Reducing Virulence of the Human Pathogen Burkholderia by Altering the Substrate Specificity of the Quorum-Quenching Acylase Pvdq

Reducing virulence of the human pathogen Burkholderia by altering the substrate specificity of the quorum-quenching acylase PvdQ

Gudrun Kocha,1, Pol Nadal-Jimeneza,2, Carlos R. Reisa,3, Remco Muntendama, Marcel Bokhoveb,4, Elena Melilloa, Bauke W. Dijkstrab, Robbert H. Coola, and Wim J. Quaxa,5

aDepartment of Pharmaceutical Biology, University of Groningen, 9713 AV, Groningen, The Netherlands; and bLaboratory of Biophysical Chemistry, University of Groningen, Nijenborgh 7, 9747 AG, Groningen, The Netherlands

Edited by Gregory A. Petsko, Weill Cornell Medical College, New York, NY, and approved December 23, 2013 (received for review June 19, 2013)

The use of enzymes to interfere with quorum sensing represents autoinducer inactive, reducing bacterial virulence and path- an attractive strategy to fight bacterial infections. We used PvdQ, an ogenesis in vivo (22–25). effective quorum-quenching enzyme from Pseudomonas aeruginosa, Recently, we have shown that the quorum-quenching acylase as a template to generate an acylase able to effectively hydrolyze PvdQ, produced by fluorescent pseudomonads, decreases the C8-HSL, the major communication molecule produced by the Burkhol- levels of the P. aeruginosa signal molecule 3-oxo-C12-HSL. deria species. We discovered that the combination of two single Consequently, by either overexpression or exogenous addition of α β mutations leading to variant PvdQL 146W,F 24Y conferredhighactivity PvdQ, expression of virulence-related genes was reduced (21, 26, α β toward C8-HSL. Exogenous addition of PvdQL 146W,F 24Y dramatically 27) in a model system measuring the survival of Caenorhabditis decreased the amount of C8-HSL present in Burkholderia cenocepacia elegans upon infection by P. aeruginosa (23). PvdQ is most ef- cultures and inhibited a quorum sensing-associated phenotype. The fective against AHLs with side chains longer than 10 carbon efficacy of this PvdQ variant to combat infections in vivo was further atoms (21), whereas showing little to no activity toward AHLs confirmed by its ability to rescue Galleria mellonella larvae upon in- with shorter acyl chains such as C8-HSL, which induces virulence fection, demonstrating its potential as an effective agent toward

in members of the Bcc (6, 9). The recently solved structure of MICROBIOLOGY Burkholderia infections. Kinetic analysis of the enzymatic activities PvdQ with a bound 3-oxo-C12 fatty acid revealed a large hy- toward 3-oxo-C12-L-HSL and C8-L-HSL corroborated a substrate drophobic substrate-binding cleft that properly accommodates switch. This work demonstrates the effectiveness of quorum- this fatty acid side chain (28). Altering the substrate range of quenching acylases as potential novel antimicrobial drugs. In ad- PvdQ toward shorter AHLs, such as C8-HSL, might therefore dition, we demonstrate that their substrate range can be easily switched, thereby paving the way to selectively target only spe- Significance cific bacterial species inside a complex microbial community.

