Specimen-specific drift of densities defines distinct subclasses of extracellular vesicles from human whole saliva

Extracellular vesicles (EVs) in body fluids constitute heterogenous populations, which mirror their diverse parental cells as well as distinct EV-generation pathways. Various methodologies have been proposed to differentiate EVs in order to deepen the current understanding of EV biology. Equilibrium density-gradient centrifugation has often been used to separate EVs based on their buoyant densities; however, the standard conditions used for the method do not necessarily allow all EVs to move to their equilibrium density positions, which complicates the categorization of EVs. Here, by prolonging ultracentrifugation time to 96 h and fractionating EVs both by floating up or spinning down directions, we allowed 111 EV-associated protein markers from the whole saliva of three healthy volunteers to attain equilibrium. Interestingly, the determined buoyant densities of the markers drifted in a specimen-specific manner, and drift patterns differentiated EVs into at least two subclasses. One class carried classical exosomal markers, such as CD63 and CD81, and the other was characterized by the molecules involved in membrane remodeling or vesicle trafficking. Distinct patterns of density drift may represent the differences in generation pathways of EVs.


Introduction
Body fluids contain large numbers of extracellular vesicles (EVs), which collectively refer to the membranous particles secreted from cells [1]. These EVs are released from various types of parental cells via distinct generation pathways, making EVs heterogeneous populations [1][2][3][4]. Numerous studies have revealed the pivotal roles of EVs in various biological activities, including normal development such as spermatogenesis [5] as well as the progression of various disorders, including cancer, neurodegenerating diseases, etc. [6][7][8]; however, a complete picture of EVs has not yet been well elucidated, and an in-depth understanding of these EV-involved biological activities requires systemized classification of the heterogeneous EV populations [1,  ]. When certain molecules (proteins, nucleic acids, lipids, or glycans) are specifically linked to a parental cell or a certain generation route, they can be used as molecular tags to concentrate or isolate the EV subclass [10][11][12]. In most cases, however, the expressions of EV markers overlap between cell types or generation routes and, therefore, the strategy does not always work for defining specific EV subclasses [1,2]. Besides the marker-based classifications, the physical-chemical properties of EVs have also been used to define subclasses of EVs. In particular, sedimentation profile of EVs in centrifugation have been frequently used to fractionate EVs [13][14][15]. For instance, a sequential centrifugation at 300, 2,000, 10,000, and 160,000 g centrifugation deposited cells, large EVs, medium EVs, and small EVs as pellets, respectively. In this differential centrifugation experiment, the pelleted fraction from 160,000 g (P160) is often called "exosomes", although the fraction contains nonvesicular entities as well as nonendosome-originated EVs [4,[15][16][17][18]. In some instances, the P160 fraction is further differentiated based on their densities running through the density-gradient medium [2,15,17,19,20]. This method is called "equilibrium density-gradient centrifugation," in which molecules are expected to stop moving at a point where the density of the particles equals the density of the solution around them [14]. However, the standard conditions used for the method do not necessarily allow all EVs to move to a state of equilibrium; that is, some EVs still remain in the process of movement through density-gradient media, complicating the categorization of EVs based on densities [2,[20][21][22]. For this reason, to determine the innate densities of subclasses of EVs, we first prolonged the ultracentrifugation time from 17 h to 96 h and allowed the EVs to move in two directions: floating up and spinning down. With this protocol, 111 proteins out of 1,420 identified from mass-spectroscopy achieved equilibrium from the whole saliva sample of three healthy volunteers, because they resided at the identical density fraction irrespective of movement directions. Interestingly, the determined buoyant densities of the markers drifted in a specimen-specific manner, and the pattern of these drifts differentiate 111 markers into two subclasses. One class contained classical exosomal markers, and the other carried molecules involved in membrane remodeling or vesicle trafficking. Distinct patterns of density drift may represent differences in the generation pathways of the EVs. Our findings would boost the development of a salivary-based liquid biopsy system for oral-associated lesions as well as for systemic diseases [23-25].

