Comparative genomic analysis of Lactobacillus …Comparative genomic analysis of Lactobacillus...

6
Comparative genomic analysis of Lactobacillus rhamnosus GG reveals pili containing a human- mucus binding protein Matti Kankainen a,b,1,2 , Lars Paulin a,1 , Soile Tynkkynen b , Ingemar von Ossowski c,1 , Justus Reunanen c , Pasi Partanen a , Reetta Satokari d , Satu Vesterlund d , Antoni P. A. Hendrickx e , Sarah Lebeer f , Sigrid C. J. De Keersmaecker f , Jos Vanderleyden f , Tuula Ha ¨ma ¨la ¨ inen a , Suvi Laukkanen a , Noora Salovuori a , Jarmo Ritari a , Edward Alatalo a , Riitta Korpela b,g , Tiina Mattila-Sandholm b , Anna Lassig h , Katja Hatakka b , Katri T. Kinnunen b , Heli Karjalainen b , Maija Saxelin b , Kati Laakso b , Anu Surakka b , Airi Palva c , Tuomas Salusja ¨ rvi b , Petri Auvinen a , and Willem M. de Vos c,i,2 a Institute of Biotechnology, c Department of Basic Veterinary Sciences, g Institute of Biomedicine Pharmacology, and h Institute of Nutrition, University of Helsinki, P.O. Box 56, FIN-00014, Helsinki, Finland; b Research and Development, Valio Ltd., P.O. Box 30, FIN-00039, Helsinki, Finland; d Functional Foods Forum, University of Turku, Ita ¨ inen Pitka ¨ katu 4 A, FIN-20014, Turku, Finland; e Department of Medical Microbiology, University Medical Center Utrecht, 3584 CX, Utrecht, The Netherlands; f Centre of Microbial and Plant Genetics, Katholieke Universiteit Leuven, B-3001, Leuven, Belgium; and i Laboratory of Microbiology, Wageningen University, 6703 HB, Wageningen, The Netherlands Communicated by Todd R. Klaenhammer, North Carolina State University, Raleigh, NC, August 6, 2009 (received for review March 26, 2009) To unravel the biological function of the widely used probiotic bacterium Lactobacillus rhamnosus GG, we compared its 3.0-Mbp genome sequence with the similarly sized genome of L. rhamnosus LC705, an adjunct starter culture exhibiting reduced binding to mucus. Both genomes demonstrated high sequence identity and synteny. However, for both strains, genomic islands, 5 in GG and 4 in LC705, punctuated the colinearity. A significant number of strain-specific genes were predicted in these islands (80 in GG and 72 in LC705). The GG-specific islands included genes coding for bacteriophage components, sugar metabolism and transport, and exopolysaccharide biosynthesis. One island only found in L. rham- nosus GG contained genes for 3 secreted LPXTG-like pilins (spaCBA) and a pilin-dedicated sortase. Using anti-SpaC antibodies, the physical presence of cell wall-bound pili was confirmed by immu- noblotting. Immunogold electron microscopy showed that the SpaC pilin is located at the pilus tip but also sporadically through- out the structure. Moreover, the adherence of strain GG to human intestinal mucus was blocked by SpaC antiserum and abolished in a mutant carrying an inactivated spaC gene. Similarly, binding to mucus was demonstrated for the purified SpaC protein. We con- clude that the presence of SpaC is essential for the mucus inter- action of L. rhamnosus GG and likely explains its ability to persist in the human intestinal tract longer than LC705 during an inter- vention trial. The presence of mucus-binding pili on the surface of a nonpathogenic Gram-positive bacterial strain reveals a previ- ously undescribed mechanism for the interaction of selected pro- biotic lactobacilli with host tissues. genome probiotics adhesion pilus lactic acid bacteria T he Gram-positive lactobacilli are commensal inhabitants of the gastrointestinal (GI) tract that also play important roles in the production and preservation of food. Based on their health-promoting effects, these bacteria are commonly marketed as probiotics (1–3). One postulated feature considered indis- pensable for some probiotic lactobacilli is adherence to human intestinal tissues, which may promote a variety of specific interactions with the host (4–6). Despite many studies demon- strating the adherent properties of lactobacilli, the molecular mechanisms governing these host–microbe interactions are only beginning to emerge as comparative and functional analyses of their genome sequences progress (7, 8). Cell surface components that promote the adherence of lactobacilli include exopolysac- charides (EPSs); teichoic acids (TAs); and surface-exposed proteins, such as the S-layer and LPXTG-like proteins (4). Several of these components also function as modulators of the immune response, such as in the Lactobacillus plantarum D- alanine–depleted TA activation of the Toll-like receptor 2 (9) and the Lactobacillus acidophilus S-layer stimulation of the DC-SIGN receptor on dendritic cells (10). Recently, it was reported that specific immune responses were elicited in human subjects by L. plantarum and Lactobacillus rhamnosus, although the molecular mechanisms of these interactions await further characterization (11, 12). A possible mechanism for adherence and colonization that has not yet been identified in lactobacilli (4) but is well established in many Gram-positive pathogens involves proteinaceous sur- face-exposed polymeric structures known as pili (13, 14). These Gram-positive pili have a narrow diameter (1–10 nm) but can project outwardly from the cell surface with lengths of 1 m or more. Unlike the Gram-negative pili, each Gram-positive pilus is an assembly of multiple pilin subunits coupled to each other covalently by the transpeptidase activity of the pilin-specific sortase (13–15). Recent studies have established that the result- ing isopeptide bonds are formed between the threonine of an LPXTG-like motif and the lysine of an YPKN pilin motif in the different pilin subunits (15, 16). Another membrane-bound transpeptidase, the housekeeping sortase, also recognizing the LPXTG-like motif on target pilins is responsible for attaching the base of the elongated pilus covalently to the peptidoglycan in the cell wall (13–15). Typically, the pilus is a heterotrimer composed of a major pilin forming the pilus shaft, a minor pilin decorating the pilus backbone, and another minor pilin with adhesive properties often situated at the pilus tip (13, 14). The genes coding for the 3 pilin subunits and the pilin-specific sortase Author contributions: S.T., J.V., R.K., T.M.-S., A.P., T.S., P.A., and W.M.d.V. designed research; I.v.O., J. Reunanen, R.S., S.V., A.P.A.H., S. Lebeer, S.C.J.D.K., A.L., K.H., H.K., and K.L. performed research; M.K., L.P., S.T., I.v.O., J. Reunanen, P.P., R.S., S.V., A.P.A.H., S. Lebeer, S.C.J.D.K., T.H., S. Laukkanen, N.S., J. Ritari, E.A., K.T.K., A.S., T.S., P.A., and W.M.d.V. analyzed data; and M.K., S.T., I.v.O., M.S., A.S., A.P., T.S., P.A., and W.M.d.V. wrote the paper. Conflict of interest statement: S.T., R.K., T.M.-S., K.H., K.T.K., H.K., M.S., K.L., A.S., and T.S. are employed by Valio Ltd, which produces and markets the L. rhamnosus GG and LC705 strains. M.K., L.P., P.A., S. Lebeer, S.C.J.D.K, J.V., and A.P. have received research funds from Valio Ltd. Freely available online through the PNAS open access option. Data deposition: The sequences reported in this paper have been deposited in the Euro- pean Molecular Biology Laboratory Nucleotide Sequence Database, www.ebi.ac.uk/embl (accession nos. FM179322, FM179323, and FM179324). 1 M.K., L.P., and I.v.O. contributed equally to this work. 2 To whom correspondence should be addressed. E-mail: matti.kankainen@helsinki.fi or [email protected]. This article contains supporting information online at www.pnas.org/cgi/content/full/ 0908876106/DCSupplemental. www.pnas.orgcgidoi10.1073pnas.0908876106 PNAS October 6, 2009 vol. 106 no. 40 17193–17198 MICROBIOLOGY Downloaded by guest on August 10, 2021