Resistance toward commonly used antibiotics is becoming computational design | enzyme engineering | antibiotic | cystic fibrosis a serious issue in the fight against bacterial pathogens. One promising strategy lies in the interference of bacterial quorum he Burkholderia cepacia complex (Bcc) comprises a group of sensing by the hydrolysis of the signaling molecules. In this T17 related bacterial species able to colonize different envi- study, we present a structure-aided computational design ap- ronmental niches (1). Over the years the Bcc has gained special proach to alter the substrate specificity of the quorum-quenching attention, as some of its members have been associated with life- acylase PvdQ. Introduction of two point mutations in residues threatening human infections (2, 3). Especially immunocom- lining the active site led to a switch in substrate specificity, ren- promised patients and people suffering from cystic fibrosis are dering the enzyme highly active toward C8-HSL and thereby re- generally infected with these pathogens; in particular, infection ducing virulence caused by Burkholderia cenocepacia.Thus,this with Burkholderia cenocepacia has been correlated with a poor work not only provides a structural insight into the substrate prognosis (1, 4). B. cenocepacia is often found cocolonizing the specificity of quorum-quenching acylases but also indicates their lungs of cystic fibrosis patients alongside the opportunistic potential in the fight against specific bacterial pathogens. pathogen Pseudomonas aeruginosa (5–9). Reports on the occurrence of these two pathogens are ap- Author contributions: G.K., P.N.-J., R.H.C., and W.J.Q. designed research; G.K., P.N.-J., C.R.R., pearing more and more frequently, underlining the difficulty R.M., M.B., E.M., and R.H.C. performed research; G.K., P.N.-J., C.R.R., R.M., M.B., E.M., B.W.D., and R.H.C. analyzed data; and G.K., P.N.-J., C.R.R., M.B., B.W.D., R.H.C., and in eradicating these pathogens with common antibiotics (10). W.J.Q. wrote the paper. Hence, novel strategies are needed to target bacterial infections The authors declare no conflict of interest. without applying too much selective pressure (11). An important bacterial Achilles’ heel is quorum sensing (QS), a cell density- This article is a PNAS Direct Submission. reliant regulatory system dependent on the secretion of N-acyl Data deposition: The crystallography, atomic coordinates, and structure factors reported in this paper have been deposited in the Protein Data Bank, www.pdb.org (PDB ID code homoserine lactones (AHLs) (12). These molecules have been 4BTH). largely associated with virulence traits, as they are pivotal for the 1Present address: Institut Für Molekulare Infektionsbiologie, Würzburg University, 97080 expression of genes involved in toxin production, motility, plas- Würzburg, Germany. mid transfer, antibiotic synthesis, and biofilm formation (13, 14). 2Present address: Bacterial Signalling Group, Instituto Gulbenkian de Ciência, 2781-901 In the last several years, many ways to interfere with QS have Oeiras, Portugal. been explored, as interference with the action of AHLs has been 3Present address: Department of Cell Biology, University of Texas Southwestern Medical demonstrated to reduce pathogenesis (15–17). The use of enzymes Center, Dallas, TX 75390-9039. in targeting QS paves a new way in combating pathogens. A major 4Present address: Department of Biosciences and Nutrition and Center for Biosciences, finding in the field was the discovery of two families of quorum- Karolinska Institutet, SE 14183 Huddinge, Sweden. quenching enzymes: the AHL lactonases and the AHL acylases 5To whom correspondence should be addressed. E-mail: [email protected]. – (18 21). Lactonases target the lactone ring, whereas acylases This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10. hydrolyze the amide bond of AHLs; both enzymes render the 1073/pnas.1311263111/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1311263111 PNAS Early Edition | 1of6 Downloaded by guest on September 26, 2021 shift the antibacterial scope of PvdQ as a therapeutic agent, by the C12 acyl chain in the C12-PvdQ crystal structure. In potentially providing an effective therapy for Bcc infections. contrast, a similar analysis using C8-HSL resulted in a substantial In this study, we report a structure-aided design approach to conformational heterogeneity within the active site of PvdQWT, modify the substrate specificity of a quorum-quenching acylase with the most energetically favorable poses showing that the car- such that it targets explicitly the signaling molecules of a re- bonyl oxygen of C8-HSL is no longer within H-bonding distance stricted range of pathogens. We have identified a PvdQ variant with Asnβ269 and the Valβ70 backbone amide (more than 3.5 Å), containing two amino acid substitutions, Leuα146Trp and and a significant distance between the carbonyl carbon of C8-HSL Pheβ24Tyr, which showed a substantially increased C8-HSL– and the catalytic Serβ1ofPvdQisobserved(6.1Å)(Fig. S2). degrading activity compared with the wild-type enzyme. Kinetic Based on the differences shown for C8-HSL and C12-HSL during analysis was in-line with a substrate switch from 3-oxo-C12-L-HSL molecular docking experiments, the following amino acid residues to C8-L-HSL caused by these mutations. Using an in vivo model were selected for in silico mutagenesis: Thrα143, Leuα146, α β β β β β β for B. cenocepacia infection (29), we demonstrate that this quorum- Gly 150, Phe 24, Leu 50, Leu 53, Asn 57, Val 158, Trp 162, β β β quenching PvdQ variant can be successfully used to attenuate Pro 185, Trp 186, and Val 187. Each amino acid was substituted pathogen virulence and increase host survival. These results by all other 19 possible amino acids, and the new model structures α β validate PvdQL 146W,F 24Y as a promising and effective potential were energy-minimized, after which C8-HSL was docked in the agent to combat emerging Bcc infections. active site of PvdQ. A final minimization was used to refine the ligand poses. The most energetically favorable substrate-docked Results and Discussion poses were analyzed with respect to the positioning of the substrate and the new distances obtained between the catalytic Serβ1Oγ and Design of PvdQ Variants for Increased C8-HSL Activity. Using the the carbonyl carbon atom of C8-HSL. Eighteen amino acid sub- recently elucidated crystal structure of PvdQ (28), we adopted stitutions out of 218 resulting in a reduction of the distance between a rational design approach to scan for PvdQ variants that would Serβ1 and the carbonyl carbon of C8-HSL of less than 4 Å were be better capable of accommodating C8-HSL than the wild-type therefore considered for further analysis. Positions Thrα143Met, enzyme. PvdQ is a heterodimeric Ntn hydrolase with an un- Thrα143Lys, Leuα146Ile, Leuα146Arg, Leuα146Trp, Pheβ24Tyr, usually large binding pocket that can accommodate the long acyl Leuβ53Phe, Leuβ53Lys, Leuβ53Ile, Leuβ53Arg, Asnβ57Arg, chain of an HSL substrate (28). The acyl chain of 3-oxo-C12- Asnβ57His, Valβ158Met, Valβ158Ile, Trpβ162Phe, Trpβ162Tyr, HSL has been structurally characterized in complex with PvdQ, Valβ187Phe, and Valβ187Tyr substitutions were therefore se- but not the homoserine lactone part. Therefore, we performed lected for site-directed mutagenesis (Table S1). The eight se- molecular docking experiments using 3-oxo-C12-HSL and C12- lected residues were also mutated to alanine. HSL in the active site of PvdQWT. The results show that the carbonyl oxygen of these substrates has good hydrogen-bonding Screening for C8-HSL Quenching. Based on this in silico screening interactions with the Nδ2 group of Asnβ269 and the Valβ70 approach, the 26 proposed site-directed mutants of PvdQ were backbone amide (Fig. 1 A and B and Fig. S1), which constitute constructed, produced, and purified. The variants containing the oxyanion hole (28). The most favorable substrate poses Valβ158Ala and Asnβ57Arg were affected in protein maturation obtained for 3-oxo-C12-HSL show the acyl chain positioned in (Fig. S3), a property often observed when mutagenizing acylases that the hydrophobic pocket in a conformation similar to that adopted rely on the same residues to perform substrate conversion and protein maturation. These were therefore excluded from further analysis. As seen with fluorescence experiments, mutations of Trpβ162 and Valβ187 had no effect on C8-HSL hydrolysis, but substitutions at the positions Thr-143 and Leu-146 of the α-subunit and Phe- 24, Leu-53, Asn-57, and Val-158 of the β-subunit resulted in increased hydrolytic activity toward C8-HSL. In particular, variants Thrα143Lys, Leuα146Trp, Pheβ24Tyr, Leuβ53Ile, Asnβ57His, Valβ158Met, and Valβ158Ile resulted in a significant increase in C8-HSL hydrolysis compared with PvdQWT,asshownbyamore than 50% decrease in mean specific fluorescence activity (Fig. 2A). A second round of computational analysis was performed, to identify possible combinatorial mutationsinPvdQwithafurther enhanced activity and specificity toward C8-HSL. The best single mutants were combined in silico and analyzed as previously for the distance between the catalytic Serβ1Oγ and the carbonyl carbon atom of C8-HSL. Based on the results obtained by this analysis, the double mutants Leuα146Trp/Pheβ24Tyr and Pheβ24Tyr/ Asnβ57His were generated and tested for their activity. Most importantly, the Leuα146Trp/Pheβ24Tyr variant displayed the α β highest hydrolytic activity on C8-HSL: 5 ng/μL PvdQL 146W,F 24Y Fig. 1. Comparative molecular docking simulations. The most favored was sufficient to quench 5 μM C8-HSL, corresponding to a five- conformations of C12-HSL (A) and 3-oxo-C12-HSL (B) in the active site of fold increased activity compared with the best single mutant PvdQ from CDOCKER, as implemented in Accelrys Discovery Studio 3.0. Hy- tested (Fig. 2A). drogen atoms were added to the protein molecule and substrates, and the Hydrolytic activity of these enzymes toward 3-oxo-C12-HSL was CHARMm force field was used to assign partial charges to the ligands. assessed using a biosensor strain (Fig. 2B and Fig. S4). Whereas Substrates were docked into PvdQ using the coordinates of 3-oxo-lauric acid WT WT PvdQ displayed high hydrolytic activity, the mutant enzyme bound in the active site of PvdQ [PDB ID code 2WYC (28)]. The residues Lα146W,Fβ24Y forming the active site of PvdQ are colored green, and the accessible solvent PvdQ was severely impaired in its capacity to hydro- surface-contoured substrates are represented in yellow and cyan sticks for lyze 3-oxo-C12-HSL under the conditions tested, rendering its C12-HSL and 3-oxo-C12-HSL, respectively. Both substrate-docking poses are function of quenching the endogenous signal from P. aeruginosa aligned for nucleophilic attack. The carbonyl oxygen forms hydrogen bonds biologically insignificant. The activities toward 3-oxo-C12-HSL and with the Asnβ269 side-chain Nδ2 and the Valβ70 backbone amide, consistent C8-HSL highlight that a substrate switch had occurred, diminishing with the proposed oxyanion hole residues involved in the stabilization of the activity toward 3-oxo-C12-HSL but substantially increasing hydro- tetrahedral transition state (28). [The amino acid numbering follows the lytic activity toward C8-HSL, as Fig. 2B clearly indicates. In- subunit composition of the mature protein; i.e., the α-chain is defined by terestingly, considering the respective single mutants, deacylase Lα146W Aspα1–Valα170 (equivalent to D24–V193 of the amino acid sequence of the activity toward 3-oxo-C12-HSL was hardly affected in PvdQ β preprotein) and the β-chain by Serβ1–Gluβ546 (equivalent to S217–E762).] and only slightly impaired in PvdQF 24Y, indicating that the