Extensive equilibrium density-gradient centrifugation
After sonication, specimens were centrifuged at 2,600 g for 30 min at 4˚C (model 5500, Kubota, Osaka, Japan). The supernatants were collected in 38.5 mL ultraclear tubes (#344058, Beckman Coulter, CA, USA) and were centrifuged at 160,000 g for 70 min at 4˚C (L-90K with SW32Ti rotor, Beckman Coulter) to prepare crude EVs. In this study, centrifugations with medium speed were omitted to analyze not only small EVs but also medium and large EVs. For downward separation, the EV pellets were resuspended in 500 μL of PBS and layered on the 8% to 47% continuous density of iodixanol in 0.02 M HEPES [4-(2-hydroxyethyl)-1-piperazine ethanesulfonic acid]/NaOH, pH 7.2, which was formed in 32 mL thickwall polycarbonate tube (#355631, Beckman Coulter) using a gradient mixer No.3 (SAN4024, Sanplatec, Osaka, Japan). For upward separation, the pellets were resuspended in 500 μL of 47% iodixanol in 0.02 M HEPES/NaOH, pH 7.2 and put at the bottom of the centrifugation tube. Subsequently, continuous density gradient was formed as described above. After centrifugation (L-90K with SW32Ti rotor) at 160,000 g for 96 h or 17 h at 4˚C, 10 individual 3-mL fractions were collected in ultraclear tubes (#344058, Beckman Coulter) from the top, and the density of each fraction was measured using a refractometer (RX-5000a, Atago Co. Ltd, Tokyo, Japan). Each fraction was centrifuged at 160,000 g for 120 min at 4˚C (L-90K with SW32Ti rotor) after adding 30 mL of PBS, and pellets were resuspended in 500 μL of PBS. These samples were stored at 4˚C until analysis.

Mass spectrometric analysis
Protein samples were purified and concentrated using the 2-D Clean-Up Kit (80648451, GE Healthcare, Chicago, IL) according to the manufacturer's instructions. Samples were then reduced in 1 × Laemmli's sample buffer (32.9mM Tris HCl, pH6.8, 1.05% SDS, 13.15% glycerol, 0.005% bromophenol blue) with 10 mM tris(2-carboxyethyl)phosphine at 100˚C for 10 min, alkylated with 50 mM iodoacetamide at ambient temperature for 45 min, and subjected to SDS-PAGE. The electrophoresis was stopped at the migration distance of 2 mm from the top edge of the separation gel. After coomassie brilliant blue-staining, protein bands were excised, destained, and cut finely prior to in-gel digestion with Trypsin/Lys-C Mix (V5071, Promega, Madison, WI, USA) at 37˚C for 12 h. The resulting peptides were extracted from gel fragments and analyzed with Orbitrap Fusion Lumos mass spectrometer (Thermo Fisher Scientific, Waltham, MA) combined with UltiMate 3000 RSLC nano-flow HPLC (Thermo Fisher Scientific). The MS/MS spectra were searched against the "Homo sapiens" protein sequence database in SwissProt using Proteome Discoverer 2.2 software (Thermo Fisher Scientific), in which peptide identification filters were set at "false discovery rate < 1%".

Analyses of the quantified values of mass spectrometric data
Mass spectrometric data assigned with human proteins were analyzed using a custom Perl script. Each entry consisted of a specimen number, the direction of the gradient movement, fraction number, an MS quantified value, and an assigned protein ID. MS quantified values were first processed with the base 10 logarithm. Since the three saliva specimens were divided into two groups for the two opposite movements, six units of samples were available for analysis. For each protein, we searched for the fraction where the protein was the most abundantly detected within a unit. The logarithm values in a unit were normalized by the maximum. The Perl script used in these analyses is available from GitHub repository (https://github.com/ yamamoto-tdc/EV-saliva).

EV-TRACK
We have submitted all relevant data pertaining to our experiments to the EV-TRACK knowledgebase (EV-TRACK ID: EV190057).

