key: cord-0277832-cqxyegui
authors: De Luca, Thomas; Stratford, Robert E.; Edwards, Madison E.; Ferreira, Christina R.; Benson, Eric A.
title: Quantification of extracellular vesicles with unaltered surface membranes using an internalized oligonucleotide tracer and droplet digital PCR applied to modeling of in vivo kinetics
date: 2020-09-22
journal: bioRxiv
DOI: 10.1101/2020.09.22.306969
sha: b1bae561653566999e7c7bc56989fd257f7f2252
doc_id: 277832
cord_uid: cqxyegui

Extracellular vesicles (EVs) continue to attract interest for their potential role in targeted therapeutics and as biomarkers of disease and drug response. In order to achieve clinical utility, it is important to determine the pharmacokinetic parameters of candidate EVs during preclinical development using in vivo animal models. To date, no methods exist for studying EV kinetics without modification to surface ligands that may affect normal behavior. Here we introduce an accessible method for labeling and quantifying EVs administered to conscious animals, without disrupting endogenous ligands. Our method relies upon established laboratory techniques and can be tailored to a variety of biological questions. Digital PCR is leveraged to detect a non-homologous oligonucleotide tracer introduced into the vesicles, allowing for quantification over a wide dynamic range. Using an application of this method, we found differences in the in vivo kinetics of EVs from three different cell types using non-linear mixed effects modeling. We propose that this method will provide a complementary approach for the of study EV ligand-receptor interactions in the context of EV uptake and targeted therapeutics.

1 Extracellular vesicles (EV) can be used to improve medical treatments if properly understood 1, 2 . Chief among 2 the EV subtypes that have captured the interest of clinical researchers are exosomes, which are small (<200 3 nm) EVs that begin as the intraluminal vesicles of the late stage endosome, where they are loaded with active 4 biological molecules such as microRNAs (miRNA), mRNA, and proteins 3 . Once secreted, they transport these 5 contents to other nearby cells or to distant tissues via the blood circulation. Targeted distribution of these 6 vesicles is governed by surface markers, the composition of which is dependent on the originating cell [4] [5] [6] . Since 7 EVs are continually secreted by virtually every eukaryotic cell, it is broadly accepted that the composition of 8 any individual vesicle reflects the status of its originating cell at a particular moment in time. This dynamic 9

heterogeneity in blood-circulating EVs makes the study of EV kinetics difficult 7, 8 . 10

In order to quantitatively decipher the complexity of circulating EVs, there is a need for an easily 11 applicable, reproducible method for determining the kinetic parameters of EVs from known origins 2 . Due to the 12 inherent difficulty of studying EV transport and distribution in humans, preclinical in vivo animal models are 13

used. Existing studies of circulating EV kinetics are limited and have involved the development of membrane-14 associated labels and companion detection methods. The use of luciferase or radiolabels anchored to 15 exogenously expressed transmembrane proteins 4,9 may provide exceptional kinetic information for the 16 evaluation of engineered targeted therapeutics, but it is not ideal for the study of unmodified EVs. To arrive at a 17 better understanding of how endogenous EV composition affects kinetics, we should measure the kinetics of 18

EVs with unmodified surface membranes. 19

To address this gap in methodology, we sought to develop an accessible and scalable approach that: 20 jugular vein was located by blunt dissection, then isolated and tied off using 4-0 silk suture (cat. no. MV683, bacterial culture to screw-top cryovials (cat. no. 430488, Corning) and adding glycerol (cat. no. 327255000, Thermo Fisher Scientific) to a final concentration of 25%. Plasmid DNA was purified from the remaining 114 cultures using a NucleoSpin Plasmid (NoLid) high-pure plasmid mini prep kit (cat. no. 740499.50, Macherey-115 Nagel, Bethlehem, PA) according to vendor instructions and quantified using the Qubit dsDNA BR Assay (cat. 116 no. Q32850, Thermo Fisher Scientific) with a Qubit Fluorimeter (cat. no. Q32857, Invitrogen, Carlsbad, CA, 117 USA). Plasmid sequence was verified by total plasmid sequencing through the Massachusetts General 118

Hospital Center for Computational & Integrative Biology (MGH CCIB) DNA Core (Cambridge, MA, USA). 119

For preparation of large amounts of plasmid DNA, a small amount of glycerol stock scraped with a 120 sterile pipet tip was used to prepare a 3 mL starter culture in LB antibiotic selection media, incubated for 12 h 121 at 37 °C with shaking. 1 mL starter culture was transferred to 160 mL LB broth with 100 µg/mL ampicillin in 122 vendor's ligation protocol. 5 µL ligation product was introduced into 50 µL Stbl3 competent E. Coli cells (cat. 143 no. C7373-03, Invitrogen) by incubation on ice for 30 min, heat shock at 42 °C for 45 s, and placement on ice 144 for 2 min. 250 µL room-temperature SOC medium was added, and the transformation was incubated for 1 h at 145 37 °C with shaking on a Thermo Forma orbital shaker at 225 rpm. Three amounts (25 µL, 50 µL, and 100 µL) 146 of the liquid culture were used for antibiotic selection as previously described. 4 colonies were selected and 147 used to prepare glycerol stocks from liquid cultures as previously described. Sanger sequencing using the H1 148 promoter ( Supplementary Fig. 1 ) was performed by ACGT, Inc. (Wheeling, IL, USA) to validate sequence 149 insertion, and total plasmid sequencing was performed by the MGH CCIB DNA Core. 150

