Conditional Epistatic Interaction Maps Reveal Global Functional ...

Resource

Conditional Epistatic Interaction Maps Reveal Global Functional Rewiring of Genome Integrity Pathways in Escherichia coli Graphical Abstract

Authors Ashwani Kumar, Natalia Beloglazova, Cedoljub Bundalovic-Torma, ..., Andrew Emili, Alexander F. Yakunin, Mohan Babu

Correspondence [email protected]

In Brief Genes functioning in bacterial DNA damage have been studied extensively in isolation, yet the interaction of underlying pathways remains unclear. Through epistatic screening under standard and DNA-damaging conditions, Kumar et al. show that the E. coli genome integrity apparatus undergoes extensive reorganization upon genomic stress and has high pan-bacterial conservation.

Highlights d

Genetic screens detect rewiring of E. coli genome integrity networks to DNA damage

d

YhbQ and YqgF are nucleases in the DNA-damage response (DDR) pathway

d

Pan-bacterial conservation suggests that DDR functional relationships are near universal

d

Epistatic interactions reveal divergent functions among bacterial gene duplicates

Kumar et al., 2016, Cell Reports 14, 648–661 January 26, 2016 ª2016 The Authors http://dx.doi.org/10.1016/j.celrep.2015.12.060

Cell Reports

Resource Conditional Epistatic Interaction Maps Reveal Global Functional Rewiring of Genome Integrity Pathways in Escherichia coli Ashwani Kumar,1,13 Natalia Beloglazova,2,13 Cedoljub Bundalovic-Torma,3,4,13 Sadhna Phanse,5,6 Viktor Deineko,6 Alla Gagarinova,5,7 Gabriel Musso,8 James Vlasblom,6 Sofia Lemak,2 Mohsen Hooshyar,9 Zoran Minic,6 Omar Wagih,5 Roberto Mosca,10 Patrick Aloy,10,11 Ashkan Golshani,9 John Parkinson,3,4,12,14 Andrew Emili,5,12,14 Alexander F. Yakunin,2,14 and Mohan Babu6,14,* 1Department

of Computer Science, University of Regina, Regina, SK S4S 0A2, Canada of Chemical Engineering and Applied Chemistry, University of Toronto, Toronto, ON M5S 3E5, Canada 3Hospital for Sick Children, 686 Bay Street, Toronto, ON M5G OX4, Canada 4Department of Biochemistry, University of Toronto, Toronto, ON M5S 1A8, Canada 5Terrence Donnelly Centre, University of Toronto, Toronto, ON M5S 3E1, Canada 6Department of Biochemistry, University of Regina, Regina, SK S4S 0A2, Canada 7Department of Biochemistry, University of Saskatchewan, Saskatoon, SK S7N 5E5, Canada 8Department of Medicine, Harvard Medical School and Cardiovascular Division, Brigham and Women’s Hospital, Boston, MA 02115, USA 9Department of Biology, Carleton University, Ottawa, ON K1S 5B6, Canada 10Joint IRB-BSC-CRG Program in Computational Biology, Institute for Research in Biomedicine (IRB Barcelona), The Barcelona Institute of Science and Technology, c/Baldiri i Reixac 10-12, Barcelona, 08028, Catalonia, Spain 11Institucio ´ Catalana de Recerca i Estudis Avanc¸ats (ICREA), Pg. Lluı´s Companys 23, Barcelona, 08010, Catalonia, Spain 12Department of Molecular Genetics, University of Toronto, Toronto, ON M5S 1A8, Canada 13Co-first author 14Co-senior author *Correspondence: [email protected] http://dx.doi.org/10.1016/j.celrep.2015.12.060 This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). 2Department

SUMMARY

INTRODUCTION

As antibiotic resistance is increasingly becoming a public health concern, an improved understanding of the bacterial DNA damage response (DDR), which is commonly targeted by antibiotics, could be of tremendous therapeutic value. Although the genetic components of the bacterial DDR have been studied extensively in isolation, how the underlying biological pathways interact functionally remains unclear. Here, we address this by performing systematic, unbiased, quantitative synthetic genetic interaction (GI) screens and uncover widespread changes in the GI network of the entire genomic integrity apparatus of Escherichia coli under standard and DNA-damaging growth conditions. The GI patterns of untreated cultures implicated two previously uncharacterized proteins (YhbQ and YqgF) as nucleases, whereas reorganization of the GI network after DNA damage revealed DDR roles for both annotated and uncharacterized genes. Analyses of pan-bacterial conservation patterns suggest that DDR mechanisms and functional relationships are near universal, highlighting a modular and highly adaptive genomic stress response.

Like all organisms, prokaryotic cells are constantly assaulted with DNA-damaging stimuli, such as UV light, ionizing radiation, and other genotoxic adducts. Bacteria cope with these threats by efficiently repairing DNA lesions by multicomponent DNA integrity pathways such as the SOS response, recombinational repair, bypass replication, and chromosomal reorganization (Schlacher and Goodman, 2007). As a result, maintenance of genome stability and cellular integrity depends on mechanistic coordination for efficient DNA repair, replication, and recombination (DRRR). These pathways are further tightly linked to other pathways ensuring bacterial cell growth and proliferation (Mirkin and Mirkin, 2007). Because of its central importance, the DNA damage response (DDR) is a common target for antibiotics (Dwyer et al., 2007, 2014; Foti et al., 2012; Kohanski et al., 2007). These include ciprofloxacin and gemifloxacin, which prevent strand rejoining and alter chromosome topology by targeting DNA gyrase and topoisomerase IV (Kohanski et al., 2010). However, acquisition of chromosomal mutations conferring antibiotic resistance is an increasing clinical threat (Kohanski et al., 2010), driving demand for innovative drug treatments. An improved understanding of functional dependencies supporting genome integrity could reveal targets for combinatorial drug strategies. Genome integrity in prokaryotes has been primarily studied in Escherichia coli, leading to detailed mechanistic investigations of individual genes and biochemical pathways involved

648 Cell Reports 14, 648–661, January 26, 2016 ª2016 The Authors

in recombinational and DNA double-strand break (DSB) repair, DNA replication, transcription-coupled repair, homology-directed strand invasion, and activation of the SOS DDR (Lesterlin et al., 2014; Schlacher and Goodman, 2007; Yeeles et al., 2013). However, it remains unclear how these pathways are functionally coordinated and how these relationships are altered in response to genotoxic insult (Uphoff and Kapanidis, 2014). Strikingly, while systematic genomic phenotyping of single mutants in E. coli has identified 549 genes involved in resistance to DNA damage (Nichols et al., 2011), one quarter (28%; 151) still lacks contextual information regarding their relevant pathway or role. Epistatic relationships (herein genetic interactions [GIs]), which indicate functional synergy between pairs of genes, can provide a pathway-scale understanding of gene function (Hartwell et al., 1999). While large-scale GI surveys in yeast have revealed extensive functional redundancy and dynamic alteration of epistatic networks in response to genotoxic insults (Bandyopadhyay et al., 2010; Gue´nole´ et al., 2013; Srivas et al., 2013), no analogous GI map of DDR in bacteria has been reported. Small-scale studies in E. coli (Babu et al., 2009b) have shown GIs among the small numbers of tested genes participating in DRRR, indicating that these systems cannot compensate for loss of certain combinations of pathways when subjected to genotoxic stress, but the extent of these dependencies is unknown. Here, we used our E. coli synthetic genetic array (eSGA) screening technology (Butland et al., 2008) to interrogate epistatic relationships among 549 genes encompassing virtually the entire known and predicted genome maintenance, stability, and integrity machineries of E. coli in both standard laboratory growth conditions and following genotoxic stress. Unbiased quantitative scoring of the resulting high-density GI networks revealed condition-specific functional rewiring within and between genome integrity processes that impact cell fitness, growth, and morphology, implicating many components and relationships not previously linked in DRRR, which were independently verified. Additionally, we show that genetically interacting genes are broadly co-conserved, defining a modular, pan-bacterial global DNA integrity paradigm. RESULTS Generating Genomic Integrity Network Maps To genetically interrogate the functional connections ensuring genome integrity in E. coli, a comprehensive survey of literature and public databases was conducted to derive a candidate set of 398 annotated genes and 151 genes predicted (Hu et al., 2009; Nichols et al., 2011) to be involved in DRRR and DNA damage or stress response signaling (Figure 1; Table S1, sheet 1). Given that genomic integrity depends on DNA metabolism, transcription, cell division, and protein degradation (Mirkin and Mirkin, 2007), we also included representative genes from these processes. The 549 targets consisted of individual deletions of 470 non-essential protein coding genes and 8 small non-coding regulatory RNAs (sRNAs) that post-transcriptionally regulate DNA repair, and 71 essential gene hypomorphs (i.e., partial loss of function alleles).

Using our eSGA screening procedure based on conjugation and homologous recombination (Butland et al., 2008), we systematically generated pairwise mutant gene combinations among all targets. In total, 458 query gene mutations, marked with a chloramphenicol (Cm)-resistance cassette, were constructed and transferred from individual E. coli K-12 ‘‘donor’’ strains into a complementary collection of 526 ‘‘recipient’’ strains bearing a non-essential gene deletion (Baba et al., 2006) or hypomorphic allele (Babu et al., 2011b; Nichols et al., 2011) marked with a kanamycin (Kan)-resistance cassette (Supplemental Experimental Procedures). After conjugation,
240,000 double mutants were selected on nutrient-rich medium (untreated, or UT) containing both antibiotics (Kan+Cm); viable double mutants were then replica pinned onto rich media supplemented with the DNA alkylating agent methyl methanesulfonate (MMS; Figure 1) to identify additional GI under genotoxic stress. After outgrowth at 32 C, the plates were digitally imaged. Colony size measurements were normalized to account for experimental variation in each condition and were used as a proxy for strain growth rates (fitness) for all viable digenic mutant combinations. After eliminating closely linked gene pairs exhibiting low recombination frequency (Figure S1A), we scored GIs using a multiplicative model (Butland et al., 2008), computing both a ‘‘static’’ epistasis (S) score (SUT, SMMS) reflecting the extent to which a particular double mutant grew better (alleviating/ suppression; positive S score) or worse (aggravating/synthetic sick or lethal; negative S score) relative to corresponding single mutants. Comparable to yeast genetic maps (Gue´nole´ et al., 2013), the average correlation of GI scores of strain replicate in the static map was high (r = 0.7; Figure S1B), suggesting a good overall reproducibility of observed interactions. Using an established enrichment method (Babu et al., 2011b), we chose statistically significant (p % 0.006; false-discovery rate [FDR] z5%) S-score thresholds (S % 2.5, S R 2.5; Figure S1C; Supplemental Experimental Procedures) to define aggravating and alleviating GI, which grossly indicate gene function in parallel or linear pathways, respectively. These two markedly different filtered static GI networks of either 23,648 (UT) or 28,885 (MMS) digenic interactions, with more GIs unique to each condition than shared in common (Figure S1D; Table S1, sheet 2). For instance,
79% of aggravating GIs observed in MMS were not detected in the UT network (p < 2.2 3 1016), thus revealing DNA damage-induced functional dependencies (Figure S1E). Topological analyses of the resulting interaction network showed essential genes with distinct patterns of connectivity compared with non-essential genes (Figures S1F and S1G), while quality-control measures (Supplemental Experimental Procedures) confirmed data quality (Figures S2A–S2C; Table S1, sheets 3 and 4). Static UT Network Reveals Functional Connections among Genome Integrity Bioprocesses To link diverse core genomic integrity processes, we examined the global functional connectivity (i.e., aggravating or alleviating GIs) between processes in the UT static network (Figure 2A; Supplemental Experimental Procedures). Nearly 13% of possible genomic integrity process combinations (737 of 5,665 process pairs tested) were significantly enriched (p % 0.008; FDR z5%) Cell Reports 14, 648–661, January 26, 2016 ª2016 The Authors 649