Extensive equilibrium density-gradient centrifugation allows more molecules to reach a state of equilibrium
In this study, to allow more EVs to attain their state of equilibrium in density gradient centrifugation, the running time was prolonged to 96 h from 17 h. Furthermore, the specimens that were collected from three healthy volunteers were fractionated using two methods: spinning down (downward) or floating up (upward) directions through a density gradient of iodixanol (Fig 1). After centrifugation, ten fractions (F1-F10) were collected and analyzed using SDS-PAGE and silver staining (S2 Fig). After 17 h of centrifugation, distributions of silver staining signals through fractions differed between samples centrifuged for fractionation in the downward direction and those centrifuged for fractionation in an upward direction, indicating that many of molecules were still en route to their corresponding density positions. In contrast, signal patterns obtained from the 96 h centrifugation were almost identical between samples centrifuged in the downward direction and those centrifuged in an upward direction, demonstrating that most molecules had already reached their state of equilibrium. To verify the states of equilibrium for specific molecules, WB analyses were conducted using anti-CD63, CD81, CD9, α-amylase, HSP70, AQP5, and TSG101 antibodies (S3 Fig). From 96 h centrifugation, all proteins except α-amylase provided the strongest signals in the identical fractions both downward and upward directions, whereas, from 17 h centrifugation, the strongest signals of CD81, HSP70, APQ5 and α-amylase were not in the identical positions between downward and upward directions, indicating that they were not in equilibrium. Although many proteins attained equilibrium in the 96 h centrifugation time, some proteins including α-amylase were still in the course of movement in the medium (S3 Fig), suggesting that these molecules themselves or their associated carriers have peculiar structures or surface properties that hinder their movement in the media.

Idiosyncratic densities of CD63 on different specimen
In the preceding analyses, it should be noted that the fractions that contained CD63 had different densities among the three specimens with 96 h centrifugation. The strongest signals in WB were observed at F7 from Specimen 1 in both the downward and upward direction, and the densities of these fractions were 1.12 g/ml (Fig 2). By contrast, from Specimens 2 and 3, the fractions containing CD63 had densities of 1.10 g/ml (F6 in both the downward and upward directions) and 1.09 g/ml (F6 in downward and F5 in upward), respectively (Fig 2). Because each fraction from different tubes had slightly different density after centrifugation, a total of 60 fractions (three specimens, 10 fractions for two sample groups, one centrifuged in the downward direction and the other centrifuged in the upward directions) were mapped on the abscissa of determined densities, which clarifies specimen specific differences in densities of CD63 ( Fig 2B, in which blue, green and red represent Specimen 1, 2 and 3, respectively). The density of CD63 from Specimen 2 was 1.10 g/ml. From Specimen 1, the strongest signal for CD63 was observed from the fraction with 1.12 g/ml but not from the one with 1.10 g/ml. Similarly, from Specimen 3, CD63 was localized at 1.09 g/ml but not at 1.12 g/ml. Thus, the data suggested that dominant EVs containing CD63 possessed specimen-specific idiosyncratic densities.

Comprehensive analyses of fractionated proteins by mass spectrometry
Western blot analyses revealed that the density of the vesicles containing CD63 differed among specimens (Fig 2). To determine if this phenomenon was specific to CD63, comprehensive analyses of the proteins of each of the 60 fractions using mass spectrometry (MS) were performed. This process identified a total of 11,749 proteins that contained 1,429 nonoverlapped proteins. The total number of proteins identified in Specimens 1, 2, and 3 were 1,097, 920 and 672, respectively. Out of them, 476 were common among all three specimens ( Fig  3A). The complete list of 1,429 proteins is provided in S1 Dataset. Among the 476 proteins that were detected from all three specimens in both sample groups (centrifuged or fractionated in the downward and upward directions), we identified those that had identical densities within a given specimen; that is, the proteins that reached a state of equilibrium. For this purpose, the total area obtained by MS was quantified with Proteome Discoverer 2.2 and used as relative quantities, which have often been employed in label-free proteomic analyses [26]. When the density of the fraction with the highest quantitative value Scheme of EV fractionation using density gradient centrifugation. Large particles in whole saliva such as desquamated epithelia and blood cells were first removed by centrifugation at 2,600 g for 30 min. Then the resultant supernatants were centrifuged at 160,000 g for 70 min to obtain the pellets of particles containing various sizes of EVs. These particles were fractionated by both spinning down (downward, D) and floating up (upward, U) through the density gradient of iodixanol at 160,000 g for 96 h. After centrifugation, ten fractions (F1 to F10) were recovered from the top and analyzed after being concentrated through centrifugation at 160,000 g for 120 min. https://doi.org/10.1371/journal.pone.0249526.g001

