Multiple Legionella Pneumophila Effector Virulence Phenotypes

Home , Legionella pneumophila

Multiple Legionella pneumophila effector virulence PNAS PLUS phenotypes revealed through high-throughput analysis of targeted mutant libraries

Stephanie R. Shamesa,1, Luying Liua, James C. Haveya, Whitman B. Schofielda,b, Andrew L. Goodmana,b, and Craig R. Roya,2

aDepartment of Microbial Pathogenesis, Yale University School of Medicine, New Haven, CT 06519; and bMicrobial Sciences Institute, Yale University School of Medicine, New Haven, CT 06519

Edited by Ralph R. Isberg, Howard Hughes Medical Institute/Tufts University School of Medicine, Boston, MA, and approved October 20, 2017 (received for review May 23, 2017) Legionella pneumophila is the causative agent of a severe pneu- poorly understood. Initial forward genetic screens aimed at identi- monia called Legionnaires’ disease. A single strain of L. pneumo- fying avirulent mutants of L. pneumophila were successful in identi- phila encodes a repertoire of over 300 different effector proteins fying essential components of the Dot/Icm system, but these screens that are delivered into host cells by the Dot/Icm type IV secretion did not identify effector proteins translocated by the Dot/Icm system system during infection. The large number of L. pneumophila ef- (10, 11). It is appreciated that most effectors are not essential for fectors has been a limiting factor in assessing the importance of intracellular replication (12), which is why the genes encoding ef- individual effectors for virulence. Here, a transposon insertion se- fector proteins that are important for virulence were difficult to quencing technology called INSeq was used to analyze replication identify by standard screening strategies that assess intracellular of a pool of effector mutants in parallel both in a mouse model of replication using binary assays that measure plaque formation or infection and in cultured host cells. Loss-of-function mutations in destruction of host cell monolayers (10, 13). Thus, new approaches genes encoding effector proteins resulted in host-specific or broad are required to systematically assess the contribution of indi- virulence phenotypes. Screen results were validated for several vidual L. pneumophila effector proteins during infection. effector mutants displaying different virulence phenotypes using High-throughput sequencing (HTS)-based phenotypic screen- MICROBIOLOGY genetic complementation studies and infection assays. Specifically, ing of bacterial transposon (Tn) mutants has become a powerful loss-of-function mutations in the gene encoding LegC4 resulted technique to assess the contribution of individual genes to bacte- in enhanced L. pneumophila in the lungs of infected mice but rial fitness during host colonization (14). Techniques such as in- not within cultured host cells, which indicates LegC4 augments sertion sequencing (INSeq) (15) and Tn sequencing (TNSeq) (16) bacterial clearance by the host immune system. The effector pro- are massively parallel HTS techniques that enable determination teins RavY and Lpg2505 were important for efficient replication of relative fitness of individual Tn mutants within a mixed pop- within both mammalian and protozoan hosts. Further analysis of ulation. These techniques have been used to generate whole- Lpg2505 revealed that this protein functions as a metaeffector genome mutant populations to identify genes that contribute to that counteracts host cytotoxicity displayed by the effector pro- virulence of several clinically important bacterial pathogens such tein SidI. Thus, this study identified a large cohort of effectors that as Campylobacter jejuni, Haemophilus influenza, Acinetobacter contribute to L. pneumophila virulence positively or negatively baumannii,andPseudomonas aeruginosa (15, 17–20). However, and has demonstrated regulation of effector protein activities by traditional whole-genome screening approaches are susceptible to cognate metaeffectors as being critical for host pathogenesis. Significance type IV secretion | transposon insertion sequencing | bacterial effectors

