Apilactobacillus kunkeei releases RNA-associated membrane vesicles and proteinaceous nanoparticles

Abstract Extracellularly released particles, including membrane vesicles, have increasingly been recognized as important for bacterial community functions and host-interaction processes, but their compositions and functional roles differ between species and also between strains of the same species. In this study, we have determined the composition of membrane vesicles and protein particles identified in the cell-free pellets of two strains of Apilactobacillus kunkeei, a defensive symbiont of honeybees. The membrane vesicles were separated from the extracellular particles using density gradient ultracentrifugation. The peaks of the RNA and protein distributions were separated from each other and the highest concentration of RNA was observed in the fractions that contained the membrane vesicles while the highest protein concentration coincided with the fractions that contained extracellular particles. A comparative proteomics analysis by LC-MS/MS showed that 37 proteins with type-I signal peptides were consistently identified across the fractionated samples obtained from the cell-free pellets, of which 29 were orthologs detected in both strains. Functional predictions of the extracellular proteins revealed the presence of glycoside hydrolases, glycosyltransferases, giant proteins and peptidases. The extracellular transcriptomes mapped to a broad set of genes with a similar functional profile as the whole cell transcriptome. This study provides insights into the composition of membrane vesicles and extracellular proteins of a bee-associated symbiont.


Introduction
Extr acellular membr ane v esicles ar e pr oduced by cells fr om all domains of life and incr easingl y thought to play important roles for the coordination of cell population r esponses, inter actions with phages and viruses, host immune system regulation, pathogenicity, and detoxification (r e vie wed in Gill et al. 2019 ).Although the production of extracellular vesicles was initially consider ed to onl y occur in Gr am-negativ e bacteria, it is now well established that also Gr am-positiv e bacteria, despite their thick peptidogl ycan cell wall, pr oduce suc h v esicles (Lee et al. 2009, Brown et al. 2014, Schrempf and Merling 2015, Vdo viko va et al. 2017 ).Commonl y r eferr ed to as outer membr ane v esicles (OMVs) in Gr am-negativ e bacteria, membr ane v esicles (MVs) in Grampositive bacteria and exosomes in eukaryotic cells, the extracellular membr ane v esicles can tr ansfer a wide r ange of car go, including nucleic acids , proteins , and metabolites .
Numerous studies have suggested that RNA is a major component of membr ane v esicles in bacteria.For example, rRNA and tRNA fr a gments wer e shown to be enriched in the MVs of Prochlorococcus (Biller et al. 2014 ) and Esc heric hia coli (Ghosal et al. 2015, Blenkiron et al. 2016 ).Small RN As (sRN A) of 80-250 nucleotides were identified in the OMVs of Pseudomonas aeruginosa (Koeppen et al. 2016 ) and microRNA-size RNA of less than 70 nucleotides were detected in the MVs of se v er al other bacteria (Choi et al. 2017, Sahr et al. 2022, Yu et al. 2022 ).A broad study of the composition of RNA molecules in the MVs of Staphylococcus aureus identified 28 sRNAs and a wide range of RNA fragments that mapped to 273 genes coding for virulence-associated factors, ribosomal pr oteins, tr anscriptional regulators and metabolic enzymes (Luz et al. 2021 ).Inter estingl y, the number and abundances of RNA fr a gments in the MVs were not static but changed with growth conditions (Luz et al. 2021 ).
Pr oteins hav e also been identified as a component of MVs, such as in Lactobacillus species that ar e commonl y used in fermentation, food production and as probiotics (Grande et al. 2017, Li et al. 2017, Dean et al. 2019, 2020, Zheng et al. 2020, Hu et al. 2021 ).For example, more than 80 proteins, including surface proteins , glucoamylase , mucus binding protein, cell-wall hydrolase and lysozyme were identified in the MVs of Lactobacillus acidophilus, Lactobacillus casei and Lactobacillus reuteri (Dean et al. 2019 ).The r elativ e fr action of pr oteins with signal peptides was more than 5-fold enriched in the MVs compared to the protein content of the bacterial cells, suggesting that the MVs pr edominantl y contain secreted proteins (Dean et al. 2019 ).It was hypothesized that an incr eased expr ession of secr eted pr oteins pr eceded the formation of vesicles, while the actual pac ka ging of proteins into the MVs was a passive process.
Metabolites and antimicrobial compounds have also been identified in the membrane vesicles of some bacterial species.For example, the induction of a 10-kb operon for the synthesis of a putative bacteriocin in L. acidophilus resulted in a more than 500-fold enrichment of the bacteriocin in the MVs (Dean et al. 2020 ).Interestingl y, molecules deliv er ed via the MVs of L. plantarum hav e been shown to enhance host immune responses against vancomycin-r esistant enter ococci, suggesting that they may serv e a r ole in host pathogen defense (Li et al. 2017 ).
The mec hanisms wher eby the v esicles ar e formed hav e not been entir el y elucidated.Ho w e v er, se v er al str essors that trigger the formation of these nanoparticles of about 20-500 nm have been pr oposed, suc h as imbalanced cell wall biosynthesis, explosive cell lysis and bubbling cell death (To y ofuku et al. 2019 ).Pr opha ge-induced cell wall damage have been shown to promote extr acellular v esicle pr oduction in Bacillus subtilis and Lactococcus lactis (To y ofuku et al. 2017, Liu et al. 2021 ), as has also antibioticinduced cell stress in Stenotrophomonas maltophilia (Devos et al. 2017 ).Besides , MVs ha ve also been shown to be produced under basal growth conditions without the use of external stressors (Rodriguez andKuehn 2020 , Luz et al. 2021 ).
Apilactobacillus kunkeei (formerly called Lactobacillus kunkeei ) colonizes the honey crop and food products of honeybees and is thought to protect the bees from infections with bee pathogens (Vásquez et al. 2012, Butler et al. 2013, Olofsson et al. 2014 ).Differ ent str ains of these bacteria gr ow r a pidl y with gener ation times of 35 to 60 minutes under laboratory conditions and have genome sizes of about 1.5 Mb (Dyrhage et al. 2022 ).Previous studies of the secretome of A. kunkeei strain Fhon2 identified 24 secreted proteins (Butler et al. 2013 ), most of whic h ar e encoded by genes located near the origin of replication (Tamarit et al. 2015 ).It was hypothesized that some of these may confer antimicrobial functions (Butler et al. 2013 ), although this has not been experimentall y demonstr ated.
It has ho w e v er been shown that A. kunkeei produces kunkecin A, a nisin-like bacteriocin, which has a strong inhibitory activity against the bee pathogen Melissococcus plutonius (Zendo et al. 2020 ).The gene cluster for kunkecin A biosynthesis is located on a 19.5 kb plasmid and can thereby be shared within the population.A compar ativ e genomics study of more than 100 isolates from sever al beehiv es sho w ed that the plasmid w as br oadl y pr esent in isolates obtained from one of the hives (Dyrhage et al. 2022 ).Based on these results, it was hypothesized that A. kunkeei serve a role as a defensive symbiont of honeybees.
Because of its antimicrobial activities and viability in solutions of high sugar concentrations, it is considered that A. kunkeei may be used as a complement in fruit pr epar ations giv en to hospitalized and imm unocompr omised patients (Ver galito et al. 2020 ).The MVs of A. kunkeei can potentially be used for the delivery of engineer ed antimicr obials, v accines or pr obiotics.It is ther efor e of gener al inter est to c har acterize the particles secr eted by A. kunkeei under normal growth conditions and determine their composition.
Here , we ha ve performed a comprehensi ve stud y of secreted particles and membrane vesicles in A. kunkeei.For the purpose of this study, we have used the r a pidl y gr owing and plasmid-less A. kunkeei strains A1401 and A0901, which were isolated from the honey-crop of honeybees sampled from the island Åland in the Baltic Sea (Dyrhage et al. 2022 ).The genomes are very similar ov er all, but in comparison to A1401, strain A0901 contained a pr edicted pr opha ge (Dyrha ge et al. 2022 ).We have used electr on micr oscopy tec hniques to examine the structur es of the secr eted particles, density-gr adient ultr acentrifugation to separ ate the MVs from other types of secreted particles, along with transcriptomics and proteomics to identify their molecular contents.The results are discussed in the context of previously proposed models for the role that secreted proteins and membrane vesicles play for bacterial communities and their hosts.

