Therapeutic Extracellular Vesicles

MANUFACTURING AND ENGINEERING OF THERAPEUTIC EXTRACELLULAR VESICLES

Kelvin S. Ng

B.S. Biomedical Engineering University of Wisconsin-Madison, 2009

Submitted to the Harvard-MIT Program in Health Sciences & Technology in Partial Fulfillment of the Requirements for the Degree of

Doctor of Philosophy

at the

MASSACHUSETTS INSTITUTE OF TECHNOLOGY

February29t

Signature redacted Signature of Author: U Hafard-MIT Program in Health Sciences & Technology February 1, 2017

Certified by: Signature redacted Jeffrey M. Karp, Ph.D. Associate Professor of Medicine Brigham & Women's Hospital, Harvard Medical School Thesis Supervisor Signature redacted/ Accepted by: Emery N. Brown, M.D., Ph.D. MASSACHUSES I NSTITUTE Director, Harvard-MIT Program in Health Sciences & Technology OF TECHNOLC Professor of Computational Neuroscience and Health Sciences & Technology FEB 2012( 19 LIBRARI ES ARCHIVES MANUFACTURING AND ENGINEERING OF THERAPEUTIC EXTRACELLULAR VESICLES

Kelvin S. Ng

Submitted to the Harvard-MIT Program in Health Sciences & Technology on February 1, 2017 in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy in Biomedical Engineering

ABSTRACT

Originally viewed as 'garbage bags' which cells release to dispose of unwanted material, extracellular vesicles (EVs) have emerged as potent messengers that package and disseminate biochemical signals. This newly recognized mode of communication between cells has brought unprecedented therapeutic opportunities; at least 8 clinical trials and 7 companies are investigating or developing EVs as therapeutic products. As the EV industry rapidly grows, there is a rising demand for strategies that facilitate EV manufacturing. In this thesis, we address several challenges in EV manufacturing. By quantifying how many EVs a cell can release before it divides, we discovered that EV output increases as cells divide more slowly, providing a new way to maximize EV output from cells. Using our mathematical description of EV output, we built a computational model to estimate costs of EV manufacturing. Selecting cells with higher EV output despite slower proliferation can drastically lower costs. Meanwhile, although ultracentrifugation is the current standard for purifying EVs, we found that ultrafiltration-specifically tangential-flow filtration-is a more economical and scalable alternative, and we experimentally determined its utility for scaling up EV purification. For quality control, we established a suite of potency assays to measure the overall inflammatory action of EVs derived from human stem cells. Significant variability in EV potency between cells of different donors was detected, substantiating the need to robustly screen for appropriate cell sources when manufacturing EVs. Towards controlling EV function, we genetically constructed a versatile, multi-domain ligand that localizes to and modifies the surface of vesicles. Integrating biological, processing, and economic aspects of EV manufacturing, this thesis recommends strategies that may accelerate commercialization and clinical translation of EV therapy.

Thesis Supervisor: Jeffrey M. Karp Associate Professor of Medicine Brigham & Women's Hospital Harvard Medical School

2 CHAPTER ONE EXTRACELLULAR VESICLES As THERAPY: RISE AND CHALLENGES

"I have had 25 years of fun with bubbles in the cell, and I wish you all 25 years of fun with bubbles outside the cell." James E. Rothman, Nobel laureate 2013 Plenary speech at 2015 meeting of ISEV*