For consistency in plasmid production at scale, starter culture stocks were prepared as follows. Two 151 starter cultures were inoculated with scrapings from the main glycerol stock and incubated for 12 h at 37 °C 152 with shaking, then scaled up in 160 mL antibiotic selection media as previously described. Cultures were 153 combined, mixed, and aliquoted into six 50 mL conical tubes (cat. no. 2231000351, Eppendorf, Hamburg, 154

Germany), then centrifuged (1180 x g, 4 °C, 5 min) to obtain bacterial pellets. Supernatants were discarded 155 and each pellet was resuspended in 25 mL LB broth with 100 µg/mL ampicillin and 25% glycerol. 1 mL aliquots 156 were transferred to cryotubes and stored at -80 °C. When needed, one aliquot was thawed to RT and added to 157 160 mL LB broth with 100 µg/mL ampicillin and incubated 20 h at 37 °C with shaking. Plasmid DNA was 158 extracted using the Qiagen HiSpeed Maxi Kit (cat. no. 12662, Qiagen) following the vendor's protocol. 159 160

For EV preparations, cells from nitrogen storage were thawed and passaged at least twice (to a maximum of 5 162 times) before use. Cells cultured in T-75 flasks (cat. no. 12565349, Thermo Fisher Scientific) were grown to 163 70-80% confluency and transfected with 40 µg plasmid DNA using Lipofectamine 3000 (cat. no. L3000-015, 164

Invitrogen), following the manufacturer's protocol. After overnight incubation, the cell culture media was 165 removed and transfected cells were washed 3 times in 1X PBS (cat. no. MT21040CV, Corning). 10 mL base 166 medium supplemented with 10-20% vacuum-filtered (cat. no. 431162, Corning), exosome-depleted FBS (cat. standard conditions for 72 h. EV-enriched cell culture media was transferred to 15 mL LoBind conical tubes (cat. no. 0030122208, Eppendorf) and centrifuged (1,000 x g, 4 °C, 10 min) in a swinging-bucket rotor to pellet 170 residual cells and large debris. All but ~ 1.5 mL of the supernatant was carefully transferred to new 15 mL 171 conical tubes, then aliquoted into 1.5 mL microcentrifuge tubes (cat. no. 022431021, Eppendorf) prior to 172 depletion of large microvesicles and other cell debris by centrifugation (10,000 x g, 4 °C, 30 min). Supernatants 173 were consolidated into 50 mL conical tubes and mixed with 0.5 volumes of Total Exosome Isolation reagent 174 (cat. no. 4478359, Invitrogen), then precipitated overnight at 4 °C. After mixing by inversion, the suspended 175 precipitate was pelleted by repeated transfer of 1.5 mL into a microcentrifuge tube, centrifugation (10,000 x g, 176 4 °C, 5 min), and discarding of the supernatant. As a general rule, one 1.5 mL microcentrifuge tube was used 177 for every T-75 flask used for EV enrichment. In this way, pellets were consolidated so that each pellet 178

represented an amount of material equivalent to one T-75 flask worth of enriched media. Pellets were gently 179 washed with 1 mL 1X PBS and then softened by overnight incubation in 100 µL 1X PBS at 4 °C. Softened 180 pellets were resuspended by vortexing briefly, and residual precipitation reagent was removed by passing the 181 were quantified using the Qubit dsDNA BR assay, and a ligation reaction was prepared in a 20 µL volume 208 using 20 ng digested plasmid DNA, 2.5 µL digested insert, 800 U T4 DNA Ligase (New England BioLabs), and 209 ultrapure water. The ligation reaction occurred for 10 min at 37 °C, then the ligase was inactivated with a 10 210 min incubation at 65 °C prior to chilling on ice. The ligation product was introduced into Stbl3 competent E. Coli 211 cells by heat shock and plated for antibiotic selection, as previously described. Five colonies were selected and 212 scaled up for Sanger sequencing by ACGT, Inc., as previously described. 213 214

EVs were evaluated for morphology and contamination by the Electron Microscopy Center at Indiana 216 University Bloomington. To prepare negative stain grid, 4 µL of sample solution was applied onto a glow-217 discharged 300-mesh copper grid coated with continuous carbon film (EMS, Hatfield, PA, USA). The sample 218 solution was left for 30 s before blotted with a piece of filter paper. The grid was quickly washed using a 4-µL 219 drop of milli-Q (MilliporeSigma) water and stained with 4 µL of negative stain solution composed of either 1% 220 (w/v) uranyl acetate (EMS) with 0.5% (w/v) trehalose (MilliporeSigma) or 1% (w/v) ammonium molybdate 221 (MilliporeSigma) with 0.5% (w/v) trehalose. The excess stain solution was removed by filter paper and the grid 222 was put aside to allow air dried. Grids were imaged on a 120-kV JEM-1400Plus (JEOL USA Inc., Peabody, CA, USA). 225

EV preparations were analyzed for size distribution with dynamic light scattering using the Particle Metrix 228

ZetaView platform (Particle Metrix, Meerbusch, Germany). Data acquisition was performed at RT using 229 dilutions of EVs in 1X PBS. Using EV preparations diluted to a protein content of 2 µg/µL, a starting dilution of 230 15 µL in 1 mL of PBS was used and then further diluted to achieve empirical particle concentrations within the 231 acceptable range of the analysis software. Nanoparticle tracking analysis measurements were recorded and 232 Where applicable, membranes were cut into strips using visible molecular weight markers as guides. The raw MS data, MRM transitions and intensities, were processed using in-house scripts in order to 303 generate a list of MRM transitions and their respective ion intensities. Comparison of the absolute ion 304 intensities for the EVs to a blank sample (injection solvent) was then assessed and the MRMs that depicted an 305 ion intensity at least 30% higher than the blank were selected. The top 200 MRMs were selected to be used in primers was prepared and aliquoted into a LoBind 96-well plate (the "supermix plate") held at 4 °C on a cold 362 block. Next, the preamplification plate was placed on a 4 °C cold block and samples were transferred to the Droplets were prepared using 20 µL of each supermix sample. 365