Study design and w orkflo w
A summary of the methods used in this study is shown in Fig. 1 .We cultivated A. kunkeei strains A0901 and A1401, which grow with generation times of ≈35 min under laboratory conditions (Dyrhage et al. 2022 ).Their cell surface structures were first examined by cryo-electr on tomogr a phy (cryo-ET), Tr ansmission Electr on Tomogr a phy (TEM) and Scanning Electr on Tomogr a phy (SEM) at exponential growth phase.Next, we pelleted the bacterial cells b y lo w-speed centrifugation and isolated secreted material smaller than 0.45 μm from the supernatant by filtration and ultracentrifugation.The secreted material in the cell-free pellet (CFP) was examined by cryo-ET and negative strain TEM (nsTEM).We separated the particles in the CFP using density gradient ultracentrifugation.Finall y, we anal yzed the composition of the secr eted material in the CFPs by collecting proteomics and transcriptomics data.For comparison, we also collected proteomics and transcriptomics data from whole cell lysates (WCL) obtained after the first low-speed centrifugation step.

Isolation of whole cell lysate and cell-free supernatant
The A. kunkeei strains A0901 and A1401 were cultivated in MRSmedium (Sigma Aldrich) according to (de Man et al. 1960 ) supplemented with 0.5% (w/v) D-Fructose (Sigma Aldrich) and 0.1% (v/v) Tween-80 (Sigma Aldrich), in the following r eferr ed to as fMRS.Batc h-cultiv ation was performed using biological triplicates at 35 • C and 5% CO 2 in batc h-cultur es of 100-150 mL fMRS medium and the cells were harvested during exponential growth phase at OD600 ≈ 0.3.
After harvesting, cells and large debris were separated from the supernatant b y lo w-speed centrifugation at 4 500 x g for 10 min at 4 • C ('P0').To obtain samples from the WCL fraction, pellets P0 w ere w ashed twice in HyClone Hypure Water, Molecular Biology Gr ade (Cytiv a) befor e being r e-suspended in 25 mM Tris, pH 8.0, 6 M urea, 1x Sigma Fast protease inhibitors (Sigma Aldrich) to ac hie v e a cell concentr ation corr esponding to OD600 ≈ 20.Cells wer e l ysed by sonication using a Vibra-Cell VCX 130 sonicator by repeated 10 s pulses (10-15 cycles) followed by 10 s breaks using a 2 mm probe and 30-40% amplitude (on ice).After sonication, the cleared cell suspensions were centrifuged at 17 000 x g for 10 min at r oom temper atur e and the supernatant was car efull y r ecov er ed for SDS-PAGE and proteomics analysis.
To obtain samples from the cell-free supernatant (CFS), any remaining cells and debris were first removed from the crude supernatant ('S0') by filtering through 0.45 μm cellulose membrane filters connected to a 50 mL syringe .T he cell-fr ee filtr ate ('F1') was subjected to ultracentrifugation at 150 000 x g for 2 h at 4 • C using a 45Ti rotor in an Optima XPN-100 ultracentrifuge (Beckman).The supernatant ('S2') was car efull y decanted and the small and tr anslucent CFPs wer e allo w ed to air-dry for ≈ 5 min before being re-suspended in 25 mM Tris, 6 M urea, pH 8.0 for proteomic analysis.For complementary analysis using nsTEM and Nanoparticle Tr ac king Anal ysis (NTA), CFPs wer e r e-suspended in 10 mM Tris, pH 7.5 and for cryo-ET in 1x PBS.Those methods are described in more detail below .T o assess the complete r emov al of whole cells, 100-500 μl aliquots of filtrate F1 were plated on fMRS agar plates and incubated over night at 35 • C, 5% CO 2 along with the remaining 10-20 ml of filtrate F1.
F igure 1. Workflo w for isolation and c har acterization of secreted particles, CFP and WCL from A. kunkeei .After batc h-cultiv ation, isolation of CFPs containing ECPs and MVs is based on low-speed centrifugation, filtration and ultracentrifugation.For further analysis, the crude CFP was fractionated by DGUC using an Optiprep gradient.For comparative analysis, pellet P0 was processed to obtain the WCL.The techniques for analysis of the different fr actions ar e indicated.

Separ a tion of crude cell-free pellets by density gr adient ultr acentrifuga tion
Optipr ep gr adients wer e used for density-gr adient ultr acentrifugation (DGUC) of CFP pr epar ations fr om A. kunkeei A1401 and A0901.For this, discontinuous gradients were prepared by overlayering 500 μl of Optiprep (Sigma Aldrich) solution (40%, 30%, 20%, 10% in 25 mM Tris, pH 7.4, 0.25 M Sucr ose).The gr adients wer e stor ed at 4 • C for 60 min and 100 μL of three independent S0 CFP pr epar ations, r e-suspended in 25 mM Tris, pH 7.4, 0.25 M sucr ose, wer e car efull y placed on top of each gradient.Ultracentrifugation was performed in a Sorvall RC M150 GX centrifuge (Thermo Fisher Scientific) using a S55-S swinging-buc ket r otor (Thermo Fisher Scientific) for 4 h at 216 000 x g (4 • C).Fractions of 200 μL wer e car efull y collected fr om top-to-bottom (F1/top-F10/bottom) and stored at 4 • C. Fractions were analyzed by SDS-PAGE and nsTEM.P article concentr ations wer e estimated by NTA, pr otein concentr ation by Br adford (Thermo Fisher Scientific), absolute RN A concentration b y the Qubit RN A HS assay (Life Technologies), r elativ e RNA concentr ation b y SYTO RN ASelect (Life Technologies) and r elativ e membr ane abundance using the lipophilic dye FM 4-64.

Nanoparticle tracking analysis
The particle concentration and size distribution (based on hydrodynamic radius) of the CFP preparations were analyzed by NTA on a NanoSight LM14 instrument (Malv ern P anal ytical, λ = 405 nm and λ = 532 nm).Particles of each sample were recorded for 5 × 30 s and quantified using the NanoSight NTA software v3.4.

Nuclease protection assay
To investigate the localization of the RNA within the secreted particles, we performed nuclease protection assays on crude CFP pr epar ations of A. kunkeei A10901 and A1401.The assays were performed using biological triplicate samples fr om eac h str ain.After diluting the CFP pr epar ations 10-fold in 10 mM Tris, samples were split into a control group ('C') and a treated group ('PS').The treated group was first subjected to proteinase K treatment (56 • C, 60 min) at a final enzyme concentration of approximately 60 mAU/mL.In a second step, SDS was added to a final concentration of 1% (v/v), follo w ed b y incubation at 56 • C for 30 min.The concentration of the control samples was adjusted accordingly using 10 mM Tris and incubated at the same temper atur es as the treated sample.Total RNA concentration was measured using the Qubit RNA HS assay before and after addition of RNAse (final concentration 50 μg/mL, 37 • C, 90 min).