Transcript of Comparative genomic analysis of Lactobacillus …Comparative genomic analysis of Lactobacillus...

Page 1: Comparative genomic analysis of Lactobacillus …Comparative genomic analysis of Lactobacillus rhamnosusGG reveals pili containing a human-mucus binding protein Matti Kankainena,b,1,2,

Comparative genomic analysis of Lactobacillusrhamnosus GG reveals pili containing a human-mucus binding proteinMatti Kankainena,b,1,2, Lars Paulina,1, Soile Tynkkynenb, Ingemar von Ossowskic,1, Justus Reunanenc, Pasi Partanena,Reetta Satokarid, Satu Vesterlundd, Antoni P. A. Hendrickxe, Sarah Lebeerf, Sigrid C. J. De Keersmaeckerf,Jos Vanderleydenf, Tuula Hamalainena, Suvi Laukkanena, Noora Salovuoria, Jarmo Ritaria, Edward Alataloa,Riitta Korpelab,g, Tiina Mattila-Sandholmb, Anna Lassigh, Katja Hatakkab, Katri T. Kinnunenb, Heli Karjalainenb,Maija Saxelinb, Kati Laaksob, Anu Surakkab, Airi Palvac, Tuomas Salusjarvib, Petri Auvinena, and Willem M. de Vosc,i,2

aInstitute of Biotechnology, cDepartment of Basic Veterinary Sciences, gInstitute of Biomedicine Pharmacology, and hInstitute of Nutrition, University ofHelsinki, P.O. Box 56, FIN-00014, Helsinki, Finland; bResearch and Development, Valio Ltd., P.O. Box 30, FIN-00039, Helsinki, Finland; dFunctional FoodsForum, University of Turku, Itainen Pitkakatu 4 A, FIN-20014, Turku, Finland; eDepartment of Medical Microbiology, University Medical Center Utrecht,3584 CX, Utrecht, The Netherlands; fCentre of Microbial and Plant Genetics, Katholieke Universiteit Leuven, B-3001, Leuven, Belgium; and iLaboratoryof Microbiology, Wageningen University, 6703 HB, Wageningen, The Netherlands

Communicated by Todd R. Klaenhammer, North Carolina State University, Raleigh, NC, August 6, 2009 (received for review March 26, 2009)

To unravel the biological function of the widely used probioticbacterium Lactobacillus rhamnosus GG, we compared its 3.0-Mbpgenome sequence with the similarly sized genome of L. rhamnosusLC705, an adjunct starter culture exhibiting reduced binding tomucus. Both genomes demonstrated high sequence identity andsynteny. However, for both strains, genomic islands, 5 in GG and4 in LC705, punctuated the colinearity. A significant number ofstrain-specific genes were predicted in these islands (80 in GG and72 in LC705). The GG-specific islands included genes coding forbacteriophage components, sugar metabolism and transport, andexopolysaccharide biosynthesis. One island only found in L. rham-nosus GG contained genes for 3 secreted LPXTG-like pilins (spaCBA)and a pilin-dedicated sortase. Using anti-SpaC antibodies, thephysical presence of cell wall-bound pili was confirmed by immu-noblotting. Immunogold electron microscopy showed that theSpaC pilin is located at the pilus tip but also sporadically through-out the structure. Moreover, the adherence of strain GG to humanintestinal mucus was blocked by SpaC antiserum and abolished ina mutant carrying an inactivated spaC gene. Similarly, binding tomucus was demonstrated for the purified SpaC protein. We con-clude that the presence of SpaC is essential for the mucus inter-action of L. rhamnosus GG and likely explains its ability to persistin the human intestinal tract longer than LC705 during an inter-vention trial. The presence of mucus-binding pili on the surface ofa nonpathogenic Gram-positive bacterial strain reveals a previ-ously undescribed mechanism for the interaction of selected pro-biotic lactobacilli with host tissues.

genome probiotics adhesion pilus lactic acid bacteria

The Gram-positive lactobacilli are commensal inhabitants ofthe gastrointestinal (GI) tract that also play important roles

in the production and preservation of food. Based on theirhealth-promoting effects, these bacteria are commonly marketedas probiotics (1–3). One postulated feature considered indis-pensable for some probiotic lactobacilli is adherence to humanintestinal tissues, which may promote a variety of specificinteractions with the host (4–6). Despite many studies demon-strating the adherent properties of lactobacilli, the molecularmechanisms governing these host–microbe interactions are onlybeginning to emerge as comparative and functional analyses oftheir genome sequences progress (7, 8). Cell surface componentsthat promote the adherence of lactobacilli include exopolysac-charides (EPSs); teichoic acids (TAs); and surface-exposedproteins, such as the S-layer and LPXTG-like proteins (4).Several of these components also function as modulators of theimmune response, such as in the Lactobacillus plantarum D-

alanine–depleted TA activation of the Toll-like receptor 2 (9)and the Lactobacillus acidophilus S-layer stimulation of theDC-SIGN receptor on dendritic cells (10). Recently, it wasreported that specific immune responses were elicited in humansubjects by L. plantarum and Lactobacillus rhamnosus, althoughthe molecular mechanisms of these interactions await furthercharacterization (11, 12).