Droplets were created using the automated droplet generator or the manual QX200 droplet generator 366 

EVs from transfected clone 9 cells expressing XMc39 tracer were used to establish a target dose amount for 380 animal experiments. EVs were prepared in bulk and quantified by protein concentration as previously 381 described. RNA extracted from 100 µg EVs (2 µg/µL) was diluted 1:100 in water and used to prepare cDNA for 382 ddPCR analysis, as previously described. Using the approximate total blood volume of a 300-450 g male 383

Sprague Dawley rat 18 , we determined that 1,000 µg EVs would result in an initial concentration (C0) near the 384 upper limit of ddPCR detection. A preliminary in vivo time course was performed to validate the calculated 385 dose amount, and to establish experimental duration. 386

High and low standards were produced to capture the batch variability of RNA extraction and analysis, 387 as follows. Citrated blood from two exsanguinated naïve animals was pooled. "Positive" plasma was prepared 388 by mixing 1,333 µg clone 9 EVs with 11.5 mL blood by gentle inversion. "Negative" plasma was prepared in 389 parallel using an identical volume of 1X PBS in place of EVs, but otherwise using the same protocol. Blood USA). Plasma was transferred to new 1.5 mL microcentrifuge tubes and further clarified by centrifugation 393 (10,000 x g, 10 min, 4 °C). "Positive" and "negative" plasma were pooled in 15 mL conical tubes (cat. no. 394 2231000349, Eppendorf), respectively. The "high" standard was developed by diluting "positive" plasma with 395 "negative" plasma and empirically determining the concentrations by ddPCR until a desirably high signal was 396 achieved. The "low" standard was prepared by diluting the "high" standard 30-fold with "negative" plasma. 397 "High" and "low" standards were evaluated by ddPCR to ensure they fell within the expected dynamic range of 398 our kinetic experiments, i.e. not to exceed the highest and lowest expected observations. 50 µL aliquots of high 399 and low standards were stored in 1.5 mL low-binding microcentrifuge tubes (Eppendorf) at -80 °C. after dosing. Each collection involved the following steps: discarding of lock solution and 0.1 mL blood, 420 collection of 0.2 mL blood, pulsatile flushing of the catheter with 0.25 mL bacteriostatic saline, and locking of 421 the catheter with 0.1 mL 4% sodium citrate. Blood plasma was separated from the blood (2,000 x g, 20 min, 4 422 °C) and then clarified (10,000 x g, 10 min, 4 °C) by centrifugation. Two 50 µL aliquots were transferred to 1.5 423 mL low-binding microcentrifuge tubes and stored at -80 °C for downstream analysis. Immediately prior to conducting an in vivo kinetic time course, detailed above, blood was collected from one 427 male Sprague Dawley rat by cardiac puncture. Briefly, the animal was euthanized by isoflurane inhalation (5% 428 induction, 5% maintenance). A laparotomy was performed, followed by a bilateral anterolateral thoracotomy. 429

One 20 mL syringe (cat. no. 309661, BD) pre-filled with 1 mL 4% sodium citrate (Fenwal) was used to obtain 430 10 mL blood from the exposed heart. The citrated blood was mixed by gentle inversion and 8 mL was EVs. Immediately after dosing the conscious rat for the in vivo time course, 150 µL (300 µg) of the same EV 434 dose preparation was spiked into the in vitro anticoagulated blood and mixed by gentle inversion. Immediately 435 subsequent to each blood collection from the conscious rat, 220 µL of blood was collected from the conical 436 tube; since the in vitro blood was previously mixed with citrate, 220 µL was collected to account for both the 437 200 µL sampled from the conscious animal and 20 µL sodium citrate added to these sampling syringes, as 438 detailed previously. After each blood collection, the citrated blood in the conical tube was mixed by gentle 439 inversion to prevent settling of the red blood cells over time. In vitro blood samples were collected at the same 440 time points, up to 240 min, and handled in the same way alongside the in vivo blood samples all the way 441 through ddPCR analysis. 442 443

Two standard curves were independently performed using serial dilutions of the high standard. Since the high copies/20 µL), a 5-fold concentration was prepared by performing Qiazol phase separation for each of 5 447 aliquots and binding the RNA precipitates to the same silica membrane column prior to elution. From this 5X 448 high standard, twofold dilutions were prepared using miRNA from naïve rat plasma as the diluent. 449

Concentrations were obtained by ddPCR. 450

To account for technical variability, high and low standards were included with every set of samples analyzed 453 by ddPCR. We normalized the data using our standard curve as follows. Standard curve copy numbers were 454 population-level parameters with associated inter-animal variability on those parameters. Initial parameter 469 estimates were made using the "initial estimates" function in Phoenix to manually create the best fit lines to the 470 observed data. Subsequently, each sequence of parameter estimation was limited to a maximum of 1,000 471

iterations. Observed concentrations were fit to the exponential form of equations describing two-compartment 472

and three-compartment model structures (Fig. 4b) . Equations were parameterized according to clearance structural parameters were included as a diagonal matrix initially. These random effects are reported as 475 percent variance from a log-normal distribution of individual subject parameter estimates, the basis of which is have 80% power to detect a 30% change in exosome clearance using an unpaired t-test and 5% type 1 error 500 rate. This is estimated based on the calculation of EV clearance to be 0.52 ml/min, and a conservative equations: 100% ID = 37 kBq; 37 kBq/100% ID x 3.2 (%ID x hr / mL) = 1.184 kBq x hr / mL = AUC; CL = d / 503 AUC; CL = 37 kBq / 1.184 kBq x hr /mL; CL = 31.25 mL / hr; thus CL = 0.52 mL / min. CL = clearance, d = 504 dose, hr = hour, AUC = area under the concentration-time curve; kBq = kilo Becquerel. We also determined 505 this sample size and sampling frequency per animal was adequate to support non-linear mixed effects 506