Electr on micr oscopy
TEM, nsTEM and SEM wer e performed at the electr on micr oscopy unit, EMiL, at Karolinska Institutet, Huddinge, Sweden, essentially as described in (Seeger et al. 2017, Mahajan et al. 2020 ).For TEM ( A. kunkeei strain A1401) and SEM analysis ( A. kunkeei strain A1401 and A0901, bacteria wer e gr own in fMRS medium until log-phase (OD600 0.2-0.6)and 1 mL cell suspension was gently pelleted by centrifugation at 500 x g for 2 min at room temperature.After removing the supernatant, the cells were fixed in 1 mL fixation solution (10 mM Phosphate buffer, pH 7.4, 2% glutaraldehyde, 1% formaldehyde) for 15 min at r oom temper atur e and stored at 4 • C until further processing.For nsTEM analysis of CFPs from A. kunkeei strains A1401 and A0901, particles were isolated as described above and processed as described in (Mahajan et al. 2020 ).Size distributions wer e gener ated by particle anal ysis in Fiji/Ima geJ (Schindelin et al. 2012, Schneider et al. 2012 ).

Cryo-transmission electron tomography
Cry o-ET w as performed on crude CFP pr epar ations and whole cells obtained from A. kunkeei strains A1401.For crude CFP prepar ations fr om A. kunkeei str ain A1401, R2/2 200 mesh grids with a 2 nm carbon support (QuantiFoil) were glow-disc har ged (20 mA 60 sec) on a Quorum GloQube. 10 nm protein-A gold fiducials (Aurion) wer e r esuspended and mixed pr operl y with the isolated particles at a 1:10 ratio.3 μL of this mixture were applied onto grids befor e plunge-fr eezing into liquid ethane in a Vitrobot Mark IV robot (FEI/Thermo Fisher Scientific) operated at 4 • C, 100% humidity and with a blot time of 5 seconds.Datasets were collected using a Talos Arctica microscope (FEI/Thermo Fisher Scientific) outfitted with a Falcon3 detector (FEI/Thermo Fisher Scientific) operated at 200 kV in nanoprobe and TEM mode, a C2 a pertur e size of 50 μm, an objectiv e a pertur e size of 70 μm.SerialEM software (Mastronar de 2005 ) w as used to acquir e eac h tilt-series using a dose-symmetric tilt scheme (Hagen et al. 2017 ) with a range of ±60 • , 2 • angular increment and a target defocus of -4 to -6 μm.Each tilt-series of 61 12-frame movies w as recor ded in counting mode with a pixel size of 1.23 Å at a dose rate of 0.8 e − /px/s for 3 sec and a total dose per tilt-series of ∼95e − / Å 2 .The initial raw movies were aligned and dose-weight filtered using 'alignframes' from the IMOD package (Mastronarde and Held 2017 ).Tilt-series were aligned using the gold fiducial markers, and tomograms were reconstructed b y w eighted bac k-pr ojection using pr ogr ams within IMOD v4.11.6 (Mastronarde and Held 2017 ).
For cryo-ET of whole cells, R2/2 200 mesh grids (QuantiFoil) w ere glo w-discharged (20 mA 60 sec) on a Quorum GloQube.10 nm protein-A gold fiducials (Aurion) were resuspended and mixed pr operl y with the cells at a 1:2 r atio.Thr ee micr oliters of this mixtur e wer e a pplied onto grids befor e plunge-fr eezing into liquid ethane in a Vitrobot Mark IV robot (FEI/Thermo Fisher Scientific) operated at 4 • C, 100% humidity and with manual back-side blotting of 5 seconds done within the Vitrobot humidified chamber.Datasets were collected using a Titan Krios G3i microscope (FEI/Thermo Fisher Scientific) outfitted with a K3 detector and Bio-Quantum imaging filter (Gatan) operated at 300 kV in nanoprobe and EF-TEM mode, a C2 a pertur e size of 50 μm, an objective aperture size of 100 μm, and an energy filter slit width of 20 eV in Zero-Loss mode.SerialEM software was used to acquire each tiltseries using a dose-symmetric tilt scheme with a range of ±60 • , 2 • angular increment and a target defocus of -3 to -6 μm.Each tilt-series of 61 10-frame movies w as recor ded in counting mode with a pixel size of 2.11 Å at a dose rate of 15 e − /px/s for 0.4 s and a total dose per tilt-series of ∼83e − / Å 2 .The initial raw movies were aligned and dose-weight filtered using 'alignframes' from the IMOD pac ka ge.Tilt-series wer e aligned using the gold fiducial markers, and tomograms were reconstructed by weighted backpr ojection using pr ogr ams within IMOD v4.11.6 (Mastronarde and Held 2017 ).Size distributions wer e gener ated by particle analysis in Fiji/ImageJ (Schindelin et al. 2012, Schneider et al. 2012 ).

Proteomics analysis
Rea gents for SDS-PAGE wer e purc hased fr om Life Tec hnologies.For SDS-PAGE of the extracted proteins from the WCL and CFP fractions, 1x NuPAGE LDS Sample Buffer and 1x NuPAGE Sample Reducing Agent were incubated at 70 • C for 10 min.About 10 μL sample and 5 μL Novex Sharp Unstained Protein Standard were loaded on a NuPAGE 4-12% Bis-Tris Protein Gel (1.0 mm) and electr ophor esis was performed in 1x MOPS SDS Running buffer supplemented with NuPAGE Antioxidant for 55 min at 200 V.The gel w as w ashed, stained and destained using SimplyBlue SafeStain based on the manufactur ers micr o w av e pr otocol and ima ged on a ChemiDoc MP imaging system (Bio-Rad).
Biological triplicates of fractions DGUC F3 and F6 from A. kunkeei strains A1401 and A0901 were subjected for proteomics analysis after in-solution digestion, essentially as described pr e viousl y (Seeger et al. 2021 ).Notably, for the F6 samples that contained a high concentration of proteins, 10 μg of proteins was used for the digestion (ca 10 μl of the sample).For the F3 samples that contained non-measurable concentration of proteins, the entire sample was used for the digestion (ca 100 μl of the sample).In addition, the gel bands from biological triplicates of both A. kunkeei A1401 and A0901 in the area between 110-150 kDa of fraction F6 were excised and subjected for proteomics analysis after in-gel digestion, similar as described pr e viousl y (Mahajan et al. 2020 ).
Extr acted pr oteins of the WCL and CFP fr actions wer e anal yzed by Label-free quantification (LFQ) by LC-MS/MS as described previously (Seeger et al. 2021 ).For the LFQ analyses, MS raw files wer e pr ocessed using MaxQuant software (Cox and Mann 2008 ) and the Andromeda search engine (Cox et al. 2011 ) against the genome sequences of A. kunkeei A0901 and A1401.For the qualitativ e anal ysis of the DGUC fr actions, database searc hes wer e performed using the Sequest algorithm, embedded in Proteome Discov er er 1.4 (Thermo Fisher Scientific) against the genome sequences of A. kunkeei A0901 and A. kunkeei A1401 that were also used for the analysis of the crude WCL and CFP fractions.Further analysis and visualization were performed in R (R Core Team 2021 ) using the pac ka ge ggplot (Wic kham 2016 ,R Cor e Team 2021 ).Tr ansmembr ane pr oteins wer e pr edicted using Phobius v1.1 (Käll et al. 2004(Käll et al. , 2007 ) ). Signal peptides (type I and type II) were predicted using the SignalP-5.0Server (Almagro Armenteros et al. 2019 ).Functional classification was based on the Clusters of Orthologous Genes, COGs (Tatusov et al. 1997, Galperin et al. 2021 ).

