that most loci don’t change. • Answer - use a constant spike-in. – Spike-in exogenic material, or – Spike-in antibody against something not changing
28. FINDING DNA-BINDING SUBTYPES
29. DIRECT TF BEFORE AFTER TAATTA TAATTA INDIRECT TF BEFORE AFTER CAGCTG CAGCTG XYZ XYZ vs. How do transcription factors recognize their binding sites in a given cell type?
30. ChIP-exo provides higher resolution protein-DNA binding information
31. CTCF ChIP-exo CTCF motif ChIP-exo has higher resolution than ChIP-seq CTCF ChIP-seq
32. 3’ 3’ exonuclease crosslink Exo denatured protein 3’ 3’ Exo Exo Exo Exo ChIP-exo captures protein-DNA crosslinking points CTCF ChIP-exo CTCF ChIP-seq
33. Motifs Reb1 (yeast) CTCF (human) p53 (human) -200 +200 -200 +200 Different protein-DNA binding events have different ChIP-exo patterns
34. Direct binding Indirect (tethered) binding A B DNA A Crosslinking Crosslinking patterns may distinguish direct and indirect protein-DNA interactions
35. TF B ChIP-exo A B TF A ChIP-exo DNA A Crosslinking Direct binding Indirect (tethered) binding Crosslinking patterns may distinguish direct and indirect protein-DNA interactions
36. A B TF A ChIP-exo DNA A TF B ChIP-exo Crosslinking Direct binding Indirect (tethered) binding Crosslinking patterns may distinguish direct and indirect protein-DNA interactions
37. Identify subtypes by clustering • Motifs • ChIP-exo tag distributions Initial tag distribution model Find peaks using mixture model Type II Type I Assign binding subtypes using hierarchical mixture II I II I ChExMix: the ChIP-exo mixture model N Yamada, WKM Lai, N Farrell, BF Pugh, S Mahony, Bioinformatics (2019) N Yamada, P Kuntala, BF Pugh, S Mahony, J Computational Biology (2020)
38. 1 : 2,831 A 2 : 9,579 3 : 3,749 1 2 C 4 : 3,009 5 : 1,411 6 : 5,345 7 : 1,999 3 4 5 6 7 Motif ERĮ ChIP-exo Subtype (sites) ERĮ tag distribution 100 0 -100 0 100 150 -100 0 100 -100 0 100 -100 0 100 10 100 0 0 0 150 200 -100 0 100 -100 0 100 0 0 0 -100 0 100 Distance from binding event (bp) 120 350 4 FoxA1 ChIP-exo FoxA1 tag distribution -100 0 100 Distance from binding event (bp) 250bp B D E Proposed model 4 Indirect binding 1-3 & 5-7 Direct binding FoxA1 ERĮ ERĮ ERĮ Fig 5 ChExMix identifies a subtype with a Forkhead motif in ERα MCF-7 ChIP-exo 3 4 5 6 7 0 -100 0 100 -100 0 100 -100 0 100 -100 0 100 10 100 0 0 0 150 200 -100 0 100 -100 0 100 0 0 0 -100 0 100 Distance from binding event (bp) 120 -100 0 100 Distance from binding event (bp) 250bp E Proposed model 4 Indirect binding 1-3 & 5-7 Direct binding FoxA1 ERĮ ERĮ ERĮ 1 2 3 4 5 6 7 100 0 -100 0 100 150 -100 0 100 -100 0 100 -100 0 100 10 100 0 0 0 150 200 -100 0 100 -100 0 100 0 0 0 -100 0 100 Distance from binding event (bp) 120 350 4 tag distribution -100 0 100 Distance from binding event (bp) E Proposed model 4 Indirect binding 1-3 & 5-7 Direct binding FoxA1 ERĮ ERĮ ERĮ ?