2of6 | www.pnas.org/cgi/doi/10.1073/pnas.1311263111 Koch et al. Downloaded by guest on September 26, 2021 α β differences were observed between PvdQL 146W,F 24Y and PvdQWT (rmsd of 0.28 Å for 710 Cα atoms). The crystal structure of the mutant protein shows that the hydrophobic substrate-binding pocket near the N-terminal nu- cleophile residue Serβ1 is in the closed state (Fig. 3A) (28). The side chains of the mutated Tyrβ24 and Trpα146 residues line this pocket, with each residue adopting at least two alternate conformations. The Leuα146Trp mutation introduces a much bulkier side chain and reduces the volume of the substrate- binding pocket from 260 to 80 Å3 for the closed conformation (Fig. 3 A and B). The cavity with Tyrβ24 in the open conformation has a volume of 140 Å3. We propose that the larger volume of the substrate-binding pocket in PvdQWT preferentially binds long fatty acid-like acyl chains, whereas the much less vol- α β uminous pocket of PvdQL 146W,F 24Y favors short-chain HSLs. α β The crystal structures of PvdQWT and PvdQL 146W,F 24Y indicate that mutating Pheβ24 to a tyrosine does not cause substantial conformational changes in the active site of PvdQ. Slightly different side-chain orientations for these aromatic residues are observed, however, with the new tyrosine moving upward relative to the phenylalanine in the active conformation (Fig. 4). This in turn creates more space at the entrance of the hydrophobic cavity (Fig. 4), and could partially contribute to the better fit of C8-HSL. Additionally, in one of the alternate side-chain conformations observed in the crystal structure, the hydroxyl group of Tyrβ24 has an interaction with the side-chain amine of Trpα146 (Fig. 4). This conformation resembles the open conformation that residue β24 Fig. 2. (A) Relative fluorescence as a measure of the presence of C8-HSL adopts in the substrate-bound state (28), and thus this interaction after incubation of PvdQ variants with the biosensor P. putida F117 (pAS-C8) may stabilize the conformation of Trpα146 and provide a better fit MICROBIOLOGY after 13 h of incubation. The protein concentration was 5 ng/μL. Values of the shorter acyl side chain of C8-HSL. reported indicate the mean specific fluorescence activity of each PvdQ variant α β normalized as a percentage of the fluorescence level in an assay with PvdQWT. PvdQL 146W,F 24Y Disrupts B. cenocepacia Signaling and Induction WT Lα146W,Fβ24Y (B) Activity of PvdQ and PvdQ toward AHLs. Enzymes were in- of Virulence. To determine whether PvdQWT and its variant, α β cubated with 5 μM 3-oxo-C12-HSL or C8-HSL. 3-Oxo-C12 levels were analyzed PvdQL 146W,F 24Y, interfere with C8-HSL accumulation and down- using E. coli (pSB1075) and C8-HSL levels with P. putida F117 (pAS-C8). Values stream signaling by B. cenocepacia, cultures of this bacterium were reported indicate the mean specific fluorescence/luminescence activity normalized incubated with either enzyme (Fig. S6). The addition of the to the units measured by AHLs only. Error bars indicate SD. enzymes did not influence bacterial growth, as indicated in Fig. S7. Hence, C8-HSL accumulation was assayed after 24 h of incubation at30°C.Fig.5A clearly shows that almost no fluorescence, and combination of both mutations is responsible for this switch in thereforenoC8-HSL,couldbedetectedinthepresenceofthe α β substrate specificity (Fig. S4). mutant acylase PvdQL 146W,F 24Y. Significant levels of fluorescence Previously, Pheβ24 had been shown to allow the entrance of WT were measured in the control culture with no enzyme addition; only 3-oxo-C12-HSL into the hydrophobic pocket of PvdQ (28). WT α a slight decrease was observed in the presence of PvdQ . Thus, Position 146 needs to be occupied by a small residue like Leu to this activity test further substantiates the results obtained in the allow binding of the C12 acyl chain. We suggest that the con- preliminary activity screen, namely the limited activity of wild-type α straints imposed by the Trp 146 side chain on the binding mode enzyme to hydrolyze C8-HSL compared with the much more α β Lα146W,Fβ24Y of C8-HSL and the stabilization of Trp 146 by the new Tyr 24 efficient PvdQ variant. contribute to the proper accommodation of C8-HSL in the active In addition, and to further substantiate that a substrate switch site of PvdQ, whereas impairing its activity toward 3-oxo-C12. had occurred, we performed the same experiment but adding the enzyme to a P. aeruginosa culture. Analysis of 3-oxo-C12-HSL Analysis of Quorum-Quenching Activity. To further characterize α β levels after incubation with the enzymes for 24 h revealed that only PvdQWT and PvdQL 146W,F 24Y, the activity of various concen- PvdQWT could significantly decrease the amounts of the signaling trations of C8-HSL and 3-oxo-C12-HSL was determined. These experiments confirmed that PvdQWT has a preference for the long- chain substrate 3-oxo-C12-HSL, whereas this preference is shifted α β to the shorter-chain substrate C8-HSL in mutant PvdQL 146W,F 24Y. Using the biosensor strain Pseudomonas putida (pAS-C8), we determined that after incubation of 1 μM C8-HSL with either 5 ng/μL α β PvdQWT or PvdQL 146W,F 24Y,only0.063μMinthecaseof α β PvdQWT or 0.0013 μM C8-HSL in the case of PvdQL 146W,F 24Y was detectable. These results indicate that using equal con- α β centrations of PvdQ proteins, PvdQL 146W,F 24Y displays WT a 48.5-fold reduction of C8-HSL levels compared with PvdQ WT (Fig. S5). Fig. 3. Residues lining the substrate-binding pocket in PvdQ and PvdQLα146W,Fβ24Y.(A) PvdQWT; the main chain is indicated by a Cα trace, the α β blue ball and sticks indicate the residues lining the substrate-binding pocket, Structural Effects of PvdQL 146W,F 24Y. To investigate the structural and the mesh shows the cavity as calculated by VOIDOO (Uppsala Software α β α β effects of the Leu 146Trp and Phe 24Tyr mutations on PvdQ, Factory). (B)PvdQL 146W,F 24Y; the main chain is indicated by a Cα trace, the we elucidated the crystal structure of the double mutant green ball and sticks indicate the residues lining the substrate-binding pocket, Lα146W,Fβ24Y PvdQ at a resolution of 1.9 Å to final Rwork and Rfree except for the mutated residues, which are indicated in orange, and the mesh values of 17.9% and 20.7%, respectively. Data collection and refine- shows the cavity as calculated by VOIDOO. The Trpα146 mutation decreases ment statistics can be found in Table S2. No major conformational the volume of the cavity, making it more suitable for short-chain fatty acids.