The contribution of individual effectors to Legionella pneumo- acteria of the genus Legionella are inhabitants of fresh water phila virulence has not been systematically examined. This study Band soil environments where they have evolved the capacity employed a parallel high-throughput transposon insertion se- to replicate in a diverse number of protozoan hosts. Although quencing technique called INSeq to probe the L. pneumophila there are over 40 species of Legionella, human infections that effector repertoire and identified multiple effectors that con- progress to a severe pneumonia called Legionnaires’ disease are ’ tribute to virulence in several host organisms, including an ani- most often caused by L. pneumophila (1). Legionnaires disease mal model of Legionnaires’ disease. Importantly, this study results from inhalation of Legionella-contaminated aerosols and demonstrates that effector proteins contribute to host virulence subsequent bacterial replication within alveolar macrophages. both positively and negatively by controlling intracellular repli- Bacterial replication occurs in a specialized Legionella-contain- cation and influencing host immune responses, which demon- ing vacuole (LCV) that evades fusion with lysosomes and asso- strates that the subtle alterations in the effector repertoire ciates intimately with the host endoplasmic reticulum (ER) (2, of a single L. pneumophila strain can greatly impact host 3). Formation of the LCV and intracellular bacterial replication pathogenicity. is dependent on the Dot/Icm type IV secretion system (T4SS) (4, 5), which translocates bacterial effector proteins into the host cell Author contributions: S.R.S., A.L.G., and C.R.R. designed research; S.R.S., L.L., and J.C.H. where they subvert normal host processes to promote pathogen performed research; W.B.S. and A.L.G. contributed new reagents/analytic tools; S.R.S., replication (6). The Philadelphia-1 strain of L. pneumophila was L.L., A.L.G., and C.R.R. analyzed data; and S.R.S. and C.R.R. wrote the paper. isolated from the eponymous Legionnaires’ disease outbreak that The authors declare no conflict of interest. occurred in 1976 (7), and this strain has been shown to encode This article is a PNAS Direct Submission. over 300 different effector proteins (8). Published under the PNAS license. Genome sequencing studies have demonstrated a high degree 1Present address: Division of Biology, Kansas State University, Manhattan, KS 66506. of plasticity in the effector repertoires encoded by between dif- 2To whom correspondence should be addressed. Email: [email protected]. ferent strains of L. pneumophila and different Legionella species This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10. (9). How the effector repertoire influences host virulence remains 1073/pnas.1708553114/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1708553114 PNAS Early Edition | 1of9 Downloaded by guest on September 25, 2021 population bottlenecks, which result in stochastic changes in mu- INSeq Analysis of the EMP Identifies Factors That Control L. pneumophila tant abundance that are unrelated to fitness (21). In most animal Virulence. A mouse model of Legionnaires’ disease was used to models of Legionnaires’ disease, these bottlenecks would likely determine whether individual mutants with known virulence occur using populations containing more than 1,000 different phenotypes could be identified after INSeq analysis of the mutants, which complicates using whole-genome INSeq ap- EMP. Specifically, the EMP was screened after intranasal in- − − proaches for assessing the contribution of effector proteins in oculation of C57BL/6 (WT) mice and NLRC4 / mice. INSeq host pathogenicity. was used to profile the output EMP population after 48 h of To circumvent the challenges associated with whole-genome infection, which is when replication of L. pneumophila peaks in mutant screening, INSeq technology was used to sequence an the lungs of mice (27). The distribution of mutants in the output arrayed L. pneumophila Tn mutant library and determine where populations was compared with the input populations (Fig. 1B individual Tn insertion mutants were located in the arrayed li- and Fig. S2). brary. From these data, mutants deficient in individual effector There was a significant increase in the proportion of flaA::Tn genes were clonally isolated to generate an effector mutant pool mutants in the lungs of C57BL/6 mice in the output population (EMP) that was used to assess the fitness of individual effector at 48 h, which indicated that the flagellin-deficient mutant had a mutants using both a mouse model of Legionnaires’ disease and competitive advantage over other mutants in the EMP (Fig. S2 cultured host cells. This systematic analysis revealed distinct and Dataset S3). By contrast, flaA::Tn mutants did not display a virulence phenotypes for individual effector mutants and a competitive advantage in NLRC4-deficient mice (Dataset S3). complex relationship between the L. pneumophila effector gene Thus, INSeq analysis of the EMP successfully determined the repertoire and host virulence. competitive advantage flaA::Tn mutants have in escaping flagellin- mediated activation of the NAIP5/NLRC4 inflammasome. Using Results C57BL/6 mice, there was a significant decrease in the output Generation of the L. pneumophila EMP. To produce a pool of population at 48 h for mutants with Tn insertions in genes en- L. pneumophila mutants where specific effector genes were coding essential Dot/Icm components (Fig. S2 and Dataset S3). inactivated, an arrayed Tn mutant library was generated and Tn The decrease in the proportion of Dot/Icm-deficient mutants in the insertion sites were mapped using INSeq technology. Briefly, output population collected from the lungs of NLRC4-deficient L. pneumophila mutants were generated using a Mariner-based mice at 48 h was more pronounced. This is because the NLRC4- Tn engineered to confer chloramphenicol resistance. Individual deficient mouse is a highly permissive host for L. pneumophila mutants were arrayed in 96-well plates until over 10,000 indi- intracellular replication, and the dramatic increase in the number vidual mutants were obtained. Combinatorial pooling and INSeq of replication-competent mutants in the output population will analysis were used to identify all Tn insertion sites and the lo- contribute to the difference in sequence reads compared with dot cation of each mutant in the arrayed library (22) (SI Materials and icm mutants that are replication-deficient (Fig. 1B and Dataset and Methods). S3). These data indicate that INSeq analysis of the EMP was able Sequencing results identified 10,163 independent insertion to detect mutants with intracellular replication defects and was events, and these data were used to determine the location of the sensitive enough to quantify the magnitude of the defects using mutants in the arrayed library. PCR analysis and phenotypic mice that varied in their permissiveness for L. pneumophila repli- analysis were used to validate the predicted location of multiple cation. Thus, using the EMP in conjunction with INSeq analysis mutants within the arrayed library (SI Materials and Methods). was successful at identifying and quantifying the predicted gene- Genome coverage did not appear to be affected by GC content, for-gene relationships between bacterial and host factors that and no insertions mapped to the lvh region because Lp01- control intracellular replication of L. pneumophila in the mouse derived strains of L. pneumophila Philadelphia-1 have a chro- model of Legionnaires’ disease. mosomal deletion that eliminated this locus (23) (Fig. 1A). Tn insertions were identified in the coding regions of 297 of the INSeq Analysis Reveals Effector Mutant Virulence Phenotypes. To 315 genes predicted to encode effector proteins (Fig. 1A and determine the contribution of individual effectors to L. pneu- Dataset S1). An EMP was generated by isolating individual ef- mophila virulence, the INSeq data obtained from the analysis of fector mutants from the arrayed library and combining them into the EMP in mice was compared with INSeq data obtained by ex a single pool. Several mutants having Tn insertions in genes vivo passaging of the EMP through bone marrow-derived mac- encoding Dot/Icm secretion system components were added to rophages (BMDMs) from NLRC4-deficient mice and the pro- the EMP to serve as internal controls for mutants that should tozoan host Acanthamoeba castellanii (Fig. 1 C and D). Mutants display severe fitness defects in both the mouse model of in- were defined as having significant fitness differences based on fection and in cultured host cells. The EMP also contained statistical analysis of INSeq data (q ≤ 0.05; see SI Materials and mutants in the flaA gene, which are deficient in production of Methods). Hierarchical clustering analysis was used to assign flagellin (Dataset S2). The flaA::Tn mutants served as controls individual mutants to distinct categories (class I–V) based on for bacteria capable of escaping detection of flagellin by the host phenotypes displayed in the cell culture and in vivo screens (Fig. NAIP5/NLRC4 inflammasome, which restricts intracellular rep- 1E). In addition to mutants deficient in the Dot/Icm system, lication of L. pneumophila in mouse macrophages (24–26). multiple effector mutants were included in the class I category, When possible, two different insertion mutants deficient in an which represents mutants that display intracellular replication effector were added to the EMP so that fitness changes identi- defects in both mammalian and amoeba hosts (Fig. 1E). In- fied by INSeq analysis could be validated independently. cluded in this group were mutants having Tn insertions in mavN The final EMP contained 528 isogenic L. pneumophila strains and sdhA, which encode effectors that have been shown pre- having independent Tn insertions that could be quantitated in viously to be important for L. pneumophila intracellular repli- parallel using INSeq. This was demonstrated by culturing the cation (28–32) (Fig. 1E and Dataset S3). Identification of these EMP axenically and sequencing the input and output pop- mutants in the INSeq analysis provided further evidence that this ulations. These data show the distribution of each mutant in the technology was successful in evaluating the role of effectors in library did not change significantly after axenic cultivation (Fig. virulence. Importantly, virulence phenotypes were detected for a S1). Thus, any changes in the distribution of a mutant following large number of uncharacterized effector mutants, which sug- cultivation in either the mouse model of infection or in cultured gested that INSeq analysis identified fitness defects displayed by host cells would indicate that the fitness of the mutant has been effector mutants that were not discovered in traditional forward altered by the host environment. genetic screens. The high-confidence hits shown in Fig. 1E,