Elimination half-life (T ½), compartment distribution half-life, and AUC were determined from the 508 Phoenix post-hoc data for the final model. Elimination T ½ for each sample ln 2/(Cl/(V + V2 + V3)), compt 2 509 distrib T ½ = ln2 / (Cl2/V2). Compt 3 distrib T ½ = ln2 /(Cl3/V3). JMP Pro 14 was used for statistical analysis. 510

Given a sample size of 10 and without the assumption of normal distribution or equal variance, Wilcoxon and 511

Kruskal-Wallis rank-sum tests were applied as a conservative non-parametric approach to determining 512 significant differences between cell lines (P < 0.05). If significance was met by the Wilcoxon/Kruskal-Wallis 513 test, then the Steel-Dwass method was applied to evaluate for significant differences between cell lines. Steel-514

Dwass makes non-parametric comparisons for all pairs and takes into account multiple comparisons similar to 515

Tukey's Method for parametric data. 516 517

In order to discriminate exogenously administered EVs from endogenous background in rats, we incorporated 520 a tracer miRNA sequence that did not share homology with known rat miRNAs. The chosen tracer miRNA was 521 expressed using a commercial lentivector which appends an exosome localization motif 19 to the resulting 522 mature miRNA. For early optimization experiments, we used a prepackaged, proprietary non-targeting 523 sequence from the vendor designated "XMIR-NT". During development, however, we encountered constraints 524 that required a known sequence. We selected C. elegans miR-39-3p (cel-miR-39) for cloning into the same 525 lentivector ( Supplementary Fig. 1) , designated "XMc39". Because of its non-homology with certain species, 526 cel-miR-39 is commonly used as a quality control spike-in for miRNA PCR experiments involving biofluids from 527 humans, rats, and other mammals 17,20,21 . The validated XMc39 plasmid was transfected into 3 established rat-528 derived cell lines (clone 9 liver hepatocyte, RFL-6 lung fibroblast, and RMC kidney mesangial cells). The with secreted vesicles. EVs were isolated from enriched media using a commercial chemical isolation reagent 531 (Fig. 1) . Compared to ultracentrifugation, chemical reagents allow for substantially greater yield when retrieving 532

EVs from biofluids and cell culture supernatants at the expense of purity 22 . Co-precipitation of medium to large 533 vesicles 3 was minimized by modifying the manufacturer's protocol with an additional clarification step 23, 24 prior 534 to addition of reagent. Remaining chemical reagent which can cause vesicle aggregation (Fig. 2a) was 535 removed by careful washing of the pellet, resuspension, and filtration through low molecular weight size 536 exclusion columns. Nanoparticle tracking analysis and transmission electron microscopy confirm that 86.5% ± 537 1.5% (mean ± S.E.) of all EVs are in the 45-195 nm size range (Figs. 2a,c; Supplementary Fig. 2) , which is 538 typical of exosome-enriched small EV (sEV) preparations, and are deprived of aggregates (Fig. 2a) . Western 539 blot analysis (Fig. 2b) non-vesicular miRNA-binding proteins (category 5 3 ) were also assayed. Argonaute1-4 were detectable in two 551 EV samples. Non-sumoylated hnRNP A2/B1 (~35 kDa), a specific miRNA-binding protein implicated in the 552 mechanism by which our tracer miRNA is selectively loaded into EVs 19 ( Supplementary Fig. 1) , was barely 553 detectable in one EV sample, whereas a higher molecular weight band was visible in all EV samples at higher 554 exposures ( Supplementary Fig. 21b ). Mass spectrometry confirmed that our EV preparations were enriched in 555 sphingolipids and cholesterols ( Supplementary Fig. 3, Supplementary Data 1) , a hallmark of exosomes 34 , and 556 indicated differences in lipid composition between EVs from each cell line (Fig. 2d) .

Ideally, pharmacokinetic studies are designed to accommodate five half-lives of the administered compound in 560 order to measure approximately 97% of its elimination. We therefore required an assay with a dynamic range 561 of five half-lives that was sensitive enough to detect very low-abundance tracer miRNA during terminal phase 562 kinetics. Droplet digital PCR (ddPCR) is more sensitive than quantitative PCR 35 , and is capable of absolute 563 copy number quantification instead of relative quantification against intra-assay standard curves. TaqMan 564 stem-loop assays are preferred for the sensitive and specific detection of low-abundance miRNAs, but the 565 additional 3' exosome localization sequence in our XMc39 tracer rendered conventional cel-miR-39-3p assays 566 unusable. Custom TaqMan assays require 3' sequence specificity within 1-2 base pairs. We empirically 567 determined the tracer miRNA sequence in labeled secreted EVs, and several variant 3' sequences were 568 identified (Fig. 2e ). After consultation with the vendor, we designed a custom TaqMan assay, but we were 569 unable to adequately detect tracer in EVs isolated from XMc39-transfected clone 9 cells (data not shown). We 570 concluded that our tracer was incompatible with TaqMan chemistry. Hence, to quantify a heterogeneous 571 population of tracer miRNA sequences, we designed an assay for use with the EvaGreen intercalating 572 fluorophore. Conditions for ddPCR were optimized as follows. 573