Koch et al. PNAS Early Edition | 3of6 Downloaded by guest on September 26, 2021 switch due to the mutations: Whereas PvdQWT preferentially α β hydrolyzes 3-oxo-C12-L-HSL, PvdQL 146W,F 24Y has a preference for C8-L-HSL. The concentration window is not sufficiently large to determine the kinetic parameters separately, but an estimation of the k :K ratio can be obtained. For 3-oxo-C12-L-HSL, these cat m − − parameters were 5.8 × 103 and 1.5 × 103 M 1s 1 for PvdQWT and α β PvdQL 146W,F 24Y, respectively, resulting in a 3.8-fold difference. For the substrate C8-L-HSL, these values were 0.8 × 103 and 3.4 × − − α β 103 M 1s 1 for PvdQWT and PvdQL 146W,F 24Y, respectively, resulting in a 4.3-fold difference. Thus, in total, the mutations result in a 16-fold difference in catalytic efficiency. Conclusions α β We identified and characterized PvdQL 146W,F 24Y, a PvdQ var- Fig. 4. Structural impression of mutations on docking of C8-HSL and 3-oxo- iant with a shifted substrate range that is highly active toward C8- α β C12-HSL. Close-up view of the two superimposed active sites of P. aeruginosa HSL. The in vivo quorum-quenching activity of PvdQL 146W,F 24Y PvdQWT (shown in green) and PvdQLα146W,Fβ24Y (in white) with the most fa- against B. cenocepacia, demonstrated by the decrease in pro- vored docked conformations for C8-HSL (A) and 3-oxo-C12-HSL (B), using the α β teolytic activity in the culture supernatant and the increase in host coordinates given by the crystal structure of PvdQL 146W,F 24Y. As shown, the survival, confirms its potential as a possible therapeutic, especially introduction of residue Trpα146 clearly reduces the hydrophobic pocket size as this protease activity has been associated with the pathogen’s and the protrusion into the interior of the enzyme, and contributes to the invasion of lung tissue (30, 31). Our results obtained with α β proper accommodation of the acyl chain of C8-HSL (A). The hydroxyl group PvdQL 146W,F 24Y together with previous results on PvdQWT (23) of the new Tyrβ24 forms a 3.2-Å hydrogen bond with the side-chain amine of Trpα146, stabilizing the conformation of Trpα146 and providing a better α β fit of the alternate acyl chain of C8-HSL. Inversely, the mutant PvdQL 146W,F 24Y no longer allows the proper accommodation of the acyl chain of 3-oxo-C12- HSL (B), with a distance between the carbonyl carbon of the substrate and the catalytic serine of 6.1 Å for the most favorable conformation.

molecule, as shown by a clear decrease in luminescence. Hardly any effect, however, on AHL levels was observed upon addition of α β PvdQL 146W,F 24Y (Fig. S6). Taken together, these AHL quantifi- α β cations show that PvdQL 146W,F 24Y affects shorter acyl chains in a more effective manner than the wild-type enzyme does. To determine the downstream effects of quorum quenching in Bcc, we monitored the proteolytic activity, as the production of extracellular protease, which plays an important role in the invasion of lung tissue by B. cenocepacia, is positively regulated by α β C8-HSL (30, 31). Addition of PvdQL 146W,F 24Y to B. cenocepacia cultures significantly decreased the number of protease units detected in culture supernatants, almost to the level of that produced by the QS-negative strain H111-I, which is unable to induce protease activity (Fig. 5A). In contrast, addition of PvdQWT resulted in protease production similar to the control culture of B. cenocepacia α β wild-type H111 (Fig. 5A). These data show that PvdQL 146W,F 24Y decreases the level of C8-HSL present in B. cenocepacia cultures, thereby reducing the expression of virulence traits.