2of9 | www.pnas.org/cgi/doi/10.1073/pnas.1708553114 Shames et al. Downloaded by guest on September 25, 2021 which represent genes where all of the insertions in that partic- responding effector protein. A virulence phenotype of particular PNAS PLUS ular gene in the EMP displayed the phenotype of interest, interest was that displayed by the two independent legC4::Tn revealed several different categories of virulence phenotypes. mutants, where these mutants had a significant fitness advantage − − These included effector mutants that displayed fitness differ- in the lungs of NLRC4 / mice but not in A. castellanii or ences in either mammalian or protozoan hosts as well as effector BMDMs (Fig. 1E and Dataset S3). These data suggested that mutants that displayed a virulence phenotype only in the context LegC4 function might decrease L. pneumophila virulence in the of the Legionnaires’ disease pulmonary infection model. context of a cellular immune response in the lung. Indeed, an isolated legC4::Tn mutant replicated to higher levels (sixfold) in − − LegC4 Enhances Pulmonary Clearance of L. pneumophila. Validation the lungs of NLRC4 / mice compared with the parental strain studies were conducted to test whether virulence phenotypes of L. pneumophila (Fig. 2A, P < 0.01). A strain containing a clean observed by INSeq analysis of individual mutants in the EMP deletion of legC4 (ΔlegC4) displayed a similar increase in the − − were the result of a loss-of-function mutation affecting the cor- bacterial burden in the lungs of NLRC4 / mice compared with MICROBIOLOGY

Fig. 1. Generation and screening of a Dot/Icm EMP. (A) Schematic diagram of the L. pneumophila genome (black). Arrayed library Tn insertions (orange) and Dot/Icm effector Tn insertions (blue) are indicated by tick marks. The lvh locus (cyan) and dot/icm loci (red) are shown. GC skewing (G−C/G+C) is shown in purple (>0) and mustard (<0). (B–D) Average normalized read counts for each insertion from INSeq analysis from input and output EMP libraries obtained − − − − from NLRC4 / mice (n = 10) (B), NLRC4 / BMDMs (n = 8) (C), and A. castellanii (n = 8) (D). Tn mutants with significant fitness differences are shown in red (q ≤ 0.05), dot/icm::Tn mutants are blue, lpg2505::Tn mutants are green, ravY::Tn mutants are cyan, and legC4::Tn mutants are orange. (E) Heat map showing log − − − − output:input ratios >1 (red), <1 (blue), and nonsignificant (white) of mutants with significant (q ≤ 0.05) fitness differences in NLRC4 / lung, NLRC4 / BMDMs, and A. castellanii. Mutants were categorized into universal defect (I), mammalian-specific defect (II), lung-specific phenotypes (III), BMDM-specific phenotypes (IV), and amoeba-specific phenotypes (V). aGenes with only one representative mutant in the EMP. Statistical analyses were performed as described in SI Materials and Methods.

Shames et al. PNAS Early Edition | 3of9 Downloaded by guest on September 25, 2021 wild-type L. pneumophila (Fig. 2A, P < 0.01). Competitive index such as neutrophils. Interleukin (IL)-12 is a proinflammatory (CI) studies, where mice are infected with a 1:1 ratio of mutant cytokine that promotes clearance of L. pneumophila in vivo but and wild-type L. pneumophila, also revealed that the legC4::Tn does not contribute to cell-autonomous defense when macro- mutant had a fitness advantage (CI > 1) over the parental strain phages are cultured ex vivo (34). To determine if LegC4 function of L. pneumophila in vivo (Fig. 2B). Importantly, complemen- may influence a proinflammatory response by infected cells, tation of the legC4::Tn mutation with a plasmid-encoded allele of BMDMs cultured ex vivo were infected with wild-type or legC4 legC4 abrogated the fitness advantage displayed by the legC4::Tn mutant L. pneumophila, and IL-12 levels were measured. Despite mutant. Indeed, this legC4::Tn mutant strain overproducing similar levels of intracellular replication, significantly less IL-12 LegC4 from a plasmid displayed a fitness defect compared with was secreted by macrophages infected with the legC4::Tn mutant the nonmutant strain containing the vector alone (Fig. 2C). compared with BMDMs infected with wild-type L. pneumophila Thus, expression of legC4 attenuates L. pneumophila virulence in (Fig. 2D, P < 0.01). IL-12 production was similar when BMDMs a mouse model of Legionnaires’ disease. infected with the complemented legC4::Tn (plegC4)strainwere Intracellular replication of legC4 mutant bacteria was assessed compared with wild-type L. pneumophila (Fig. 2D). Thus, using isolated BMDMs and A. castellanii to determine if L. pneumophila producing a functional LegC4 protein may be LegC4 function stimulates a cell-autonomous immune defense less virulent for animals because this effector has the capacity to pathway capable of restricting L. pneumophila intracellular enhance a proinflammatory response that stimulates cellular replication. Intracellular replication of the legC4::Tn mutant immunity. over 72 h was similar to the parental strain of L. pneumophila in − − NLRC4 / BMDMs (Fig. S3A). In A. castellanii, the legC4::Tn The Effectors Lpg0086, Lpg2505, and RavY Are Important for mutant displayed a modest defect in replication at 24 h, but no L. pneumophila Intracellular Replication. Tn insertions in several defect in replication was detected at later times (Fig. S3B). Thus, uncharacterized effector genes resulted in decreased L. pneumophila LegC4 function does not attenuate replication of L. pneumophila intracellular replication. Loss-of-function mutations in the genes in isolated host cells, which suggests that LegC4 is detrimental to lpg0086, lpg2505, and ravY resulted in fitness defects in BMDMs L. pneumophila virulence only in the context of animal infection. cultured ex vivo (Fig. 1E, class I and III), which suggested these Proinflammatory cytokines stimulate clearance of L. pneu- mutants may have intracellular replication defects. Additionally, mophila in vivo (33). Thus, the in vivo-specific phenotype ob- lpg2505::Tn and ravY::Tn strains displayed growth defects in served for legC4 mutant bacteria could result from enhanced A. castellanii and the mouse model of Legionnaires’ disease. In- production of inflammatory cytokines from infected macro- tracellular replication of lpg0086::Tn, lpg2505::Tn, and ravY::Tn phages that could enhance clearance by bystander immune cells mutants was measured in both BMDMs and A. castellanii.