Vesicular Messengers Between Cells The fluid mosaic model, first proposed in 1972 [1], underscores two qualities of cellular membranes. As tightly packed, thermodynamically stable mosaics of phospholipids and proteins, membranes can sequester biomolecules, delineating boundaries that define a cell, and breaking down the interior of a cell into compartments with distinct functions. Meanwhile, fluidity enables fission and fusion of membranes, which are essential for subcellular compartments to communicate with each other via smaller membranous sacs called vesicles [2]. Recognition of how membranes and vesicles enable cell biology formed the basis of the Nobel Prizes in Physiology or Medicine awarded in 1974 and 2013 [3]. Vesicular trafficking is, however, not limited to intracellular compartments. Since the 1950s, synaptic vesicles have been observed to transfer biomolecules between neighboring neurons across a tightly enclosed region of extracellular space [4]. Since the 1960s, cells dying through a process now known as apoptosis have been shown to disintegrate into large vesicles, or apoptotic bodies, to facilitate engulfment and clearance by phagocytes [5,6]. Perhaps shrouded by the generality of apoptotic bodies across cell types, when vesicles were first reported in the 1980s to be naturally released by live, healthy cells into the open extracellular space [7,8,9,10], they were thought to be a mechanism by which cells dispose of unwanted biomolecules during development or homeostasis. Although these extracellular vesicles (EVs) were soon demonstrated to display a variety of functional proteins on the surface [11], the possibility that a dynamic network of vesicular trafficking exists not only within the cell but also between cells was not supported until B lymphocytes were reported to present and disseminate functional antigens via EVs to other immune cells [12]. Seminal work published in 2007 uncovered the ability of EVs to transfer RNA between cells [13], indicating that EVs can fuse with cells and deliver biomolecules into the cytosol. EVs are now understood to contain diverse biomolecules including RNA and proteins (Figure 1). Early evidence from electron microscopy established two subcellular origins of EVs in live cells: multivesicular endosomes [7,8,9,10,14] and plasma membrane including microvilli [15,16,17,18]. Based on these origins, EVs have been respectively termed as exosomes and microvesicles [19,20,211 (Figure 1). Currently, these subsets are difficult to distinguish in practice, since exosomes and microvesicles are polydisperse and overlap in size, and no subset-specific markers have been conclusively identified. Isolated EVs can be separated into more than two fractions each bearing a different protein composition [22,23,24,25]. When a EV population is profiled for its RNA composition, even the most abundant RNA sequence occurs at less than 0.01 copy per EV [26]. Hence, there is significant compositional variation between individual EVs released by a given cell population (Figure 1). A prevailing question in EV biology concerns how composition determines the fate of EVs. Intracellularly, vesicular trafficking is tightly coordinated not only by selective sorting of membrane components where vesicles are generated, but also by regulation of membrane components where vesicles are consumed [27,28,29]. Although EV biogenesis is partly understood [20,30,31], how trafficking of EVs between cells may be regulated, given their heterogeneity, remains to be investigated.

t International Society for Extracellular Vesicles

3 Glycolipid A / mRNA Transmembrane protein A mRNA Glycoprotein Intracellular

Microvesicle Exosome

------Lipid /Extracellular vesicles Figure 1. Cells release EVs which contain diverse biomolecules. According to traditional nomenclature, exosomes are EVs secreted from the endolysosomal compartment, while microvesicles are EVs that bud from the plasma membrane. Due to the lack of specific markers, exosomes and microvesicles are difficult to distinguish in practice. A population of experimentally isolated EVs are typically polydisperse and heterogeneous in composition.

PUBMED entries bearing relevance to exosomes, microvesicles, and extracellular vesicles 900

800

700 First EV diagnostic kit commercialized 600 First EV isolation kit commercialized * 500

y 400 E Z 300 First EV diagnostics trial reported

200 First EV therapy trial reported 100 J 0 1967 1973 1979 1985 1991 1997 2003 2009 2015 Year of publication

Figure 2. EV research has grown exponentially in the last decade or so and gained commercial interest. The plot shows results of a search on PUBMED for EV-related literature, using "(exosome* OR microvesicle* OR extracellular vesicle) NOT (exosome complex*)". Exosome complex refers to the intracellular assembly of RNases that regulate RNA degradation.

Therapeutics in a Class of Its Own

Regardless of our limited understanding of how individual EVs operate physiologically, EVs collectively have shown unique clinical value. As membrane-bound carriers, EVs not only disseminate membrane and cytosolic constituents of their parent cells, but also protect the biomolecules within. The ability of EVs to report aberrant molecular make-up (e.g. genetic mutations, misfolded proteins) of their parent cells, coupled I with the abundance of EVs in circulation, brings unprecedented diagnostic opportunities [20,321. From the therapeutic standpoint, there is tremendous potential in harnessing the endogenous functions of EVs to reverse etiologies, and using EVs to deliver exogenous drugs [20,33,341. Exponential growth in EV research since the 2000s, along with a rise in human trials and commercial interest, signifies a rapidly emerging industry (Figure 2).