A possible mechanism for adherence and colonization that hasnot yet been identified in lactobacilli (4) but is well establishedin many Gram-positive pathogens involves proteinaceous sur-face-exposed polymeric structures known as pili (13, 14). TheseGram-positive pili have a narrow diameter (1–10 nm) but canproject outwardly from the cell surface with lengths of 1 m ormore. Unlike the Gram-negative pili, each Gram-positive pilusis an assembly of multiple pilin subunits coupled to each othercovalently by the transpeptidase activity of the pilin-specificsortase (13–15). Recent studies have established that the result-ing isopeptide bonds are formed between the threonine of anLPXTG-like motif and the lysine of an YPKN pilin motif in thedifferent pilin subunits (15, 16). Another membrane-boundtranspeptidase, the housekeeping sortase, also recognizing theLPXTG-like motif on target pilins is responsible for attachingthe base of the elongated pilus covalently to the peptidoglycanin the cell wall (13–15). Typically, the pilus is a heterotrimercomposed of a major pilin forming the pilus shaft, a minor pilindecorating the pilus backbone, and another minor pilin withadhesive properties often situated at the pilus tip (13, 14). Thegenes coding for the 3 pilin subunits and the pilin-specific sortase

Author contributions: S.T., J.V., R.K., T.M.-S., A.P., T.S., P.A., and W.M.d.V. designedresearch; I.v.O., J. Reunanen, R.S., S.V., A.P.A.H., S. Lebeer, S.C.J.D.K., A.L., K.H., H.K., andK.L. performed research; M.K., L.P., S.T., I.v.O., J. Reunanen, P.P., R.S., S.V., A.P.A.H., S.Lebeer, S.C.J.D.K., T.H., S. Laukkanen, N.S., J. Ritari, E.A., K.T.K., A.S., T.S., P.A., and W.M.d.V.analyzed data; and M.K., S.T., I.v.O., M.S., A.S., A.P., T.S., P.A., and W.M.d.V. wrote thepaper.

Conflict of interest statement: S.T., R.K., T.M.-S., K.H., K.T.K., H.K., M.S., K.L., A.S., and T.S.are employed by Valio Ltd, which produces and markets the L. rhamnosus GG and LC705strains. M.K., L.P., P.A., S. Lebeer, S.C.J.D.K, J.V., and A.P. have received research funds fromValio Ltd.

Freely available online through the PNAS open access option.

Data deposition: The sequences reported in this paper have been deposited in the Euro-pean Molecular Biology Laboratory Nucleotide Sequence Database, www.ebi.ac.uk/embl(accession nos. FM179322, FM179323, and FM179324).

1M.K., L.P., and I.v.O. contributed equally to this work.

2To whom correspondence should be addressed. E-mail: [email protected] [email protected].

This article contains supporting information online at www.pnas.org/cgi/content/full/0908876106/DCSupplemental.

www.pnas.orgcgidoi10.1073pnas.0908876106 PNAS October 6, 2009 vol. 106 no. 40 17193–17198

MIC

ROBI

OLO

GY

Dow

nloa

ded

by g

uest

on

Aug

ust 1

0, 2

021

Page 2: Comparative genomic analysis of Lactobacillus …Comparative genomic analysis of Lactobacillus rhamnosusGG reveals pili containing a human-mucus binding protein Matti Kankainena,b,1,2,

are usually localized at the same locus as a gene cluster and arefrequently flanked on both ends by transposable elements,suggesting an origin by horizontal gene transfer (13).

L. rhamnosus GG, originally cultured from a healthy humanintestinal source, has been thoroughly studied and used safely asa probiotic strain in a variety of functional foods for nearly 20years (2, 3, 17–20). In our efforts to decipher the molecularmechanisms involved in the interaction between strain GG andthe human host, we determined the complete genome sequencesfor L. rhamnosus GG and L. rhamnosus LC705, an industrialstrain used routinely as an adjunct starter culture in dairyproducts (21). Several studies have compared the adhesionproperties of these two strains and concluded that GG isconsiderably more adherent than LC705 to human tissue (22,23). In this study, the presence of a cluster of pilus-encodinggenes (spaCBA) associated with the genome of strain GG isreported, and the expression of a pilin subunit (SpaC) wasconfirmed by immunogold electron microscopy. Employing anti-SpaC immune serum and an L. rhamnosus GG mutant carryingan inactivated spaC gene, it was demonstrated that SpaC is a keyfactor for adhesion between strain GG and human intestinalmucus. These findings and the comparative genomic analysiswith strain LC705 provide an important framework for under-standing a molecular mechanism for the interaction of L.rhamnosus GG with human host tissues.

ResultsComparative Genomics of the L. rhamnosus GG and LC705 Genomes.The complete genomes of the L. rhamnosus GG and LC705strains were sequenced and annotated. Both genomes are amongthe largest sized of the lactobacilli genomes, which typicallyaverage 2 Mbp (7, 8). The genome of strain GG contains asingle circular chromosome 3.01 Mbp in size, whereas strainLC705 contains a slightly smaller chromosome (2.97 Mbp) anda 64.5-kb circular plasmid, pLC1 [Table 1 and supportinginformation (SI) Fig. S1]. Between the two L. rhamnosus strains,3,000 predicted proteins are conserved with a high averageamino-acid identity (98%) (Table S1). However, a detailedcomparison of the predicted protein sequences equal to or longerthan 100 aa revealed that strain GG contains 331 strain-specificproteins, somewhat less than the 383 proteins for LC705.

The genomes of strains GG and LC705 exhibit a high degreeof synteny that is unexpectedly interrupted by several genomicislands (Fig. S1 and Fig. S2 A). Five genomic islands in strain GG(designated GGISL1-5) are predicted to encode 80 uniqueproteins with lengths of 100 residues or more, whereas in strainLC705, 4 islands (designated LCISL1-4) encode 72 uniqueproteins. All the genomic islands were identified as those DNAsequences deviating in codon usage, base composition, anddinucleotide frequency from the rest of the genome (24).