α β PvdQL 146W,F 24Y Attenuates Burkholderia Virulence upon in Vivo Infection of Galleria mellonella Larvae. The promising results obtained in degrading C8-HSL and protease activity after exogenous α β addition of PvdQL 146W,F 24Y to B. cenocepacia in vitro prompted us to investigate the quorum-quenching effects of this enzyme in vivo. As depicted in Fig. 5B, injection of larvae of the great wax moth α β Galleria mellonella with B. cenocepacia H111 culture kills nearly all Fig. 5. Effects of PvdQWT and PvdQL 146W,F 24Y on B. cenocepacia QS. (A) larvae, whereas injection with the QS-negative strain B. cenocepacia Cells of B. cenocepacia were incubated for 24 h without PvdQ (first set of α β H111-I does not affect survival, in accordance with the importance of bars) or in the presence of PvdQWT (indicated as WT) or PvdQL 146W,F 24Y QS in B. cenocepacia infection and pathogenesis. Whereas pre- (indicated as MUT) and tested for C8-HSL levels with the P. putida F117 (pAS- B. cenocepacia WT C8) biosensor strain (gray bars). Fluorescence units were calculated relative incubation of H111 bacteria with PvdQ hardly WT affected the survival rates of the larvae, preincubation with to the activity in the presence of PvdQ (equal to 1). Cultures were also α β PvdQL 146W,F 24Y led to a nearly complete attenuation of bacterial analyzed for the production of protease by the activity on skim milk (dashed virulence and increased the overall survival of the larvae (Fig. 5B). bars). One unit of protease was defined as the activity that produced Lα146W,Fβ24Y WT a change in the OD600 of 0.1 per h (30). Lower protease units were found to This result demonstrates that PvdQ , but not PvdQ ,is α β be produced when PvdQL 146W,F 24Y was added to the cultures (dashed α β able to diminish the virulence of B. cenocepacia H111. bars). (B)PvdQL 146W,F 24Y protects for B. cenocepacia H111 infection in the insect model. G. mellonella larvae were injected with B. cenocepacia H111 Kinetic Analysis of the Enzymatic Activities of the PvdQ Enzymes. wild-type cells either untreated or treated with PvdQWT or PvdQLα146W,Fβ24Y. Enzymatic activities were determined by an end-point assay and After 48 h of incubation, larval survival was assessed. G. mellonella injected derivatization with ortho-phthaldialdehyde. The low solubility of with buffer only resulted in 100% survival, B. cenocepacia H111 had a lethal 3-oxo-C12-L-HSL allowed us to screen for activity in the range of outcome for the larvae, whereas pretreatment of the bacterial cultures with α β 0–0.2 mM, whereas the activity toward C8-L-HSL was measured up PvdQL 146W,F 24Y rescued survival. For all determinations, a quorum-sensing to 0.6 mM. The activity plots (Fig. S8) clearly show a substrate negative-strain H111-I served as control. Error bars indicate SD.

4of6 | www.pnas.org/cgi/doi/10.1073/pnas.1311263111 Koch et al. Downloaded by guest on September 26, 2021 provide the first steps toward the development of future antimi- Initial Screening of Mutants. An initial screening for C8-HSL degradation was crobial therapies aiming to effectively combat P. aeruginosa and B. conducted using the bioreporter strain P. putida F117 carrying the plasmid cenocepacia infections, the two most important Gram-negative pAS-C8 (39). This plasmid encodes a cepR-PcepI::gfp fusion that drives gfp pathogens isolated from the lungs of cystic fibrosis patients. Fur- expression in response to C8-HSL, allowing quantification of the amount of thermore, we illustrate how the design approach used in this study remaining C8-HSL after exposure to the PvdQ mutants. An overnight culture of this strain was diluted 100 times in LB medium (38) containing gentamicin can be applied to successfully produce quorum-quenching variants μ μ with specific substrate ranges to target a selected group of bacteria. (20 g/mL) and a final concentration of 0.5, 1, or 5 M C8-HSL. The concentration of tested proteins used in this assay was either 5 or 10 ng/μL; at these The application of highly active enzymes as potential treatment can WT have a number of beneficial effects not only with respect to lung concentrations, PvdQ does not affect the amount of GFP fluorescence produced by the biosensor. Reaction conditions without enzyme or AHLs alone infections but also to those occurring in the gastrointestinal tract. were used as controls. GFP expression was monitored every 30 min during Specific targeting of the quorum-sensing systems of pathogens a 20-h experiment in a multifunctional microplate reader (FLUOstar Omega; would leave the beneficial microbiota unharmed. Decreasing the BMG Labtech; excitation wavelength, 485 nm; emission wavelength, 520 nm). deleterious side effects on the normal host microbiota is a major Screenings were conducted in at least three independent experiments. factor contributing to host protection against invading competitive, 3-Oxo-C12-HSL degradation was assayed as previously described (40) using opportunistic bacteria and increasing the recovery of the host the biosensor strain E. coli JM109 containing the plasmid pSB1075 (lasR- after infection (32, 33). We aim with our findings to substantiate the PlasI::luxCDABE) (41) that produces luminescence in the presence of this high potential of quorum-quenching enzymes in targeting bacterial autoinducer. All PvdQ variants that caused a 50% decrease in fluorescence pathogens. correlating to an increase in activity toward C8-HSL were scored positive.

α β Materials and Methods Analysis of Quorum-Quenching Activity of PvdQWT and PvdQL 146W,F 24Y. Enzy- Structure-Based Design of PvdQ Mutants by Molecular Docking of C8-HSLs. matic activity was investigated using the above-mentioned biosensor systems. Based on the crystal structure of the P. aeruginosa quorum-quenching acy- Luminescence and fluorescence, respectively, were followed over a period in the lase (PvdQ) in complex with 3-oxo-lauric acid [Protein Data Bank (PDB) ID presence of various concentrations of AHL substrates (0–10 μM) with and without code 2WYC (28)], molecular docking experiments were performed using the the addition of 5 ng/μL enzyme. Linearity of the reaction was checked. 3-oxo-C12-L-HSL and C8-L-HSL substrates. Molecular docking simulations of the 10 lowest-energy poses were done using the grid-based approach Data Collection, Crystal Structure Determination, and Refinement. Purified CDOCKER, a molecular dynamics simulated annealing-based algorithm in PvdQLα146W,Fβ24Y was crystallized as previously described (28). Crystals were which the receptor is held rigid while the ligands are allowed to flex during flash-cooled in liquid nitrogen using mother liquor supplemented with 25% the refinement (34, 35), implemented in Discovery Studio 3.0 (Accelrys). All of (vol/vol) glycerol as cryoprotectant. One hundred and twenty degrees of data