4 By virtue of their size, membranous structure, ability to carry diverse biomolecules, and nature as a cell- derived but cell-free product, EVs are distinct from most other classes of therapeutics: small molecules, proteins, cells, tissues, synthetic drug carriers, medical devices. One class of therapeutics that may resemble EVs is viruses, particularly retroviruses which carry RNA and proteins within a lipid envelope [35]. However, EVs do not multiply and propagate per se even within a host, and bear constituents that are much less precisely regulated (given the diversity of RNA in EVs). Packaging EVs with defined molecules like how viruses can be designed is possible but challenging, since mechanisms that sort biomolecules into EVs are still poorly understood, in addition to the fact that EVs have at least two intracellular origins. Likewise, while EVs may be seen as protein-laden liposomes and be used as drug carriers, endogenous content of EVs may interfere with the exogenous drug and may need to be precisely controlled. Alike how cultured cells do not exactly reproduce physiological phenomena [36,37,38] but can still be therapeutic in vivo [39,401, EVs harvested from cultured cells have been used for treatment preclinically and clinically [33,34], despite incomplete understanding of the physiological roles of EVs. Especially in regenerative medicine, EVs can shift physiological states through complex, pleiotropic effects including reprogramming and epigenetic changes [41,42,43,44] that are previously observed only in the medicinal use of higher-order biologics such as cells and tissues. Such potent effects are likely due to the ability of EVs to co-deliver multiple signals: each EV, with its plethora of surface proteins, can trigger signaling in a recipient cell by binding multiple surface receptors, or even fuse with the recipient cell to insert RNA and proteins into the cytosol. Hence, EV therapy may not befit the paradigm of targeting specific molecules, genes, or pathways in diseases, which is often the case when small molecules, antibodies, or viruses are used. Rather, at least for now, EV therapy more closely resembles cell therapy, where multiple mechanisms of action are expected. Therefore, EVs represent a new class of therapeutics. Recognizing the unique challenges associated with EV therapy, regulatory changes have been proposed [45]. A major regulatory concern is how EVs may be manufactured robustly, consistently, and cost-effectively.

Box 1. Early companies developing EVs as products.

Anosys Inc., established in 1997, started with the aim of manufacturing autologous dendritic cell-derived EVs as cancer vaccines and needed to simultaneously establish bioprocesses for both cells and EVs. After showing promise in two phase 1 clinical trials, the company was acquired at the end of 2005 by Chromos Molecular Systems Inc. Subsequent development of its EV pipeline remains unknown. ReNeuron Group PLC, established in 1997, uses neural stem cells (NSCs) to treat neurological and ischemic conditions and is conducting several clinical trials for NSC therapy. EVs from their NSCs appear to promote wound repair, as presented at the Second International Meeting of the International Society for Extracellular Vesicles (Boston, MA, 2013). The company has listed EVs as a product in preclinical development. Systems Biosciences, Inc. (SBI), established in 2000, is a leader in developing technologies to isolate EVs from biological fluids and cell culture supernatants. The company's proprietary ExoQuick reagent contains a polymer that precipitates EVs from fluids and can be removed using spin columns. SBI supplies EV-depleted serum for cell culture in EV research and EVs isolated from a range of cells cultured in media supplemented with the company's own EV-depleted serum. Capricor Inc., established in 2013, specializes in cell therapy for cardiac applications. Based on cardiosphere-derived cells (CDCs), the company's lead product is being tested in clinical trials. Recently, EVs were reported to be the principal mechanism of action by which CDCs exert their therapeutic effects. Capricor has since secured exclusive rights to use CDC-derived EVs for tissue regeneration.

Manufacturing for Clinical Translation Despite its rapid growth, the EV field is relatively young. Literature discussing industrial manufacturing of EVs is very limited. Nevertheless, at least 7 companies are currently developing EVs as products for clinical or research use, the earliest of which are listed in Box 1 [46]. As the EV market continues to expand, demand

5 for EV products is expected to increase; tools that facilitate EV manufacturing and its scale-up will be increasingly needed. Since 2011 (Figure 2), many companies have started to commercialize tools to aid EV research. However, majority of these tools are suited for EV diagnostics but not EV therapeutics (Figure 3). Given the sensitivity of current analytical techniques (e.g. sequencing, microarrays), EV diagnostics requires smaller EV samples and can tolerate destructive isolation or analysis of EVs as long as the molecules of interest remain intact. In contrast, EV therapeutics entails processing of functionally intact EVs at clinical scale, placing more stringent specifications that perhaps led to the current dearth of commercial tools pertaining to EV therapeutics. This thesis seeks to provide tools that address the following challenges in manufacturing EV therapeutics. Detailed background information will be provided at the beginning of each chapter. 1. Maximizing EV output from cultured cells. Chapter 2 will demonstrate how EV output from cultured cells may be unambiguously quantified, and how culture conditions may be used to manipulate EV output. 2. Scaling up isolation of EVs. Chapter 3 will show how ultracentrifugation as the current standard for EV isolation is not scalable, and examine the utility of a more scalable alternative.