Moreover, these genomic islands were not conserved betweenthe closely related strains (Fig. S2 A), suggesting that they hadoriginated by horizontal gene transfer. In general, horizontallytransferred gene islands display special biological functions, andthis appeared to be the case for the genomic islands found instrains GG and LC705. Genomic islands with the capacity totransport or metabolize sugars (GGISL1, LCISL1, LCISL2, andLCISL4) and produce specific EPSs (GGISL5 and LCISL3)were observed, as were 2 islands in strain GG enriched withphage-associated genes (GGISL3 and GGISL4). One of theprophages (GGISL3) extending from LGG_01086 toLGG_01143 resembles the Lactobacillus casei American TypeCulture Collection (ATCC) 334 prophage (8, 25, 26) to someextent. Remarkably, the GGISL2 island appears to encode a setof genes for 3 pilin proteins (SpaCBA) and a sortase, anobligatory requirement for the assembly of pilus structures,which are considered important for colonization and host inter-action in some Gram-positive bacteria (13, 14).

L. rhamnosus belongs to a taxonomic group (L. casei, Lacto-bacillus paracasei, and Lactobacillus zeae) that is evolutionarydistant from other lactobacilli (1). Genomic comparisons withother lactic acid bacteria (LAB) confirmed this taxonomicgrouping, and our analysis showed that L. casei ATCC334 (8) isthe nearest relative of L. rhamnosus GG and LC705. Themajority of the deduced proteomes (68–72%) for the two L.rhamnosus strains and L. casei are conserved and display a highdegree of average amino-acid identity (85%) (Table S1).Moreover, overall genomic comparisons showed that the colin-earity between each of the strains was interrupted by thegenomic islands described previously (Fig. S2 A). Additionalcomparisons between the predicted proteomes of differentLactobacillus spp. indicated that, apart from the L. casei strain,strains GG and LC705 were distinctly different from otherlactobacilli, showing an average amino-acid identity of only52–59% (Table S1 and Fig. S2B). In total, 143 and 176 proteinswere identified in strains GG and LC705, respectively, withoutequivalent proteins in other lactobacilli. Many of these predictedproteins (24 in GG and 22 in LC705) were related to carbohy-drate transport and metabolism (Fig. S3).

Carbohydrate utilization assays showed that strains GG andLC705 both use a variety of mono- and disaccharide substrates(Table S2), a metabolic activity considered advantageous forbacteria residing in the carbohydrate-rich proximal region of thesmall intestine (4). Both genomes encode a ubiquitous set ofphosphotransferase system (PTS) transporters (Table S3) as wellas many carbohydrate-metabolizing enzymes depicted in a met-abolic network reconstruction (Fig. S4A). An important meta-bolic feature commonly exploited in industrial applications is theability to use lactose, which is lost in strain GG because offrameshifts in the antiterminator (lacT) and 6-phospho--

Table 1. General genomic features of L. rhamnosus GG and LC705 and selected Lactobacillus spp.

Organism L. rhamnosus GG L. rhamnosus LC705 L. casei ATCC334 L. plantarum WCFS1 L. acidophilus NCFM L. johnsonii NCC 533

Genome size, Mbp 3.01 3.03 2.92 3.35 1.99 1.99No. genes 2,944 2,992 2,771 3,057 1,862 1,821Plasmids 0 1 1 3 0 0rRNA operons 5 5 5 5 4 6tRNA genes 57 61 59 70 61 79GC content, % 47 47 47 44 35 35Coding efficiency, % 85 85 82 84 88 89Average gene size, bp 873 868 867 917 945 977Transposases 69 29 110 35 19 18Prophage clusters 3 3 2 3 0 2CRISPR 1 0 1 0 1 0

GC content, frequency of G and C nucleotides.

17194 www.pnas.orgcgidoi10.1073pnas.0908876106 Kankainen et al.

Dow

nloa

ded

by g

uest

on

Aug

ust 1

0, 2

021

Page 3: Comparative genomic analysis of Lactobacillus …Comparative genomic analysis of Lactobacillus rhamnosusGG reveals pili containing a human-mucus binding protein Matti Kankainena,b,1,2,

galactosidase (lacG) genes (Fig. S4B). Other experimentallyverified metabolic differences, including the inability of strainGG to use rhamnose, ribose, and maltose, were explained bygenetic variations in enzymes or transporters (Fig. S4A). We alsoidentified 40 and 49 genes predicted to encode potential glyco-sidases in the genomes of strains GG and LC705, respectively.Because of their annotation and predicted cellular location,several of these (10 in GG and 9 in LC705) may participate inpeptidoglycan hydrolysis and conversion of complex polysaccha-rides and prebiotics to simple carbohydrates.

Unlike L. rhamnosus LC705, strain GG is unable to catabolizelactose or casein, the major nutritional components of milk.Given that nitrogen utilization would be important for growth ofL. rhamnosus within the host GI tract (4), we compared thepredicted protein metabolism of both strains. Biosynthesis andtransport of amino acids and peptides and the proteolyticmachineries in strains GG and LC705 are similar but differ inone major aspect (Fig. S4C and Table S3). Only strain LC705harbors a plasmid (pLC1) carrying the cysE and cysK genes,which permits the biosynthetic conversion of serine to cysteine(Fig. S4C). No obvious differences were observed in the pre-dicted enzymatic machinery for casein breakdown between thetwo L. rhamnosus strains, despite the observation that only strainLC705 can degrade this type of milk protein (Fig. S5). Bothstrains encode a cell envelope serine protease (PrtP), maturationprotein (PrtM), and proteinase (PrtR), in addition to a similarset of 25 peptidases. However, within the LC705 genome, a genefor an additional secreted subtilisin-like serine protease(LC705_02680) was predicted, which may be involved in caseindegradation.

Molecules Involved in Host–Microbe Interaction and Adhesion. Spe-cific secreted and small-sized soluble proteins produced by L.rhamnosus GG have been reported to modulate epithelial cellgrowth and immune responses (27, 28). We predicted thatsecreted or cell surface-exposed proteins (Table S3) encom-pass close to 7% of the deduced proteomes in both L.rhamnosus strains and that 85 proteins associated with eachgenome are soluble and may possibly contribute to bacterium–host interactions.

Other cell surface components putatively involved in host–cell interactions (29, 30) and potentially implicated in pro-moting beneficial health effects are the cell envelope-bound or-secreted EPSs. The EPS biosynthetic gene cluster (LCILS3)in the genome of strain LC705 (Fig. S1B) was nearly identicalto the clusters present in 4 other L. rhamnosus strains(ATCC9595, R, RW-9595M, and RW-6541M) producingrhamnose-rich EPSs (29). However, these other clusteredgenes are all genetically different from the EPS gene cluster inL. rhamnosus GG (GGISL5), which has recently been shownto produce a long galactose-rich EPS that reportedly modu-lates biofilm formation (30).