− − Fig. 2. LegC4 function attenuates L. pneumophila replication in the mouse lung. (A) Enumeration of WT, legC4::Tn, or ΔlegC4 from lungs of NLRC4 / mice. (B)CIoflegC4::Tn versus WT from the lungs of NLRC4−/− mice. (C)CIoflegC4::Tn (pV) or legC4::Tn (plegC4) versus WT (pV) in the lungs of NLRC4−/− mice. Each symbol represents an individual animal, and asterisks denote statistical significance by Mann–Whitney U test (*P < 0.05, **P < 0.01). (D) ELISA for IL- − − 12 p40 secretion from NLRC4 / BMDMs infected with the indicated strains. Data are presented as mean ± SD, and asterisks denote statistical significance (**P < 0.01; n.s., not significant). Data are representative of at least two independent experiments.

4of9 | www.pnas.org/cgi/doi/10.1073/pnas.1708553114 Shames et al. Downloaded by guest on September 25, 2021 Consistent with the class I phenotype observed in the EMP was introduced in trans to the respective chromosomal deletion PNAS PLUS screens, the isolated lpg2505::Tn and ravY::Tn mutants were at- mutant (Fig. 3 B and C and Fig. S4 C and D). tenuated for intracellular replication in BMDMs and A. castellanii CI studies confirmed the fitness defects displayed by the (Fig. 3A and Fig. S4 A and B). The lpg0086::Tn mutant had a lpg2505::Tn and ravY::Tn mutants in the INSeq analysis. Spe- significant intracellular replication defect in BMDMs, and in- cifically, the parental strain of L. pneumophila outcompeted the tracellular replication of lpg0086::Tn was enhanced signifi- Δlpg2505 mutant and the ΔravY mutant, but not the complemented − − cantly upon introduction of the wild-type lpg0086 gene on a strains, in the lungs of NLRC4 / mice (Fig. 3 D and E). Thus, plasmid (Fig. S5). By contrast, the lpg0086::Tn mutant did not RavY and Lpg2505 are effectors important for L. pneumophila have an intracellular replication defect in A. castellanii or a intracellular replication and host virulence, further indicat- fitness defect in mice, which validates the class III phenotype ing that INSeq screening of the EMP was successful in displayed by this mutant (Fig. S5). After 72 h of infection, identifying novel L. pneumophila effectors that contribute to L. pneumophila strains mutated in lpg2505 or ravY displayed host pathogenesis. roughly a 2-log decrease in cfu counts compared with the isogenic wild-type strain (Fig. 3). In-frame deletions in lpg2505 or Lpg2505 Is a Metaeffector That Regulates SidI-Mediated Cytotoxicity. ravY were created in the parental strain of L. pneumophila to Bioinformatic analysis indicated that lpg2505 is encoded down- further validate the growth defects observed for the Tn insertion stream of the effector gene sidI in a predicted operon (35, 36) mutants. Indeed, the Δlpg2505 and ΔravY mutants displayed in- (Fig. 4A). SidI inhibits host cell protein synthesis and is toxic tracellular replication defects, which were complemented when when produced ectopically in eukaryotic cells (37). Several the plasmid-encoded allele encoding wild-type effector protein L. pneumophila effectors have been characterized that have MICROBIOLOGY

Fig. 3. Lpg2505 and RavY are effectors important for intracellular replication. (A) Growth of WT, lpg2505::Tn, ravY::Tn, and dotA::Tn mutant strains in NLRC4−/− BMDM. (B) Growth of WT, Δlpg2505 (pV), Δlpg2505 (plpg2505), and dotA::Tn in NLRC4−/− BMDM. (C) Growth of WT ΔravY (pV), ΔravY (pravY), or − − dotA::Tn in NLRC4 / BMDM. Asterisks denote statistical significance by Student’s t test (**P < 0.01). (D)CIofΔlpg2505 (pV) or Δlpg2505 (plpg2505) versus WT − − − − in the lungs of NLRC4 / mice at 48 h postinfection. (E) CI of WT versus ΔravY (pV) or ΔravY (pravY) in the lungs of NLRC4 / mice at 48 h postinfection. Each point represents a single mouse and data shown are mean ± SD. Asterisks denote statistical significance by Mann–Whitney U test (**P < 0.01). Data are representative of at least two independent experiments.

Shames et al. PNAS Early Edition | 5of9 Downloaded by guest on September 25, 2021 biochemical functions that modulate the activity of other effec- defect (Fig. 4C), which indicates that Lpg2505 is important for tor proteins after delivery of the proteins into the host cell (38, virulence in strains producing a cytotoxic SidI protein. Lastly, the 39). These modulating effectors have been called metaeffectors. ability of Lpg2505 to suppress the toxic activity of SidI was tested in Because many characterized L. pneumophila metaeffectors are yeast. Consistent with previous studies (37), expression of a func- encoded near the effector protein they modulate (38), the possi- tional SidI protein was toxic to yeast and interfered with colony bility that Lpg2505 functions as an metaeffector that regulates SidI formationonagarplates(Fig.4D). SidI toxicity was suppressed function was tested. Consistent with Lpg2505 being an effector when Lpg2505 was coexpressed in yeast, but expression of other that could modulate SidI-mediated toxicity, the sidI::Tn mutants in effector proteins encoded nearby did not suppress SidI cytotoxicity the EMP did not display a fitness defect in the INSeq analysis even (Fig. 4D). Thus, Lpg2505 regulates SidI activity to prevent host though these mutants were likely to be defective in lpg2505 ex- damage that is detrimental to L. pneumophila intracellular in- pression. In addition, previous studies demonstrated that a loss-of- fection and limits virulence. function mutation in the sidI gene did not affect L. pneumophila replication in cultured host cells (37). This suggested that lpg2505 Discussion might be essential for intracellular replication only if a functional This study utilized INSeq technology to generate and screen a SidI protein was produced. targeted pool of effector mutants for virulence phenotypes in To determine if the virulence defect displayed by the Δlpg2505 cultured host cells and the mouse model of Legionnaires’ dis- mutant required SidI function, intracellular replication of the ease. Compared with whole-genome screening approaches, the Δlpg2505 mutant was compared with an isogenic mutant where use of a targeted mutant library decreased the chances of pop- both sidI and lpg2505 were deleted (Δoperon). Indeed, the Δoperon mutant did not have a detectible intracellular replica- ulation bottlenecks and enabled identification of mutants with tion defect, which indicates the elimination of sidI suppressed significant virulence phenotypes. Importantly, loss-of-function the intracellular replication defect resulting from a lpg2505 loss- mutations in multiple effector genes resulted in high-confidence of-function mutation (Fig. 4B). hits where significant fitness differences in at least one of the host screens was observed for all of the individual mutants in the EMP. The mutant SidIR453P protein has a single amino acid sub- stitution that eliminates the cytotoxic effect of SidI on eukaryotic This screen also identified a large collection of effector genes host cells (37). To address whether Lpg2505 function is important where one of two independent Tn insertions displayed significant for neutralizing SidI-mediated cytotoxicity during L. pneumophila fitness phenotypes (Dataset S3). Thus, INSeq analysis identified infection, allelic exchange was used to replace the wild-type sidI effector mutants with replication and virulence phenotypes that gene with the mutant sidIR453P allele in the Δlpg2505 mutant. The were not easily detected using more traditional screening methods sidIR453P, Δlpg2505 mutant did not have an intracellular replication for intracellular replication.