526 F- ‘recipient’ mutants

X

Target inclusion (549) Tr an Tr scr an ipt D spo ion N A rte (21 R rec r (2 ) N A- om 1) Fo rela bin t a ld in ed tion D g o pro (2 N r d ce 7) A m egr ss et ( D ab ada 34) N A t re olis ion m pl ( D i (3 23) am ca ag tion 6) e ( 39 D or N ) st A r re pa ess C ir (5 el ( 6) 43 ld ) iv is i o O n th (4 er 6) s (5 2) U nc le ar (1 51 )

458 Hfr ‘donor’ mutants

+

i Δ :: Cm

j Δ:: Kan Double mutant (DM) selection on LB (Kan+Cm)

+Methymethane sulfonate (MMS) DMs on LB rich medium (Untreated; UT)

DMs on MMS

Viable DMs replica pinned onto MMS

+

Quantify colony sizes and assign GI scores

MMS recA

recA

UT

recB

dinI

recD

dinI

recD

F- ‘recipient’ gene mutants (526) recC

recX

recX

recC

Differential (DF) network

Subtract scores, threshold, and define significance recA

DF recB dinI

recC

recD

Alleviating/suppression/ positive epistasis (Δi Δj) > (Δi) (Δj)

recX

Aggravating

deaD

dbpA

rhlB srmB

Static UT

rhlE

dbpA

rhlE

deaD

rhlB

Static MMS

DF aggravating DF alleviating

deaD

dbpA

srmB

Alleviating

DF

Functional divergence of paralogous genes in static and DF networks

rhlE

Aggravating / synthetic sick lethal / negative epistasis (Δi Δj) < (Δi) (Δj)

UT

recB

Quantify colony sizes and assign GI scores

MMS

Static network

Hfr ‘donor’ gene mutants (458)

Rich medium (UT)

rhlB srmB

DF

Figure 1. Schematic of the Static and DF GI Networks Genome integrity genes assigned into 12 representative bioprocesses (see Table S1 for details), shown in terms of the number of F recipient and Hfr donor mutants used for constructing double mutants. The DF map was generated by identifying differential growth effects between nutrient-rich (Luria-Bertani, LB) and DNA-damaging (MMS-induced) conditions and was employed to assess condition-specific epistasis (see also Figures S1 and S2), notably among duplicated genes. Edge (line) thickness indicates the S-score magnitude for each GI.

for high-confidence GIs, with half (
48%; 351 of 737) exhibiting predominantly aggravating and the other half (
52%; 386 of 737) alleviating relationships (Table S1, sheet 5). The map recapitulates known functional dependencies (e.g., DNA mismatch repair with recombination; Junop et al., 2003), but unexpected relationships were also evident (Figure 2A). For example, transporters were enriched (p = 1. 8 3 106; Table S1, sheet 5) for alleviating GIs with DNA repair (or recombination) factors, which may reflect the close coupling of ATP metabolism to DNA repair. Similarly, an enrichment (p = 2.2 3 103) for aggravating interactions was seen in the UT static network between the cell division and DNA repair machineries (Figure 2A). These dependencies likely result from impaired cell division and DSB repair and recombination (Zahradka et al., 1999). Additionally, enrichment (p = 1.3 3 103) for aggravating interactions was observed 650 Cell Reports 14, 648–661, January 26, 2016 ª2016 The Authors

between DNA metabolism and recombination genes (Figure 2A), consistent with impaired metabolic precursors causing DNA damage (Kanaar et al., 2008). These interprocess connections were independently verified using chemical-genetic profiles (Nichols et al., 2011). Specifically, E. coli single-gene mutants deficient in DNA repair, recombination and cell division, or DNA recombination and metabolism (Figure 2B; Table S1, sheet 6) were hypersensitive to antibiotics, such as trimethoprim (TMT), puromycin (PUR), and clarithromycin (CLART), that inhibit bacterial replication and protein synthesis. Moreover, all three drugs had correlated (r = 0.2–0.4) single-gene deletion mutant sensitivity profiles, underlining that functional coupling of these processes ensures genome integrity and the likely efficacy of combination drug therapies. Epistatic interactions were also observed between several unannotated genes and the DNA mismatch repair and

N

O O

Cell division

Metabolism

.3 r=0

DNA repair

DNA recomb

N

DNA recomb.

N

∆helD ∆radA ∆rdgC ∆recA ∆recB ∆rep ∆rusA ∆ruvA ∆ruvB ∆ruvC ∆sbcD

r=0 .2

∆glvG ∆mrp ∆nudG ∆rhlB ∆smtA ∆tdk ∆yfaE ∆yffH ∆yggF ∆yggV ∆yjbQ ∆ymfB ∆yrdD

Drug-drug correlation (Nichols et al) Enriched bioprocesses (FDR ≈ 5%; this study) p ≤ 0.05 Enrichment between drug and bioprocess (this study) p ≤ 0.1

O

N

∆radA ∆rdgC ∆recA ∆recB ∆recT ∆rusA ∆ruvA ∆ruvB ∆sbcC ∆sbcD ∆ycaJ

Trimethoprim (TMT)

CLART

B

Unclear ∆stfR ∆yaeQ ∆ybfP ∆ybgJ ∆ybjD ∆ydaE ∆ydaT ∆ydaU ∆ydcK ∆ydcS ∆yebS ∆yeeS ∆yfaA ∆yfcH ∆yggC ∆yhbQ ∆yhdE ∆yhfS ∆yhfX ∆yhhZ ∆yicR ∆ymgE ∆ynjF ∆yqhH

DNA metabolism

N

DNA recomb.

A

O

O O

O

O

N

O

O

N

O

O

O

O

N

r = 0.4

O

O

N

N

N

Clarithromycin (CLART)

N

Puromycin (PUR) PUR

Unclear Damage/ stress

Enriched bioprocess pairs (P ≤ 0.008; FDR ≈ 5%)

DNA replication

alleviating neutral

∆ybbM

∆aqpZ ∆emr ∆modF ∆tatA ∆tatB ∆tatC ∆ybbL

∆recF ∆recO ∆recR

D

Predicted target

P ≥ 0.01 (FDR > 5%)

*mukB ∆yagA

Known target

5 4 3

**

ΔisrA *ssb

Control (- drug) ΔisrA

**

**

rng

ssb

L tfa Q

sRNA

**

gro

Alleviating (Static UT) P ≤ 0.006 (FDR < 5%)

gE

∆dicF

ob

*ftsZ

∆tfaQ

rG

∆yagS

mu

∆rybA

e)

∆rng

** **

2

*ssb

+v

*holB

∆zapA

**

WT

1

Validation (qRT-PCR and morphology)

E

ΔmicF

0

Transcript level relative to WT

∆isrA

pA (

DNA repair (or recomb.)

∆abgT

Transporters

za

∆nohB

-0.4-0.2 0.0 0.2 0.4 0.6 0.8 1.0 TMT

CLART

∆dcm ∆exoX ∆mutL ∆mutS ∆mutY ∆nfi ∆vsr ∆ybcN ∆micF

TMT

CLART 1.0 r(PUR,CLART) = 0.4 0.8 0.6 0.4 PUR 0.2 0.0 - 0.2 - 0.4 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0 PUR

≥2.5

aggravating

∆ninE ∆yaiV ∆ydbA ∆ydeK ∆ydfD ∆yfaA ∆yiaT

∆glmY

≤-2.5 -1 -0.5 0 0.5 1

*obgE

*murG

r(TMP, PUR) = 0.3

S-score Folding or degradation

Cell division

Mismatch repair

∆groL

PUR

1.0 0.8 0.6 0.4 0.2 0.0 -0.2 -0.4

P < 10-39 to P < 10-3

Alleviating Aggravating

DNA repair

RNA related

Transcription

Unclear

C

O

O

DNA metabolism

Transporter

∆dam ∆dcm *ligA ∆mfd ∆mutM ∆mutS ∆nei ∆polB ∆recG ∆recO ∆recR ∆sbcB ∆ung ∆uvrY ∆vsr ∆xthA ∆ygjF

O

O

∆cedA ∆damX ∆mraZ ∆nlpL ∆sdiA ∆yejH ∆yqjA DNA repair

Others

O

DNA recombination

Cell division

CLART 1.0 r(TMP,CLART) = 0.2 0.8 0.6 0.4 0.2 TMT 0.0 -0.2 -0.4 -0.4-0.2 0.0 0.2 0.4 0.6 0.8 1.0 TMT

MMC (2 μg/ml)