3. Developing potency assays to assure consistency in manufacturing.Chapter 4 will address variability in EV quality between different cell sources.

4. ControllingEV action. Chapter 5 will discuss how the surface of vesicles may be modified using genetically encoded exogenous ligands.

5. Minimizing costs of manufacturing. Chapter 6 will examine costs of goods and identify the cheapest bioprocess at different scales.

EV research tools EV diagnostics (Z32 trials) e invitrogen- exosomex part of technologies Isolated EVs may be disrupted ...:::::: Sciences

- Alnylam H. Med QWGEN

gCELL EV therapeutics (28 trials) MBLdance systems

Inenainl oprain TCopricor Exdvita NORGEN Theropeutica- biosciences s CORP. Isolad Es must EXUQON BIOTEK remain intact Seek Wfle- ReNeuron K npR J cc""a~ Figure 3. More commercial tools are available for developing EV diagnostics than for EV therapeutics. Majority of tools currently available for EV research are designed for isolation and destructive analysis of small EV samples (i.e. less than 10 mL), facilitating the progress of EV diagnostics. Much less tools are amenable for EV therapeutics, which involves larger volumes (i.e. more than 1 L) and requires EVs to be recovered intact.

6 CHAPTER TWO

EV PRODUCTIVITY OF CULTURED CELLS

Summary . Because cells are constantly dividing as they release EVs, accounting for rate of cell proliferation when quantifying rate of EV output permits distinction between a true boost in EV biogenesis pathways versus an apparent increase due to increased cell proliferation. . EV output varies between cells derived from different donors, and increases as cells divide more slowly.

Introduction Cells release extracellular vesicles (EVs) which contain RNA, proteins, and lipids. As complex assemblies of biomolecules, EVs can perform a variety of functions such as targeting distant cells, triggering signaling pathways, and transferring cytosolic RNA [20,31]. EVs from some cell types are naturally therapeutic: stem cell EVs can be regenerative [47,48], while dendritic cell EVs can serve as vaccines [49]. An emerging application utilizes EVs for targeted delivery of therapeutic RNA or protein [50,51,52]. For EV therapy to meet preclinical and clinical doses, maximizing EV output from cultured cells is imperative. Methods that have been reported to increase EV output include adding cytokines [53,54,55] and changing culture conditions such as type of culture vessel [561, atmosphere [57,58], and serum deprivation [58]. However, comparisons of EV output between conditions have typically been based on end-point EV yield, occasionally normalized to initial cell number [59]. Rarely has the kinetics of EV output been reported [60]. Moreover, without tracking cell growth, whether a condition truly increases EV output or merely increases cell growth without increasing EV output remains unclear, complicating the elucidation of pathways that govern EV biogenesis and release. A parameter widely used in biologics manufacturing to assess the effectiveness and consistency of cell culture is cell-specific productivity (qp), defined by Equation 1, where k, is the rate constant for exponential cell growth [61,62]. The ratio qp/kc indicates the increase in product output per cell between population doublings, which can be viewed as the total amount of product a cell makes before it divides. It isolates the rate of product output from the rate of cell growth, and can therefore decouple effects of the perturbation in question on product output versus cell growth. This is necessary to lay conclusive claims on whether a particular pathway regulates product output.

(Product at time 2 - Product at time 1) k (Cell density at time 2 - Cell density at time 1) ( The use of qp to describe EV output has yet to be reported. In this study, we aim to characterize the kinetics of EV output and superimpose it on cell growth to validate the applicability of qp to EVs. We choose mesenchymal stem cells (MSCs) as the model cell type given its therapeutic utility in multiple disease applications [63]. We use k, in place of qp to emphasize our focus on EVs and the assumptions from which Equation 1 is derived, namely, that cell growth is exponential and that the rate of change in EV number (V) is monophasic and linearly proportional to the instantaneous cell number (N): dN = kc N