Bacteriocins are small antimicrobial peptides secreted bymany Gram-positive bacteria (31). Although some studies havereported the detection of a bacteriocin-like activity in L. rham-nosus GG, it is still unclear whether this strain actually producesan active bacteriocin (4, 23). The genome of strain GG containsan 8.7-kb putative type IIb bacteriocin operon (LGG_02385–LGG_02392) that encodes the various components required forbacteriocin synthesis, including the export protein, ABC/C39-type peptidase, 2-component signal transduction system, immu-nity protein, and bacteriocin. The predicted bacteriocin fromstrain GG consists of 2 short peptides, both containing thebacteriocin type II leader motif required for C39 peptidase-mediated recognition. An identically organized operon(LC705_02382–LC705_02390) was also detected in the genomeof strain LC705, although the export protein and the histidineprotein kinase both possess a single frameshift mutation, which

likely results in the translation of truncated and nonfunctionalproteins.

L. rhamnosus GG adheres to mucus and epithelial cell linesabout 10-fold more efficiently than strain LC705 (22, 23). Ahuman intervention study showed that strain GG persisted athigher levels than LC705 throughout the study and for 7 dayslonger in the GI tract of healthy volunteers (Fig. 1). Because themucus-binding capacity of strain GG appears to be sensitive toprotease (22), we speculated that a protein-mediated strain-specific adhesion was responsible for the high binding propertiesof strain GG. The comparative genomic analysis identifiedmultiple types of proteins (31 in GG and 37 in LC705) withdomains related to adhesion and host colonization (Fig. S6).Several of these proteins contained fibronectin-binding domains,but, surprisingly, only a single protein in both genomes(LGG_02337 in GG and LC705_02328 in LC705) containedmucus-binding domains (Table S3). It was noteworthy that 8proteins in strain GG, including the proteins encoded by thespaCBA gene cluster (see below), and 14 proteins in strain LC705were strain specific.

Demonstration of Mucus-Binding Pili in L. rhamnosus GG. Pili are cellsurface-localized protrusions that have been well characterizedin several Gram-positive pathogens. In general, two or threesubunit genes encoding a Gram-positive pilus are organized intoan operon, along with at least one sortase gene (13, 14). Genesencoding pilin subunits and pilin-specific sortases were identi-fied in the genomes of strains GG and LC705, an observationthat has not been documented previously in the genomes of otherLactobacillus species. Specifically, GG harbors 2 separate pilusloci in its genome: the spaCBA (GGISL2) and spaFED geneclusters (Fig. S1 A). In contrast, only 1 pilus locus identical to thespaFED gene cluster in strain GG was detected in the genomeof strain LC705 (Fig. 2 A and Fig. S1B). Aside from the closelyrelated Lactobacillus strains, a homology search did not yieldmatches to other bacterial pilins that exhibited high amino-acididentities. However, a low degree of amino-acid sequence iden-tity (30–40%) was observed for pilins found in related Gram-positive species, such as Enterococcus faecalis and Enterococcusfaecium (32). Based on these analyses, SpaA and SpaD werepredicted to be the major pilin subunits forming the pilus shaftor backbone, SpaB and SpaE the ancillary minor pilin subunits,and SpaC and SpaF the large-sized minor pilin subunits likelyfunctioning as adhesins. Moreover, several motifs found typically

0.00

0.25

0.50

0.75

1.00

-7 0 1 3 5 7 10 14 17 21

Per

sist

ence

in h

um

an v

olu

nte

ers

Detection day

Fig. 1. The colonization properties of L. rhamnosus GG and LC705 in thehuman GI tract. A short-term intervention study involving 12 healthy adultsassessed the colonization properties of two L. rhamnosus strains in the GI tract.Fecal samples were collected during the intervention (7 days) and postint-ervention (0–21 days) periods and analyzed for the presence of strains GG(black) and LC705 (white). The fraction of subjects containing the L. rhamno-sus strains above a detection threshold is presented (refer to SI Text for furtherdetails.)

Kankainen et al. PNAS October 6, 2009 vol. 106 no. 40 17195

MIC

ROBI

OLO

GY

Dow

nloa

ded

by g

uest

on

Aug

ust 1

0, 2

021

Page 4: Comparative genomic analysis of Lactobacillus …Comparative genomic analysis of Lactobacillus rhamnosusGG reveals pili containing a human-mucus binding protein Matti Kankainena,b,1,2,

in the primary structure of pilin were identified (Fig. 2 A). Eachpilin subunit contains a Sec-dependent secretion signal; anLPXTG-like motif; and, excluding SpaE, an E box. Moreover,the YPKN pilin-like motif (13, 14), which contains an essentiallysine residue whose side-chain amino group has been implicatedin isopeptide bonding, was also detected in the SpaA, SpaB, andSpaD pilins (Fig. 2 A and SI Text).

To confirm the expression of pili on the cell surface, cell wallprotein extracts from strains GG and LC705 were immunoblot-ted using polyclonal antibodies specific for Escherichia coli-expressed SpaC pilin (Fig. 2B). In addition to the 90-kDamonomeric SpaC subunit, a ladder of high molecular weight(HMW) protein complexes (lane 3) representing the variousextended lengths of pili was detected in the cell wall-associatedprotein fraction from strain GG, as typically observed forpiliated Gram-positive pathogens (14). Cell wall protein extractsfrom cells of strain LC705 (lane 4) served as a negative control,because the spaCBA cluster is not present in the LC705 genome.To establish whether the SpaC pilin is a component of the pilusstructure, the spaC gene was inactivated using genetic techniquesdescribed previously (30) and the resulting insertional mutantstrain (GG-spaC) was then analyzed by immunoblotting. Weobserved that bands for the HMW protein complex were notpresent in the mutant strain (lane 2), suggesting that SpaC hadnot polymerized into the pilus structure. However, we did detecta protein band that corresponds to the predicted size (56 kDa)for a C-terminal truncated SpaC protein produced because ofthe insertional inactivation.