Tr anscriptomics anal ysis
For the transcriptomics analysis, the RN A w as extr acted fr om WCL and CFP fr actions following cultiv ation of A. kunkeei str ain A1401 in biological triplicates as described abo ve .After separation of the cell pellet P0 from the crude supernatant S0, P0 was resuspended in RNAlater, stored at 4 • C overnight before storage at -20 • C. The cell-free pellet (CFP) was re-suspended in Hy-Clone Hypure Water, Molecular Biology Grade (Cytiva) and stored at -20 • C.
The RNA extraction was performed using the RiboPure-Bacteria Kit including DNAse I treatment (LifeTechnologies).Sequencing libr aries wer e pr epar ed fr om 500 ng total RNA using the Illumina Str anded Total RNA Pr ep, Ligation with Ribo-Zer o Plus kit (Illumina, cat#20 040 525).The libr ary pr epar ation was performed according to the manufacturers' protocol (# 1000000124514-00).Sequencing was performed on a MiSeq system using v3 sequencing chemistry (Illumina) and a paired-end 75 bp read length.
For the tr anscriptomics anal yses, the obtained sequence reads wer e ma pped to the genome sequence of A. kunkeei strain A1401 using hisat2 v2.2.1 (Kim et al. 2019 ).Read counts for the 1423 predicted coding sequences, including 9 predicted extrachromosomal coding sequence (Dyrhage et al. 2022 ), were determined using FeatureCounts v2.0.1 (Liao et al. 2014 ).For filtering genes with low expression levels and for allowing comparisons between genes , ra w read counts were transformed to transcripts per million (TPM) (Wagner et al. 2012 ).Only genes with transcripts with TPM values ≥ 10 in either the CFP or the WCL fr action wer e r etained for further analysis, similar as described pr e viousl y (Srikumar et al. 2015 ).Differential enrichment analysis using raw read counts was performed in R using the DESeq2 method (Love et al. 2014 ).The standard DESeq2 workflow includes normalization of r aw r ead counts to account for sequencing depth and compositional bias as well as correction for gene dispersion as a means to measure variation.Genes with significant differences in their abundance between the CFP and WCL fractions (adjusted p-value < 0.05) and an absolute log2-ratio larger than the standard deviation of the determined log2-ratios of the dataset were classified as differ entiall y enric hed.

Results
The w orkflo w of this study is sc hematicall y depicted in Fig. 1 .A. kunkeei strains A1401 and A0901 were cultivated in fMRS medium and the bacterial cells were harvested during the exponential growth phase.We examined the morphologies of particles detected on the surfaces of the bacterial cells as well as the morphologies and composition of secreted particles and vesicles obtained from the pelleted supernatant, after r emov al of the bacterial cells.We also examined the proteomes and transcriptomes of the whole cell lysates (WCL) and the pellet of the cell free supernatant (CFP).

Morphological char acteriza tion of cell surface particles
Ima ges pr oduced b y SEM sho w ed that A. kunkeei strains A1401 and A0901 formed r egularl y r od-sha ped cells as well as corkscr e wed cells, both of which contained particles on the cell surfaces (Fig. 2AB).The particles of A. kunkeei strain A1401 were examined in more detail by TEM, which sho w ed that the majority of particles wer e contr ast-ric h with spec kled surfaces (Fig. 2 C).Like wise, tomogr a phy anal ysis based on cryo-ET of A. kunkeei str ain A1401 sho w ed the presence of particles on the bacterial cell surfaces, some of which were located in close vicinity to each other (Fig. 2 D).The mean (median) diameter of the particles was estimated to 39.9 nm (40.5 nm) based on analyses of the TEM images (n = 105) (Fig. 2 C, and to 36.2 nm (36.2 nm) based on analyses of the cryo-ET micr ogr a phs (n = 12) ( Table S1 ).

Morphological char acteriza tion of extr acellular particles and vesicles
Next, we studied the morphology of the extracellular material in the cell-fr ee pellet, whic h was obtained after centrifugation of the medium in which the bacterial cells had been cultivated, and as a negativ e contr ol we also pelleted the fMRS medium.The negative stain TEM micr ogr a phs of the dissolved pellets from the two A. kunkeei strains (Fig. 3AB) sho w ed the presence of two morphologically distinct particle types, one of which resembled membrane vesicles (MVs) while the other type consisted of irr egularl y sha ped particles, whic h we r efer to as extr acellular particles (ECPs).No suc h particles wer e visible in the media contr ol samples .T he micr ogr a phs also r e v ealed the pr esence of a pha ge in A. kunkeei strain A0901 (Fig. 3 B), consistent with a prophage in the genome of A0901 (Dyrhage et al. 2022 ).Cryo-ET tomography of the extracellular material obtained from strain A1401 sho w ed MVs containing lipid bilayers separated by ≈ 4 nm as well as contr ast-ric h ECPs with speckled surfaces (Fig. 3 C).The negative stain TEM (Fig. 3 A) and cryo-ET images (Fig. 3 C) indicated that some of the ECPs aggregated.The mean (median) diameter of particles that resembled ECPs in morphology in the nsTEM images were estimated to 39.1 nm (39.2) nm in strain A0901 (n = 211) (Fig. 3 A, Table S1 ) and to 48.9 nm (52.2 nm) in strain A1401 (n = 223) (Fig. 3 B, Table S1 ).Like wise, the cryo-ET ima ges also sho w ed that the ECPs w ere homogeneous in size (29.4-39.3nm), with a mean (median) diameter of 39.3 (37.3 nm) in strain A1401 (n = 12) ( Table S1 ).A few particles of variable sizes that resembled MVs were identified in the nsTEM micr ogr a phs, but they wer e too fe w to allow accur ate size estimates .T he cryo-ET images indicated that the MVs were of variable sizes (23-162 nm) with an estimated mean (median) diameter of 64.7 nm (46.7 nm) (n = 13),

Biochemical and biophysical char acteriza tion of secreted particles
The particle concentrations in the CFPs derived from A. kunkeei strains A0901 and A1401 were determined by NTA, and the RNA and protein contents were quantified using Qubit fluorometric assays ( Figure S1 , Table S2 ).The particle, protein and RNA concentr ations wer e consistentl y higher in the CFP fr om A. kunkeei str ain A0901 than from strain A1401 despite similar cell density during batc h-cultiv ation.The size distribution of the isolated particles, based on the hydr odynamic r adius measur ed by NTA, indicated a polydisperse sample .T he mode diameter (major peak in the distribution) of the secreted particles was estimated to 84 nm (n = 3) for A. kunkeei strain A0901 and 68 nm (n = 3) for A. kunkeei strain A1401 ( Figure S2 , Table S2 ).Total RNA concentrations in the crude CFP samples wer e measur ed in untr eated contr ol samples and in samples after treatment with proteinase K and SDS.No significant differences in the RNA concentrations were found between the untreated and the treated samples ( Figure S3 ).Ho w ever, the RNA concentr ation decr eased below the detection limit in both samples after treatment with RNAse .T hese results suggest that the RNA is located extr av esicularl y and/or is onl y loosel y associated with the particles ( Figure S3 ).