Fig. 4. Lpg2505 is a metaeffector that inhibits SidI toxicity. (A) Schematic diagram of the putative operon encoding sidI and lpg2505.(B) Growth of WT, −/− −/− Δlpg2505, Δoperon, and dotA::Tn in NLRC4 BMDM over 72 h. (C) Growth of WT, Δlpg2505, Δoperon::sidIR453P, and dotA::Tn in NLRC4 BMDM over 72 h. Data shown are mean ± SD, and asterisks denote statistical significance by Student’s t test (**P < 0.05). (D) Yeast expressing vector, lpg2498, lpg2505, lpg2508, or lpg2509 were transformed with vector, sidI, lpg2505, lpg2508, or left untransformed, and growth was assessed on selective media. Data are representative of at least two independent experiments.

6of9 | www.pnas.org/cgi/doi/10.1073/pnas.1708553114 Shames et al. Downloaded by guest on September 25, 2021 Many of the previous screens to identify genes important for other L. pneumophila strains in the mouse model of infection. PNAS PLUS L. pneumophila virulence have used death of the host cell as an The observation that the legC4 mutant did not have a replication indicator of intracellular replication (10, 13), which is problem- advantage in cultured macrophages but did replicate to higher atic because mutants that replicate less efficiently but still cause levels in the lungs of mice infected indicates that this mutant is host toxicity would elude detection. By contrast, sequence reads not simply a “cheater” receiving a benefit from other L. pneu- provided by this INSeq approach correlate with the intracellular mophila in the EMP. The modest replication defect observed for replication capacity of each mutant in the pool so that effector the legC4 mutant in A. castellanii implicates this effector as mutants having intracellular replication defects that were not as having a beneficial role during L. pneumophila infection of severe as the defects displayed by dot or icm mutants were protozoan hosts in their natural environment, but the activity of identified. Additional studies will be necessary to validate the LegC4 is detrimental when these bacteria enter the lungs of the virulence phenotypes associated with the uncharacterized effec- mammalian host. The observation of decreased IL-12 produc- tor mutants listed in Fig. 1E and Dataset S3. For example, tion by macrophages infected with the legC4 mutant suggests lpg1751::Tn mutants and lpg1752::Tn mutants both displayed that LegC4 function may enhance innate immune detection of intracellular growth defects in BMDMs, but these two genes are L. pneumophila, which could accelerate clearance through the predicted to be encoded in an operon. Thus, it is possible that recruitment of neutrophils to the infected area and/or enhance the phenotype displayed by the lpg1751::Tn mutants results from clearance through IFNγ production by natural killer (NK) cells polar effects on the expression of lpg1752. There were also ex- (40, 41). As L. pneumophila is an accidental pathogen of humans amples, such as the lpg0717::Tn mutant, where the effector and rarely transmits person-to-person, effectors with functions mutant added to the EMP had multiple insertions so that the that attenuate bacterial replication in the context of a cellular virulence phenotype displayed by these mutants requires addi- immune response have likely not been selected against over time. tional validation to ensure that the phenotype is the result of an Validation of the legC4::Tn mutant phenotype revealed by INSeq effector loss-of-function. analysis demonstrates the utility of this approach in identifying Although most of the effector mutants in the EMP that were virulence phenotypes displayed by effector-deficient L. pneu- defective for intracellular replication in cell culture also dis- mophila during infection of a mammalian host. − − played fitness defects in NLRC4 / mice, there were exceptions INSeq analysis revealed that previously uncharacterized effec- where replication defects were detected using cultured host cells tors, such as RavY and Lpg2505, were essential for L. pneumo- but these defects were not apparent in the animal model of phila virulence. Interestingly, RavY is an effector that is highly