Predicted target mRNAs

Figure 2. Functional Dependencies between Genome Integrity Genes and Bioprocesses (A) Global view of the aggravating and alleviating GIs between different genome integrity processes (representative bioprocess crosstalk shown as a heat map) in the UT growth condition. Node size represents the number of genes in each process, while edge thickness is inversely proportional to the p value of the GI as assessed using the hypergeometric p value distribution (following a Benjamini-Hochberg multiple testing correction). For clarity, functional crosstalk is shown for major processes (Table S1, sheet 5 for all enriched bioprocess pairs; *hypomorphs). (B) Hypergeometric enrichment of drug and epistatic links (black lines) among cell division, metabolism, DNA repair, and recombination. Scatter plots show the correlation between drugs (brown lines in network) based on E. coli single gene mutant fitness (Nichols et al., 2011). (C) Alleviating GIs between known or predicted mRNA targets and their cognate sRNA regulators in the UT network. (D) Quantitative real-time (qRT) PCR analysis of candidate transcript levels in the indicated mutants. Values indicate mean ± SD (error bars; **p < 0.01 by Fisher’s exact test) of biological triplicates normalized to glyceraldehyde phosphate dehydrogenase. (E) Representative micrographs of the wild-type (WT) and indicated double mutants before and after DNA damage (MMC; T = 2 hr). See Figure S2D for full micrographs. Scale bar represents 10 mm.

recombination systems (p = 1.3 3 1012 and p = 2.7 3 104, respectively; Figure 2A), implicating roles for unknown genes. For example, ybjD showed aggravating GIs with DNA helicases acting in recombination (recB, ruvB). All three encode a nucleoside triphosphate hydrolase motif essential for DNA binding and recombination (Calloni et al., 2012), suggesting coordination of biochemical activities underlies genome maintenance. The Static GI Network Reveals Regulatory Relationships Antisense transcripts that base pair with mRNA via short, imperfect complementarity have been implicated in the posttranscrip-

tional regulation of a wide range of cellular processes, yet targets linked to genome integrity are not fully known (Modi et al., 2011). To address this, we performed eSGA screens using eight donor strains deleted for an sRNA (dicF, glmY, isrA, micF, oxyS, rybA, ryeE, or ryhB), ranging from 53 to 340 nt in length, that are encoded in intergenic regions of the E. coli chromosome, to define potential GI with known and predicted targets in the UT network (Table S1, sheet 7). Consistent with expectations, known and predicted mRNA targets showed predominantly alleviating interactions with their cognate sRNA regulators (Figure 2C; Table S1, sheet 7). For Cell Reports 14, 648–661, January 26, 2016 ª2016 The Authors 651

example, micF had an alleviating interaction with its cell division factor zapA (Modi et al., 2011) and other targets (Figure 2C), including rng, an endoribonuclease RNase G involved in RNA maturation and decay (Deana and Belasco, 2004). Similarly, isrA displayed alleviating interactions with ssb, which encodes a single-stranded DNA binding protein involved in DRRR (Shereda et al., 2008). These regulatory relationships were confirmed by quantitative PCR (Figure 2D), which showed a significant (1.5to 4-fold) increase in candidate transcript levels in isrA or micF mutants as compared with WT cells grown in rich medium, comparable to that previously reported for zapA (Modi et al., 2011). Given that SOS-induced filamentous growth to gyrase inhibitors is indicative of DNA damage (Modi et al., 2011), we challenged strains lacking isrA and its predicted target ssb with norfloxacin and mitomycin C (MMC; Figures 2E and S2D). Whereas isrA and ssb single mutants grew similar to WT cells, filamentation was markedly increased when both isrA and ssb were jointly deleted, suggesting that they post-transcriptionally inhibit cell division following DNA damage, analogous to SOS genes (Campoy et al., 2005). Hence, the inferred GI network captured known sRNA targets and predicted candidates in other bioprocesses, providing a basis for mechanistic exploration of sRNA-mediated gene regulation. YhbQ and YqgF Function as Nucleases in DRRR Global GI patterns from eSGA can define the biological role of uncharacterized cellular components (Babu et al., 2014). We therefore examined subnetworks in the UT genetic map to reveal the function of unannotated genes. We noted that deletion of yhbQ and a hypomorph of yqgF displayed aggravating and alleviating GIs, respectively, with DRRR components (Figure 3A), which we confirmed using liquid culture growth assays (Figure S3A; representative yhbQ-uvrD and yqgF-polA double mutant shown). YhbQ contains a GIY-YIG endonuclease domain involved in repair-recombination in both eukaryotes and bacteria (Aravind and Koonin, 2001; Dunin-Horkawicz et al., 2006), whereas YqgF has an RNase H-like domain, found in a large nuclease superfamily implicated in DRRR, transposition, and RNAi (Aravind et al., 2000; Majorek et al., 2014). To explore the role of YhbQ and YqgF in DNA repair, we examined purified recombinant proteins for metaldependent nuclease activity against ssDNA and found 30 to 50 exonuclease activity in both proteins (Figures 3B and S3B). In contrast to YqgF, YhbQ showed metal-dependent (Mg2+, Mn2+, Co2+; data not shown) endonuclease activity against both circular ssDNA (M13mp18 phage) and linear dsDNA (phage l), but not against circular dsDNA (pUC19) (Figures S3C–S3J) or RNA (Figures S3K and S3L), whereas YqgF effectively cleaved both ssRNA and RNA/DNA hybrids (Figures S3M and S3N). Next, we analyzed nuclease activity against model DNA substrates representing DNA repair intermediates. YhbQ showed high activity against several substrates related to nucleotide excision repair (NER), including dsDNA substrates containing a single nt mismatch or 1 to 20 nt ssDNA loops (Figures 3C and 3D), as well as against DNA recombination intermediates, including flap and splayed arm substrates (Figures 3E and 3F). YqgF bound strongly to ssRNA, ssDNA, and DNA recombination intermediates, as well as cleaved Holliday junction (HJ), replication fork (RF), and flap substrates (Figures 3G and S3O–S3R). 652 Cell Reports 14, 648–661, January 26, 2016 ª2016 The Authors

Next, we examined the unpublished crystal structure of YqgF (PDB: 1NMN, 1NU0) and located a potential active site at the end of a central b sheet encompassing conserved Asp9, Glu96, and Asp122 positioned close to Ser125, with the fourth carboxylic acid (Asp41) further away (Figure 3H). Alanine replacement mutagenesis of Asp9, Asp41, Glu96, Asp122, or Ser125 inactivated catalytic activity, consistent with YqgF active site (Figure 3H). Similarly, sequence analysis and structural modeling of YhbQ (PDB: 1ZG2; Figure 3I) revealed a putative active site near the GIY-YIG motif (Figure 3I). Site-directed mutagenesis of YhbQ confirmed that Phe5, Tyr7, Tyr17, Gly19, Arg27, and Glu64 are required for nuclease activity (Figure 3I). Consistent with processing DRRR intermediates, affinity-purified YhbQ and YqgF physically interacted with other proteins, including DNA polymerases and RecA-dependent recombination factors, involved in binding and unwinding DNA substrates (Figure S4A). In addition, the physical association of YqgF with transcription termination factors NusA, NusG, and Rho, recently implicated in DNA damage tolerance and RF-transcription collisions (Iwamoto et al., 2012), implies participation in transcription-coupled DNA repair. Since YhbQ showed aggravating GIs with Uvr (NER DNA helicase II; Figure 3A), we examined the sensitivity of yhbQ and uvr deletion strains to MMC, 4-nitroquinoline-1-oxide (4NQO), or UV irradiation, which cause DNA lesions such as pyrimidine dimers, interstrand crosslinks, and bulky DNA adducts removed by NER (Batty and Wood, 2000). Like uvr mutants (Epshtein et al., 2014), strains lacking yhbQ showed hypersensitivity (Figure S4B), but inactivation of both yhbQ and uvr did not enhance this effect, consistent with a cooperative (epistatic) repair function, with YhbQ possibly functioning upstream of UvrD. Similarly, alleviating GI between yqgF and DNA replication genes (e.g., translesion DNA polymerase Pol I, polA; Figure 3A) was reflected by hypersensitivity to the SOS-inducing nalidixic acid, which inhibits bacterial DNA replication (Kohanski et al., 2010), while yqgF-polA double mutants showed no additive effect (Figure S4C), suggesting joint participation of YqgF with PolA in DNA replication. Together, the epistatic profiles of two nucleases, YhbQ and YqgF, implied distinct functions in DRRR. Specially, aggravating GI seen between yhbQ and uvr or several rec recombination mutants supports the notion that YhbQ functions redundantly as a catalytic nuclease required for repair of damaged DNA substrates. In contrast, the alleviating phenotype seen upon loss of both yqgF and any of several DRRR gene products suggests joint participation in DNA repair. Differential Genetic Dependencies under DNADamaging Conditions Given that E. coli has evolved an adaptive response to genotoxins (Kohanski et al., 2010), we subjected all of single and double mutants surviving the initial selection on rich medium to a DNAdamage (MMS). Condition-specific GIs were detected via changes in growth on MMS using a previously established scoring procedure (Bandyopadhyay et al., 2010). Specifically, the difference in S score between MMS and UT (SMMS-SUT) was computed, and each gene pair was assigned a p value

A

D

DRRR profile (Static UT)

*

∆yhbQ *yqgF

∆uvrC

∆uvrD

∆uvrB

∆recR

NER ∆uvrA

∆recN

∆recO ∆recQ

∆recG ∆recJ

∆recB

∆recA

∆polB

Recombination (or repair)

∆polA

∆dnaT

*dnaX

*dnaE *dnaN

Replication

nt M 80 70 60 50 40

E

8nt

20nt

*

-

-

HJ

5'F

3'F

SA

RF

3/

[YhbQ]

* nt M 40

-

* 0.2 0.4

-

5/ 0.2 0.4

-

*

*

* -

0.2 0.4

0.2 0.4

-

0.2 0.4 YhbQ (μg) ssDNA

30

30

∆yhbQ

20

*yqgF

20 aggravating

alleviating

S-score

neutral

≤-2.5 -1 -0.5 0 0.5 1 ≥2.5

F

20 nt

3'

nt 50 40

10 nt

3'

*5' 22 nt

5' 22 nt 3'

5'

-

-

G

3'

22

3'

nt

30 nt

22

nt

[YhbQ] 3*ssDNA 46 nt nt

C

22

M C

nt

5*ssDNA 45 nt

22

B

5' 22 nt 3'

*5' 22 nt 3'

5'*

5' nt 50

[YhbQ]

-

-

3'

5'*

[YqgF]