Separ a tion of membrane vesicles from protein particles
In order to investigate whether the membrane vesicles and extracellular particles in the CFP could be separated, we performed density gr adient ultr acentrifugation of the CFP deriv ed fr om A. kunkeei strain A1401 using an Optiprep gradient.We analyzed 10 fr actions fr om the gr adient for particle concentr ation, pr otein content, RNA concentr ation and r elativ e membr ane abundance.
The absolute concentrations and fluorescence intensities of the different assays are summarized in Figure S4 .The r elativ e particle abundance (determined by NTA) across the 10 fractions correlated strongly with the protein profile (determined by Bradford).Both types of analyses showed that fractions F5 to F7 contained the majority of particles and proteins in both strain A1401 (Fig. 4 A) and strain A0901 ( Figure S5A ).Consistently, analysis by SDS-PAGE showed the highest protein abundance in fraction F6 (Fig. 4 B; Figure S5B ).The sizes and intensities of the strongest bands in fraction F6 (density: 1.19 g/mL) were identical to those in the crude CFP-pr epar ation (Fig. 4 C).Bands of similar sizes were visible also in the flanking fractions F4-F5 and F7-F8, but at gr aduall y lo w er intensities, in accordance with the lo w er total abundance of proteins in these fractions.Interestingly, the  S5C ).Fraction F2 and F3 contained mostly MVs and particles larger than 20 nm wer e onl y r ar el y observ ed.Vice v ersa, fr action F6 contained a large number of ECPs and only very few MVs.Some of the observed particles in fraction F6 appeared to be connected by a filamentous structure of approximately 6 nm in diameter.We also observed single particles that displayed small protrusions of morphologicall y similar structur es .T he size distribution of the indicated a mean (median) diameter of 42.2 nm (39.0 nm) in fraction F6 from A. kunkeei strain A1401 (n = 206) and of 30.3 nm (29.8 nm) in frac-tion F6 fr om str ain A0901 (n = 100) ( Table S1 ).Taken together, this suggests that the membrane vesicles can be separated from the speckled particles, and that the particles mostly consist of proteins while the RNA fractionated with the membrane vesicles.

T he extr acellular proteome
To examine the composition of proteins in the CFP sample, we performed two sets of compar ativ e pr oteomics studies in A. kunkeei strain A1401 and strain A0901.
In the first study, we compared the protein composition in the crude WCL and CFP samples.Analyses by SDS-PAGE showed that the band profiles differed extensively between the WCL and CFP samples for A. kunkeei str ain A1401 ( Figur e S6A ) and strain A0901 ( Figure S6B ).A pelleted media control used as a negative control sho w ed no bands, which confirmed that the protein bands wer e deriv ed fr om the bacterial cells and not from the media ( Figure S6C ).In total, 591 proteins were identified by LC-MS/MS in three biological replicates of at least one sample in A. kunkeei strain A1401, of which 504 proteins were only identified in the WCL sample, 9 only in the CFP sample and 45 proteins were identified in three replicates in both samples ( Table S4 ).In A. kunkeei strain A0901, a total of 710 proteins were identified in three biological replicates in at least one sample, of which 566 proteins were only identified in the WCL sample, 23 were exclusively found in the CFP sample, while 61 proteins were identified in three replicates in both samples ( Table S4 ).Type-I signal peptides were predicted for 50% of the proteins in the CFP samples (Fig. 5 ) as compar ed to onl y 6-7% of the pr oteins in the WCL samples ( Table S4 ).In total, 44 protein families comprising 46 proteins with type-I signal peptides were detected in the CFP samples of the two strains, of which 27 families comprising 29 homologous proteins were identified in both strains (Fig. 5 ).
In the second study, we compared the protein compositions in the DGUC fractions F3 and F6.The concentration of proteins was estimated to 0.04-0.11μg/ μl in fraction F6, but was so low that it was unmeasurable in fraction F3 ( Table S3 ).Consistently, the F6 and F7 fractions contained the highest abundance of proteins according to the SDS/PAGE analyses, while no or only faint bands were visible in the F1, F2 and F3 fractions ( Figure S7 ).Ho w ever, despite the different intensities of the bands, the visible band pro-files were similar across lanes ( Figure S7 ).Analyses by LC-MS/MS sho w ed that 50 and 100 protein families contained proteins identified in 3 biological replicates in the F6 fractions of A. kunkeei strains A0901 and A1401, respectively.Of these, 37 proteins identified in one or both strains contained predicted type-I signal peptides (Fig. 5 ; Table S5 ).Ho w e v er, no peptides that matched the signal peptides of these proteins were detected in the LC/MS analyses .T his suggests that the proteins with type-I signal peptides are mature and secreted although it should be cautioned that not all peptides of a protein are identified in LC/MS analyses.
The proteins identified in the F3 fractions were also examined, although the identification of protein in this fraction may be less reliable due to the ov er all lo w er abundance of pr oteins.Ne v ertheless, we found that 80% of the CFP proteins with type-I signal peptides identified in the F6 fractions in three biological replicates from both strains were also identified in the F3 fractions ( Table S5 ).No protein was uniquely identified in three biological replicates in the F3 fraction in both strains .T his suggests that the F3 and F6 fractions contain a similar set of surface-associated and secreted proteins, albeit in different abundances.
Finall y, we compar ed the identity of proteins with type-I signal peptides across the different datasets .T he comparison sho w ed that all 37 pr oteins wer e detected in three biological replicates in the CFP sample as well as in the F3 and F6 fractions of at  A0901, and for homologs present in both strains (Core).Crude CFP refers to the number of protein families that contain proteins identified in three biological replicates in the crude CFP samples.Fraction F6 refer to the number of protein families that contain proteins identified in three biological replicates in fr actions F6.Fr action F3 r efers to the number of pr otein families that contain pr oteins identified in thr ee biological r eplicates in fr action F3.The table shows the number of proteins in which signal peptides (SPI and SPII) were predicted by SignalP5.0.No signal peptides were predicted for proteins classified as least one A. kunkeei str ain, of whic h 29 orthologs were detected in both strains ( Table S6 ).Functional predictions of this core set of extracellular proteins with type-I signal peptides indicated the presence of soluble proteins with glycosyl hydrolase, peptidase and nuclease enzymatic functions ( Table S6 ).The giant proteins, whic h ar e encoded by a str etc h of fiv e co-located genes, wer e also part of the core set of extracellular proteins.Glycosyl hydrolases of the GH20 protein family have predicted β-hexoseamidase activity, while enzymes with the GH25 domains have predicted l ysozyme activity.Notabl y, the cor e extr acellular pr oteome also contained gl ycosyltr ansfer ases of the GH70 pr otein famil y, putativ el y involv ed in the synthesis of α-D-glucans, such as dextran.The sizes of two strongest bands in the SDS-PAGE gels were in the range of 110-160 kDa ( Figure S7 ), which corresponded well with the estimated molecular weights of the 114 kDa glycosyltr ansfer ases (A0901_05 380 and A1401_04 720) and the 150-158 kDa gl ycosyltr ansfer ases (A0901_13 270 and A1401_12 750).In gel digestion of proteins embedded in this section of the SDS-PAGE gel confirmed that the high-intensity bands in this segment of the gel contained the GH70 family of gl ycosyltr ansfer ases ( Table S5 ).
In addition, lipoprotein signal peptides were predicted for 5 proteins in the F6 fraction, including oligopeptide and dipeptidebinding proteins.In Gram + bacteria, lipoproteins are anchored in the cytoplasmic membrane and span the cell wall, but may also be released into the en vironment.T he number of proteins without signal peptides were much more variable than those with signal peptides, r anging fr om 8 to 55 pr oteins in fr action F6 fr om str ains A0901 and A1401, r espectiv el y (Fig. 5 ).All of these proteins were also detected in the WCL samples and included some of the most highl y expr essed intr acellular pr oteins suc h as for example ribosomal pr oteins, RNA pol ymer ase subunits and gl ycer aldehyde-3phosphate dehydrogenase ( Tables S4, S5 ).Finally, the proteomics analyses confirmed the expression of 13 phage proteins in A. kunkeei strain A0901, in agreement with the prediction of a prophage in the genome (Dyrhage et al. 2022 ) and the identification of phage particles in the nsTEM micr ogr a phs (Fig. 3 B).