PLOS ONE
Drift of densities defines subclasses of EV in saliva was the same in both the sample group centrifuged in the downward direction and the sample group centrifuged in the upward direction for a given specimen, the protein was regarded as having attained equilibrium, and this density was defined as "peak fraction density (PFD)." Out of 476 proteins, the PFD values of 111 were determined; that is, 111 proteins achieved equilibrium in all of three specimens (Fig 3A). These 111 proteins were categorized as Group I.

Specimen-specific PFD values of 111 proteins outline two EV subpopulations
In agreement with the results of WB analyses (Fig 2), PFD values of CD63 in MS analyses were determined to be 1.12 g/ml, 1.10 g/ml, and 1.09 g/ml for Specimen 1, Specimen 2, and Specimen 3, respectively. Furthermore, out of 111 proteins, we identified 16 proteins (including CD63) that possessed an identical pattern of PFD values observed in CD63; these proteins were classified into IA-1 ( Fig 3B). The remaining 95 proteins were sorted into two major groups (IA-2 and IB) and 5 minor groups (IA-3, IA-4, IA-5, IA-6 and IA-7). Subgroup IA-2 consisted of 59 proteins, and it also demonstrated idiosyncratic PFD values (1.12 g/ml, 1.13 g/ ml and 1.09 g/ml for Specimens 1, 2 and 3, respectively), the pattern of which was distinct from subgroup IA-1. Unlike the proteins categorized into Group IA, 29 proteins had identical PFD values among the three specimens and were classified into IB (Fig 3B). Proteins in IB included many globulin group proteins, such as Igα1 and Igα2 (S1 Dataset). In summary, Groups IA-1 and IA-2 showed a specimen-specific drift of densities, and Group IB showed an identical high density among the specimens.

Group IA-1 and Group IA-2 subclasses outline two EVs having distinct functionalities
Sixteen proteins classified into Group IA-1 included CD63, CD81, CD9, and others, most of which were membrane proteins or membrane anchored (via lipidic modifications) proteins, except for CIB1, LITAF, and UBC (Table 1). Among them, CD63, CD81, and CD9 have often been referred to as classical or canonical exosome markers.
Group IA-2 was composed of 59 proteins ( Table 2), many of which are involved in membrane dynamics and membrane trafficking. Group IA-2 was also characterized by the presence of the proteins involved in the innate immune system.