ctkN V(t) = !- Ni[ek(t-tl) -- + V, V2 - V1 = (N 2 - N1) (2) d = kVN c c dt

7 Methods Selection of culture medium Presence of particles and proteins in StemPro serum-free medium (Gibco) versus minimum essential medium a (MEMu; Gibco) containing 10% fetal bovine serum (Atlanta Biologicals) was assessed by dynamic light scattering (ZetaSizer Nano ZS, Malvern). The refractive index of EVs was assumed to be 1.37 as previously estimated [64]. For each sample, measurements were made for 100 reads at 250C, and accepted only when the attenuation index was < 11. To isolate protein signals from EV signals, a detergent 0.1% Triton X-100 was added to lyse EVs. Cell culture Human MSCs immortalized with human telomerase reverse transcriptase (hTERT) were procured from ATCC, while primary human MSCs were procured from Lonza. All MSCs were cultured on CELLstart substrate (Gibco) in StemPro serum-free medium supplemented with penicillin/streptomycin (Gibco), and subcultured upon reaching 80% confluency. hTERT-MSCs were used unless otherwise stated. To track cell growth and EV output over time, MSCs were seeded in CELLstart-coated 6-well plates at the reported densities with 2 mL of serum-free medium per well. Each plate corresponded to one time point and contained up to six different conditions. After 6-8 hours, attached cells were rinsed twice with calcium- and magnesium-containing phosphate-buffered saline (PBS*/') to commence a time point where EV concentration is minimal, and fresh medium equilibrated at 37*C was added at 3 mL per well. Measurement of cell and EV numbers At each time point, conditioned medium was removed at 500- and 50-pL increments to estimate the volume of medium per well. Cells were immediately rinsed twice with cold PBS'/' to remove dead cells and frozen at -80*C to permeabilize attached cells. At the end of the time course, frozen cells in all plates were thawed, lysed, and measured for their DNA content using CyQUANT cell proliferation assay kit (Invitrogen). Complete cell lysis was verified by inspection using a bright-field microscope. At each time point, the conditioned medium was centrifuged once at 500g for 10 min to remove dead cells and twice at 2,000g for 20 min to remove cell debris including apoptotic bodies, at 4*C throughout [651. The final supernatant was kept on ice until its particle concentration was measured at 25*C by nanoparticle tracking analysis (NanoSight NS300, Malvern). The timeframe from collection to measurement was kept within 3 hours since MSC EVs were previously found to lose some physical features within 3 hours at 37*C and within 24 hours at 4*C [66]. For each sample, three 60-second videos were taken from which an average particle concentration was calculated. Between samples, the NS300 was rinsed at least once with detergent and at least five times with ultrapure water until particles were undetectable. In a separate experiment, the amount of DNA per MSC was estimated. A suspension of MSCs with unknown density was diluted to different extents. Four hemocytometers-two reusable and two disposable- were used to determine the cell number for each diluted sample. Concurrently, a sample at each dilution was frozen for the CyQUANT assay. The amount of DNA per MSC was finally determined by the increase in DNA content per increase in the consensus cell number among the hemocytometer readings.

Results StemPro contains minimal EVs and is suitable for tracking EV output from cultured MSCs. Dynamic light scattering (DLS) tracks fluctuations in the intensity of the aggregate light scattered from an illuminated population of particles. Smaller particles undergo faster Brownian motion and cause more fluctuations in the aggregate scattered light. DLS detected two particle populations in MEMx (Figure 4A),

.2 one of which was detergent-sensitive, had a modal size near 90 nm, and hence represented EVs (Figure 4B). The detergent-resistant population represented proteins, whose modal size was 10 nm. StemPro showed only one population which was detergent-resistant (Figure 4C-D). Unlike DLS, nanoparticle tracking analysis (NTA) tracks individual particles, reports particle concentration, and calculates the size of each particle from its Brownian motion. Since NTA cannot resolve particles below 80 nm, it excludes most proteins and tracked predominantly EVs in MEM(X, recording a modal size near 165 nm (Figure 4E). NTA could not detect enough signal in StemPro to generate a histogram. Our measured values for serum EVs as well as the discrepancy between DLS and NTA readings are in agreement with published reports [67,681. Collectively, DLS and NTA indicated that StemPro contains minimal EVs. To ensure that particles measured when tracking EV output originated only from cultured cells, we monitored the particle concentrations of MEMx and StemPro at 37*C for up to 4 days. No new particles were generated within each medium or from CELLstart substrate (Figure 5). We conclude that StemPro is a suitable culture medium for tracking EV output from MSCs.

