Overview Functional Genomics Dissections of Transcriptional Networks Rani

Overview: Functional Genomics Dissections of Transcriptional Networks Rani Elkon Ron Shamir, Yossi Shiloh

I. Reverse-Engineering of Transcriptional Networks

‘Reverse engineering’ of transcriptional networks n Infers regulatory mechanisms from gene expression data ¨ Assumption: co-expression → transcriptional co-regulation → common cis-regulatory promoter elements n n n Step 1: Identification of co-expressed genes using microarray technology (clustering algs) Step 2: Computational identification of cisregulatory elements that are over-represented in promoters of the co-expressed genes Such methodologies were first demonstrated in yeast

Reverse-engineering of the Yeast Cell-Cycle n n Expression profiles were recorded in synchronized yeast cells in 10 min intervals over 2 cell cycles. ~ 500 ORFs showed a periodic expression pattern

Reverse-engineering of the Yeast Cell-Cycle (II)

Reverse-engineering of the Human Cell-Cycle n n n Whitfield et al. recorded expression profiles during the progression of human cell cycle. 874 genes showed periodic expression patterns, and were partitioned into five clusters (G 1/S, S, G 2/M and M/G 1). We applied promoter analysis to these 5 clusters

p = 1. 2 x 10 -8 (true positive) 78 promoters (92 hits) p = 1. 2 x 10 -11 (152, 203) p = 8 x 10 -4 (20, 25)

Enriched TFs in cell cycle-regulated promoters

II. Cis-Regulatory Modules

Transcriptional Modules I: Co-occurrence n Transcriptional regulation is combinatorial Promoter #1 Promoter #n

• Defining transcriptional modules: • Co-occurrence • Positional bias (distance) • Orientational bias (order)

Positional and Orientational Bias for the RRPE-PAC Module

Modules II: Expression Coherence

III. Comparative Genomics

Conservation of Regulatory Elements Gene “DNA replication licensing factor MCM 6”: (G 1/S)

Human Cell Cycle Revisited n We detected global enrichments that pointed to major TFs in human cell cycle regulation. n However, we did not report on specific target genes due to high rate of false positive hits. n Comparative Genomics greatly boosts the specificity of in-silico detection of regulatory elements. n It now allows us to pinpoint TF targets with high confidence.

E 2 F Human-Mouse Conserved Hits 16, 299 human-mouse ortholog promoters (Ensembl) Total E 2 F hits % Enrichment Factor P-value Cell Cycle Promoters 697 Rest of Genome 15, 602 75 525 11% 3. 3 E-17

E 2 F Conserved Hits: Phase Distribution Num of targets 75 Enrich. Factor P-val x 3. 3 E-17 G 1/S 33 x 7. 3 E-18 S 15 x 3. 9 E-5 G 2+M 18 --- M/G 1 9 --- Cell Cycle – Total

CHR Regulatory Element n Cell-cycle Homology Region n To date, CHR was experimentally identified on 7 cell cycle-regulated promoters: • including CDC 2, CCNB 1, CCNB 2 and CDC 25 C (major regulators of G 2 -M)

Transcriptional Modules Promoter #1 CHR and NF-Y elements show significant cooccurrence rate (p<10 -11) Promoter #n

CHR-NF-Y Module 16, 299 Hs-Mm ortholog promoters; NFY-CHR putative targets: 71 Num of targets 42 Enrich. Factor P-val x 32 E-39 --- --- G 2+M 40 x 64 E-49 M/G 1 1 --- Cell Cycle – Total G 1/S S CHR-NFY: novel transcriptional module with 1 a pivotal---role in G 2 -M --regulation

CHR-NFY Module Dictates Expression that is Specific to G 2/M G 1/S

CHR-NFY Module – False Positive Rate Total CHR-NFY conserved hits Cell cycle promoters Rest of genome (negative set) 697 15, 602 42 29 False positives 0. 19%*697 = 1. 3 True positives ~40/42 = 95% 29/15602 = 0. 19% Comparative genomics yields highly specific identification of novel CHR-NFY cell-cycle targets

Regulation of Cyclin. BCDC 2 activity Novel CHR-NFY Targets in the G 2 -M Network Rho GTPases pathways Regulation of the mitotic spindle assembly Cytokinesis Regulation of the kinetochore apparatus