40 30

30

20

20

12 5*ssDNA 45 nt

M C

C

[YqgF] 3*ssDNA 46 nt

30 nt

H

YqgF

I

K13

YhbQ F5

D9

E96

Y7 D122

D41

2nt

4nt

-

*

*

-

30 20

Y17

S125

bp

[YhbQ] K13

C

wt

A

-

*

ssDNA 45 nt

C wt

L6

-

1nt

F5

*

9A K1 3A D 41 A I5 8A E9 6A R 97 A D 12 2 S1 A 25 A

-

mm

D

nt M 80 70 60 50 40

*

H31

A19

E64

R97

C

R27

A Y7 A Y1 7A T1 8A G 19 R A 27 H A 31 E6 A 4A

20 nt

ssDNA 45 nt

Figure 3. E. coli Nucleases Linked to DRRR (A) Aggravating and alleviating GIs of the yhbQ deletion and yqgF hypomorph (*), respectively, with DRRR genes in the UT network. (B) 50 to 30 exonuclease activity of purified YhbQ and YqgF against the 50 - or 30 -[32P]-labeled ssDNA (46 nt). Samples were incubated at 37 C without (lane C) or with 25, 50, 100, or 200 ng of YhbQ for 15 min or 50, 100, 150, 200, or 300 ng of YqgF for 30 min, and the products were analyzed by denaturing gel electrophoresis. (C and D) YhbQ cleaves dsDNA substrates containing one mismatched (mm) nucleotide (nt); 1, 2, or 4 nt loops. Perfect dsDNA (bp), a 20 nt bubble structure, or a 8 nt loop was incubated at 37 C for 25 min in the absence () or presence of YhbQ (100, 200, or 300 ng, C; 25, 50, 100, 200, or 300 ng, D). For all substrates, the top strand was [32P] labeled at the 50 end (*), and modifications are shown on the substrate models. (E) YhbQ cleaves 50 -[32P]-labeled (*) Holliday junction (HJ), 50 -flap (50 F), 30 -flap (30 F), splayed arm (SA), or replication fork (RF) substrates. (F and G) Cleavage of splayed arm substrates by YhbQ or YqgF (30 min of incubation at 37 C; 25, 50, 100, 200, or 300 ng protein). See also Figure S3. (H and I) Crystal structure of YqgF (H) and the structural model of YhbQ (I) showing the active site residues. The amino acid side chains are shown as green sticks and labeled along a protein ribbon (gray). The indicated residues were mutated to Ala, and nuclease activity of purified mutant proteins (0.3 mg) against 50 -[32P]labeled ssDNA (45 nt) was analyzed by denaturing gel electrophoresis (shown below the structures).

based on the null distribution of score differences obtained from replicate GI screens from the same condition (differential, or DF, network; Figure 1). After applying DF score thresholds corresponding to two SDs (jS scorej R 2; p % 0.05; Figures 4A and S4D), we identified 4,589 significant DF aggravating (i.e., MMS-induced synthetic sickness/lethality) and 3,638 DF alleviating (i.e., MMS suppression) GIs. As in yeast (Bandyopadhyay et al., 2010), most (64%, 5,262 of 8,227) DF GIs showed low epistasis scores in the static UT network (Figure S4E; Table S1, sheet 8), yet became pronounced under stress.

To quantitatively identify interactions specific to DDR (e.g., GIs in MMS only), we used a computational framework to visualize changes between the static UT and MMS maps (Figure S4F; Supplemental Experimental Procedures). We found 373 gene pairs (Table S1, sheet 9) with a significant (p % 0.05) increase or decrease in S score, either gaining (e.g., mutT-yhfG and ada-obgE) or losing (e.g., nudG-recC and soxS-yegP) GIs after DNA damage, respectively; Figure S4F). These results reveal compensatory changes underlying the DDR. Cell Reports 14, 648–661, January 26, 2016 ª2016 The Authors 653

-20

-10

Static UT (S-score)

B

UT

MMS

0

10

DF (S-scores) DF

DNA (R-M) system

0 (no) 140

2

10-3

10-4

P = 4.9 x 10-5 + +

+

+ +

+

Growth defect

Phenotypic score

None

>0

Moderate

< 0 and >-2.0

High

< -2.0

groL

ftsZ

20 High (n = 36)

MMS single mutant sensitivity (Nichols et al)

≥2 10

alleviating 12 (high)

Cell division DRRR, Damage/stress Transcription Others Unclear Non-hub genes

zipA uvrB yegP thyA gyrA ygeP ftsQ ftsK ftsB dnaG cedA tag mutY xerD yaiUdnaA dnaN dnaE nei umuD ygcL yfaA ygcK yadD alkB

0

Moderate (n = 186)

8

modF

0

None (n =213)

6

120

10-2

4

ruvC

|MMS single mutant sensitivity| (Nichols et al)

100

10-1

80

100

60

Enrichment (p-value)

80

120

100

40

Number of DF GIs

10-1

40

10-2

160

C

10-3

Number of significant aggravating DF GIs

10-4

ruvA recC

Hub genes

E

DNA damage (SOS)

xerD

rA uvrY marA u seqA hfq helD recG alkB ihfB alkA

DNA repair

DNA recomb. & repair

recA

S-score

0.6 0.0

P = 0.05

Cell division

recB xerC

dcm

Base-excision repair

RNA processing

neutral

0.8

1.0 20

Others ihfA cspI ylcG

≤-2 -1 -0.5 0 0.5 1

20

Unclear

aggravating

10

Transcription

0.4

0

DRRR, Damage/stress

P = 1.1 x 10-9 r = -0.21

0.2

-10

AC (r = 0.58)

Autocorrelation (AC; MMS vs UT)

Density 0.00

-10 -20

AC (r = 0.25)

ylcG (Static, UT)

Alleviating

Aggravating

0.05

0

Significant (P ≤ 0.05) cut-off selection

ruvC (Static, UT) ruvC (Static, MMS)

ylcG (Static, MMS)

0.10

≤ 0.05 ≤ 0.025 ≤ 0.005 ≤ 0.0005

10

D

0.15

P-value

-20

Static MMS (S-score)

20

A

0

20

40

P =7.7x10-4 r =-0.06 recR

ihfB ihfA yhbQ fis ruvA 60

dnaJ recG

hfq

alkA 80

ruvC

100

uvrD

120

140

Number of significant alleviating DF GIs

Figure 4. MMS Induced DF Interaction Network (A) Scatterplot of S scores with distinct p value-based cutoffs between UT and MMS networks. (Right) The histogram of DF S scores filtered at p % 0.05 (see also Figure S4), indicating tails with significant (jS scorej R 2) GIs. (B) Hypergeometric enrichment of GIs in the static and DF networks with genes functioning in the indicated processes (R-M, restriction-modification). p > 0.05 considered insignificant. (C) Box plot showing the distribution of the number of DF GIs against the MMS single mutant sensitivity (Nichols et al., 2011). p value by Student’s t test. See also Figure S5. (D) Correlation of GI profiles for each query gene (ruvC and ylcG shown as an example) between the UT and MMS networks (autocorrelation) plotted against the gene’s absolute MMS single mutant sensitivity (Nichols et al., 2011). p value by Student’s t test. (E) Scatterplot of the hub genes from the indicated processes in the DF network, shown with the number of aggravating and alleviating DF GIs (see Table S2 for complete list). Negative slope indicates significant hub genes with either more DF aggravating or alleviating GIs, respectively.

Further inspection of all three networks indicated that, in contrast to the DF network, the static MMS network was enriched for GIs by genes functioning in DNA restriction modification (Figure 4B; Table S2, sheet 1). Conversely, genome integrity genes involved in cell division, base-excision repair, SOS DDR, DNA recombination and/or repair, or RNA processing (Figure 4B) showed greater enrichment in DF network than at least one of the static networks. Moreover, deletion of genes that exhibited the largest number of DF GIs (Figures 4C and S5A; Table S2, sheet 2), and genes in the static GI profiles that were conditionally disrupted 654 Cell Reports 14, 648–661, January 26, 2016 ª2016 The Authors

by MMS (i.e., autocorrelation; Figure 4D; Table S2, sheet 3) were also sensitive to MMS in a large-scale phenomics screen (Nichols et al., 2011). For example, deletion of the HJ resolvase ruvC (required for recombinational repair), and a protein of unknown function ylcG (previously linked with the HJ resolvase rusA; Nichols et al., 2011), which showed low and high autocorrelation in the GI patterns between conditions, respectively, conferred growth defects in MMS (Figure 4D). Consistent with this, genes showing DF interactions were also more likely to be conditionally essential for growth and survival in the presence of a variety of other drugs and environmental stressors

C ∆nth ∆ada ∆ogt ∆dcm ∆mutH ∆mutY ∆radA ∆recF ∆recO ∆recR ∆xerC ∆ligB ∆mfd *dnaE ∆dnaG ∆pioO *nrdA

20

40

GIs with ∆yegP

60

80

MMS

∆yegP

All genes

LB (- drug)

Base-excision repair Damage/stress

WT

∆recR ∆recO DNA recombination

∆recF ∆recR ∆yegP ∆recO ∆yegP

DNA repair/ replication

∆recF ∆yegP 10-1 10-2 10-3 10-4 10-5

S-score

E

≤-2 -1 -0.5 0 0.5 1

LB (+MMS; 0.05%)

aggravating

D

≥2

alleviating

neutral

WT

LB (- drug)

DRRR

F

HJ

ΔyegP ΔrecF

Miscell.

6

0.8 0.6 ∆recF

0.4

∆recO

0.2 0.0 -0.2 ∆recR

-0.4

-0.6 Chr. partitioning

3

2

Nucl. metabolism

Disulfide oxidoreduc.

Nucl. reductase

CD sep. inhibition

(i)

1

0.0

0.2

0.4

yafN

minD

yafO

minE

1

dinB

dicB

mRNA degrad. and helicase

Metabolism Transcrip., DRRR

topB (ii)

sbcB priA

xerD

ΔyegP ΔrecR

5 Number of intermodule DF GIs

Module size (based on the proportion of interacting genes)

1

Moderate

9

DF Aggravating Alleviating

MMS sensitivity

High P-value

(iii)

6

7 hepA

8

recJ nudG

smtA

zipA

mukB

ftsZ

nupG

mukF

rpoD

34

ftsA

mltC

yggX

0.6

2

minC

2

4

-0.2

Correlation with ∆yegP (UT)

DNA pol. III

mRNA degrad. and DNA pol IV

4

-0.4

1

8

6

LipidA biosynth. RNA degrad.

1.0

-0.6

Chr. repl. 1

7 RNAP

10-1 10-2 10-3 10-4 10-5

5

DNA 3 recomb.

Chaperones

CD sep. formation

SOS resp. and HR

Alkyl. damage

Phenylalan. syn. and trans. repress.