Subsequently, direct localization of pili on the cell surface ofL. rhamnosus GG was demonstrated by immunogold transmis-sion electron microscopy (Fig. 3). Briefly, strain GG cellscultivated to stationary phase were treated with anti-SpaCpolyclonal antibodies, labeled with protein A-conjugated goldparticles (10 nm) and then negatively stained. As shown in theelectron micrograph (Fig. 3), multiple pili, averaging 10 to 50 percell and with lengths of up to 1 m, extend outwardly from thesurface of the cells, with the majority situated predominantlynear the cell poles (Fig. 3B). Significantly, the gold particles arenot merely confined to the pilus tip but are also found through-out the length of the pilus, indicating that multiple copies ofSpaC can be incorporated within the pilus structure.

To assess the receptor specificity of SpaC-containing pili, weexamined the adhesion properties between E. coli-expressedSpaC protein and human intestinal mucus. Binding of radiola-beled (125I) SpaC to mucus was observed and was approximately6-fold more than the background level of mucus binding byradiolabeled ovalbumin (Fig. 4). Competition with unlabeledSpaC protein showed that the mucus binding was inhibited in adose-related manner (P 2.0 107), suggesting that the SpaCpilin may recognize mucosa-related components. Moreover,binding of L. rhamnosus GG cells to mucus has been determinedpreviously (6, 22), but we observed a 10-fold reduction in mucus

*

1 2 3

220 -

120 -100 -

80 -

60 -50 -

40 -

30 -

20 -

kDa

220 -

120 -100 -

80 -

60 -50 -

40 -

30 -

20 -

kDa

4

HMW

B

GG: spaCBA

GG: spaFED

LC705: spaFED

22% 14% 20% 33%

97% 98% 97% 98%

A

Fig. 2. The unique spaCBA pili cluster of L. rhamnosus GG and its expression.(A) Schematic illustration of the spaCBA and spaFED gene clusters. Depictedare the sortase (dark gray arrows), pilin subunits (white arrows), Cna proteinB-type domain (gray), E box (black), YPKN pilin-like motif (black diagonalstripes), LPXTG-like motif (light gray arrowheads), and secretion signal (lightgray). The von Willebrand type A domain is indicated by black and gray verticalstripes. The best homology match between proteins (gray arrows) and %amino-acid identity are indicated. (B) Detection of cell wall-associated pro-teins by immunoblotting. E. coli-purified SpaC (lane 1), cell wall proteinextracts of strains GG-spaC (lane 2), GG (lane 3), and LC705 (lane 4) wereprepared from cells grown in Man–Rogosa–Sharpe broth, separated by SDS/PAGE, electroblotted to a membrane, and probed with SpaC antiserum. Thepositions of the monomeric SpaC protein (*), HMW protein complexes, andmolecular weight standards (kDa) are indicated.

A

B

Fig. 3. Identification of pili in L. rhamnosus GG by immunogold electronmicroscopy. L. rhamnosus GG was grown to stationary phase, treated withanti-SpaC serum, labeled with protein A-conjugated gold particles (10 nm),negatively stained, and examined by transmission electron microscopy. (A)High-resolution electron micrograph showing multiple pili and an isometricbacteriophage (black arrow). Also included is a panel inset adjusted forheightened contrast and darkness to highlight the pilus ultrastructure (whitearrow). (B) Electron micrograph showing pili clustered at the cell poles. (Bars:A, 200-nm; B, 500-nm.)

17196 www.pnas.orgcgidoi10.1073pnas.0908876106 Kankainen et al.

Dow

nloa

ded

by g

uest

on

Aug

ust 1

0, 2

021

Page 5: Comparative genomic analysis of Lactobacillus …Comparative genomic analysis of Lactobacillus rhamnosusGG reveals pili containing a human-mucus binding protein Matti Kankainena,b,1,2,

binding when cells were pretreated with SpaC antiserum (Fig. 5).To rule out steric hindrance by the antibodies bound to the pilusstructure as the reason for the reduced mucus adherence, wetested the mucus-binding properties of an insertional mutantstrain (GG-spaC). Our results showed that insertional inacti-vation of the spaC gene caused mucus binding to be nearlyeliminated (Fig. 5), providing strong evidence for a role of theSpaC pilin in mucin binding and likely contributing to retentionof strain GG during transit through the GI tract.

DiscussionWe determined the complete genome sequences of probioticstrain L. rhamnosus GG and adjunct starter culture strainLC705 and performed an in-depth comparative genomic sur-vey complemented by functional analyses. Our comparativeanalysis of the two genomes, each 3.0 Mbp in size, revealeda high level of sequence homology and genomic synteny.However, several differences were observed in the number ofstrain-specific genes and the genomic islands encoding pro-teins for EPS biosynthesis (29, 30), specific sugar utilization(Table S2), and bacteriophage production. Considering thatonly a few studies have addressed bacteriophages in L. rham-

nosus strains (26), it is noteworthy that an isometric bacterio-phage was observed in the electron micrographs of GG cells,possibly indicating that the phage components encoded by oneof the genome islands are functional (Fig. 3A).

An unexpected observation in the GG and LC705 genomeswas the presence of genes for 3 canonical pilus subunits and adedicated sortase typically required for assembling pili, protein-aceous appendages characterized previously in Gram-positivepathogens for facilitating host–cell attachment (13, 14). Twopilus gene clusters (spaCBA and spaFED) were identified withinthe GG genome, whereas only the spaFED cluster was predictedin strain LC705. To characterize the unique spaCBA cluster, weestablished SpaC pilin expression and verified the presence ofmultiple SpaCBA pili on the surface of GG cells. Moreover,using a mutant with an insertionally inactivated spaC gene andby treating GG cells with SpaC antiserum, it was shown thatSpaCBA pili mediate strain GG adherence to human intestinalmucus. Because the spaC insertional mutant used here hadsecreted a C-terminal truncated SpaC protein, the most likelyexplanation for a reduction in mucosal adhesion is that thetruncated form was missing the consensus sequences (E box,YPKN pilin-like motif, and sortase recognition site) consideredimportant for pilus assembly, and thus could not be coupledwithin the polymerizing pilus structure. Alternate explanationsfor the reduced mucus adherence may include a polar effect ofthe spaC mutation, the disruption of the pilus assembly processby incorporating a truncated SpaC, or the assembly of a trun-cated SpaC with an impaired mucus-binding site. However, thelatter explanation is less likely, because the most probableadhesion domain for SpaC is located within the translatedN-terminal region of the truncated protein. Interestingly, astretch of SpaC sequence (residue 137–262) is similar to the typeA domain of the von Willebrand factor (vWFA), which isconsidered a prerequisite for the adherence of Streptococcusagalactiae pili to epithelial cells (33). Despite interacting pri-marily with human extracellular matrix proteins (34), a region ofthe vWFA domain (residue 201–299) demonstrates weak ho-mology with a fucose-binding lectin domain, suggesting that thevWFA-like domain in SpaC may bind in a lectin-type manner, adistinct plausibility considering that heavily glycosylated mucinglycoproteins are the main component of mucus (35). Becausethe L. rhamnosus GG and LC705 genomes only encode a limitedset of strain-specific adhesins, it is reasonable to speculate thatthe prolonged intestinal persistence of strain GG found duringan intervention study (Fig. 1) may be attributed to the mucus-binding capacity of SpaCBA pili.