Discussion
In this study, soluble fractions of human whole saliva after 2,600 g centrifugation were spun at 160,000 g to obtain sediment of particles. Then, these particles were fractionated using density gradient centrifugation, which has often been used to separate vesicular particles (which have low density) from nonvesicular entities, such as supramolecular complex that display high density [4]. The methodology has also been used to subtype EVs based on their slight differences in density [2,15,17,19,20]. However, the kinetics of particles in media are influenced by the size, shape, and frictional characteristics of particles [52], and therefore it is very difficult to configure the conditions that allow all particles to achieve equilibrium, especially for EV populations that display a wide range of diversity. Standard conditions for density gradient centrifugation (e.g., 160,000 g for 17 h) have been shown to be insufficient for achieving equilibrium [2,[20][21][22]. In order to determine the innate density of subclasses of EV, a prolonged time (96 h) for density gradient centrifugation was employed in this study. In addition, particles were centrifuged for fractionation both in downward and upward directions, and if a given cargo protein was recovered from the same density fraction irrespective of movement directions, the particle carrying the protein was regarded as being in equilibrium. We triplicated these analyses using whole saliva from three healthy volunteers, and we discovered that 111 proteins (Group I) achieved equilibrium in all three specimens (Fig 3A). By using the total area obtained by MS as the index of quantitative evaluation, the fraction that contained the Venn diagram shows the number of proteins that were detected common to Specimens 1 and 2; the number of proteins that were detected common to Specimens 1 and 3; the number of proteins that were detected common to Specimens 2 and 3; the number of proteins that were detected only in Specimen 1; the number of proteins that were detected only in Specimen 2; and the number of proteins that were detected only in Specimen 3. The blue-colored area represents the number of proteins that were detected only in Specimen 1; the green-colored area, the number of proteins that were detected only in Specimen 2; the red-colored area, the number of proteins that were detected only in Specimen 3, and the pink-colored area, the numbers of proteins that were detected common to Specimens 1 and 2, Specimens 1 and 3, and Specimens 2 and 3, respectively. The size of each circle corresponds to the number of proteins in each category. The list of proteins in each group is provided in S1 Dataset. (B) Graphs showing the densities of the fractions of each specimen in Groups IA-1 to IA-7, IB and some proteins in Group II. The horizontal axis indicates densities that were determined using the refractometer. The circles and squares indicate the fraction numbers having PFD (see text) from the downward and upward direction, respectively. The number of proteins contained in each group is indicated under the name of the group.
https://doi.org/10.1371/journal.pone.0249526.g003 largest quantity of a given protein was defined as "peak fraction density (PFD)" and was mapped for 333 (111 x 3) proteins (S4 Fig). Patterns of PFD were primarily classified into two types, Group IA and Group IB. Group IB contained immune-globin, HSPB1, S100A14, and other similar proteins (S1 Dataset), and displayed identical density among the three specimens, which was relatively high (1.16 g/ml) compared to the reported density of EVs. With the current available data, we do not know whether the proteins in Group IB are associated with EVs or not. Atomic force microscopy (AFM) observation detected numerous tiny particles in the denser fractions of EVs, which are not protected from protease treatment as in the case of EVincluded proteins, suggesting many proteins in the high-density fraction are not included in the cargo of EVs, but are instead the extraneous entities of vesicles [15]. Recent reports by other groups have also concluded that most molecules recovered from high density fractions in density gradient separation were nonvesicular components [4,18]. Whereas Group IA proteins had identical PFD values, the ones from Group IA exhibited specimen-dependent drifts; that is, each specimen had idiosyncratic PFD values, the patterns of which were classified mainly into two types. One type (Group IA-1) exhibited the pattern of 1.12 g/ml, 1.10 g/ml and 1.09 g/ml, and was composed of 16 proteins, which included classical exosome markers, CD63, CD9 and CD81. This group also contained proteins related to vesicle formation as well as components of the immune system (Table 1). Group IA-2, whose PFD values were 1.12 g/ml, 1.13 g/ml and 1.09 g/ml for Specimens 1, 2 and 3, respectively, was composed of 59 proteins, and many of them are involved in membrane dynamics and the innate immune system (Table 2). IA-2 also contained many proteins responsible for this innate pathogen protecting system, including lysozyme C [53], lactoperoxidase [54], cystatin S, cystatin SA, and cystatin SN [55], BPI fold-containing families (BPIFA1, BPIFA2, BPIFB2) [56], cathelicidin antimicrobial