39. 1 : 2,831 A 2 : 9,579 3 : 3,749 1 2 C 4 : 3,009 5 : 1,411 6 : 5,345 7 : 1,999 3 4 5 6 7 Motif ERĮ ChIP-exo Subtype (sites) ERĮ tag distribution 100 0 -100 0 100 150 -100 0 100 -100 0 100 -100 0 100 10 100 0 0 0 150 200 -100 0 100 -100 0 100 0 0 0 -100 0 100 Distance from binding event (bp) 120 350 4 FoxA1 ChIP-exo FoxA1 tag distribution -100 0 100 Distance from binding event (bp) 250bp B D E Proposed model 4 Indirect binding 1-3 & 5-7 Direct binding FoxA1 ERĮ ERĮ ERĮ Fig 5 1 2 C 3 4 5 6 7 Motif ERĮ ChIP-exo e ERĮ tag distribution 100 0 -100 0 100 150 -100 0 100 -100 0 100 -100 0 100 10 100 0 0 0 150 200 -100 0 100 -100 0 100 0 0 0 -100 0 100 Distance from binding event (bp) 120 350 4 FoxA1 ChIP-exo FoxA1 tag distribution -100 0 100 Distance from binding event (bp) 250bp B D E Proposed model 4 Indirect binding 1-3 & 5-7 Direct binding FoxA1 ERĮ ERĮ ERĮ Fig 5 1 2 C 3 4 5 6 7 Motif ERĮ ChIP-exo ERĮ tag distribution 100 0 -100 0 100 150 -100 0 100 -100 0 100 -100 0 100 10 100 0 0 0 150 200 -100 0 100 -100 0 100 0 0 0 -100 0 100 Distance from binding event (bp) 120 350 4 FoxA1 ChIP-exo FoxA1 tag distribution -100 0 100 Distance from binding event (bp) 250bp B D E Proposed model 4 Indirect binding 1-3 & 5-7 Direct binding FoxA1 ERĮ ERĮ ERĮ Fig 5 3 4 5 6 7 -100 0 100 -100 0 100 10 100 0 0 150 200 -100 0 100 -100 0 100 0 0 0 -100 0 100 Distance from binding event (bp) 120 -100 0 100 Distance from binding event (bp) 250bp E Proposed model 4 Indirect binding 1-3 & 5-7 Direct binding FoxA1 ERĮ ERĮ ERĮ FoxA1 ChIP-exo ERα Subtype 4 -100 0 100 -100 0 100 0 0 35 0 10 0 FoxA1 binds to a subset of ERα sites
40. Motif 1 : 2,666 A FoxA1 ChIP-exo 2 : 2,648 3 : 24,749 1 2 3 FoxA1 tag distribution C 40 -100 0 100 25 0 0 0 500 Subtype (sites) ERĮ ChIP-exo CTCF ChIP-exo 250bp 250 -100 0 100 -100 0 100 E ERĮ tag distribution -100 0 100 1 0 CTCF tag distribution -100 0 100 40 Distance from binding event (bp) 2 Distance from binding event (bp) Proposed model ERĮ FoxA1 1 Indirect binding CTCF FoxA1 2 Indirect binding 3 Direct binding FoxA1 B D Fig 4 500 0 Distance from binding event (bp) -100 0 100 Subtype 1 2,666 sites Subtype 2 2,648 sites Subtype 3 24,749 sites FoxA1 direct binding 30bp 2 3 00 0 100 00 0 100 00 0 100 -100 0 100 0 CTCF tag distribution -100 0 100 40 Distance from binding event (bp) 2 Distance from binding event (bp) CTCF FoxA1 2 Indirect binding 3 Direct binding FoxA1 ChExMix discovers three subtypes in FoxA1 MCF-7 ChIP-exo 40 -100 0 100 0
41. 40 -250 0 250 Distance from binding event (bp) Nuclear hormone receptor motif -100 0 100 30bp 2,666 sites ERĮ FoxA1 ? FoxA1 Subtype 1 0 FoxA1 may be tethered to ERα in subtype 1
42. 40 0 -250 0 250 Distance from binding event (bp) -100 0 100 30bp -100 0 100 250 ERα ChIP-exo 2,666 sites -250 0 250 ERĮ FoxA1 0 Nuclear hormone receptor motif FoxA1 Subtype 1 FoxA1 may be tethered to ERα in subtype 1
43. ASSAYING 3-D CHROMATIN STRUCTURE
44. The genome is topologically organized in 3-dimensional space • 3-D structure of chromatin is somewhat consistent between cells of same type. • Chromatin split up into topological domains that are active, repressed, etc. • Little is known about how 3-D chromatin structure is established.
45. Processes leading to close spatial proximity of loci
46. Chromatin interactions
47. Chromatin interactions
48. Modeling genome organization 3-D genome “structures” chr1 chr1 Chromatin interactions
49. 3-D genome structure corresponds to regulatory activity levels Active Inactive Active states HMM states Histone modifications Ernst, J., & Kellis, M. (2012). ChromHMM: automating chromatin-state discovery and characterization. Nature Methods, 9(3), 215.
50. Further reading • Park PJ “ChIP-seq: advantages and disadvantages of a maturing technology”, Nature Reviews Genetics (2009) 10(10):669-680 • Mahony S & Pugh BF “Protein-DNA binding in high resolution”, Critical Reviews in Biochemistry and Molecular Biology (2015) 4:269-283 • Dekker, et al., “Exploring the three-dimensional organization of genomes: interpreting chromatin interaction data”, Nature Reviews Genetics (2013)