Following the initial reported health benefit of L. rhamnosusGG in 1987 (36), strain GG has become one of the mostcomprehensively studied probiotic cultures in use today (2, 3).However, a detailed understanding of the molecular mecha-nisms governing the interplay between the human host and strainGG has been hindered and largely undiscovered because ofinsufficient genomic information on L. rhamnosus GG. Bydetermining the complete genome sequence of strain GG and bycomparing it with the closely related strain LC705, the molecularframework has been expanded for the discovery of cell surfacecomponents and other factors involved in host–microbe inter-actions. Our findings represent the previously unreported ob-servation of pili in probiotic LAB, indicating how the probioticstrain GG may persist in the host and possibly compete withpathogens for residence sites in the human intestinal tract (6). Aswell, because pili from Gram-positive pathogens have estab-lished immunostimulatory effects (37), it is tempting to speculatethat the observed pilus structures could function as immunestimuli that may support some health-promoting properties of L.rhamnosus GG (17–19).

0

5

10

15

20

25

SpaC 2x 10x 100x Background

Fm

ol b

ou

nd

Sp

aC

Fig. 4. In vitro competitive binding of L. rhamnosus GG SpaC pilin to humanintestinal mucus. Recombinant SpaC pilin was radiolabeled (125I), and thebinding to human intestinal mucus was competitively inhibited by 2-, 10-, and100-fold increases in unlabeled SpaC protein (in molar amounts). Binding byradiolabeled ovalbumin defined the background level of mucus binding.Binding results are an average of 4 to 6 measurements (standard deviationsare depicted), and all dataset comparisons were considered significant (P 0.05). Competition with unlabeled SpaC caused dose-related inhibition ofmucus binding (P 2.0 107).

0

1

2

3

4

5

6

7

8

9

10

GG GG with SpaC antiserum

Ad

hes

ion

(%

) to

hu

man

inte

stin

al

mu

cus

GG-ΩspaC

Fig. 5. Adhesion of L. rhamnosus GG and GG-spaC to human intestinalmucus. Radiolabeled (3H) cells of the L. rhamnosus GG strain (with or withoutSpaC antiserum pretreatment) and the GG-spaC mutant were tested forbinding to human intestinal mucus. Cell-binding results are averaged from 4to 6 measurements, and the standard deviations are depicted. Comparisonsbetween strain GG in the absence or presence of SpaC antiserum and betweenstrain GG and the GG-spaC mutant were considered significant (P 0.05).

Kankainen et al. PNAS October 6, 2009 vol. 106 no. 40 17197

MIC

ROBI

OLO

GY

Dow

nloa

ded

by g

uest

on

Aug

ust 1

0, 2

021

Page 6: Comparative genomic analysis of Lactobacillus …Comparative genomic analysis of Lactobacillus rhamnosusGG reveals pili containing a human-mucus binding protein Matti Kankainena,b,1,2,

Materials and MethodsL. rhamnosus GG and LC705 strains were manipulated, sequenced, and ana-lyzed as described in SI Text. The genomic sequences of strains GG and LC705and the plasmid pLC1 have been deposited in the European Molecular BiologyLaboratory (EMBL) database under accession numbers FM179322, FM179323,and FM179324, respectively.

Detailed descriptions of methods for cloning, immunoblotting, adhesionassays, insertional mutant construction, intervention trials, and immunogoldelectron microscopy are provided in SI Text. The GI colonization study wasapproved by the Coordinating Ethics Committee for the Hospital District ofHelsinki and the Uusimaa region. Resected human intestinal tissue was thesource of mucus, and its recovery and use were approved by the Ethics

Committee for the Hospital District of Southwest Finland and with the fullyinformed written consent of the patients.

ACKNOWLEDGMENTS. We thank Docent Ilkka Palva for helpful discussions onpilus-related topics. We are grateful to the technician team at the Institute ofBiotechnology and thank Juha Laukonmaa and Tuula Vahasoyrinki at ValioLtd., Tine Verhoeven and Lindsey Broos at Katholieke Universiteit Leuven, andKatariina Kojo and Marko Sutinen at the University of Helsinki for their skillfultechnical assistance. Heikki Huhtinen, MD, PhD, from the Department ofSurgery, Turku University Central Hospital, is acknowledged for collectinghuman intestinal tissue samples. The Finnish Funding Agency for Technologyand Innovation (Grants 40274/06 and 259/05) and Academy of Finland (Grants118165 and 118602), including the funding for the Center of Excellence inMicrobial Food Safety Research, are acknowledged for supporting this work.

1. Felis GE, Dellaglio F (2007) Taxonomy of Lactobacilli and Bifidobacteria. Curr IssuesIntest Microbiol 8:44–61.

2. Bernardeau M, Guguen M, Vernoux JP (2006) Beneficial lactobacilli in food and feed:Long-term use, biodiversity and proposals for specific and realistic safety assessments.FEMS Microbiol Rev 30:487–513.

3. Saxelin M, Tynkkynen S, Mattila-Sandholm T, de Vos WM (2005) Probiotic and otherfunctional microbes: From markets to mechanisms. Curr Opin Biotechnol 16:204–211.

4. Lebeer S, Vanderleyden J, De Keersmaecker SC (2008) Genes and molecules of lacto-bacilli supporting probiotic action. Microbiol Mol Biol Rev 72:728–764.

5. Mack DR, Ahrne S, Hyde L, Wei S, Hollingsworth MA (2003) Extracellular MUC3 mucinsecretion follows adherence of Lactobacillus strains to intestinal epithelial cells in vitro.Gut 52:827–833.