A. MEMa without Triton X-100 (DLS) C. StemPro without Triton X-100 (DLS)

Proteins A ~EVs ftqf-

B. MEMa with 0.1% Triton X-100 (DLS) D. StemPro with 0.1% Triton X-100 (DLS) ~ I Good

E. MEMa without Triton X-100 (NTA)

Figure 4. StemPro contains minimal EVs. (A,B) Dynamic light scattering (DLS) detected two populations the of particles in serum-containing medium (MEMt), one of which disappeared upon addition of detergent 0.1% Triton X-100. The detergent-resistant population was identified as proteins, while the detergent-sensitive population was identified as EVs. (C,D) Detergent-sensitive particles were not detected in serum-free medium (StemPro), indicating the absence of EVs. (E) Nanoparticle tracking analysis (NTA) was unable to detect particles below 80 nm and therefore measured predominantly EVs.

9 Particles In MSC culture medium 40 -5-MEMa wIh 10% serum 3-e-StemPro with CELLStart 30- -StemnPro wthout CELLStart

15 S

0 0 12 24 36 48 60 72 84 96 108 lime from plating (hours) Figure 5. Particle concentrations of MSC culture media remain stable over time. Prolonged incubation of serum-containing or serum-free medium at 37*C, with or without CELLstart substrate, did not generate new particles.

The average DNA content of a MSC is about 9.4 pg.

To track cell growth, we use DNA content as a surrogate for cell count. Attempts with direct cell counting using hemocytometers yielded highly variable results. Reasons include: low cell numbers from 6- well plate cultures; incomplete enzymatic dissociation; and variability between hemocytometers. To determine the average amount of DNA in a MSC, a suspension of unknown MSC concentration was split into five samples at different dilutions. For each sample, cell number was estimated using four different hemacytometers, and DNA content was measured using the CyQUANT assay. Readings from all hemacytometers were pooled and correlated with DNA content, yielding an approximate DNA content of 9.4 pg per MSC (Figure 6), which was used in subsequent experiments to estimate cell number.

Hemocytometer cell count versus DNA content Correlation between cell count and DNA content 25,000 400 300,000 , - Reusable #1

- Reusable 62 350 250,000 20,000 - Dsposable 1 300 - Disposebe 02

---- CyQUANT 06200,000 15,000 250

200 $ 150,000 o 0 y - 9.40x + 13,470.30 U U 0 10,000 1 0.96 50 4C Z< R' 100,000

100 5,000 50,000 50 din

0 0 0 0% 20% 40% 60% 80% 100% 0 5,000 10,000 15,000 20,000 25,000 30,000 Percentage of Initial cell concentration Cell count Figure 6. MSCs contain about 9.4 pg of DNA per cell. (Left) Four hemocytometers from two different brands were used to directly count cells in samples diluted from a stock MSC suspension. DNA content was determined for the same samples using the CyQUANT assay. (Right) Cell count from different hemocytometers for each sample were pooled and correlated with DNA content. The line of best fit indicated an increase of 9.4 pg DNA per increase in cell count.

10 MSCs can grow exponentially in a 6-well plate. One of the two assumptions in Equation 2 states that cell growth is exponential. MSCs seeded between 2) were able to 50,000 and 200,000 cells per well in a 6-well plate (i.e. between 5,200 and 21,100 cells/cm 2 Not all seeded grow exponentially (R > 0.95), at slower growth rate with increasing seeding density (Figure 7). cells attached: only 35.0 2.25% of seeded cells attached by 8 hours after seeding, regardless of seeding density. Doubling times ranged between 20 and 45 hours. Other studies similarly employing StemPro and CELLstart 2 in MSC culture reported 46.3 7.1 hours at a seeding density of 10,000 cells/cm [691, 48-96 hours at 5,000- 2 2 time with 10,000 cells/cm [701, and 48-120 hours at 5,000 cells/cm [711. The increase in doubling increasing seeding density corroborated with previous reports using MEMa for MSC culture [721. In these studies, MSCs were cultured in T-flasks instead of multi-well plates.