IV. Regulation of Gene Expression by Micro-RNAs

He and Hannon, 2004

§ Mature mi. RNA (~ 22 bp) tend to: n n Start with a “U” base Bind their target m. RNAs at sites of length 8 bp. Target site is complementary to positions 1 -8 of the mature mi. RNA. Assumed to play major regulatory function during development (many show tissuespecific expression pattern)

n Transfected two mi. RNAs into Hela human cells and examined changes in m. RNA expression profiles: mi. R-1: expressed in skeletal muscle ¨ mi. R-124: expressed in brain ¨ n n 96 and 174 genes were significantly down-regulated by mi. R-1 and mi. R-124, respectively Comparison with human tissue expression atlas: Genes down-regulated by mi. R-1 are expressed at lower levels in skeletal muscle and heart than in other tissues ¨ Genes down-regulated by mi. R-124 are expressed at lower levels in the brain than in other tissues ¨ n Searching for enriched signals in the 3’-UTRs of the down-regulated genes discovered the cognate binding sites

n n n Computational identification of putative mi. RNA targets – scan 3’-UTRs for putative target sites Anti-Correlation between the expression pattern of mi. Rs and their putative targets (using the human tissue gene expression atlas) Genes expressed at the same time and place as a mi. RNA evolved to avoid sites matching the mi. RNA

n n Comparative analysis of promoter and 3’-UTR regulatory motifs using the human, mouse, rat and dog genomes. Search for highly conserved motifs (degenerate strings, 6 -18 bases) ¨ Motif Conservation Score (MCS): Z score of the proportion of the conserved occurrences of a motif relative to the conservation rate of comparable random motifs. n Promoters (-2 kb to + 2 kb relative TSS): ¨ 174 highly conserved motifs (MCS > 6): n 69 – known (out of 123 TRANSFAC motifs, 56%) n 105 potentially novel regulatory elements

n Demonstrating biological function for the discovered motifs: ¨ Correlate the occurrence of a motif with tissue -specific gene expression (using data from the human tissue expression atlas) Target sets of 86% (59 out of 69) of the known motifs showed significant tissue-specific expression n 53 out of the 105 (50%) novel motifs n ¨ Examine positional bias of the motif hits

n 3’ UTR signals: ¨ 106 highly conserved motifs (MCS > 6) ¨ Hypothesis: function as binding sites for mi. RNAs ¨ Many of the discovered motifs show features of mi. RNA binding sites: n Strong strand bias of the conservation rate ¨ n n n consistent with a role in post-transcriptional regulation, acting at the RNA rather than DNA level Biased length distribution: strong peak at 8 bp High rate of “A” in position #8 Search for matches of the 8 -mer motifs to the known human mi. Rs: ¨ in 95% of the cases the matches begins at position 1 or 2 of the mature mi. RNAs.

V. Integrative Analysis

n Systems-level analysis of the DNA damage response in yeast by an integrated approach that combines: Genome-wide profiling of TF-promoter binding (Ch. IP-chip data) 2. Expression profiling (in deleted and w. t. strains) 3. Phenotyping sensitivity to DNA damage in deleted strains 4. Wide scale protein-protein interaction data 1.

1. Systematic screen for TFs involved in the DNA damage response: Ø 30 (out of 141) TFs based on either: n n n 2. Expression: differentially expressed after DNA damage Binding: bind promoters of genes induced by DNA damage Sensitivity: TF-mutant strain is hyper-sensitive to DNA-damaging agent TF-promoter binding profiling (Chipchip) for each of these 30 TFs, without and after exposure to DNA damaging agent

3. Validation of functional roles of the measured TF-promoter binding interactions: Ø Ø Gene expression profiling in w. t. and deleted strains (27 out of the 30 are non-essential) without and after exposure to DNA damaging agent “Deletion Buffering”: genes that respond to the damage in w. t. but become unresponsive in a specific TF-deleted strain Ø Ø Only 11% (37 out of 341) of the observed deletion-buffering events could be explained by direct TF-promoter interaction The rest are probably mediated by longer, indirect, regulatory pathways linking the deleted TF and the buffered gene

4. Physical pathways that explain indirect deletion-buffering events were searched for using Bayesian modeling procedure Ø Utilized various data sources: n n n ¨ TF-promoter binding data measured in this study Tf-promoter binding data measured for all yeast TFs (in nominal conditions) 14 K high-throughput protein-protein interactions (in nominal conditions) The inferred network explains a total of 82 deletion-buffering events.

and in human cells? Small-interfering RNA (si. RNA)

He and Hannon, 2004

RNA Interference (RNAi( Ø A major technological breakthrough in biomedical research Ø Allows rapid establishment of mammalian cell lines which are stably knocked-down for any gene of interest – pivotal tool in functional genomics Ø Efforts to establish cell lines in which specific genes are silenced, eventually spanning most of the genome

Ø The combination of RNAi and microarrays holds promise as a powerful tool for a systematic, genome-wide, dissection of transcriptional networks in human cells

Experiment Goal Ø Proof of principle that RNAi+microarrays can "deliver" Ø Focus on transcriptional network induced by DNA damage as a test case

Transcriptional network induced by DNA double strand breaks DNA Double Strand Breaks AP-1 ATM CREB g 13 g 12 g 11 g 10 E 2 F 1 NF-k. B g 9 g 8 g 7 g 6 g 5 p 53 g 4 g 3 g 2 g 1

Heatmap colors: Red – above average induction Black – average induction Green – below average induction 26 genes whose activation is: • Strongly reduced in the absence of ATM and Rel-A • Partially reduced in the absence of p 53 Ø ATM-NFκB-dependent cluster, partial role for p 53