6. Vesterlund S, Karp M, Salminen S, Ouwehand AC (2006) Staphylococcus aureus adheresto human intestinal mucus but can be displaced by certain lactic acid bacteria. Micro-biology 152:1819–1826.

7. Ventura M, et al. (2009) Genome-scale analyses of health-promoting bacteria: pro-biogenomics. Nat Rev Microbiol 7:61–71.

8. Makarova K, et al. (2006) Comparative genomics of the lactic acid bacteria. Proc NatlAcad Sci USA 103:15611–15616.

9. Grangette C, et al. (2005) Enhanced anti-inflammatory capacity of a Lactobacillusplantarum mutant synthesizing modified teichoic acids. Proc Natl Acad Sci USA102:10321–10326.

10. Konstantinov SR, et al. (2008) S layer protein A of Lactobacillus acidophilus NCFMregulates immature dendritic cell and T-cell functions. Proc Natl Acad Sci USA105:19474–19479.

11. van Baarlen P, et al. (2009) Differential NF-kappaB pathways induction by Lactobacillusplantarum in the duodenum of healthy humans correlating with immune tolerance.Proc Natl Acad Sci USA 106:2371–2376.

12. Di Caro S, et al. (2005) Effects of Lactobacillus GG on genes expression pattern in smallbowel mucosa. Dig Liver Dis 37:320–329.

13. Mandlik A, Swierczynski A, Das A, Ton-That H (2008) Pili in Gram-positive bacteria:Assembly, involvement in colonization and biofilm development. Trends Microbiol16:33–40.

14. Proft T, Baker EN (2009) Pili in Gram-negative and Gram-positive bacteria—Structure,assembly and their role in disease. Cell Mol Life Sci 66:613–635.

15. Budzik JM, Oh SY, Schneewind O (2009) Sortase D forms the covalent bond that linksBcpB to the tip of the Bacillus cereus pili. J Biol Chem 284:12989–12997.

16. Kang HJ, Coulibaly F, Clow F, Proft T, Baker EN (2007) Stabilizing isopeptide bondsrevealed in gram-positive bacterial pilus structure. Science 318:1625–1628.

17. Hatakka K, et al. (2001) Effect of long term consumption of probiotic milk on infectionsin children attending day care centres: Double blind, randomised trial. Br Med J322:1327-1329.

18. Kalliomaki M, et al. (2001) Probiotics in primary prevention of atopic disease: Arandomised placebo-controlled trial. Lancet 357:1076–1079.

19. Guarino A, Lo Vecchio A, Canani RB (2009) Probiotics as prevention and treatment fordiarrhea. Curr Opin Gastroenterol 25:18–23.

20. Salminen MK, et al. (2002) Lactobacillus bacteremia during a rapid increase in probioticuse of Lactobacillus rhamnosus GG in Finland. Clin Infect Dis 35:1155–1160.

21. Suomalainen T, Mayra-Makinen A (1999) Propionic acid bacteria as protective culturesin fermented milks and breads. Lait 79:165–174.

22. Tuomola EM, Ouwehand AC, Salminen SJ (2000) Chemical, physical and enzymaticpre-treatments of probiotic lactobacilli alter their adhesion to human intestinal mucusglycoproteins. Int J Food Microbiol 60:75–81.

23. Jacobsen CN, et al. (1999) Screening of probiotic activities of forty-seven strains ofLactobacillus spp. by in vitro techniques and evaluation of the colonization ability offive selected strains in humans. Appl Environ Microbiol 65:4949–4956.

24. Tu Q, Ding D (2003) Detecting pathogenicity islands and anomalous gene clusters byiterative discriminant analysis. FEMS Microbiol Lett 221:269–275.

25. Ventura M, et al. (2006) Comparative genomics and transcriptional analysis of proph-ages identified in the genomes of Lactobacillus gasseri, Lactobacillus salivarius, andLactobacillus casei. Appl Environ Microbiol 72:3130–3146.

26. Durmaz E, Miller MJ, Azcarate-Peril MA, Toon SP, Klaenhammer TR (2008) Genomesequence and characteristics of Lrm1, a prophage from industrial Lactobacillus rham-nosus strain M1. Appl Environ Microbiol 74:4601–4609.

27. Tao Y, et al. (2006) Soluble factors from Lactobacillus GG activate MAPKs and inducecytoprotective heat shock proteins in intestinal epithelial cells. Am J Physiol290:C1018–C1030.

28. Yan F, et al. (2007) Soluble proteins produced by probiotic bacteria regulate intestinalepithelial cell survival and growth. Gastroenterology 132:562–575.

29. Peant B, et al. (2005) Comparative analysis of the exopolysaccharide biosynthesis geneclusters from four strains of Lactobacillus rhamnosus. Microbiology 151:1839–1851.

30. Lebeer S, et al. (2009) Identification of a gene cluster for the biosynthesis of a longgalactose-rich exopolysaccharide in Lactobacillus rhamnosus GG and functional anal-ysis of the priming glycosyltransferase. Appl Environ Microbiol 75:3554–3563.

31. Klaenhammer TR (1993) Genetics of bacteriocins produced by lactic acid bacteria. FEMSMicrobiol Rev 12:39–85.

32. Hendrickx AP, et al. (2008) Expression of two distinct types of pili by a hospital-acquiredEnterococcus faecium isolate. Microbiology 154:3212–3223.

33. Konto-Ghiorghi Y, et al. (2009) Dual role for pilus in adherence to epithelial cells andbiofilm formation in Streptococcus agalactiae. PLoS Pathog 5:e1000422.

34. Whittaker CA, Hynes RO (2002) Distribution and evolution of von Willebrand/integrinA domains: Widely dispersed domains with roles in cell adhesion and elsewhere. MolBiol Cell 13:3369–3387.

35. Robbe C, Capon C, Coddeville B, Michalski JC (2004) Structural diversity and specificdistribution of O-glycans in normal human mucins along the intestinal tract. BiochemJ 384:307–316.

36. Gorbach SL, Chang TW, Goldin B (1987) Successful treatment of relapsing Clostridiumdifficile colitis with Lactobacillus GG. Lancet 2:1519–1519.

37. Barocchi MA, et al. (2006) A pneumococcal pilus influences virulence and host inflam-matory responses. Proc Natl Acad Sci USA 103:2857–2862.

17198 www.pnas.orgcgidoi10.1073pnas.0908876106 Kankainen et al.

Dow

nloa

ded

by g

uest

on

Aug

ust 1

0, 2

021