T he extr acellular tr anscriptome
Transcriptomic data was collected using rRNA-depleted extraction pr otocols fr om the solubilized pellets obtained from the whole cell lysate and the cell free supernatant of A. kunkeei strain A1401.The CFP sample was highl y concentr ated after the ultracentrifugation step, such that it contained approximately 99% of the original cell suspension, while the WCL sample was pr epar ed fr om onl y 1% of the gr owth media.Ne v ertheless, after the concentration and sample preparation steps there was no significant difference between the total number of reads obtained from the WCL and CFP samples either before or after rRNA depletion (p less than 0.05) ( Table S7 ).In both samples, the rRNA reads were reduced to about 4% of the total reads after rRNA depletion, as compared to about 94% of the total reads before depletion ( Table S7 ).
For comparison of the number of reads that mapped to proteincoding genes within each sample, TPM (transcripts per million) was used to estimate transcript abundance levels ( Figure S8 ).After mapping and quantification of reads, only transcripts with TPM values ≥ 10 in either the CFP or the WCL gr oup wer e in-cluded for further anal ysis, r esulting in a data set that r epr esented 1276 protein-coding genes ( Table S8 ).For this set, the number of ma pped r eads per gene differed by more than three orders of magnitudes within each sample, ranging from a few genes associated with highly abundant transcripts with TPM values > 1000 to the large majority of genes with moder atel y abundant tr anscripts (100-1000 TPM) and lowly abundant transcripts with TPM values < 100.
For comparisons of the r elativ e tr anscript abundance le v els in the CFP and WCL samples, DESeq2 analysis based on raw read counts was used (Fig. 6 , Table S9 ).The results sho w ed that 356 genes displayed differential enrichment patterns (adjusted pvalues < 0.05, absolute log2-ratios ≥ 0.7), which corresponds to 28% of the analyzed gene set.Of these, 180 genes attracted relativ el y mor e r eads fr om the CFP sample, while 176 genes were r elativ el y mor e abundant in the WCL sample (Fig. 6 A).The functional profiles of genes that sho w ed no or differential enrichment w ere similar (Fig. 6 B).Ho w ever, the distribution of transcript abundances for the enriched genes in the WCL and CFP samples differed.While the enriched genes in the WCL sample sho w ed a normal abundance distribution profile (Fig. 6 C), the enriched genes in the CFP sample a higher fraction of the most abundant transcripts (Fig. 6

D).
A more detailed comparison of the most abundant transcripts sho w ed that they mapped to clusters of genes for ribosomal proteins and the ATP synthase complex ( Figure S9 ).Ribosomal protein genes in the spc and alpha gene-clusters were highly enriched in the CFP sample and genes in the S10 gene-cluster were moder atel y enric hed.In contr ast, ribosomal pr otein genes in the upstream str operons were not enriched in either sample and genes for the ribosomal proteins RplM and RpsI, whic h wer e located else wher e wer e enric hed in the WCL sample.Beyond similar tr anscript abundance le v els for genes that wer e cluster ed and likel y part of the same operon, no other feature could be identified that accounted for the different enrichment profiles.
Finally, we used the transcriptomics data to gain an estimate of the r elativ e abundances of transcripts for the secreted proteins with signal I peptides in the CFP sample ( Table S6 ).Genes for the 33 kDa cysteine peptidase, the 37 kDa LPxTG cell wall binding protein and the 43 kDa matrixin import protein attracted a high abundance of reads, with TPM values in the range of 400-1000.The gene for the 158 kDa GH70 gl ycosyltr ansfer ase (A1401_12 750) recruited most reads with TPM values of 917 in the CFP sample and 1137 in the WCL sample, which supports the conclusion from the pr oteomics anal ysis that this is one of the most highly abundant proteins in the CFP sample.