Growth curves of MSCs in 6-well plate Growth rates of MSCs in 6-well plate 160,000 50 2 .50,000 R = 0.980 2 140,000 1.75,000 R = 0.991 .45 2 &100,000 R = 0.991 120,000 *150,000 R1 = 0.958 40 2 200,000 R = 0.983 100,000 ... 0

E 80 000 C , ---- 30

60,000 * 0 25 40,000 -..

20,000 20

15 0 150 200 250 0 10 20 30 40 50 60 0 50 100 3) Time from seeding (hours) Number of cells seeded per well (x10 plates at Figure 7. MSCs can grow exponentially in a 6-well plate. (Left) MSCs were seeded in 6-well medium densities between 50,000 and 200,000 cells per well, and incubated without changing culture cell number. for up to 3 days. At each time point, DNA content was determined for each well to estimate MSCs were able to grow exponentially at all densities. (Right) Since MSC growth was exponential, doubling time could be calculated, revealing slower growth rate with increasing seeding density.

EV output tracks cell growth at rates dependent on seeding density. time NTA was able to detect a monophasic and approximately exponential increase in EV number over cell and EV in medium conditioned by MSCs exponentially growing in a 6-well plate (Figure 8). Fitting both 2 The rate of EV kinetics data to Equation 2 revealed an R between 0.697 and 0.914 (on average 0.83 0.081). calculated as output per cell (k,) as well as the EV output per cell per population doubling (k,/kc) were per shown in Table 1. On average, each MSC added approximately 301-381 EVs to the culture medium hour. Taking cell growth into account, these rates indicate that, on average, each MSC released approximately cannot be 11,035-24,693 EVs before it divided. Such estimates quantifying specific biological phenomena cells, or to the derived from EV yields, which can be a normalization of EV number to the number of seeded instantaneous cell number (Figure 9). between Furthermore, instantaneous EV yield is unable to consistently discern a difference in EV output EVs, the different seeding densities (Figure 9B). While it reports that a lower seeding density yielded more less difference in instantaneous EV yield between different seeding densities becomes increasingly to the distinguishable over time. Such time-dependent differences are also seen when EV yield is normalized that can be drawn number of cells seeded, albeit to a smaller extent (Figure 9A). This means that conclusions from EV yields will depend on the timeframe of data collection.

11 Interestingly, instantaneous EV yield decreased over time for lower seeding densities, but increased for higher seeding densities (Figure 9B). This suggests that at lower seeding densities, cell number increased at a higher rate than EV number, and the converse occurred at higher seeding densities. Indeed, the rate of EV output per cell (k.), or cell-specific productivity, did not vary dramatically between seeding densities (340 31.4) (Figure 9C), whereas cell growth slows with increasing seeding density (Figure 7). This is consistent with the observation that EV output per population doubling increased with increasing seeding density (Figure 9D). Hence, rate parameters defined in Equation 2, which account for time, can report time- independent differences as long as the assumptions remain valid (i.e. exponential cell growth and monophasic EV output).

50,000 cellstwell 75,000 cellstwell 100,000 cellslwell 160,000 35 160,000 35 160.000 35

140,000 140,000 30 30 140,000 30

120,000 120,000 25 120,000 25 25 100,000 100,000 100,000 20j 20 20 E ~180,000 U80000 C C 15* 15 60,000 i U 60,000

10 10 40,000 40,000 10 40.000

20,000 5 5 20,000 20,000 5 0 0 0 0 0 0 12 24 36 48 60 0 12 a 24 36 48 0 60 0 12 24 36 48 60 Time from seeding (hours) Time from seeding (hours) Time from seeding (hours) 150,000 cellswell 200,000 cellsiwell 160,000 35 160,000 35

140.000 30 140,000 30

120,000 120.000 25 25 100,000 100,000 20j 20j 80,000 80,000 C 15 * I4, 15 U 60.000

10 10 Ir 40,000

20,000 5 20.000 5

0 0 0E&0 12 0 24 36 48 60 0 12 24 36 48 80 Time from seeding (hours) Time from seeding (hours)

Figure 8. EV output tracks cell growth at rates dependent on seeding density. hTERT-MSCs were seeded at different densities in StemPro. Cell numbers were tracked over time by DNA content, and EV numbers by nanoparticle tracking analysis. Generally, both cell and EV numbers increased exponentially over time in a monophasic fashion.