ΔyegP ΔrecO

LB (+MMS; 0.05%)

∆yegP

Mismatch repair

Correlation with ∆yegP (MMS)

DF

B

UT

UT

MMS

P = 4.2 x 10-35

0

Significant DF GIs (%)

100

A

ruvC

selD topA

4 rnhA

≤ 0.05 ≥ 0.05 to ≤ 0.1

recQ

Figure 5. DF Interactions Identify DNA Damage-Dependent Genes and Pathways (A) DF GIs between the UT and MMS networks for YegP. p value using Student’s t test. (B) GI profile of yegP with MMS-specific aggravating interactions highlighted for a gene subset. (C) Growth sensitivity of mutants and WT cells grown on MMS. (D) Correlation coefficients of yegP mutants in the UT and MMS conditions. (E) Representative phase-contrast micrographs of mutants and WT cells with or without MMS. See Figure S5C for full micrographs. Scale bar represents 10 mm. (F) Enrichment of functional modules connected by DF GIs (only portion of module GIs with p % 0.05 shown for clarity; full list of module-module interactions is provided in Table S3). Top fifth (p % 0.05; thick line) and tenth (p R 0.05 to % 0.1; dashed line) percentile of select module set shown in periphery.

(Figure S5B; Table S2, sheet 4) from the E. coli phenomics screen (Nichols et al., 2011). Given that many DF interaction hubs (i.e., genes with R50 GIs) were annotated to participate in DRRR and other genomic integrity pathways (Figure 4E; Table S2, sheet 5), we investigated whether genes of unknown function that were highly connected in the DF GI network were DDR components. An illustrative example is yegP, an uncharacterized DF ‘‘hub’’ gene (Figure 4E), which showed a high number (p = 4.2 3 1035; Figure 5A) of aggravating GIs in response to MMS (
89%; 71 of 80), including with genes required for DDR (ada, ogt), replication (dnaEG), mismatch repair (mutHY), and base-excision repair (nth; Figure 5B). In addition, genes encoding subunits of the recFOR

recombination and repair complex displayed strong DF aggravating GIs with yegP, which we confirmed by spot dilution assay (Figure 5C). Strikingly, after MMS treatment, the yegP GI profile became correlated with the profile of recO, which is involved in the repair of single-strand gaps (Figure 5D), suggesting similar roles in DSB repair. To confirm this association, we tested whether mutation of yegP, as with recF and recO, impairs cell division and chromosome segregation upon exposure to genotoxic agents (Babu et al., 2011a). Aberrant morphology, consisting of long, nonseptate, multinucleate filaments, occurs in the presence of DSBs or unresolved chromosome dimers prior to division. After MMS treatment, yegP recF/recO/recR double mutant exhibited longer Cell Reports 14, 648–661, January 26, 2016 ª2016 The Authors 655

(
9.2 mm average cell length), more filamentous cells compared with WT or single mutants (
5.4 mm; Figures 5E and S5C). This suggests that YegP functions in parallel with the RecFOR in DSB repair pathway. Genomic Integrity Modules Enriched for DF Interactions We integrated the DF GI data with additional supporting evidence of functional associations, such as protein-protein interactions to improve understanding of how biological pathways and protein complexes (i.e., functional modules) are reconfigured in response to DNA damage (Bandyopadhyay et al., 2010; Gue´nole´ et al., 2013). We cross-mapped the relevant subset (92 of 316) of previously reported E. coli functional modules (Peregrı´n-Alvarez et al., 2009), containing about half (253 of 549) of genomic integrity genes targeted in this study. Strikingly, DF GIs were not enriched within modules (Figure S6A), but rather between modules (Figure S6B). These results suggest that, as in yeast (Bandyopadhyay et al., 2010), bacterial protein complexes are stable across conditions, whereas GIs are more context dependent. Permutation testing (Babu et al., 2014) revealed a global map consisting of 479 module-module interactions consisting of 84 (of 92 modules) significantly enriched (jZ scorej R 2.5; p % 0.006) modules (Figure S6C; Supplemental Experimental Procedures) showing substantial change in response to DNA damage (Figure 5F; Table S3). Close inspection recapitulated known relationships (i.e., crosstalk of DNA-damage induced GIs among modules) and identified unique testable hypotheses. For example, DNA polymerase IV dinB, and downstream genes encoding the toxin yafO and cognate antitoxin yafN, which are co-regulated by the SOS response to DNA damage (Yamaguchi and Inouye, 2011), showed alleviating DF GIs with a module containing dicB and minCDE, which physically interact to regulate cell division and septum formation through disruption of ftsZ polymerization (Johnson et al., 2002). This association of yafOyafN with septum inhibition is consistent with an analogous TA pair, cbtA-yeeU, wherein the toxin CbtA has been reported to likewise inhibit FtsZ function (Masuda et al., 2012). Another prominent module pair with alleviating DF GIs included the HJ resolvase ruvC with factors involved in end processing (recJ, sbcB), RF re-establishment (priA), HJ dissolution (topB), DNA damage avoidance (nudG), and chromosome segregation (xerD). These links underscore synergistic roles in DSB repair via homologous recombination (Ayora et al., 2011). Aggravating DF GIs were more frequent (63%, 300 of 477 module pairs) than alleviating interactions in the module map (Figure S6D). These intermodule interactions included those between RNA polymerase (RNAP; hepA, rpoD) and genes involved in chromosome partitioning (mukBF), DNA base-damage recognition (smtA), nucleoside transport (nupG), peptidoglycan hydrolase (mltC), and iron-sulfur cluster formation (yggX, which functions upstream of nupG). While the mechanism behind this coupling remains unclear (Figure 5F), it is possible that impaired RNAP function after DNA damage may block chromosome segregation, DNA damage recognition, and cell integrity, which manifests as an aggravating phenotype (Kruse et al., 2006). Additionally, aggravating DF GIs between DNA metabolism, DRRR, and cell division proteins in Z-ring formation (ftsAZ, 656 Cell Reports 14, 648–661, January 26, 2016 ª2016 The Authors

zipA; Figure 5F) are consistent with bacteria responding to DNA damage by impeding cell division, linking chromosome segregation to DSB repair. Functional Crosstalk and Prokaryotic Evolution As cross-species gene co-conservation can infer functional relatedness (Pellegrini et al., 1999), we explored whether conservation of high-confidence GI relationships (jS scorej R 2.5j) in the static and DF networks could yield functional insights into other, diverse bacteria. As a result, we constructed a phylogenetic profile for each gene from our target index, determining orthology across 747 species in 11 major bacterial phyla (Table S4, sheet 1). These phylogenetic profiles were then used to determine the average conservation of interacting gene pairs, as well as the likely conservation of GI-defined modules or bioprocesses. As noted before in yeast and bacteria (Babu et al., 2014; Ryan et al., 2012), genes having GIs and functioning within the same modules, in both the static and DF networks, tended to be more co-conserved than genes with GIs to different modules (Figure S6E). This suggests that within-module GI pairs reflect functional relationships that are more fundamental to all prokaryotic species. We also found that, for some bioprocesses, genes interacting in the static and DF GI networks tended to show consistent patterns of co-conservation across the 11 major bacterial phyla considered, whereas others varied more substantially (Figure 6A; Table S4, sheet 2). For instance,
3% (145 of 4,706) of the GIs detected among E. coli genes involved in recombination in the static networks showed a strong tendency for co-conservation of their interactors across bacterial phyla, of which only 11% (16 of 145) were altered in response to DNA damage in the DF network. Conversely, genes in or interacting with other bioprocesses, like cell adhesion, metabolism, and SOS DDR, showed poor co-conservation, with less than 1% (29 of 5,384) co-occurring in other bacterial phyla (i.e., conservation restricted to proteobacteria; Figure 6A). This suggests that E. coli has evolved systems to counter genotoxic stresses unique to its specialized environmental niche. Next, we examined the co-conservation of genes comprising high-confidence interacting gene pairs, restricting our focus to proteobacteria, as it is the largest (360 genera) and most diverse phylum among naturally occurring bacterial communities (Hoppe et al., 2015). The top 5% (p < 0.05) of most co-conserved interprocess gene pairs exhibiting GI within each of the static and DF networks displayed notable differences in bioprocess annotation (Figure 6A), such as relative enrichment for conserved DF GIs (larger yellow wedges in Figure 6A) between bacterial DRRR and cell division, as well as between DRRR and transcription and RNA processing or translation (Table S4, sheets 3–5). This suggests that despite high conservation of core processes, species-specific functional adaptations exist, likely to accommodate specialized environmental challenge. Epistatic Interactions Reveal Divergent Functions among Bacterial Gene Duplicates Given that GIs between paralogous gene pairs can indicate functional redundancy post-duplication (Musso et al., 2008; VanderSluis et al., 2010), we probed epistatic relationships among

Translation Unclear

Stress response Transcription

RNA proccessing

Protein folding Ribosome assembly

Protein degradation

PG biosynthesis

DNA replication DNA metabolism

DNA recombination

DNA repair

DNA degradation

UT MMS DF Not Enriched

DNA damage (SOS)

0.6

Cell adhesion Cell division DNA damage (SOS) DNA modification DNA recombination DNA repair DNA replication DNA metabolism PG biosynthesis Protein folding RNA processing Stress response Transcription Translation Unclear

Cell division

0.3

Conservation of bioprocesses across proteobacteria

Enriched (P < 0.05)

DF 0

Thermotogae (9) Chloroflexi (11) Chlamydiae (11) Deino -Thermus (12) Spirochaetes (18) Tenericutes (21) Cyanobacteria (27) Bacteroidetes (46) Actinobacteria (96) Firmicutes (136) Proteobacteria (360)

MMS

Thermotogae (9) Chloroflexi (11) Chlamydiae (11) Deino -Thermus (12) Spirochaetes (18) Tenericutes (21) Cyanobacteria (27) Bacteroidetes (46) Actinobacteria (96) Firmicutes (136) Proteobacteria (360)

Proportion of conserved genes having GIs UT

Thermotogae (9) Chlamydiae (11) Chloroflexi (11) Deino-Thermus (12) Spirochaetes (18) Tenericutes (21) Cyanobacteria (27) Bacteroidetes (46) Actinobacteria (96) Firmicutes (136) Proteobacteria (360)

A

Conservation of bioprocesses across proteobacteria

Cell division

Static UT

Static MMS

DNA damage (SOS) DNA degradation DNA recombination DNA repair DNA replication DNA metabolism PG biosynthesis Protein degradation Protein folding