The goal of ddPCR optimization is to maximize discrimination between positive and negative droplets 574 ( Fig. 3a) while also maximizing the sensitivity of target detection. Although EvaGreen is selectively fluorescent 575 in the presence of double-stranded DNA molecules, it is sensitive to high concentrations of single-stranded 576 DNA molecules such as primers. Using cDNA synthesized from a known quantity of purified RNA template, we 577 optimized two critical parameters: primer concentration and annealing temperature (T a) of the PCR reaction. 578

Early optimization used a synthetic RNA oligonucleotide of the proprietary XMIR-NT sequence and 579 corresponding forward primer, supplied by the vendor. Using a conservative starting primer concentration of 580 100 nM, we determined that 60 °C was the highest Ta to yield a discriminate band of positive droplets (Fig. 3b) . 581

Using 60 °C as an upper limit Ta to avoid non-specific amplifications that can occur at lower-than-optimal 582 temperatures, we then tested a primer concentration gradient (Fig. 3c) . The effect of primer concentration on 583

EVAgreen fluorescence is evident in this figure, as higher primer concentrations clearly increase the median 584 fluorescence of negative droplets. We determined the optimal primer concentration fell between 200-250 nM, established 58 °C as optimal. A final primer concentration gradient (Fig. 3e) using this Ta indicated an optimal 588 primer concentration of 200-225 nM. We determined 200 nM was preferable in order to minimize nonselective 589

EvaGreen fluorescence. At this stage of development, the proprietary XMIR-NT sequence was replaced with 590 cel-miR-39-3p (XMc39). Based on similarity in sequence length with the XMIR-NT forward primer, we used the 591 same primer concentration of 200 nM to perform a final temperature gradient for XMc39 (Fig. 3f) . We 592 determined 56 °C was the optimal Ta. 593

Commercial cDNA synthesis kits typically include single-stranded oligo(dT) adapters in the 2-5 µM 594 range. These kits are not optimized for use with ddPCR, and the high amount of oligo(dT) carryover is enough 595 to cause excessive EvaGreen fluorescence in droplets. We tested two strategies for minimizing oligo(dT) 596 carryover into the ddPCR reaction. First, we varied the input amount of oligo(dT) by using an alternative "flex" 597 cDNA synthesis kit from the same vendor which included separately packaged master mix components (data 598 not shown). Second, we diluted the cDNA product 1:10 in ddPCR supermix and performed PCR 599 preamplification immediately prior to droplet generation and ddPCR (Fig. 3g) . The rationale for preamplification 600 was twofold: lost sensitivity caused by the dilution of cDNA could be partially restored, and carryover oligo(dT) 601 functions as a reverse primer that is consumed during preamplification. We found that both strategies 602 minimized the "rain" (droplets that fall between negative and positive) (Fig. 3g ). For applicability with the 603 simpler non-"flex" protocol, we incorporated a 5 cycle preamplification into our standard protocol. 604

In order to establish linearity of the assay in a biologically relevant context, a synthetic RNA 605 oligonucleotide representing the ideal XMc39 sequence was purchased and two-fold serial dilutions were 606 prepared in miRNA extracted from naïve rat plasma. We then calculated expected ddPCR copy number values 607 from known input amounts of oligonucleotide template and compared them to analytical data. As shown in 608 We noted during the course of these experiments that negative controls consisting of naïve plasma 611 produced low, variable numbers of false positive droplets. To explore this, we prepared miRNA from two naïve 612 plasma samples and analyzed replicate aliquots of each. These samples produced a random signal ranging ddPCR amplification, however, consistently yielded no more than 1 positive droplet. Similar to the standard curve in Figure 3h , we analyzed lower concentrations of XMc39 and found that very low amounts of positive 616 control RNA template reduced the number and variability of false positive droplets found in our negative 617 controls ( Supplementary Fig. 4) . Taking this into consideration, we decided to use negative controls as a 618 measure of quality control rather than a hard threshold for data exclusion. Sample sets with negative controls 619 greater than 200 were reanalyzed using RNA as the starting material. Otherwise, we did not define a lower 620 limit of quantification and allowed the model to use all the data. 621

To ensure that experimentally observed concentrations of tracer were not artifactual, we assessed for 624 possible catheter contamination after dosing, as well as RNAse digestion of tracer in the blood. When dosing 625 animals with highly concentrated EVs through catheters, laminar flow might cause EVs to be retained within 626 the lumen. The interior volume of the jugular vein catheters used in this study was determined to be 627 approximately 40 µL. We filled a catheter with 2 µg/µL labeled clone 9 EVs and then injected PBS in sequential 628 40 µL volumes, collecting the outflow each time. Less than 0.1% tracer was detectable in the 4 th sample 629 relative to the 1 st (data not shown). We adopted the practice of using a pulsatile action when depressing the 630 plunger of a syringe in order to create turbulent flow when flushing. Furthermore, we decided to use a volume 631 of 500 µL when flushing the catheter with saline after dosing and 250 µL after sample collection. 632

Next, we performed a final validation of our EV preparation by testing the stability of incorporated tracer 633 miRNA. Since blood plasma is rich in RNases that degrade unprotected circulating RNAs (data not shown), we 634 needed to ensure that any observed elimination kinetics of tracer miRNA was specific to the behavior of its 635 encapsulating vesicles and not due to RNAse degradation of free-floating molecules. For the stability assay, 636 labeled EVs were intravenously administered to a live rat; in parallel, a proportional amount of the same 637 labeled EV preparation was spiked into anticoagulated whole blood. The whole blood was incubated in vitro in 638 a DNA LoBind tube at 37 °C. During the course of the experiment, in vitro blood samples were drawn 639 immediately after in vivo blood samples at pre-specified time intervals (Fig. 3j) . Tracer miRNA was stable for at 640 least 4 h in vitro, whereas it exhibited marked elimination over the same 4 h in vivo. From this, we concluded that the detectable tracer miRNA in our EV preparations was protected from RNAse degradation in the blood 642 and that observed in vivo kinetics are representative of the labeled EVs. 643