Discussion
Extr acellular MVs hav e been described in both Gr am-negativ e and Gr am-positiv e bacteria, arc haea and eukaryotes, and numer ous r eports hav e suggested that they serv e v arious functional r oles in for example cell-cell comm unication pr ocesses, host ada ptation and antimicrobial defense by mediating the transfer of RNA, proteins and metabolites across cell boundaries (Brown et al. 2015, Liu et al. 2018, Gill et al. 2019, Yu et al. 2022 ).An alternative hypothesis, particularly for eukaryotic MVs, is that their basic function is to secrete material out from the cell for off-site degradation and recycling of membrane molecules (Vidal 2019 ).Here, we report the results from a broad survey of the RNA and protein cargo of the secreted material from A. kunkeei , a defensive symbiont of honeybees .T he starting point for our analyses was the observation that A. kunkeei strain A1401 contained surface-associated structures on the bacterial cells.For comparison, we also included strain A0901 which seemed to have fewer such structures.Howe v er, anal yses by TEM, proteomics and other techniques sho w ed no major differences of the extracellular material produced by the two strains.Based on distinct morphological and biochemical features of the extracellular particles, we infer that they represent two different types; membrane vesicles and a large proportion of non-membr anous pr otein particles of 30-40 nm in size.
Before discussing the contents of the secreted molecules, let us first briefly comment on the methods used for pr epar ation of the particles.Various isolation protocols have been used in the past to examine the content of MVs, and these can r oughl y be categorized into ultr acentrifugation-, filtr ation-, a ggr egation-and affinity-based a ppr oac hes.Appr oac hes based on ultr acentrifugation have most commonly been used, sometimes combined with differ ential gr adient ultr acentrifugation to separ ate the tar geted particles primaril y fr om loosel y bound pr oteins (Konoshenk o et al. 2018 ).Our and many other protocols include a low-speed centrifugation to deplete the cell suspension fr om lar ge debris and intact cells follo w ed b y filtr ation thr ough 0.22-0.45μm filters to completel y r emov e r esidual cells and debris.Gr owth contr ol is important at this stage to ensure that the filtrate is cell-free.Following filtr ation, the secr eted particles in the cell-free supernatant were pelleted by ultracentrifugation to yield what we refer to as cell-free pellet (CFP).Omics and imaging analysis were performed on the crude CFP, essentially as described before for other Lactobacillus species (Dean et al. 2019, Hu et al. 2021 ).
We used a combination of electron microscopy, including SEM, TEM, nsTEM and cryo-ET, to obtain a compr ehensiv e and detailed view on the morphology of isolated protein particles and MVs from A. kunkeei .F or bioph ysical analysis, we used Nanoparticle Tr ac king Anal ysis (NTA), commonl y emplo y ed for c har acterizing exosomes and membrane vesicles.We observed a discrepancy between the size distributions obtained from TEM, nsTEM and cryo-ET data in comparison to the NTA data, where the size appeared to be overestimated by NTA.This has been observed befor e (Bac hurski et al. 2019 ) and might be the result of the hydr odynamic r adius measur ed by NTA being lar ger than the actual particle size.It has also been noted that the detection limit for NTA is 60-70 nm (Bachurski et al. 2019 ), which is larger than the speckled particles and small MVs that we clearly detected by electr on micr oscopy.This can either indicate that, in the pol ydisperse and heterogenous CFP, NTA detects aggregated particles and MVs, or that the hydrodynamic radius of the isolated particles is larger than the actual particle radius.Despite those uncertainties, the possibility to obtain an estimate of the particle concentration makes NTA a valuable method for quality control, and in the present study particle concentration correlated very well the pr otein concentr ation.In isolated particle pr epar ations, they wer e observed as single particles as well as associated to form smaller chains .T his was also observed in tomograms of whole bacterial cells indicating that chain formation is a biologicall y r ele v ant pr ocess and not an experimental artefact due to a ggr egation upon ultracentrifugation.
For more detailed analyses of the CFP proteomes and transcriptomes, w e also emplo y ed density gr adient ultr acentrifugation using an Optipr ep gr adient, similarl y as described for the isolation of bacterial and plasma extr acellular v esicles (Karimi et al. 2018, Onódi et al. 2018 ).In those pr e vious and our studies, the crude CFP-pr epar ation was loaded on top of the density gr adient.In other studies where bacterial membrane vesicles were isolated in an Optiprep gradient, the crude pellet was re-suspended in the highest density solution, loaded in the bottom of the cen-  Hu et al. 2021 ).Such an approach is based on the flotation of membrane vesicles against the centrifugal force until the vesicles reach the corresponding density, while loose proteins, due to their high density, will stay pelleted at the bottom of the tube (Graham 2002 ).Choosing the top-loading a ppr oac h seemed mor e r ational for our pur pose as we observ ed a nov el and a ppar entl y non-membr anous type of secr eted particles in the crude CFP, together with MVs, without a priori information on the corresponding particle densities.
The results of the DGUC-fractionation provided further evidence that the secreted material consists of distinct particle types with differ ent physicoc hemical pr operties.Membr ane-ric h particles were enriched at a density of approximately 1.13 g/mL, peaking in DGUC fraction F3.Negative stain TEM analysis of this fraction demonstrated the presence of typical membrane vesicles of variable sizes in addition to large mesh-like structures .T he membrane-less and speckled particles, representing the major-ity of detected particles in the crude CFP-pr epar ation wer e separated at a density of 1.19 g/mL, peaking in DGUC fractions F6 and F7.The RN A sho w ed a similar fractionation pattern as the membranes with a peak in DGUC fraction F3, while the proteins sho w ed a similar fractionation pattern as the speckled particles with a peak in DGUC fractions F6 and F7.A comparison of the protein composition in the crude CFP samples and DGUC fractionated samples, as inferred from the intensities and sizes of the bands on an SDS-PAGE gel as well as fr om anal yses by MS/MS-LC, suggest that all fractions contain a similar set of proteins, albeit in different abundances.Based on this observation, we conclude that the large majority of proteins identified in the CFP sample fractionated with the particles, rather than with the MVs.Howe v er, we consider ed the possibility that a less abundant subset of the CFP pr oteins, whic h would go undetected on the SDS-PAGE gel, were associated with the MVs in DGUC fraction F3.To test this hypothesis, we analyzed and compared the protein compositions of fractions F3 and F6, but were unable to identify proteins uniquely associated with fraction F3 in both strains .T his suggests that the MVs produced by the A. kunkeei cells do not pac ka ge a specific subset of proteins.
A comparison of the CFP proteomes sho w ed that 50 proteins wer e pr esent in thr ee biological r eplicates in both str ains, about half of which contained type-I signal peptides and are thus likely to correspond to secreted proteins in A. kunkeei.The CFP proteomes contained most of the 24 proteins identified in a previous study of the cell-free supernatant of A. kunkeei strain Fhon2 under LPS-stress (Butler et al. 2013 ).The cor e CFP pr oteins included the giant pr oteins, whic h ar e 3 000 to 9 000 amino acids in length and encoded by a 90 kb long cluster of genes in the A. kunkeei genome (Tamarit et al. 2015 ).All five giant proteins were identified in strain A0901, and the proteins encoded by the first two genes in the cluster were also detected in str ain A1401.Thr ee of the proteins contain tr ansmembr ane domains, two of whic h wer e also identified in the WCL sample, suggesting that the giant proteins are surface-associated.Their function is unknown but like other v ery lar ge extr acellular surface pr oteins , they ma y be involved in surface adherence.
The identified core CFP proteins additionally included large membr ane-associated gl ycoside hydr olases of 114 to 158 kDa, whic h ar e likel y to serv e important functions in the synthesis or modification of pol ysacc harides suc h as dextr an.In gel-digestion confirmed that the high molecular weight gl ycosyltr ansfer ases r epr esent some of the most highly abundant proteins in the CFP samples.In addition to the type-I signal peptide, these enzymes also contained the KxYKxGKxW signal peptide, which is a characteristic signal sequence of serine-ric h, lar ge gl ycosylated surface adhesins in Gr am-positiv e bacteria (Ga gic et al. 2013 ).One of the gl ycosyltr ansfer ases (A1401_12750) sho w ed sequence similarity to the A. kunkeei enzyme GtsZ, which contains two catalytic cor es, CD1 and CD2, pr edicted to hav e glucansucr ase and br anc hing sucr ose specificity, r espectiv el y (Meng et al. 2018 ).Biochemical studies of GtsZ-CD2 have confirmed the pr edicted br anc hing sucrose specificity and sho w ed that this protein catalyzes the production of structur all y complex alpha-glucans (Meng et al. 2018 ).Ho w e v er, GtsZ-CD1 could not be expressed and was ther efor e not c har acterized further.Inter estingl y, the two catal ytic cor es ar e split into two genes in strain A1401, of which the protein encoded by A1401_12750 corresponds to GtsZ-CD1, whereas A1401_12760, whic h corr esponds to GtsZ-CD2 could not be detected in neither the CFP nor the WCL pr oteome.Consistentl y, the TPM v alues for A1401_12760 corresponded to only 1%-2% of the TPM values for A1401_12750, whic h attr acted a high abundance of RNA reads in both the CFP and the WCL samples .T he demonstrated high abundance of GtsZ-CD1 in this strain now enables a more detailed biochemical study of the glucan sucrase activity.
Proteomic studies on membrane vesicles from other bacterial species have shown release of cytoplasmic proteins without predicted signal peptides or other export signals.It should be noted though that se v er al of those studies did not compare the extracellular proteome to the whole-cell pr oteome, whic h is important in order to conclude whether a protein is uniquely present or enriched in the extracellular milieu or simply present as a result of the detection of trace amounts of a highly expressed protein, such as EF-Tu.In our dataset, EF-Tu was r eliabl y detected in both the WCL and CFP samples, as were also ribosomal pr oteins, RNA pol ymerase subunits and ATP synthase subunits.Ho w ever, in contrast to the consistent detection of same set proteins with signal peptides in multiple extractions from two different strains, the identification of cytoplasmic proteins was mor e v ariable and the r esults wer e not r epr oducible between either str ains or samples.
Mor eov er, it should be recalled that the CFP samples are highly concentr ated compar ed to the WCL samples, and it can ther efor e not be excluded that intracellular proteins derived from a small fraction of lysed cells may be detected in the CFP samples.
Importantly, our study indicated that the secreted proteins were not embedded within the MVs, which instead sho w ed the same fractionation pattern as the RNA molecules.We used a rRNA depletion pr otocol, whic h depleted mor e than 95% of the rRNA pool of both CFP and WCL samples but also pr e v ented an y compar ativ e anal ysis based on rRNA abundances.We detected tRNA sequences in our data, but compar ativ e anal yses of tRNA sequencing data are hampered by the modifications and rigid structures of these molecules.Accurate estimates of tRNA levels in the cells would have required targeted library preparation protocols, which was not employed in this study (Schwartz et al. 2018, Pinkard et al. 2020 ).T hus , although we r ecov er ed differ ent types of RNA from the total RNA pool of both the CFP and WCL fractions, including rRNA and tRNA, the subsequent anal yses wer e focused on the mRNA transcripts.
Ov er all, we estimated that the total amount of secreted RNA molecules r epr esented no mor e than about 1% of the amount of cellular RNA, and is thus only a tiny fraction of the intracellular RNA.Unlike the secreted proteins which were mostly uniquely present in the extracellular material, all transcripts detected in the CFP sample were also identified in the WCL sample.Sequence r ead cov er a ge for eac h data set v aried by se v er al orders of ma gnitude for the individual protein-coding genes.Ho w ever, w e noted a r elativ el y higher abundance of RNA fr a gments for the most highly expressed genes in the CFP samples compared to the WCL samples, such as those coding for ribosomal proteins, RNA polymerase subunits and ATP synthase subunits .Likewise , a biased composition of the RNA content of the MVs wer e observ ed in S. aureus (Luz et al. 2021 ).As in our study, the RNA tr anscripts cov er ed colocated genes, pr esumabl y belonging to the same oper ons, whic h code for highly expressed genes involved in translation, energy production and carbohydrate metabolism, but the previous study sho w ed that the compositional bias depended on the environmental conditions (Luz et al. 2021 ).
T his lea v es open the question of the biological r ole that the extracellular RNA sequences play, if any.One hypothesis is that the transfer of functional RNA to recipient cells that are lacking a corresponding gene might confer a transient horizontal phenotype (Luz et al. 2021 ), which seems unlikely in A. kunkeei since the most abundant transcripts in the CFP samples were obtained fr om the univ ersal cor e genes.Another hypothesis is that the secreted RNA molecules mediate bacterial interactions with the host, such as for example serving host immunomodulatory functions (Tsatsaronis et al. 2018, Rodriguez and Kuehn 2020, Kurata et al. 2022 ).While small, non-coding RNAs have been directly associated with immunomodulatory responses (Koeppen et al. 2016, Yu et al. 2022, Sahr et al. 2022 ), the specificity of such effects by bulk-RNA ar e not full y r esolv ed.Furthermor e, pr e vious studies hav e shown that RNAs are secreted as fragments, arguing against a biological role for the large majority of secreted RNA molecules deriv ed fr om long mRNA molecules (Rodriguez and Kuehn 2020, Luz et al. 2021, Kurata et al. 2022 ).Selective explanations also require some mec hanism wher eb y the RN As that modulate the proposed functions in the recipient cells are selected for secretion and packaging, ho w ever, no such targeting sequencing or selective packaging mechanism have yet been identified.
We acknowledge the possibility that the mRNAs in the CFP samples may be r andoml y associated with or embedded inside the MVs, for example upon cell lysis, like the intracellular proteins.If so, the observ ed differ ences in the composition of the RNA cargo in the CFPs compared to WCL samples might be due to a r elativ el y higher r ecov ery of mRNAs with long half-lives in the CFP sample and/or to the use of differ ent pr otocols for pr epar ation of RNA from the two samples.Howe v er, e v en if the RNA fr a gments associated with the MVs would result from lysis of a small fraction of bacterial cells, this does not exclude the possibility that some of the short RNA fr a gments ma y ha v e e volv ed to serv e specific r oles in specific recipient cells.It remains to be shown in future studies whether the mRNA molecules identified in the CFP samples of A. kunkeei ar e involv ed in inter action and comm unication pr ocesses among the bacterial cells or with the honeybees.In particular, the secretion and putative role of small RNAs in A. kunkeei need further investigation.
In summary, we have demonstrated the presence of distinct extracellular particle types in two Apilactobacillus kunkeei strains .T he analyses suggest that the CFP fraction consists of secreted protein complexes, cell surface proteins and phage proteins as well as MVs that are associated with cellular RNA.To the best of our knowledge, this is the first time that the RNA and protein content of extracellular samples from a bacterial source have been separated into two biochemically and morphologically distinct fractions .So, although both MVs , RNA and protein particles may be part of the same extracellular material, they may not necessarily be associated with each other.Future studies should be directed to w ar ds elucidating the structure, function and evolution of the surface-associated and secreted proteins, particularly the giant proteins and the glycoside hydrolases .P ossible functional roles could involve a defensive mechanism by the secreted peptidases and GH25 hydrolases and, at the same time, the production of an exopol ysacc haride layer by the action of the GH70 hydrolases.