disease (Fig. 1E), which was a phenotype displayed by the conserved in all strains of L. pneumophila but does not appear to MICROBIOLOGY lpg0086::Tn mutant. This likely reflects a difference in sensitivity be encoded in many other Legionella species (9). Thus, RavY is a between the two assays. Using cultured host cells, it is typical to candidate virulence factor that contributes to the enhanced viru- observe a 3-log increase in bacterial numbers resulting from in- lence potential displayed by L. pneumophila for humans. tracellular replication over a 72-h assay, whereas in the mouse The metaeffector paradigm, where the primary function of an model of disease the increase in bacterial numbers is typically effector is to counteract the specific activity of another effector only 1-log because innate immune defenses that limit bacterial after both have been translocated into the host cell, was estab- replication and dissemination in the lung are activated within the lished for L. pneumophila when the function of the ubiquitin ligase first 24 h of infection (Dataset S3). Thus, lower levels of ex- LubX was shown to target the effector protein SidH for degra- pansion of the EMP in the lungs of infected mice likely prevent dation by the host proteasome (39). In this example, the meta- detection of virulence defects that are caused by subtle differ- effector interacts directly with the effector it modulates. There are ences in the levels of intracellular replication. also metaeffectors that do not interact directly with other effectors There were significant advantages to screening the EMP in a but instead balance the activity of the effector by reversing a mouse model of Legionnaires’ disease. Testing 300 different modification to a host protein mediated by that effector. An ex- effector mutants of Legionella individually using an animal ample is the de-AMPylation activity displayed by SidD, which model of disease is impractical, which is why there has not been a reverses DrrA(SidM)-mediated AMPylation of the host protein comprehensive analysis of the role individual effectors play in Rab1 (42, 43). It is now appreciated that metaeffectors are used disease. In this study, it was demonstrated that INSeq screening frequently to modulate the activity of other effectors in the host of the EMP in a mouse model of Legionnaires’ disease enabled cell (31, 39, 42–45) and a recent study that used yeast to identify the fitness of over 500 different mutants to be measured in- effector two-hybrid interactions and suppression of host cytotox- dividually in parallel in a single animal. Although INSeq tech- icity identified a large cohort of effector–metaeffector pairs (38). nology was used here to address the fitness of mutants deficient The metaeffector SidJ has also been implicated in virulence, and in Dot/Icm-translocated effector proteins, insertion site mapping recent data indicate that SidJ reverses phosphoribosyl-linked data obtained from the arrayed library will now enable the rapid protein ubiquitination of host proteins mediated by the SidE construction of other mutant pools that could be used to eval- family of effectors (46, 47). The SidI–Lpg2505 effector–meta- uate virulence phenotypes for L. pneumophila defective in other effector pair that was revealed here using INSeq screening of the pathways, such as mutants defective in type II secreted proteins EMP, however, eluded detection in these previous studies, which or mutants defective in metabolic pathways that could be im- provides additional support for the power of this unbiased ap- portant for intracellular survival. proach in revealing new effector activities. Another advantage of using INSeq technology to screen the In summary, INSeq analysis of a targeted L. pneumophila Tn EMP in an animal model of disease was that fitness advantage or mutant library provided a quantitative assessment of fitness de- cost of an individual effector could be assessed in the context of a fects resulting from inactivation of potential virulence determi- complex cellular immune response. Accordingly, there were nants. Using a targeted EMP in conjunction with screens in both categories of effector mutants that had fitness defects or fitness animal and cultured host cells, it was revealed that inactivation advantages that were revealed only in the mouse model of Le- of a single effector has the potential to influence virulence of gionnaires’ disease. The legC4::Tn mutant was selected for fur- L. pneumophila both positively and negatively. Thus, addressing ther analysis to validate an effector mutant virulence phenotype the role of effectors through systematic analysis of mutant phe- that was revealed only in the animal infection model. In- notypes revealed a complex interrelationship of effector activi- dependent Tn insertion mutants and a strain containing a clean ties that ultimately impact the ability of L. pneumophila to cause deletion of the legC4 gene confirmed that L. pneumophila de- disease. Determining the biochemical function of effectors with ficient in the effector LegC4 had a competitive advantage over virulence phenotypes that were revealed in this analysis will

Shames et al. PNAS Early Edition | 7of9 Downloaded by guest on September 25, 2021 further our understanding of how L. pneumophila is able to placed in a unique subset of pools (22). INSeq libraries were generated from promote human disease and how effector proteins contribute to each of the 24 pools using a different 6-bp barcode for each pool as de- innate immune control of L. pneumophila in healthy hosts. scribed (22). Adapter sequences are listed in Dataset S4. Adapter-ligated li- Lastly, INSeq mapping data obtained for the arrayed library can braries were normalized to 10 nM, and equal volumes were combined before sequencing on an Illumina HiSeq 2500 (Yale Center for Genome be used to assemble additional libraries to address other func- Analysis, Yale University). Data were analyzed via a data analysis package tional classes of genes in L. pneumophila, which include genes described previously and modified to align reads to the L. pneumophila regulating metabolic pathways or additional secretion pathways. Philadelphia-1 genome (22). A total of 10,164 mutants were mapped to wells of the library (Dataset S1), and mapping was validated by PCR and NaCl Materials and Methods sensitivity and found to be accurate (10). Bacterial Strains and Culture Conditions. L. pneumophila serogroup-1 strains were cultured on supplemented charcoal N-(2-Acetamido)-2-aminoethanesulfonic Generation of the EMP. The EMP was generated by isolation and subsequent acid (ACES)-buffered yeast extract (CYE) and grown at 37 °C as described (48, pooling of individual clones from the arrayed library as described in SI Ma- 49). Liquid cultures were grown at 37 °C in supplemented ACES-buffered yeast terials and Methods. extract (AYE) as described. All L. pneumophila were grown in the presence of μ · −1 − − 100 g mL streptomycin. When necessary, media were supplemented with Production of INSeq Libraries. BMDMs from C57BL/6 NLRC4 / mice and μ · −1 μ · −1 5 g mL chloramphenicol (Tn mutants), 10 g mL chloramphenicol (plasmid A. castellanii were derived and cultured as described in SI Materials and μ · −1 μ · −1 maintenance), 10 g mL kanamycin (allelic exchange), or 25 g mL kana- Methods. Infections were performed by growing the EMP on CYE media for mycin (plasmid maintenance). Plasmids used in this study are listed in Table S1. 72 h. Bacteria were scraped from the plate and grown overnight in liquid For molecular cloning and strain construction, please refer to SI Materials AYE media until the culture reached OD600 = 3.2–3.8. A sample of this in- and Methods. oculum was plated into CYE and used to generate input library gDNA. BMDMs and A. castellanii were infected as described in SI Materials and Mice. Six-week-old female C57BL/6 mice were purchased from the Jackson Methods. Bacteria were harvested and plated on CYE from cell lysates at −/− Laboratories. C57BL/6 NLRC4 mice were obtained from Richard Flavell, 48 h postinfection and used to generate output library gDNA. INSeq library Yale School of Medicine, New Haven, CT. All procedures using mice were preparation and data analysis were performed as described (15, 22) (SI performed with approval of the Yale Institutional Animal Use and Care Materials and Methods). − − Committee and in accordance with the Animal Welfare Act. Six- to 8-wk-old female NLRC4 / or wild-type C57BL/6 mice were infected with the EMP for INSeq analysis. The EMP was prepared as described above, Arrayed Tn Mutant Library Production. Electrocompetent L. pneumophila and mice were infected by the intranasal route at 5 × 106 bacteria per ani- were transformed with pSRS_CM1, plated on selective media, and grown at mal. Input libraries were prepared as described above, and output libraries 37 °C. pSRS_CM1 encodes an MmeI-modified Mariner Tn and requires pir were prepared from the lungs of mice following 48 h of infection (SI Ma- from phage lambda (λpir) to replicate. Since L. pneumophila lacks λpir, any terials and Methods). INSeq library preparation and data analysis were resistant colonies resulted from incorporation of the Tn into the chromo- performed as described (15, 22) (SI Materials and Methods). some, which was confirmed by Southern blot analysis. Single colonies of Please refer to SI Materials and Methods for more information regarding chloramphenicol-resistant L. pneumophila were patched on 48-well agar infection of mice, BMDM and A. castellanii growth curves, ELISA, yeast ex- plates and grown at 37 °C for 48 h. Chloramphenicol-resistant bacteria were periments, and statistical analysis. resuspended in sterile d2H2O, transferred into a sterile 96-well plate containing 2× freezing medium (4% peptone, 10% glycerol), and stored at −70 °C. This − − ACKNOWLEDGMENTS. We thank Dr. Richard Flavell for providing NLRC4 / procedure was repeated until 106 96-well plates were stocked. mice and Aline Gozzi for technical assistance in library preparation. This The library of L. pneumophila Tn mutants was arrayed as previously de- work was supported by NIH Grants AI041699 and AI048770 (to C.R.R.) and scribed (22). An EpMotion 5075 robot was used as previously described with GM118159 (to A.L.G.), a CIHR postdoctoral fellowship (to S.R.S.), the China indicated programs to generate 24 pools of strains, in which each strain was Scholarship Council (L.L.), and the Burroughs Wellcome Fund (A.L.G.).