DF

1.5

∆rhlB

≥2.5 ≤-2.5 -1 -0.5 0 0.5 1

S-score

∆srmB ∆deaD ∆dbpA ∆rhlE ∆rhlB

neutral

∆srmB ∆deaD ∆dbpA ∆rhlE ∆rhlB

alleviating

Static UT ∆deaD

-5

Static MMS ∆dbpA

∆srmB ∆deaD ∆dbpA ∆rhlE ∆rhlB

DF

** ** **

∆srmB

1.5

∆rhlE

∆rhlB

1.0

∆deaD

0.5 ∆umuC ∆dbpA

∆rhlE

∆rhlB

∆deaD

∆dbpA

1.0

**P = 1.17 x 10

0.0

∆srmB

-0.5

-0.5

Random singleton pairs (II)

∆rhlE

0.6

0.8

Unclear (III)

r(∆dinB, = 0.2 din ∆umuC)

C

∆srmB

0.4

Static UT

Static MMS

Correlation of GI profiles

0.2 0.0

Translation

r(∆ = 0.4 ((∆dinB, ∆umuC) u

aggravating

1.0

P < 2.2x 10-16

Duplicate pairs

Proportion of GIs (Aggr + Allev)

Transcription r(∆dinB, ∆umuC) = 0.6

∆dinB 0.5

P = 5.12x 10-3

**

Stress response UT MMS DF

0.0

0.5 0.4

-2

0.2

0.3

** ** ** P = 3.25 x 10

RNA processing (I)

0.0

Proportion of aggravating GIs

D

UT MMS DF

0.1

B

0.6

Ribosome assembly

∆dbpA ∆deaD

UT MMS Duplicates

Singletons

∆rhlB ∆rhlE

DF -0.2

∆srmB

1 DF

Figure 6. GIs Conserved across Species and among Duplicates in Static and DF Networks (A) Evolutionary conservation heatmap of static and DF GIs based on co-occurrence of orthologs across bacterial phyla (dashed line indicates conservation patterns of GIs among E. coli genes involved in DNA recombination between networks). Number of fully sequenced genomes of the bacterial species within each phyla (left; see Table S4 for details) is shown in parentheses. Conserved interprocess GIs in static and DF networks (right) are shown only for proteobacteria (vertical dotted line); node size is proportional to the number of conserved GIs, and color indicates network source. (B) Proportion of aggravating GIs among duplicate pairs versus random singleton pairs. (C) Number of aggravating and alleviating GIs plotted for duplicates and singletons; **p value similar in all networks (C) and computed using Fisher’s exact test (B and C). (D) Scatterplot of correlated GI profiles (I, II) or GI patterns (III; hierarchically clustered using Cluster 3.0) for SOS stress-inducible Y-family polymerases (dinBumuC) or among the five DEAD-box RNA helicase duplicated gene pairs in static and DF networks. See also Figure S6 and Table S5.

duplicated genes in our GI networks. Of 33 putative duplicate pairs, encompassing 45 genomic integrity genes (Table S5, sheets 1 and 2), interparalog aggravating GIs occurred significantly more frequently compared with randomly drawn singleton gene pairs, particularly in the MMS and DF networks (p = 5.1 3 103 and p < 2.2 3 1016, respectively; Figure 6B), consistent with the notion that functional redundancy between paralogs is adaptive for specific stress conditions (Musso et al., 2008; VanderSluis et al., 2010). Additionally, as in yeast (VanderSluis et al., 2010), duplicated genes displayed fewer GIs with other genes,

on average, than non-duplicated singletons (Figure 6C), further reflective of functional buffering among bacterial paralogs. Given that correlated GI patterns are a strong indicator of shared functionality (Costanzo et al., 2010), we assessed the functional relationship of paralogs in the static and DF networks based on GI profile similarity (Figure S6F). We found nearly half of paralog pairs had distinct GI profiles in the static UT network, but greater similarity in the MMS and DF networks. For example, paralogs of SOS stress-inducible Y-family polymerases dinB (Pol IV) and umuC (Pol V), which bypass DNA lesions (Sutton and Cell Reports 14, 648–661, January 26, 2016 ª2016 The Authors 657

Walker, 2001), showed high GI correlation specific to the MMS and DF networks (r = 0.6 and 0.4, respectively; Figure 6DI; Table S5, sheet 3). Similarly, positive correlations (r =
0.3 to 0.6; Figure 6DII) were observed after DNA-damage between duplicates of the DEAD-box RNA helicases (dbpA, deaD, srmB, rhlB, rhlE) involved in RNA metabolism. Additionally, the GI profiles of many helicase duplicate pairs, including deaD-dbpA, were markedly altered in the MMS network (Figure 6DIII), reflecting their role in the adaptation of bacteria to changing environments. A subset of paralog pairs that had positively correlated UT GI profiles showed anti-correlated profiles in MMS and vice-versa (Figure S6F; Table S5, sheet 3). For example, anti-correlation was observed between hupA and ihfB, which have suggested opposing roles in nucleoid structure and DNA supercoiling (Grove, 2011). In contrast, anti-correlated profiles were observed between the superoxide dismutases sodA and sodB in UT only, suggesting buffering capability specifically induced under DNAdamage conditions. DISCUSSION The majority (
85%) of genes maintaining genome integrity in E. coli are dispensable for viability under standard laboratory growth conditions (Baba et al., 2006), yet are required to resist genotoxic stress and are frequent targets of antibiotics. In this study, we identified unanticipated gene-pathway dependencies, including components not been previously linked to well-studied genome integrity systems, that provide rationale avenues for mechanistic follow-up studies and candidate targets for drug combination therapies. For example, GI connections centered on DNA repair, recombination, cell division, and metabolism revealed metabolitic drugs (TMT, CLART, PUR) targeting distinct but functionally coupled processes markedly sensitize E. coli to DNA damage. Another important finding is that the global pathway architecture in the UT network allows for detailed functional characterization of genes, including two nucleases (yhbQ and yqgF) linked to annotated DDR genes. As well, the GI patterns of sRNA illuminate their regulatory roles in genome integrity, such as between isrA and its putative target ssb in the SOS response. As in yeast (Bandyopadhyay et al., 2010), we identified many GIs only in DNA-damaging conditions, including unannotated E. coli genes not previously linked to DNA repair. For example, the DF interaction network identified uncharacterized gene YegP as functioning in parallel to the RecFOR in DSB repair pathway. To expedite similar findings, all high-confidence GIs are available via a dedicated web portal (http://ecoli.med. utoronto.ca/eMap/DNA). By partitioning the DF GI network with respect to known E. coli functional modules (Peregrı´n-Alvarez et al., 2009), we found that while bacterial genome integrity systems are highly integrated, they are unable to compensate for the loss of certain combinations of pathways, particularly when subjected to genotoxic stress. For example, epistasis was observed between RNAP/ chromosome segregation and genes regulated by oxidative stress or transcribed under SoxS control (yggX and mltC), consistent with a model in which SoxS transcription is activated through binding of RNAP (Zafar et al., 2010). 658 Cell Reports 14, 648–661, January 26, 2016 ª2016 The Authors

Although the underlying evolutionary basis for nascent GIs is unclear, our conservation analysis allowed us to project condition-specific genome integrity functions from E. coli to distantly related bacterial species. For instance, pairwise co-conservation of epistatic genes in DRRR and cell division across proteobacterial species suggests selection based on adaptive functionality and condition-specific action. Also, as seen in yeast (VanderSluis et al., 2010), we report that bacterial duplicates functioning in DNA damage behave differently based on condition, likely due to stress-specific buffering. In summary, antimicrobial therapies targeting genes involved in repair of dsDNA breaks or inhibiting DRRR may counter the increasing prevalence of antibiotic resistance (Kohanski et al., 2007). In this regard, while large-scale phenomics screening (Nichols et al., 2011) has generated condition-based single mutant fitness information, this alone cannot capture the functional dependencies required to survive DNA-damaging conditions, similar to how the growth phenotype of single gene deletion mutants under normal growth condition cannot identify gene pairs that are synthetic sick or lethal (Martin et al., 2015). By undertaking an unbiased, quantitative network-based screening approach, our work exemplifies the dynamic cellular responses of bacterial genomic integrity systems to DNA damage. Our static and DF GI networks have revealed condition-specific functional relationships at high resolution, which we anticipate will spur deeper mechanistic investigation into the molecular biology of genome integrity pathways, and motivate drug combination screening strategies for more effective therapeutic interventions. EXPERIMENTAL PROCEDURES Strains and Growth Conditions The bacterial strains used in this study are listed in Table S6. For epistatic screening, F-recipient single gene deletion mutant strains were derived from the Keio knockout library (Baba et al., 2006). Hfr Cavalli query donor mutants were constructed using l-Red recombination, and hypomorphic alleles were generated by integrating the selection cassette into the 30 -UTR of essential genes to perturb transcript abundance as previously described (Butland et al., 2008). Luria-Bertani (LB) broth or agar medium was supplemented with Kan (50 mg/ml), Cm (34 mg/ml), and/or ampicillin (100 mg/ml), as required. Detailed methods on target gene selection, genetic screens, computational processing of epistatic data, and other bioinformatics analyses are described in the Supplemental Experimental Procedures. Growth Curve and Phenotypic Assays Growth curve analyses were performed by inoculating the overnight cultures of the parental and mutant strains into 96-well microtiter plates containing 100 ml of LB medium. The plates were then incubated with shaking at 32 C, with the absorbance of the culture measured at OD600 every 15 min for over 24 hr using a Tecan Sunrise microplate reader. Small-scale drug assays were performed by growing the E. coli WT and deletion mutant cultures to OD600
0.3 in LB medium and then manually spotting onto LB-agar plates in serial dilutions in the absence and presence of the drugs. Assays were in the following concentrations: MMC (2 mg/ml), MMS (0.05%), norfloxacin (62 ng/ml), nalidixic acid (2 mg/ml), and 4NQO (1 mM). For UV assays, cells spotted on LB agar plates in serial dilutions were irradiated to a dose of 35 J/m2 using the XL-1000 UV cross-linker (Spectroline) containing a UV (254 nm) lamp. Both the untreated control and treated plates were incubated overnight at 32 C, and the sensitivity of the agent was assessed after 36 to 48 hr of incubation in the dark. Cell morphology was examined by growing the WT and mutant strain cultures to an OD600
0.3–0.4 in LB medium at 32 C. Prior to imaging, the