the structures were further energy-minimized using CHARMm (36), consisting were collected at 100 K with an oscillation range of 0.2° at the PX-1 beamline, MICROBIOLOGY of 150 steps of steepest descent followed by 2,000 iterations of the Swiss Light Source (Villigen) supplied with a Pilatus detector. Data were in- adopted basis set Newton–Raphson algorithm using an energy tolerance tegrated and scaled using XDS (42) and SCALA (43) using the CCP4 interface − − of 0.01 kcal·mol 1·Å 1. To obtain a structural overview of the new docked (44). Initial phase information of the double mutant was obtained with Phaser substrate conformations with regard to the structure of 3-oxo-lauric acid (45) using PDB ID code 2WYE (28) as a search model. Model building and bound to PvdQWT, an overlay was made of the structures containing the refinement were done with Coot (46) and REFMAC5 (47) using translation/ most favorable ligand poses for 3-oxo-C12-HSL and C8-HSL. Analysis of libration/screw refinement (48). Structure validation was done with Mol- ligand-binding pattern tools in Discovery Studio 3.0 allowed us to char- Probity (49). acterize and compare ligand-binding poses in PvdQ and visualize specific interactions between protein residues and bound substrates, such as res- Virulence Assays of B. cenocepacia H111. Purified proteins were tested for re- idues involved in hydrogen bonding, charge or polar interactions, and van duction of C8-HSL levels present in B. cenocepacia H111 cultures. Overnight der Waals interactions. Analysis of the residues involved in interactions cultures of wild-type B. cenocepacia H111 (50) and the synthase-negative strain with the fatty acid chain of C12-HSL and 3-oxo-C12-HSL permitted us to B. cenocepacia H111-I (51) were diluted 100-foldinLBmedium.Recombinant further curtail our search and select for determinant residues involved in PvdQWT or PvdQLα146W,Fβ24Y was added to the cultures at a final concentration interactions with the substrates of interest. Selected amino acid residues of 0.045 mg/mL, cultures containing buffer only served as controls, and were were mutated in silico into all other possible 19 amino acid residues and incubated at 30 °C for 24 h. Bacterial growth was measured in a Tecan plate energy-minimized as described above. The new energy-minimized struc- reader over a period. In addition, 2-mL aliquots were collected by centrifuga- tures were then used to perform docking as described above using C8-HSL tion at 5000 × g and supernatants were filtered through a 0.2-μmfilterand as substrate. The five most favorable ligand poses were inspected for stored at −20 °C until further analysis. To determine AHL concentrations, 900 μL binding energy and distance between the carbonyl carbon of the substrate of supernatant was first acidified with 1 M HCl (100 μL) and incubated at 37 °C and the catalytic Serβ1oftheβ-subunit of PvdQ. Finally, targeted amino for 18 h to revert spontaneous hydrolysis of the AHLs. Detection was per- acid substitutions were selected for site-directed mutagenesis. formed as mentioned above with the biosensor strain F117 (pAS-C8) by incubating 180 μL bacterial strain with 10 μL of bacterial supernatant. Fluo- Mutagenesis of pvdQ. The plasmid pMCT-pvdQ was under the control of iso- rescence was measured every 30 min for 20 h. propyl β-D-1-thiogalactopyranoside-inducible expression from the lacZ pro- Detection of bacterial protease activity was measured upon mixing 300 μL moter/operator (21). Point mutations were generated using the MEGAWHOP of bacterial supernatant with 700 μL of skim milk (2%) in LB medium. The method as described in ref. 37. Briefly, a megaprimer (200–500 bp) containing OD600 was measured every 30 min for 18 h. In this assay, proteolytic activity the desired mutation was generated using Phusion polymerase (Finnzymes). results in a decrease of absorbance due to the breakdown of milk proteins. After purification, the megaprimer was used to amplify the whole plasmid. The One unit of protease was defined as the activity that produced a change in PCR mixtures were subsequently digested with DpnI (Fermentas) to remove the the OD600 of 0.1 per h (30). template. All constructs were verified by DNA sequencing (Macrogen). P. aeruginosa AHL Analysis. The same procedure was followed as described Protein Expression and Purification. Escherichia coli strain DH10B was used for previously for B. cenocepacia. In brief, an overnight culture of P. aeruginosa expression of recombinant PvdQ protein as described previously (21). Briefly, ΔpvdQ (40) was diluted 100-fold in LB medium. Recombinant PvdQWT or α β E. coli DH10B cells harboring pMCT-pvdQ and variants were grown for 48 h PvdQL 146W,F 24Y was added to the cultures at a final concentration of 0.045 at 30 °C in 2× trypton-yeast medium (38) supplemented with chloramphenicol mg/mL, cultures containing buffer only served as controls, and were in- (50 μg/mL). After harvesting cells by centrifugation at 5000 × g, pellets were cubated at 30 °C for 24 h. To determine AHL concentrations, 900 μLofsu- resuspended in Tris·EDTA buffer (50 mM Tris·HCl, pH 8.8, 2 mM EDTA) and pernatant (of a given time point) was first acidified with 1 M HCl (100 μL) lysed by sonication. PvdQ was purified by a two-step procedure as previously and incubated at 37 °C for 18 h to revert spontaneous hydrolysis of the AHLs. described (28). In short, the flow-through of an anion-exchange chromatog- Detection was performed as mentioned above with the biosensor strain raphy column (Q-Sepharose; GE Healthcare) was collected and adjusted to E. coli JM109 containing the plasmid pSB1075 (lasR-PlasI::luxCDABE) (41). a 0.7 M final concentration of ammonium sulfate and loaded onto a phenyl- Sepharose column. PvdQ and variants eluted at a final concentration of 0% G. mellonella Infection Assay. Infection assays were performed as previously ammonium sulfate in T50 (50 mM Tris·HCl, pH 8.8) with a purity of ≥95% as described (29). Briefly, bacterial overnight cultures were diluted 1:100 in LB

shown by SDS/PAGE (Invitrogen) and Coomassie staining. medium and grown to an OD600 of 0.6–0.8. Cultures were collected and adjusted