1. Newton HJ, Ang DKY, van Driel IR, Hartland EL (2010) Molecular pathogenesis of 16. van Opijnen T, Bodi KL, Camilli A (2009) Tn-seq: High-throughput parallel sequencing infections caused by Legionella pneumophila. Clin Microbiol Rev 23:274–298. for fitness and genetic interaction studies in microorganisms. Nat Methods 6:767–772. 2. Swanson MS, Isberg RR (1995) Association of Legionella pneumophila with the 17. Skurnik D, et al. (2013) A comprehensive analysis of in vitro and in vivo genetic fitness macrophage endoplasmic reticulum. Infect Immun 63:3609–3620. of Pseudomonas aeruginosa using high-throughput sequencing of transposon li- 3. Horwitz MA (1983) Formation of a novel phagosome by the Legionnaires’ disease braries. PLoS Pathog 9:e1003582. bacterium (Legionella pneumophila) in human monocytes. J Exp Med 158:1319–1331. 18. Wang N, Ozer EA, Mandel MJ, Hauser AR (2014) Genome-wide identification of 4. Vogel JP, Andrews HL, Wong SK, Isberg RR (1998) Conjugative transfer by the viru- Acinetobacter baumannii genes necessary for persistence in the lung. MBio 5: lence system of Legionella pneumophila. Science 279:873–876. e01163–e14. 5. Segal G, Purcell M, Shuman HA (1998) Host cell killing and bacterial conjugation re- 19. Gao B, Lara-Tejero M, Lefebre M, Goodman AL, Galán JE (2014) Novel components of – quire overlapping sets of genes within a 22-kb region of the Legionella pneumophila the flagellar system in epsilonproteobacteria. MBio 5:e01349 e14. genome. Proc Natl Acad Sci USA 95:1669–1674. 20. Wong SM, Bernui M, Shen H, Akerley BJ (2013) Genome-wide fitness profiling reveals 6. Nagai H, Kagan JC, Zhu X, Kahn RA, Roy CR (2002) A bacterial guanine nucleotide adaptations required by Haemophilus in coinfection with influenza A virus in the murine lung. Proc Natl Acad Sci USA 110:15413–15418. exchange factor activates ARF on Legionella phagosomes. Science 295:679–682. 7. Fraser DW, et al. (1977) Legionnaires’ disease: Description of an epidemic of pneu- 21. Abel S, Abel zur Wiesch P, Davis BM, Waldor MK (2015) Analysis of bottlenecks in experimental models of infection. PLoS Pathog 11:e1004823. monia. N Engl J Med 297:1189–1197. 22. Goodman AL, Wu M, Gordon JI (2011) Identifying microbial fitness determinants by 8. Burstein D, et al. (2009) Genome-scale identification of Legionella pneumophila ef- insertion sequencing using genome-wide transposon mutant libraries. Nat Protoc 6: fectors using a machine learning approach. PLoS Pathog 5:e1000508. 1969–1980. 9. Burstein D, et al. (2016) Genomic analysis of 38 Legionella species identifies large and 23. Samrakandi MM, Cirillo SLG, Ridenour DA, Bermudez LE, Cirillo JD (2002) Genetic and diverse effector repertoires. Nat Genet 48:167–175. phenotypic differences between Legionella pneumophila strains. J Clin Microbiol 40: 10. Sadosky AB, Wiater LA, Shuman HA (1993) Identification of Legionella pneumophila 1352–1362. genes required for growth within and killing of human macrophages. Infect Immun 24. Zamboni DS, et al. (2006) The Birc1e cytosolic pattern-recognition receptor contrib- 61:5361–5373. utes to the detection and control of Legionella pneumophila infection. Nat Immunol 11. Berger KH, Isberg RR (1993) Two distinct defects in intracellular growth com- 7:318–325. – plemented by a single genetic locus in Legionella pneumophila. Mol Microbiol 7:7 19. 25. Molofsky AB, et al. (2006) Cytosolic recognition of flagellin by mouse macrophages ’ 12. O Connor TJ, Adepoju Y, Boyd D, Isberg RR (2011) Minimization of the Legionella restricts Legionella pneumophila infection. J Exp Med 203:1093–1104. pneumophila genome reveals chromosomal regions involved in host range expan- 26. Ren T, Zamboni DS, Roy CR, Dietrich WF, Vance RE (2006) Flagellin-deficient Le- sion. Proc Natl Acad Sci USA 108:14733–14740. gionella mutants evade caspase-1- and Naip5-mediated macrophage immunity. PLoS 13. Andrews HL, Vogel JP, Isberg RR (1998) Identification of linked Legionella pneumo- Pathog 2:e18. phila genes essential for intracellular growth and evasion of the endocytic pathway. 27. Mascarenhas DPA, Pereira MSF, Manin GZ, Hori JI, Zamboni DS (2015) Interleukin Infect Immun 66:950–958. 1 receptor-driven neutrophil recruitment accounts to MyD88-dependent pulmonary 14. van Opijnen T, Camilli A (2013) Transposon insertion sequencing: A new tool for clearance of Legionella pneumophila infection in vivo. J Infect Dis 211:322–330. systems-level analysis of microorganisms. Nat Rev Microbiol 11:435–442. 28. Laguna RK, Creasey EA, Li Z, Valtz N, Isberg RR (2006) A Legionella pneumophila- 15. Goodman AL, et al. (2009) Identifying genetic determinants needed to establish a translocated substrate that is required for growth within macrophages and pro- human gut symbiont in its habitat. Cell Host Microbe 6:279–289. tection from host cell death. Proc Natl Acad Sci USA 103:18745–18750.