genotoxic agent was added to the cultures at the indicated concertation and incubated for 120 min, and the cells were pelleted and re-suspended in cold PBS. Roughly 2–3 ml of the suspended cells were spotted onto a precleaned glass slide for imaging. Cell morphology images were digitally captured using a Zeiss AxioVert.A1 inverted epifluorescence microscope with 633/1.4 Plan Apo objective. Cell length was measured using Carl Zeiss ZEN blue software. Quantitative Real-Time PCR To quantify the candidate mRNA levels in the isrA or micF deletion mutants, target RNA was extracted using an Aurum total RNA mini kit (Bio-Rad) from cells grown to stationary phase in complete media and diluted according to the manufacturer’s instructions. Following RNA extraction, 1 mg of total RNA was used to synthesize cDNA using a cDNA synthesis kit (Bio-Rad). iQ SYBR green supermix (Bio-Rad) was used for q-RT PCR reactions. Briefly, each reaction contained 300 nM forward and reverse primers, 100 ng cDNA, 5ml iQ SYBR green supermix, and 20 ml nuclease free water. Primer sequences for the targeted candidate genes, and the glyceraldehyde phosphate dehydrogenase (gapA) internal control, are indicated in Table S6. The mRNA expression levels were measured by calculating the difference in threshold cycle (Ct) value for each candidate gene in their respective deletion mutant background and relative to an internal control. This value was then normalized to the value from WT. Protein Affinity Purification and Mutagenesis For biochemical assays, WT E. coli YhbQ, YqgF, and mutant proteins were overexpressed in E. coli and purified as 63 His-tag fusions using affinity and size-exclusion chromatography as previously described (Babu et al., 2011a). For affinity purification combined with mass spectrometry, both YhbQ and YqgF were C-terminally SPA (sequential peptide affinity) tagged and purified according to the previously established procedure (Babu et al., 2009a). The trypsin-digested purified protein was subjected to an Orbitrap mass spectrometer, and the resulting MS/MS spectra were searched against the protein coding sequences of the E. coli W3100 derivative strain using the SEQUEST search engine. High-confidence matches were then evaluated using spectral (i.e., peptide) counts and probability scores generated by the STATQUEST algorithm, as previously described (Babu et al., 2009a). Site-directed mutagenesis of YhbQ and YqgF was performed using a protocol based on the QuikChange site-directed mutagenesis kit (Stratagene). Preparation of DNA and RNA Substrates ssDNA and ssRNA oligonucleotide substrates (34–83 nt; Table S6) were purchased from IDT. The oligonucleotides were 50 labeled using [g-32P] ATP (3,000 Ci/mmol; Perkin Elmer) and T4 polynucleotide kinase or 30 labeled using [a-32P] dATP (3,000 Ci/mmol) and calf thymus terminal transferase. The labeled oligonucleotides were purified using PAGE (15% polyacrylamide [PAA] and 8 M urea). The synthetic RNA/DNA complex was prepared by annealing the oligonucleotides as shown in Table S6. Nuclease Assays The reaction mixture for DNase assays contained 25 mM Tris-HCl (pH 8.5), 25 mM KCl, 2 mM MgCl2, 0.1 mM 50 - or 30 -[32P]-labeled ssDNA and the protein at indicated concentrations in a final volume 10 ml. The reaction mixture for RNase assays was comprised of 25 mM Tris-HCl (pH 7.5), 50 mM KCl, 5 mM MgCl2, 0.1 mM 50 -[32P]-labeled ssRNA, and YhbQ as indicated. In both assays, the solutions were incubated at 37 C for the indicated time and quenched by the addition of an equal volume of formamide loading buffer. The reaction products were separated using electrophoresis of 12%–15% PAA/8 M urea gels in TBE running buffer and visualized by autoradiography. An imidazole ladder produced by partial RNA cleavage by 2 M imidazole was used as a nucleotide size marker. Endonuclease assays were performed at 37 C for 15 min using the circular ssDNA of the M13mp18 phage (New England BioLabs) as a substrate, with the reaction mixtures containing 25 mM Tris-HCl (pH 8.5), 20 mM KCl, 2 mM MgCl2, 5 nM DNA (M13mp18), and 50–500 nM YhbQ. The reactions were stopped by the addition of agarose gel loading buffer (10% glycerol, 0.025% bromophenol blue, 10 mM EDTA [pH 8.0], 0.5% SDS), separated by electrophoresis in 0.9% agarose gels, and visualized by SYBR green staining.

DNA/RNA Binding Assays Reaction mixtures (10ml) for RNA binding contained 25 mM Tris-HCl (pH 7.0), 20 mM KCl, 0.5 mM MgCl2, and 0.05 mM [32P]-labeled RNA, whereas the DNA binding activity buffer was comprised of 25 mM Tris-HCl (pH 8.0), 1 mM MnCl2, 1 mM DTT, and 0.05 mM [32P]-labeled DNA. These mixtures were then incubated at 25 C for 5 min, analyzed in native 6%–10% PAA/8 M urea gels, and visualized by phosphorimaging.

SUPPLEMENTAL INFORMATION Supplemental Information includes Supplemental Experimental Procedures, Figures S1–S6, and Tables S1–S6 and can be found with this article online at http://dx.doi.org/10.1016/j.celrep.2015.12.060. AUTHOR CONTRIBUTIONS A.K. and M.B. designed the study. N.B., V.D., A. Gagarinova, S.L., M.H., and Z.M. contributed to the experiments. A.K., C.B.-T., S.P., G.M., J.V., O.W., and R.M. analyzed the data. S.P. developed the web portal. A.K., A.Y., and M.B. wrote the manuscript with critical input and edits from C.B.-T., G.M., J.V., P.A., A. Golshani, J.P., and A.E. All authors read and approved the manuscript. ACKNOWLEDGMENTS We thank members of the M.B. and A.E. laboratories for technical assistance. This work was supported by grants from the Canadian Institutes of Health Research (CIHR) to A.E. (MOP-82852); Natural Sciences and Engineering Research Council (NSERC) Strategic Network grant IBN to A.F.Y.; NSERC Discovery to A. Golshani, J.P. (DG-06664), and M.B. (DG-20234); as well as the Canada Foundation for Innovation to M.B. J.V. is supported by the Saskatchewan Health Research Foundation Postdoctoral Research Fellowship, and M.B. holds a CIHR New Investigator award. Received: September 1, 2015 Revised: November 8, 2015 Accepted: December 10, 2015 Published: January 7, 2016 REFERENCES Aravind, L., and Koonin, E.V. (2001). Prokaryotic homologs of the eukaryotic DNA-end-binding protein Ku, novel domains in the Ku protein and prediction of a prokaryotic double-strand break repair system. Genome Res. 11, 1365– 1374. Aravind, L., Makarova, K.S., and Koonin, E.V. (2000). SURVEY AND SUMMARY: holliday junction resolvases and related nucleases: identification of new families, phyletic distribution and evolutionary trajectories. Nucleic Acids Res. 28, 3417– 3432. Ayora, S., Carrasco, B., Ca´rdenas, P.P., Ce´sar, C.E., Can˜as, C., Yadav, T., Marchisone, C., and Alonso, J.C. (2011). Double-strand break repair in bacteria: a view from Bacillus subtilis. FEMS Microbiol. Rev. 35, 1055–1081. Baba, T., Ara, T., Hasegawa, M., Takai, Y., Okumura, Y., Baba, M., Datsenko, K.A., Tomita, M., Wanner, B.L., and Mori, H. (2006). Construction of Escherichia coli K-12 in-frame, single-gene knockout mutants: the Keio collection. Mol. Syst. Biol. 2, 0008. Babu, M., Butland, G., Pogoutse, O., Li, J., Greenblatt, J.F., and Emili, A. (2009a). Sequential peptide affinity purification system for the systematic isolation and identification of protein complexes from Escherichia coli. Methods Mol. Biol. 564, 373–400. Babu, M., Musso, G., Dı´az-Mejı´a, J.J., Butland, G., Greenblatt, J.F., and Emili, A. (2009b). Systems-level approaches for identifying and analyzing genetic interaction networks in Escherichia coli and extensions to other prokaryotes. Mol. Biosyst. 5, 1439–1455.

Cell Reports 14, 648–661, January 26, 2016 ª2016 The Authors 659

Babu, M., Beloglazova, N., Flick, R., Graham, C., Skarina, T., Nocek, B., Gagarinova, A., Pogoutse, O., Brown, G., Binkowski, A., et al. (2011a). A dual function of the CRISPR-Cas system in bacterial antivirus immunity and DNA repair. Mol. Microbiol. 79, 484–502. Babu, M., Dı´az-Mejı´a, J.J., Vlasblom, J., Gagarinova, A., Phanse, S., Graham, C., Yousif, F., Ding, H., Xiong, X., Nazarians-Armavil, A., et al. (2011b). Genetic interaction maps in Escherichia coli reveal functional crosstalk among cell envelope biogenesis pathways. PLoS Genet. 7, e1002377.

Iwamoto, A., Osawa, A., Kawai, M., Honda, H., Yoshida, S., Furuya, N., and Kato, J. (2012). Mutations in the essential Escherichia coli gene, yqgF, and their effects on transcription. J. Mol. Microbiol. Biotechnol. 22, 17–23. Johnson, J.E., Lackner, L.L., and de Boer, P.A. (2002). Targeting of (D)MinC/ MinD and (D)MinC/DicB complexes to septal rings in Escherichia coli suggests a multistep mechanism for MinC-mediated destruction of nascent FtsZ rings. J. Bacteriol. 184, 2951–2962. Junop, M.S., Yang, W., Funchain, P., Clendenin, W., and Miller, J.H. (2003). In vitro and in vivo studies of MutS, MutL and MutH mutants: correlation of mismatch repair and DNA recombination. DNA Repair (Amst.) 2, 387–405.

Babu, M., Arnold, R., Bundalovic-Torma, C., Gagarinova, A., Wong, K.S., Kumar, A., Stewart, G., Samanfar, B., Aoki, H., Wagih, O., et al. (2014). Quantitative genome-wide genetic interaction screens reveal global epistatic relationships of protein complexes in Escherichia coli. PLoS Genet. 10, e1004120.

Kanaar, R., Wyman, C., and Rothstein, R. (2008). Quality control of DNA break metabolism: in the ‘end’, it’s a good thing. EMBO J. 27, 581–588.