This optimized method was applied to test our hypothesis that EVs from cultured cell lines of different origin 646 exhibit different kinetics. Three Sprague Dawley-derived cell lines were selected for this study: clone 9 647 hepatocytes, RFL-6 lung fibroblast, and RMC mesangial cells. Liver, lungs, and kidneys have been identified 648 as major organs of exosome clearance [36] [37] [38] [39] [40] . Labeled EVs were isolated from each of these cultured cell lines 649 and intravenously administered to Sprague Dawley rats at a target bolus dose of 1,000 µg protein equivalent 650 (range: 935-1,000 µg). EV preparations from each cell line were administered to 10 animals; thus, 30 animals 651 were used in total. Blood samples were collected from each rat (clone 9, n = 10; RFL-6, n = 9; RMC, n = 9) at 652 2, 7.5, 15, 30, 60, 120, 240, 480, 960, and 1440 min following EV administration. Samples were processed in 653 sets, analyzed by ddPCR, normalized to high and low standards between assays, and normalized to dose 654 aliquots within assays ( Supplementary Fig. 5, Supplementary Data 2) . Two animals were excluded from 655 analysis. One animal from the RFL-6 group was removed for concern of cross-sample contamination, and one 656 animal from the RMC group for repeatedly failing quality control according to the pre-defined negative control 657

Normalized observed concentrations were plotted against the ideal collection time; differences between 659 cell lines were visually apparent on a semi-log plot (Fig. 4a ) and appeared to be multi-exponential, likely tri-660 exponential. Using ideal instead of actual collection times allowed for an analysis of standard error around the 661 mean. Using industry standard pharmacokinetic modeling software, for compartmental analysis, we used first 662 order conditional estimation -extended least squares (FOCE ELS) to estimate pharmacokinetic parameters. A 663 one-compartment model would not execute in the modeling software. As reported in Table 1 , a three-664 compartment model with one elimination from the central compartment ("3 compt model") results in a much 665 lower Akaike information criterion (AIC) value, which is a measure of model goodness of fit, than a two-666 compartment model with one elimination from the central compartment ("2 compt model") (Table 1) . Models 667 with elimination from the central compartment are the simplest models 41 (Fig. 4b) , and likely exhibit the lowest circulation. 670

Covariates of cell line, weight, and batch were incorporated into the three-compartment model, and only 671 the cell line covariate resulted in a meaningful decrease in the AIC value and change in eta-covariate 672 comparisons. Our protocol required animals to fall within a narrow weight range, so it is not surprising that 673 weight had little effect as a covariate. Using a shotgun approach of applying the cell line covariate to each 674 parameter, we found that applying the covariate to Volume 2 (V2), Volume 3 (V3), Clearance (Cl), and 675

Clearance 3 (Cl3) (Fig. 4b) resulted in the lowest AIC value (Table 1) . Code for execution of the model can be 676 found in Supplementary Note. 677 time. The simulated three-compartment model with covariates contains the observed data within the shaded 696 confidence interval, which suggests a good model description. 697

Notably, the volume of distribution for the central compartment (28 mL) is similar to the mean calculated total 700 blood volume of a male Sprague Dawley rat 18 with an average weight of 372 ± 6 g, or 26 ± 0.4 mL (mean ± 701 S.E.). As shown in Table 2 RMC and clone 9, and between RMC and RFL-6. 707

A bootstrap analysis using 1,000 simulations was performed to evaluate the likelihood of achieving 708 similar results if the experiment was replicated. Overall, the bootstrapped results mirrored the actual 709 experiments. One exception is that the clearance of elimination from the central compartment (Cl) was similar 710 for all 3 cell lines ( Table 2 , Supplementary Fig. 8 ). This suggests that cell line differences in EV kinetics are 711 due to differences in EV distribution to the peripheral compartments. circulation, yet modalities for the study of in vivo EV kinetics are limited to modifications of membrane 716 composition that provide a partial picture of how composition affects kinetics. Here, we detail an accessible 717 method for measuring the in vivo kinetics of EVs derived from cultured cells. Conceptually, we integrate 718 several major techniques in this approach. First, an expression vector is used to encode a non-homologous 719 tracer miRNA that is selectively packaged into exosomes. Next, labeled EVs are harvested from enriched cell 720 culture media and injected into rats. Lastly, ddPCR is used for the sensitive detection of cDNA prepared from 721 low-volume blood samples. Compared to other reported techniques 4-6 , our method bypasses the need for 722 modifying EV surfaces with external ligands. We also designed our protocol so that an entire kinetic time course may be performed using a single animal, minimizing the variability of pooling different time points 724 across multiple animals. 725

We successfully applied our method to test a hypothesis that EVs from different cell lines exhibit 726 different kinetics in vivo. These studies quantitatively described for the first time, though with important 727 caveats, significant differences in kinetic parameters between EVs derived from liver, lung, and kidney cells. 728