peptide (CAMP) [57], and toll-interacting protein (TOLLIP) [58] ( Table 2). Interestingly, it has been suggested that lysozyme, peroxidase, and lactoferrin are secreted from the acinic cells of the salivary gland via exocytosis transport as free proteins [59]. Our observation suggested the following possibilities: 1) secretory vesicles that contain these proteins are released from acinic cells; or 2) these proteins attached themselves to the exterior side of EVs in saliva after they are released as free form proteins. It is also noteworthy that Group IA-2 contained some of glycolytic enzymes; that is, GAPDH, TPI1, ENO1, PKM2, PGAM1, PGK1, and ALDOA ( Table 2). Further studies including morphological observation are required to determine the role of these glycolytic enzymes in EVs; however, some of these enzymes have been shown to regulate membrane dynamics [60,61] and some of them have been reported to be recovered from extracellular space [4].
Our study also identified 21 proteins that did not attain a state of equilibrium in any of the three specimens under the conditions used (Group II). They included a -amylase, apolipoprotein (a) and others, which may have particular forms that retard their movement in media. The results emphasize the importance of two direction (upward and downward) analyses in density-gradient centrifugation. The list of proteins in Group II is found in S1 Dataset.
To elucidate the molecular mechanism that causes specimen-specific density drift, further studies are required; however, it is possible that environments of oral cavity (such as pH or osmotic pressure) could alter the physiochemical properties of particular subclasses of EV that have certain channel proteins. In this regard, it is noteworthy that the osmotic pressure of saliva is correlated with the rate of saliva secretion (or the amount of secretion) [62]. It should be interesting to investigate the effects of forced environmental disturbances on the properties of subsets of EVs in vitro.
The observations obtained here would offer new insight into EV-biology and contribute to the development of salivary EV diagnostic systems. The 111 proteins contained diagnostic markers, such as SLC44A4 [63], PMSA [64], CEACAM8/CD66b [65], and Serum amyloid A-1 protein (SAA1) [66]. Because these diagnostic markers have a probability of coboarding with the other proteins within the same subgroup, the information obtained in this study would help to select coboarded molecules that are used in sandwich ELISA [67] or affinity-based preconcentration [10][11][12].
Supporting information S1 Fig. The effects of a sonicating pretreatment of whole saliva on the properties of crude EV fraction. (A) Thirty mL of whole saliva from a healthy volunteer was divided into two portions, one of which was sonicated as shown in the Material and methods section. The larger materials were removed by centrifuging at 2,600 g for 30 min at 4˚C, and the resultant supernatants were centrifuged at 100,000 g for 110 min at 4˚C (Avanti JXN-30 with JS-24.38 rotor, Beckman Coulter (equivalent to 160,000 g for 70 min)) to obtain the crude 500 μL of crude EV fraction. (B) Ten μL of the crude EV fraction were placed on formvar-coated (Nissin EM, Tokyo, Japan) nickel grids (S-300 square mash, Gilder, Grantham, UK). After staining with 2% uranyl acetate (Wako, Tokyo, Japan) for 1 min, transmission electron microscopy (TEM) images were obtained with a H-7650 instrument (Hitachi, Co., Tokyo, Japan). (C) Aliquots of the samples were smeared on glass slides (Superfrost, Matsunami Glass Inc., Osaka, Japan) and stained with Papanicolaou's solution. Optical microscopic images were taken using a 100X object lens (Plan Apochromat, 1.45 x 0.13 mm, Keyence, Osaka, Japan) equipped on a BZ-X800 (Keyence), and the images were processed in Z-stack mode by using a BZ-800 Analyzer ver. 1 ; the logarithm values in a unit were normalized by the maximum. The fraction with the maximum was colored with red. According to the amount of each protein detected in every unit, heat maps were drawn from red to green using 256 steps in SVG format. If the protein was not detected, the corresponding fraction was colored with gray. Whereas experimentally 10 fractions were available, their densities varied depending on the units of samples. As illustrated in Fig 2, F1 to F10 were spread out actual density fractions, which resulted in many gray lanes appearing in the heat maps. The Perl script used in these analyses is available from GitHub repository (https://github.com/ yamamoto-tdc/EV-saliva). CD63 is shown in the top of the figure as the representative. D and U denote downward and upward separation, respectively. (PDF) S1 Dataset. A list of proteins found from human whole saliva in Excel data format. Data 1 column shows specimen numbers in which the protein was identified in no less than one from 20 fractions (10 fractions x two directions). Data 2 column represents the subclasses of the protein (see text). Density values of PFD (see text) for each specimen are shown for three specimens in the right three columns, in which "X" indicates that the protein was not detected from either or both directions, and "N. E." denotes that the protein has not achieved equilibrium in the specimen.