Bandyopadhyay, S., Mehta, M., Kuo, D., Sung, M.K., Chuang, R., Jaehnig, E.J., Bodenmiller, B., Licon, K., Copeland, W., Shales, M., et al. (2010). Rewiring of genetic networks in response to DNA damage. Science 330, 1385–1389.

Kohanski, M.A., Dwyer, D.J., Hayete, B., Lawrence, C.A., and Collins, J.J. (2007). A common mechanism of cellular death induced by bactericidal antibiotics. Cell 130, 797–810.

Batty, D.P., and Wood, R.D. (2000). Damage recognition in nucleotide excision repair of DNA. Gene 241, 193–204.

Kohanski, M.A., Dwyer, D.J., and Collins, J.J. (2010). How antibiotics kill bacteria: from targets to networks. Nat. Rev. Microbiol. 8, 423–435.

Butland, G., Babu, M., Dı´az-Mejı´a, J.J., Bohdana, F., Phanse, S., Gold, B., Yang, W., Li, J., Gagarinova, A.G., Pogoutse, O., et al. (2008). eSGA: E. coli synthetic genetic array analysis. Nat. Methods 5, 789–795.

Kruse, T., Blagoev, B., Løbner-Olesen, A., Wachi, M., Sasaki, K., Iwai, N., Mann, M., and Gerdes, K. (2006). Actin homolog MreB and RNA polymerase interact and are both required for chromosome segregation in Escherichia coli. Genes Dev. 20, 113–124.

Calloni, G., Chen, T., Schermann, S.M., Chang, H.C., Genevaux, P., Agostini, F., Tartaglia, G.G., Hayer-Hartl, M., and Hartl, F.U. (2012). DnaK functions as a central hub in the E. coli chaperone network. Cell Rep. 1, 251–264. Campoy, S., Salvador, N., Corte´s, P., Erill, I., and Barbe´, J. (2005). Expression of canonical SOS genes is not under LexA repression in Bdellovibrio bacteriovorus. J. Bacteriol. 187, 5367–5375. Costanzo, M., Baryshnikova, A., Bellay, J., Kim, Y., Spear, E.D., Sevier, C.S., Ding, H., Koh, J.L., Toufighi, K., Mostafavi, S., et al. (2010). The genetic landscape of a cell. Science 327, 425–431. Deana, A., and Belasco, J.G. (2004). The function of RNase G in Escherichia coli is constrained by its amino and carboxyl termini. Mol. Microbiol. 51, 1205–1217. Dunin-Horkawicz, S., Feder, M., and Bujnicki, J.M. (2006). Phylogenomic analysis of the GIY-YIG nuclease superfamily. BMC Genomics 7, 98. Dwyer, D.J., Kohanski, M.A., Hayete, B., and Collins, J.J. (2007). Gyrase inhibitors induce an oxidative damage cellular death pathway in Escherichia coli. Mol. Syst. Biol. 3, 91. Dwyer, D.J., Belenky, P.A., Yang, J.H., MacDonald, I.C., Martell, J.D., Takahashi, N., Chan, C.T., Lobritz, M.A., Braff, D., Schwarz, E.G., et al. (2014). Antibiotics induce redox-related physiological alterations as part of their lethality. Proc. Natl. Acad. Sci. USA 111, E2100–E2109. Epshtein, V., Kamarthapu, V., McGary, K., Svetlov, V., Ueberheide, B., Proshkin, S., Mironov, A., and Nudler, E. (2014). UvrD facilitates DNA repair by pulling RNA polymerase backwards. Nature 505, 372–377. Foti, J.J., Devadoss, B., Winkler, J.A., Collins, J.J., and Walker, G.C. (2012). Oxidation of the guanine nucleotide pool underlies cell death by bactericidal antibiotics. Science 336, 315–319.

Lesterlin, C., Ball, G., Schermelleh, L., and Sherratt, D.J. (2014). RecA bundles mediate homology pairing between distant sisters during DNA break repair. Nature 506, 249–253. Majorek, K.A., Dunin-Horkawicz, S., Steczkiewicz, K., Muszewska, A., Nowotny, M., Ginalski, K., and Bujnicki, J.M. (2014). The RNase H-like superfamily: new members, comparative structural analysis and evolutionary classification. Nucleic Acids Res. 42, 4160–4179. Martin, H., Shales, M., Fernandez-Pin˜ar, P., Wei, P., Molina, M., Fiedler, D., Shokat, K.M., Beltrao, P., Lim, W., and Krogan, N.J. (2015). Differential genetic interactions of yeast stress response MAPK pathways. Mol. Syst. Biol. 11, 800. Masuda, H., Tan, Q., Awano, N., Wu, K.P., and Inouye, M. (2012). YeeU enhances the bundling of cytoskeletal polymers of MreB and FtsZ, antagonizing the CbtA (YeeV) toxicity in Escherichia coli. Mol. Microbiol. 84, 979–989. Mirkin, E.V., and Mirkin, S.M. (2007). Replication fork stalling at natural impediments. Microbiol. Mol. Biol. Rev. 71, 13–35. Modi, S.R., Camacho, D.M., Kohanski, M.A., Walker, G.C., and Collins, J.J. (2011). Functional characterization of bacterial sRNAs using a network biology approach. Proc. Natl. Acad. Sci. USA 108, 15522–15527. Musso, G., Costanzo, M., Huangfu, M., Smith, A.M., Paw, J., San Luis, B.J., Boone, C., Giaever, G., Nislow, C., Emili, A., and Zhang, Z. (2008). The extensive and condition-dependent nature of epistasis among whole-genome duplicates in yeast. Genome Res. 18, 1092–1099. Nichols, R.J., Sen, S., Choo, Y.J., Beltrao, P., Zietek, M., Chaba, R., Lee, S., Kazmierczak, K.M., Lee, K.J., Wong, A., et al. (2011). Phenotypic landscape of a bacterial cell. Cell 144, 143–156.

Grove, A. (2011). Functional evolution of bacterial histone-like HU proteins. Curr. Issues Mol. Biol. 13, 1–12.

Pellegrini, M., Marcotte, E.M., Thompson, M.J., Eisenberg, D., and Yeates, T.O. (1999). Assigning protein functions by comparative genome analysis: protein phylogenetic profiles. Proc. Natl. Acad. Sci. USA 96, 4285–4288.

Gue´nole´, A., Srivas, R., Vreeken, K., Wang, Z.Z., Wang, S., Krogan, N.J., Ideker, T., and van Attikum, H. (2013). Dissection of DNA damage responses using multiconditional genetic interaction maps. Mol. Cell 49, 346–358.

Peregrı´n-Alvarez, J.M., Xiong, X., Su, C., and Parkinson, J. (2009). The Modular Organization of Protein Interactions in Escherichia coli. PLoS Comput. Biol. 5, e1000523.

Hartwell, L.H., Hopfield, J.J., Leibler, S., and Murray, A.W. (1999). From molecular to modular cell biology. Nature 402 (6761, Suppl), C47–C52.

Ryan, C.J., Roguev, A., Patrick, K., Xu, J., Jahari, H., Tong, Z., Beltrao, P., Shales, M., Qu, H., Collins, S.R., et al. (2012). Hierarchical modularity and the evolution of genetic interactomes across species. Mol. Cell 46, 691–704.

Hoppe, B., Krger, K., Kahl, T., Arnstadt, T., Buscot, F., Bauhus, J., and Wubet, T. (2015). A pyrosequencing insight into sprawling bacterial diversity and community dynamics in decaying deadwood logs of Fagus sylvatica and Picea abies. Sci. Rep. 5, 9456. Hu, P., Janga, S.C., Babu, M., Dı´az-Mejı´a, J.J., Butland, G., Yang, W., Pogoutse, O., Guo, X., Phanse, S., Wong, P., et al. (2009). Global functional atlas of Escherichia coli encompassing previously uncharacterized proteins. PLoS Biol. 7, e96.

660 Cell Reports 14, 648–661, January 26, 2016 ª2016 The Authors

Schlacher, K., and Goodman, M.F. (2007). Lessons from 50 years of SOS DNA-damage-induced mutagenesis. Nat. Rev. Mol. Cell Biol. 8, 587–594. Shereda, R.D., Kozlov, A.G., Lohman, T.M., Cox, M.M., and Keck, J.L. (2008). SSB as an organizer/mobilizer of genome maintenance complexes. Crit. Rev. Biochem. Mol. Biol. 43, 289–318. Srivas, R., Costelloe, T., Carvunis, A.R., Sarkar, S., Malta, E., Sun, S.M., Pool, M., Licon, K., van Welsem, T., van Leeuwen, F., et al. (2013). A UV-induced

genetic network links the RSC complex to nucleotide excision repair and shows dose-dependent rewiring. Cell Rep. 5, 1714–1724.

Yamaguchi, Y., and Inouye, M. (2011). Regulation of growth and death in Escherichia coli by toxin-antitoxin systems. Nat. Rev. Microbiol. 9, 779–790.

Sutton, M.D., and Walker, G.C. (2001). Managing DNA polymerases: coordinating DNA replication, DNA repair, and DNA recombination. Proc. Natl. Acad. Sci. USA 98, 8342–8349.

Yeeles, J.T., Poli, J., Marians, K.J., and Pasero, P. (2013). Rescuing stalled or damaged replication forks. Cold Spring Harb. Perspect. Biol. 5, a012815.

Uphoff, S., and Kapanidis, A.N. (2014). Studying the organization of DNA repair by single-cell and single-molecule imaging. DNA Repair (Amst.) 20, 32–40. VanderSluis, B., Bellay, J., Musso, G., Costanzo, M., Papp, B., Vizeacoumar, F.J., Baryshnikova, A., Andrews, B., Boone, C., and Myers, C.L. (2010). Genetic interactions reveal the evolutionary trajectories of duplicate genes. Mol. Syst. Biol. 6, 429.

Zafar, M.A., Shah, I.M., and Wolf, R.E., Jr. (2010). Protein-protein interactions between sigma(70) region 4 of RNA polymerase and Escherichia coli SoxS, a transcription activator that functions by the prerecruitment mechanism: evidence for ‘‘off-DNA’’ and ‘‘on-DNA’’ interactions. J. Mol. Biol. 401, 13–32. , K., Petranovic , M., and Petranovic , D. (1999). ChromoZahradka, D., Vlahovic some segregation and cell division defects in recBC sbcBC ruvC mutants of Escherichia coli. J. Bacteriol. 181, 6179–6183.

Cell Reports 14, 648–661, January 26, 2016 ª2016 The Authors 661