Koch et al. PNAS Early Edition | 5of6 Downloaded by guest on September 26, 2021 7 to a final concentration of 4 × 10 cfu/mL (OD600 0.125) using 10 mM MgSO4. reagent (Sigma-Aldrich). The absorbance at 340 nm was recorded after Before injection, cultures were incubated at 30 °C for 1 h with or without 0.15 15–20 min in a microplate reader (FLUOstar Omega; BMG Labtech). A cali- mg/mL enzyme, after which the cultures were immediately transferred to ice. An bration curve was made with 0–0.5 mM homoserine lactone (Sigma-Aldrich) insulin pen (HumaPen Luxura; Lilly Nederland) was used to inject 10-μL aliquots and showed linearity up to an absorbance of 1.4. The absorbance values into the hindmost proleg of G. mellonella. Fifteen healthy larvae were injected were plotted as a function of time and the initial rates were calculated

per strain and incubated at 30 °C. Animals only injected with MgSO4 served as using the slope of the calibration curve. Finally, the initial rates were controls. Larvae were monitored after 24 and 48 h, respectively, and were scored plotted as a function of the substrate concentration. For each substrate dead if they did not respond to touch or had turned black. Assays were per- concentration at least three, but mostly five, experiments were performed in three independent experiments. formed. Controls with only enzyme or only substrates were performed, and all steps of the experimental setup were checked for effectiveness and Kinetic Analysis. The enzymatic activities of PvdQWT and PvdQLα146W,Fβ24Y introduction of artifacts. were tested with an end-point assay using a derivatization with ortho- phthaldialdehyde (52). Stocks of the substrates 3-oxo-C12-L-HSL and C8-L- ACKNOWLEDGMENTS. We thank Leo Eberl and Kathrin Riedel for providing B. HSL (Bio-Connect) were made in methanol. These substrates were added to cenocepacia H111 and H111-I strains as well as theC8biosensorstrainF117(pAS- reaction vials, after which the methanol was removed by evaporation. The C8). We gratefully acknowledge Rien Hoge for helpful discussions regarding Galleria infection assays, Eli Lilly Nederland for providing empty insulin cartridges, substrate was then carefully solubilized in PBS at 30 °C. Enzyme was added μ μ and Rita Setroikromo, Ronald van Merkerk, and Putri Dwi Utari for technical to 5 g/mL and samples of 60 L were taken immediately and at 2- to 5-min assistance with the assays. We thank the beamline staff of PX-1 (Swiss Light intervals. Each sample was immediately heat-inactivated and stored on ice. Source) for their assistance. Jessica A. Thompson is thanked for carefully reading At the end of the assay, 50 μL of each sample was transferred to a well of a and correcting the manuscript. This research was partly funded by European 384-well plate (Greiner Bio-One) and mixed with 50 μL of phthaldialdehyde Union Grant Antibiotarget MEST-CT-2005-020278 (to G.K. and P.N.-J.).