The known biology of EVs suggests that multiple routes of elimination are likely to exist (e.g. tissue 729 sequestration, intracellular degradation, and excretion). We did not have enough data to support a model with 730 elimination from a peripheral compartment, however a three-compartment model supports the idea that EVs 731 circulate in the vasculature and then move between shallow and deep peripheral compartments potentially 732

representing tissue distribution and intracellular metabolism. 733

A three-compartment model best described the observed kinetics of EVs derived from all three cell 734 lines used in this study. While there was a high degree of reproducibility between EV preparations from 735 multiple passages of the same cell line, one caveat is that differences between cell lines should be interpreted 736 with some caution. Cell lines were grown under optimal conditions recommended by ATCC, which 737 necessitated the use of different base culture mediums and different concentrations of FBS. While it is possible 738 that differences in nutrient and serum concentrations might generally influence EV composition, and thus 739 kinetics, our intention was to preserve the intended characteristics of each cell line without subjecting them to 740 the possible stress of suboptimal culture conditions. 741

As for the second caveat, we must note that the polyethylene glycol (PEG)-based EV isolation method 742 used in our study, while often chosen by researchers for its relative ease of use compared to 743 ultracentrifugation-based methods, is known to provide high yield at the expense of purity 25, 29, 42 . Using 744 standard protocols, PEG isolation is best described as providing a crude source of sEVs with the possibility of 745 co-isolating non-EV contaminants. Since this was the first application of our method to measure in vivo kinetics 746 of EVs, we chose to use PEG isolation of EVs for its two main advantages: high yield and reproducibility 747 across studies. In order to minimize the influence of potential contaminants, three steps were included in 748 addition to the standard protocol. First, we subjected EV-enriched media to an additional centrifugation step in 749 order to remove larger microvesicles and small debris 23, 24 . Second, the EV pellet was gently washed to remove column to remove small molecules such as unbound oligonucleotides and residual PEG. The resulting size distributions, transmission electron micrographs, and western blots are representative of sEV preparations 753 using other methods 37, 38, 40, 43 , and lipidomic characterization of our EV preparations provides further support by 754 demonstrating enrichment of endosomal lipids. Interestingly, the EVs prepared for this study did not induce any 755 noticeable inflammatory response in recipient animals; this may be due to the utilized cell lines originating 756 historically from Sprague Dawley rats. 757

The expression vector used in our study appends a localization motif to the encoded stem-loop tracer 758 miRNA sequence, and is expected to selectively enrich EVs of endosomal origin with the mature sequence 19 . 759

Although we demonstrated that tracer miRNA in our EV preparations was resistant to RNAse degradation and 760 took this as evidence for its encapsulation within vesicles, our chosen method of EV isolation implicates two 761 miRNA-binding proteins, Ago2 and hnRNPA2/B1, as possible co-isolated contaminants. Ago2 protects bound 762 miRNA from RNAse degradation, evidenced by the stability of non-vesicular circulating miRNAs in plasma 44 , 763 but evidence for hnRNP proteins conferring similar resistance is less convincing 45 . Ago2-bound miRNA is 764 abundant in plasma 46 and has been detected with EVs isolated by PEG from plasma 27 , but several studies 765 have failed to identify Ago2 in conditioned media 47 or EVs prepared from cultured cell lines [47] [48] [49] [50] , with two 766 exceptions 33, 51 . These differences in observations may be attributable to any number of experimental 767 conditions, including the type of cell lines used. We assayed our EV preparations by immunoblotting with a 768 pan-Argonaute (Ago1-4) antibody and found variable amounts of protein, including distinct bands in two of 769 three of our EV samples (Fig. 2) . In the case that co-precipitated protein-bound miRNA may have resulted in a 770 measurable signal, we have no reason to believe that the proteins of greatest concern (e.g. Ago2, hnRNPs) 771 would cause cell-specific differences in elimination kinetics, nor introduce significant artifacts in our model that 772 falsely attribute differences in elimination kinetics to the cell line covariate. However, since the method 773 presented in this report is applicable to measuring the kinetics of EVs isolated by any method, including 774 ultracentrifugation, it would indeed be possible to quantify the influence of different EV isolation methods on EV 775 kinetics in vivo and to determine the best-suited methods for such characterizations. 776

In this report, we demonstrate how the rich, quantitative data obtained by our method may be used with 777 nonlinear mixed-effects modeling to produce kinetic models of EVs secreted from cultured cells in vitro and including good practices for establishing ddPCR assay conditions and analytical reproducibility. One limitation of our study is the inability of our biological negative control (naïve plasma) to accurately define a lower limit of 781 quantification. As demonstrated, there was higher variability of false positive droplets in samples prepared from 782 naïve plasma than from samples prepared using very dilute positive controls. This suggests random off-target 783 amplification that is reduced when target template is present, even at very low concentrations. 784

Our findings demonstrate the ability of our method to parse the kinetic differences between EVs 785 isolated from different cellular origins or grown under different conditions. We expect this to be a powerful tool 786 for developing systems pharmacology-based and physiologically-based EV kinetic models that elucidate 787 heretofore unknown characteristics of EV biology. This will make it possible to develop and use in vivo kinetic 788 models to discover EV function and better design studies for therapeutics and circulating biomarkers. 789

Aspects of our method may be easily modified to explore different biological questions. For instance, 790

EVs of cultured cells may be interrogated for differences in response to various treatments, compared to 791 untreated controls. Additionally, labeling and vesicle isolation protocols may be altered to study other EVs such 792 as microvesicles and apoptotic bodies. Ultracentrifugation-based methods may be used to further isolate 793 subpopulations of EVs. By using conventional techniques and reagents, our method can be tailored to address 

Extracellular vesicles: a new communication paradigm?