8of9 | www.pnas.org/cgi/doi/10.1073/pnas.1708553114 Shames et al. Downloaded by guest on September 25, 2021 29. Isaac DT, Laguna RK, Valtz N, Isberg RR (2015) MavN is a Legionella pneumophila 43. Neunuebel MR, et al. (2011) De-AMPylation of the small GTPase Rab1 by the path- PNAS PLUS vacuole-associated protein required for efficient iron acquisition during intracellular ogen Legionella pneumophila. Science 333:453–456. growth. Proc Natl Acad Sci USA 112:E5208–E5217. 44. Mukherjee S, et al. (2011) Modulation of Rab GTPase function by a protein phos- 30. Liu Y, Luo Z-Q (2007) The Legionella pneumophila effector SidJ is required for effi- phocholine transferase. Nature 477:103–106. cient recruitment of endoplasmic reticulum proteins to the bacterial phagosome. 45. Goody PR, et al. (2012) Reversible phosphocholination of Rab proteins by Legionella – Infect Immun 75:592 603. pneumophila effector proteins. EMBO J 31:1774–1784. 31. Havey JC, Roy CR (2015) Toxicity and SidJ-mediated suppression of toxicity require 46. Qiu J, et al. (2017) A unique deubiquitinase that deconjugates phosphoribosyl-linked distinct regions in the SidE family of Legionella pneumophila effectors. Infect Immun protein ubiquitination. Cell Res 27:865–881. 83:3506–3514. 47. Qiu J, et al. (2016) Ubiquitination independent of E1 and E2 enzymes by bacterial 32. Jeong KC, Sexton JA, Vogel JP (2015) Spatiotemporal regulation of a Legionella effectors. Nature 533:120–124. pneumophila T4SS substrate by the metaeffector SidJ. PLoS Pathog 11:e1004695. 48. Feeley JC, et al. (1979) Charcoal-yeast extract agar: Primary isolation medium for 33. Archer KA, Roy CR (2006) MyD88-dependent responses involving toll-like receptor – 2 are important for protection and clearance of Legionella pneumophila in a mouse Legionella pneumophila. J Clin Microbiol 10:437 441. model of Legionnaires’ disease. Infect Immun 74:3325–3333. 49. Saito A, Rolfe RD, Edelstein PH, Finegold SM (1981) Comparison of liquid growth – 34. Brieland JK, Remick DG, LeGendre ML, Engleberg NC, Fantone JC (1998) In vivo regula- media for Legionella pneumophila. J Clin Microbiol 14:623 627. tion of replicative Legionella pneumophila lung infection by endogenous interleukin-12. 50. Luo Z-Q, Isberg RR (2004) Multiple substrates of the Legionella pneumophila Dot/Icm Infect Immun 66:65–69. system identified by interbacterial protein transfer. Proc Natl Acad Sci USA 101: 35. Dam P, Olman V, Harris K, Su Z, Xu Y (2007) Operon prediction using both genome- 841–846. specific and general genomic information. Nucleic Acids Res 35:288–298. 51. Nagai H, Roy CR (2001) The DotA protein from Legionella pneumophila is secreted by 36. Mao F, Dam P, Chou J, Olman V, Xu Y (2009) DOOR: A database for prokaryotic a novel process that requires the Dot/Icm transporter. EMBO J 20:5962–5970. operons. Nucleic Acids Res 37:D459–D463. 52. Bardill JP, Miller JL, Vogel JP (2005) IcmS-dependent translocation of SdeA into 37. Shen X, et al. (2009) Targeting eEF1A by a Legionella pneumophila effector leads to macrophages by the Legionella pneumophila type IV secretion system. Mol Microbiol inhibition of protein synthesis and induction of host stress response. Cell Microbiol 11: 56:90–103. – 911 926. 53. Lara-Tejero M, et al. (2006) Role of the caspase-1 inflammasome in Salmonella ty- 38. Urbanus ML, et al. (2016) Diverse mechanisms of metaeffector activity in an in- phimurium pathogenesis. J Exp Med 203:1407–1412. tracellular bacterial pathogen, Legionella pneumophila. Mol Syst Biol 12:893. 54. Case CL, Roy CR (2013) Analyzing caspase-1 activation during Legionella pneumophila 39. Kubori T, Shinzawa N, Kanuka H, Nagai H (2010) Legionella metaeffector exploits infection in macrophages. Methods Mol Biol 954:479–491. host proteasome to temporally regulate cognate effector. PLoS Pathog 6:e1001216. 55. Ivanov SS, Roy CR (2013) Pathogen signatures activate a ubiquitination pathway that 40. Spörri R, Joller N, Albers U, Hilbi H, Oxenius A (2006) MyD88-dependent IFN-gamma production by NK cells is key for control of Legionella pneumophila infection. modulates the function of the metabolic checkpoint kinase mTOR. Nat Immunol 14: – J Immunol 176:6162–6171. 1219 1228. 41. Brown AS, et al. (2016) Cooperation between monocyte-derived cells and lymphoid 56. Choy A, et al. (2012) The Legionella effector RavZ inhibits host autophagy through cells in the acute response to a bacterial lung pathogen. PLoS Pathog 12:e1005691. irreversible Atg8 deconjugation. Science 338:1072–1076. 42. Tan Y, Luo Z-Q (2011) Legionella pneumophila SidD is a deAMPylase that modifies 57. Moffat JF, Tompkins LS (1992) A quantitative model of intracellular growth of Le-

Rab1. Nature 475:506–509. gionella pneumophila in Acanthamoeba castellanii. Infect Immun 60:296–301. MICROBIOLOGY

Shames et al. PNAS Early Edition | 9of9 Downloaded by guest on September 25, 2021