1. Mahenthiralingam E, Urban TA, Goldberg JB (2005) The multifarious, multireplicon 28. Bokhove M, Nadal Jimenez P, Quax WJ, Dijkstra BW (2010) The quorum-quenching Burkholderia cepacia complex. Nat Rev Microbiol 3(2):144–156. N-acyl homoserine lactone acylase PvdQ is an Ntn-hydrolase with an unusual sub- 2. Goldmann DA, Klinger JD (1986) Pseudomonas cepacia: Biology, mechanisms of vir- strate-binding pocket. Proc Natl Acad Sci USA 107(2):686–691. ulence, epidemiology. J Pediatr 108(5 Pt 2):806–812. 29. Uehlinger S, et al. (2009) Identification of specific and universal virulence factors in 3. Isles A, et al. (1984) Pseudomonas cepacia infection in cystic fibrosis: An emerging Burkholderia cenocepacia strains by using multiple infection hosts. Infect Immun problem. J Pediatr 104(2):206–210. 77(9):4102–4110. 4. Jones AM, et al. (2004) Burkholderia cenocepacia and Burkholderia multivorans:In- 30. Kooi C, Subsin B, Chen R, Pohorelic B, Sokol PA (2006) Burkholderia cenocepacia ZmpB fluence on survival in cystic fibrosis. Thorax 59(11):948–951. is a broad-specificity zinc metalloprotease involved in virulence. Infect Immun 74(7): 5. Coenye T, LiPuma JJ (2003) Molecular epidemiology of Burkholderia species. Front 4083–4093. Biosci 8:e55–e67. 31. McClean S, Callaghan M (2009) Burkholderia cepacia complex: Epithelial cell-patho- 6. Gotschlich A, et al. (2001) Synthesis of multiple N-acylhomoserine lactones is wide-spread gen confrontations and potential for therapeutic intervention. JMedMicrobiol among the members of the Burkholderia cepacia complex. Syst Appl Microbiol 24(1):1–14. 58(Pt 1):1–12. 7. Govan JR, Deretic V (1996) Microbial pathogenesis in cystic fibrosis: Mucoid Pseudo- 32. Sekirov I, Russell SL, Antunes LC, Finlay BB (2010) Gut microbiota in health and dis- – monas aeruginosa and Burkholderia cepacia. Microbiol Rev 60(3):539–574. ease. Physiol Rev 90(3):859 904. 8. Govan JR, Hughes JE, Vandamme P (1996) Burkholderia cepacia: Medical, taxonomic 33. Sekirov I, Finlay BB (2009) The role of the intestinal microbiota in enteric infection. – and ecological issues. J Med Microbiol 45(6):395–407. J Physiol 587(Pt 17):4159 4167. 9. Lewenza S, Conway B, Greenberg EP, Sokol PA (1999) Quorum sensing in Burkholderia 34. Erickson JA, Jalaie M, Robertson DH, Lewis RA, Vieth M (2004) Lessons in molecular cepacia: Identification of the LuxRI homologs CepRI. J Bacteriol 181(3):748–756. recognition: The effects of ligand and protein flexibility on molecular docking accu- – 10. Davies JC, Rubin BK (2007) Emerging and unusual Gram-negative infections in cystic racy. J Med Chem 47(1):45 55. fibrosis. Semin Respir Crit Care Med 28(3):312–321. 35. Wu G, Robertson DH, Brooks CL III, Vieth M (2003) Detailed analysis of grid-based 11. Njoroge J, Sperandio V (2009) Jamming bacterial communication: New approaches molecular docking: A case study of CDOCKER-A CHARMm-based MD docking algo- – for the treatment of infectious diseases. EMBO Mol Med 1(4):201–210. rithm. J Comput Chem 24(13):1549 1562. 12. Fuqua WC, Winans SC, Greenberg EP (1994) Quorum sensing in bacteria: The LuxR-LuxI 36. MacKerell ADJ, et al. (1998) CHARMM: The energy function and its parameterization with an overview of the program. Encyclopedia of Computational Chemistry,ed family of cell density-responsive transcriptional regulators. JBacteriol176(2):269–275. Schleyer PV (John Wiley, Chichester, UK), pp 271–277. 13. Antunes LC, Ferreira RB, Buckner MM, Finlay BB (2010) Quorum sensing in bacterial 37. Miyazaki K, Takenouchi M (2002) Creating random mutagenesis libraries using virulence. Microbiology 156(Pt 8):2271–2282. megaprimer PCR of whole plasmid. Biotechniques 33(5):1033–1034, 1036–1038. 14. Ng WL, Bassler BL (2009) Bacterial quorum-sensing network architectures. Annu Rev 38. Sambrook J, Fritsch EF, Maniatis T (2008) Molecular Cloning: A Laboratory Manual Genet 43:197–222. (Cold Spring Harbor Lab Press, Cold Spring Harbor, NY). 15. Rasmussen TB, Givskov M (2006) Quorum-sensing inhibitors as anti-pathogenic drugs. 39. Steidle A, et al. (2001) Visualization of N-acylhomoserine lactone-mediated cell-cell Int J Med Microbiol 296(2-3):149–161. communication between bacteria colonizing the tomato rhizosphere. Appl Environ 16. Rasmussen TB, Givskov M (2006) Quorum sensing inhibitors: A bargain of effects. Microbiol 67(12):5761–5770. Microbiology 152(Pt 4):895–904. 40. Nadal Jimenez P, et al. (2010) Role of PvdQ in Pseudomonas aeruginosa virulence 17. Sokol PA, Malott RJ, Riedel K, Eberl L (2007) Communication systems in the genus Bur- under iron-limiting conditions. Microbiology 156(Pt 1):49–59. kholderia: Global regulators and targets for novel antipathogenic drugs. Future Microbiol 41. Winson MK, et al. (1998) Construction and analysis of luxCDABE-based plasmid sen- 2(5):555–563. sors for investigating N-acyl homoserine lactone-mediated quorum sensing. FEMS 18. Dong YH, Wang LY, Zhang LH (2007) Quorum-quenching microbial infections: Microbiol Lett 163(2):185–192. Mechanisms and implications. Philos Trans R Soc Lond B Biol Sci 362(1483):1201–1211. 42. Kabsch W (1993) Automatic processing of rotation diffraction data from crystals of 19. Dong YH, Zhang LH (2005) Quorum sensing and quorum-quenching enzymes. initially unknown symmetry and cell constants. J Appl Crystallogr 26(Pt 6):795–800. J Microbiol 43(1):101–109. 43. Evans P (2006) Scaling and assessment of data quality. Acta Crystallogr D Biol Crys- 20. Uroz S, Dessaux Y, Oger P (2009) Quorum sensing and quorum quenching: The yin tallogr 62(Pt 1):72–82. – and yang of bacterial communication. ChemBioChem 10(2):205 216. 44. Collaborative Computational Project, Number 4 (1994) The CCP4 suite: Programs for 21. Sio CF, et al. (2006) Quorum quenching by an N-acyl-homoserine lactone acylase from protein crystallography. Acta Crystallogr D Biol Crystallogr 50(Pt 5):760–763. – Pseudomonas aeruginosa PAO1. Infect Immun 74(3):1673 1682. 45. McCoy AJ, Grosse-Kunstleve RW, Storoni LC, Read RJ (2005) Likelihood-enhanced fast 22. Dong YH, et al. (2001) Quenching quorum-sensing-dependent bacterial infection by translation functions. Acta Crystallogr D Biol Crystallogr 61(Pt 4):458–464. – an N-acyl homoserine lactonase. Nature 411(6839):813 817. 46. Emsley P, Lohkamp B, Scott WG, Cowtan K (2010) Features and development of Coot. 23. Papaioannou E, et al. (2009) Quorum-quenching acylase reduces the virulence of Acta Crystallogr D Biol Crystallogr 66(Pt 4):486–501. Pseudomonas aeruginosa in a Caenorhabditis elegans infection model. Antimicrob 47. Murshudov GN, Vagin AA, Dodson EJ (1997) Refinement of macromolecular structures – Agents Chemother 53(11):4891 4897. by the maximum-likelihood method. Acta Crystallogr D Biol Crystallogr 53(Pt 3):240–255. 24. Reimmann C, et al. (2002) Genetically programmed autoinducer destruction reduces 48. Winn MD, Murshudov GN, Papiz MZ (2003) Macromolecular TLS refinement in virulence gene expression and swarming motility in Pseudomonas aeruginosa PAO1. REFMAC at moderate resolutions. Methods Enzymol 374:300–321. Microbiology 148(Pt 4):923–932. 49. Davis IW, et al. (2007) MolProbity: All-atom contacts and structure validation for 25. Uroz S, et al. (2003) Novel bacteria degrading N-acylhomoserine lactones and their proteins and nucleic acids. Nucleic Acids Res 35(Web Server issue):W375–W383. use as quenchers of quorum-sensing-regulated functions of plant-pathogenic bacte- 50. Römling U, et al. (1994) Epidemiology of chronic Pseudomonas aeruginosa infections ria. Microbiology 149(Pt 8):1981–1989. in cystic fibrosis. J Infect Dis 170(6):1616–1621. 26. Huang JJ, Han JI, Zhang LH, Leadbetter JR (2003) Utilization of acyl-homoserine lac- 51. Huber B, et al. (2001) The cep quorum-sensing system of Burkholderia cepacia H111 tone quorum signals for growth by a soil pseudomonad and Pseudomonas aerugi- controls biofilm formation and swarming motility. Microbiology 147(Pt 9):2517–2528. nosa PAO1. Appl Environ Microbiol 69(10):5941–5949. 52. Wahjudi M, et al. (2011) PA0305 of Pseudomonas aeruginosa is a quorum quenching 27. Koch G, et al. (2010) The acylase PvdQ has a conserved function among fluorescent acylhomoserine lactone acylase belonging to the Ntn hydrolase superfamily. Pseudomonas spp. Environ Microbiol Rep 2(3):433–439. Microbiology 157(Pt 7):2042–2055.

6of6 | www.pnas.org/cgi/doi/10.1073/pnas.1311263111 Koch et al. Downloaded by guest on September 26, 2021