46 genes whose activation is: • Strongly attenuated in the absence of ATM and p 53 • Not affected by the absence of Rel-A Ø ATM-p 53 -dependent cluster

Response of known NF-κB targets Gene C Lac. Z NF-κB p 53 ATM NFKBIA 4. 61 5. 4 1. 26 2. 67 1. 02 RELB 3. 7 2. 89 0. 82 2. 95 0. 91 TNFAIP 3 8. 26 5. 34 1. 15 3. 02 1. 18 TNFRSF 9 4. 01 3. 5 1. 1 2. 07 1. 21 CD 83 3. 45 2. 98 1 1. 73 1. 06 IER 3 4. 43 5. 13 1. 43 2. 35 1. 44 • Knocking down Rel-A subunit of NF-κB abolished the induction of known NF-κB targets • ATM is required for the activation of the NF-κB mediated transcriptional response • p 53 plays a positive role in the activation of NF-κB targets (? )

Response of known p 53 targets Gene C Lac. Z NF-κB p 53 ATM GADD 45 A 2. 35 2. 07 2 1. 08 1. 21 ATF 3 3. 43 3. 75 7. 05 1. 54 1. 47 FOS 1. 71 1. 42 2. 22 1. 06 1. 21 JUND 1. 67 1. 98 2. 67 1. 3 1. 02 JUN 2. 01 1. 45 2. 71 1. 36 1. 26 • Knocking down p 53 attenuated the induction of its known targets • ATM is required for the activation of p 53 targets • NF-κB plays an inhibitory role in the induction of some components in the p 53 pathway (? )

PRIMA - Results

p 53 response is mediated by the activation of the ATF pathway Gene C Lac. Z NF-κB p 53 ATM GADD 45 A 2. 35 2. 07 2 • Its partners are ATF 3 and Jun 1. 08 1. 21 ATF 3 3. 43 3. 75 7. 05 1. 54 1. 47 FOS 1. 71 1. 42 2. 22 1. 06 1. 21 JUND* 1. 67 1. 98 2. 67 1. 3 1. 02 JUN* 2. 01 1. 45 2. 71 1. 36 1. 26 • ATF 2 is an hetrodimeric transcription factor

Conclusions ATM NF-k. B p 53 Others Jun ATF 3 ATF 2

VI. Gene Expression and Cancer Treatment

Some patient are responsive and other are resistant to certain chemotherapy modalities (although suffering from the same cancer) n Probably due to different deregulated oncogenic pathways underlying the cancer n Goal: Personalized/targeted therapies n

n n Various oncogenes (e. g. , Ras, Myc, E 2 F) were expressed in otherwise quiescent cells Expression signatures characteristic of each oncogenic pathway were identified Tested on various mouse cancer models: the oncogenic-pathway signatures successfully predicted the deregulated pathways Prediction of oncogene-pathway deregulation leads to prediction of cancer sensitivity to therapeutic agents

Ø Ø Ionizing radiation (IR) is the most common cancer treatment IR causes cell death mainly via the induction of DNA double strand breaks (DSBs) The ATM protein is the master regulator of the cellular responses to DSBs. We examined expression profiles in response to IR in w. t. and Atm-/- mice (lymph node).

Major. Atm-dependent Gene Clusters –early Irradiated Lymph node responding genes

Major Gene Clusters 2–nd. Irradiated Lymph node Atm-dependent wave of responding genes

PRIMA - Results

PRIMA - Results Transcription factor Enrichment factor P-value NF- B 5. 1 3. 8 x 10 -8 p 53 4. 2 9. 6 x 10 -7 STAT-1 3. 2 5. 4 x 10 -6 Sp-1 1. 7 6. 5 x 10 -4

Biological endpoints of p 53 - and NF- Bmediated arms Red – induction Yellow – No change Gray – N. A. • In lymphoid cells proand anti-apoptotic pathways are activated in parallel in an Atmdependent manner • The pro-apoptotic arm is mediated by p 53 • The pro-survival arm is mediated by NF-κB

Responses to IR in B-CLL Ø Stankovic et al. (Blood, 2004) used microarrays to profile IR responses in: q ATM-mutant B-CLLs q p 53 -mutant B-CLLs q ATM+/+, p 53+/+ B-CLLs

B-CLLs Response to IR (Stankovic et al. Blood, 2004) ATM (79 genes) (61 genes) p 53 Pro-apoptotic signals NF-κB ? Pro-survival signals • Blocking NF-κB arm → increase radiosensitivity of lymphoid tumors

Acknowledgements Ø Ø Sharon Rashi-Elkeles Yaniv Lerenthal Tamar Tenne Yossi Shiloh Ø Ø Ø Ø Chaim Linhart Roded Sharan Ron Shamir Ø Ø Rita Vesterman Nira Amit Giora Sternberg Ran Blechman Jackie Assa Nir Weisman Ari Barzilai Ninette Amariglio Gidi Rechavi