Figure 2 .
Figure 2. Cell surface particles of A. kunkeei .Particles on the surfaces of bacterial cells are shown by (A, B) SEM, (C) TEM and (D) cryo-ET micr ogr a phs of A. kunkeei strains (A, C, D) A1401 and (B) A0901.(C) The size distribution is based on particle measurements performed in Fiji/ImageJ based on analyses of the TEM images.(D) The raw (left), a segmented tomogr a ph (middle) and the final segmentation of the cryo-ET tomogram are are shown (scale bar: 100 nm).

Figure 3 .
Figure 3. P articles secr eted by A. kunkeei.P articles secr eted by A. kunkeei str ains (A) A0901 and (BC) A1401 ar e sho wn b y (AB) nsTEM and (C) cry o-ET micr ogr a phs (scale bar: 100 nm).(AB) The size distributions were estimated from measurements of the sizes of all particles in the nsTEM images.Black arrows and arrowheads point to MVs and particles, respectively.A phage particle is also indicated.

Figure 4 .
Figure 4. Separation of CFP components by DGUC.Crude CFP (n = 3, biological replicates) from A. kunkeei strain A1401 were separated by Optiprep-based DGUC into 10 fractions.(A) Analysis of DGUC fractions for particles (NTA), protein (Bradfor d), RN A (Qubit RNA HS assa y, SYT O RNASelect), membranes (FM4-64) and density.Relative concentrations and intensities are shown based on average and standard deviation from three biological r eplicates.Corr esponding absolute v alues ar e shown in Figur e S3 .(B) SDS-PAGE anal ysis of DGUC fr actions (F1-F10) of r epr esentativ e replicate sample .T he crude CFP sample (10x diluted) was loaded for comparison with the DGUC fractions .No vex Sharp Unstained marker was used as the molecular weight standard and the corresponding molecular weights (in kDa) are shown next to the gel.(C) Comparison of r elativ e band intensities of 6 selected bands from crude CFP fraction and DGUC fraction F6.The band numbers on the x-axis correspond to the band numbers in (B) .(D) nsTEM analysis of DGUC fractions F2, F3 and F6.Black arrows point towards MVs, orange arrowheads point towards protrusions from and connections between ECPs .T he size distribution is based on particle measurements performed in Fiji/ImageJ.

Figure 5 .
Figure5.Number of protein families in the CFP proteomes.Data is shown for the CFP proteomes of A. kunkeei strains A1401 and A0901, and for homologs present in both strains (Core).Crude CFP refers to the number of protein families that contain proteins identified in three biological replicates in the crude CFP samples.Fraction F6 refer to the number of protein families that contain proteins identified in three biological replicates in fr actions F6.Fr action F3 r efers to the number of pr otein families that contain pr oteins identified in thr ee biological r eplicates in fr action F3.The table shows the number of proteins in which signal peptides (SPI and SPII) were predicted by SignalP5.0.No signal peptides were predicted for proteins classified as

Figur e 6 .
Figur e 6. T he whole-cell WCL and extracellular CFP transcriptome of A. kunkeei A1401.(A) Negative log10 p-values are shown as a function of the log2-ratio for the differentially enriched genes in the CFP and WCL fractions.Genes that are not classified to be enriched in either the WCL or CFP fr action ar e color ed in light gr ay.Dashed v ertical and horizontal lines indicate the thr esholds for determining enric hment in either the CFP (positiv e log2-ratios) or WCL fraction (negative log2-ratios).(B) The frequencies of differentially enriched genes in the CFP and WCL transcriptomes are shown for the corresponding COG categories.Histograms and density plots illustrating the gene-frequency distribution as a function of the log10-transformed TPM-normalized read counts of the genes in the (C) CFP and (D) WCL fractions.Colors indicate enrichment in either the CFP (orange) or WCL fraction (blue) or whether genes are not enriched ('none', gray).