From Endogenous Compounds as Biomarkers to Plasma-Derived Nanovesicles as Liquid Biopsy; Has the Golden Age of Translational Pharmacokinetics-Absorption, Distribution, Metabolism, Excretion-Drug-Drug Interaction Science Finally Arrived?

Minimal information for studies of extracellular vesicles 2018 (MISEV2018): a position statement of the International Society for Extracellular Vesicles and update of the MISEV2014 guidelines

Quantitative analysis of tissue distribution of the B16BL6-derived exosomes using a streptavidin-lactadherin fusion protein and iodine-125-labeled biotin derivative after intravenous injection in mice

CD169 mediates the capture of exosomes in spleen and lymph node

Analysis and Control of in Vivo Kinetics of Exosomes for the Development of Exosome-based DDS

Extracellular Vesicle Heterogeneity: Subpopulations, Isolation Techniques, and Diverse Functions in Cancer Progression

Cells release subpopulations of exosomes with distinct molecular and biological properties

Visualization and in vivo tracking of the exosomes of murine melanoma B16-BL6 cells in mice after intravenous injection

Guide for the care and use of laboratory animals

Public Health Service Policy on Humane Care and Use of Laboratory Animals. (National Institutes of Health

Improving the Patency of Jugular Vein Catheters in Sprague-Dawley Rats by Using an Antiseptic Nitrocellulose Coating

Comprehensive lipid profiling of early stage oocytes and embryos by MRM profiling

Histologic analysis and lipid profiling reveal reproductive age-associated changes in peri-ovarian adipose tissue

A rapid method of total lipid extraction and purification

Incubation of whole blood at room temperature does not alter the plasma concentrations of microRNA-16 and -223

Carboplatin with Decitabine Therapy, in Recurrent Platinum Resistant Ovarian Cancer, Alters Circulating miRNAs Concentrations: A Pilot Study

Gender differences in the blood volume of conscious Sprague-Dawley rats

Sumoylated hnRNPA2B1 controls the sorting of miRNAs into exosomes through binding to specific motifs

Variability in microRNA recovery from plasma: Comparison of five commercial kits

Evaluation of several methodological challenges in circulating miRNA qPCR studies in patients with head and neck cancer

Comparative analysis of exosome isolation methods using culture supernatant for optimum yield, purity and downstream applications

ExtraPEG: A Polyethylene Glycol-Based Method for Enrichment of Extracellular Vesicles

Isolation of extracellular vesicles: Determining the correct approach (Review)

The Regenerative Potential of Amniotic Fluid Stem Cell Extracellular Vesicles: Lessons Learned by Comparing Different Isolation Techniques

Effects of different separation methods on the physical and functional properties of extracellular vesicles

Different EV enrichment methods suitable for clinical settings yield different subpopulations of urinary extracellular vesicles from human samples

Comparison of isolation methods of exosomes and exosomal RNA from cell culture medium and serum

Procoagulant and immunogenic properties of melanoma exosomes, microvesicles and apoptotic vesicles

Extracellular histones are the ligands for the uptake of exosomes and hydroxyapatitenanoparticles by tumor cells via syndecan-4

Exosomes maintain cellular homeostasis by excreting harmful DNA from cells

Reassessment of Exosome Composition

Mass-spectrometry-based molecular characterization of extracellular vesicles: lipidomics and proteomics

Droplet Digital PCR versus qPCR for gene expression analysis with low abundant targets: from variable nonsense to publication quality data

Lung-derived exosome uptake into and epigenetic modulation of marrow progenitor/stem and differentiated cells

Characterization and comprehensive proteome profiling of exosomes secreted by hepatocytes

Characterisation of adipocyte-derived extracellular vesicles released pre-and post-adipogenesis

Microvesicles released from human renal cancer stem cells stimulate angiogenesis and formation of lung premetastatic niche

Exosomal transmission of functional aquaporin 2 in kidney cortical collecting duct cells

Reproducibility and efficiency of serum-derived exosome extraction methods

Extracellular vesicle measurements with nanoparticle tracking analysis -An accuracy and repeatability comparison between NanoSight NS300 and ZetaView

Argonaute2 complexes carry a population of circulating microRNAs independent of vesicles in human plasma

Extensive association of HuR with hnRNP proteins within immunoselected hnRNP and mRNP complexes

Quantitative and stoichiometric analysis of the microRNA content of exosomes

Proteomic Analysis of Cancer-Associated Fibroblasts Reveals a Paracrine Role for MFAP5 in Human Oral Tongue Squamous Cell Carcinoma

Proteomic Profiling of Secreted Proteins, Exosomes, and Microvesicles in Cell Culture Conditioned Media

Ultracentrifugation versus kit exosome isolation: nanoLC-MS and other tools reveal similar performance biomarkers, but also contaminations

Associated parameter stability from a bootstrap analysis (1,000 simulations) are also presented. Statistical analysis results of between-source comparisons of half-life and total exposure (AUC) are also summarized. V refers to volume of distribution of a compartment; Cl refers to elimination clearance from compartment 1 (Cl1), or between compartment 1 and 2 (Cl2) or

The authors report no conflict of interest. 817

We compared goodness of fit scatterplots between the two-and three-compartment models ( Fig. 6 ). The addition of the covariates to the three-compartment model 689 further improved upon the base model (Fig. 4 c, Supplementary Fig. 6 ) with individual model fits in 690 Supplementary Fig. 7 . 691We performed an observation-based simulated posterior predictive evaluation with prediction-corrected 692 visual predictive check (pcVPC, Fig. 4d) . The pcVPC used a log-additive error model to prevent simulating 693 negative concentrations. The pcVPC simulates concentration data from the three-compartment model with 694 covariates and plots the distribution of observations and the distribution of the predicted concentrations over