12 A. Normalization to number of seeded cells B. Normalization to instantaneous cell number 35,000 60,000 -- 50,000 -e-50,000 --- 75,000 -+-100,000 -- 30,000 75,000 50,000 150,000 -- 200000 -- 100,000 25,000 -150,000 U 40,000 -1-200,000 'U 20,000 3.' 30,000

V 15,000 0 Z 20,000 10,000 a ac C 10,000 5,000

0 0 0 12 24 36 48 60 0 12 24 36 48 60 Time from seeding (hours) Time from seeding (hours)

C. Cell-specific productivity D. EV output per population doubling 450 30,000

400 Z 25,000 350

300 ' 20,000 Ai 250 15,000 200

150 & 10,000

100 U o 5,000 1. 50

0 0 0 50,000 100,000 150,000 200,000 250,000 0 50,000 100,000 150,000 200,000 250,000 Number of cells seeded per well Number of cells seeded per well

Figure 9. EV rate parameters can explain time-dependent differences in EV yield between conditions. (A,B) Normalization to cell number (i.e. EV yield), be it the number of cells seeded or the instantaneous cell number, showed time-dependent differences between seeding densities. In particular, instantaneous EV yield suggests that the rate of cell growth exceeded the rate of EV output at lower cell densities, but conversely at higher cell densities. This may be explained by (C) a relatively consistent rate of EV output per cell (i.e. cell-specific productivity) coupled with cell growth that decreased with increasing seeding density. It follows that (D) cells at higher densities released more EVs per population doubling.

13 Donor I in 100% StemPro Donor I in 1% StemPro

160,000 35 160,000 35

140,000 30 140,000 30

120,000 120,000 25

100,000 100,000 20 1

1 80,000 80,000 X 15e 25101Q 60,000

10 0 40,000 40,00

5 20,000 20,000

0 I 1 0 0 0 12 24 36 48 60 72 84 98 108 120 0 12 24 36 48 60 72 84 98 108 120 Time from seeding (hours) Tne from seeding (hours) Donor 2 in 100% StemPro Donor 3 in 100% StemPro

160,000 35 100,000 35 01 140,000 30 140,000 30

120,000 120,000 25 25

100,000 100.000 20j 1 80000 E 80,000 15 15 0 60,000 0 60,000

10 I. 10 40.000 40,000

5 20,000 5 20,000

0 0 0 0 0 12 24 36 48 60 0 12 24 38 48 80 lime from seeding (hours) Time from seeding (hours) Figure 10. Choice of medium and cell source affects EV output. (A,B) Primary MSCs of the same donor from RoosterBio were cultured in StemPro with 100% or 1% supplement. Cell growth was affected to a greater extent than EV output. (C,D) Primary MSCs of two donors from Lonza produced less EVs than the RoosterBio MSCs. All primary MSCs were seeded at 50,000 cells/well.

Correlation between k, and kJk 50

45 Donor 1 i 40

N 30 hTERT 200k

25 hTERT 150k 20 .1 hTERT 50k 15 Donor 3 - a * hTERT 100k 10 I 5 Donor 2 - hTERT 75k 0 0 10 20 30 40 50 Doubling time (hours) Figure 11. Cells generally release more EVs with slower growth. The relationship between doubling time and EV output per population doubling shows a positive correlation. Conditions shown were all in 100% StemPro. Inclusion of the condition in 1% StemPro will retain the trend (see Table 1).

14 Table 1. Measured and calculated values for cell growth and EV output parameters. 2 MSC source Donor Culture medium Seeding density R for fit to Rate of cell Doubling time Rate of EV output, EV output per (cells/well) Equation 2 growth, k, (hours) kv doubling, k,/kc /hour) (EVs/cell/hour) (EVs/cell)

ATCC hTERT 100% StemPro 50,000 0.861 0.0313 22.1 361 11,524 100% StemPro 75,000 0.914 0.0272 25.4 301 11,035 100% StemPro 100,000 0.697 0.0210 33.0 325 15,489 100% StemPro 150,000 0.849 0.0172 40.2 330 19,125 100% StemPro 200,000 0.834 0.0154 45.0 381 24,693

RoosterBio 1 100% StemPro 50,000 0.852 0.0184 37.7 713 38,782 1 1% StemPro 50,000 0.833 0.00171 404 668 389,947

Lonza 2 100% StemPro 50,000 0.792 0.0366 18.9 351 9,589 3 100% StemPro 50,000 0.859 0.0429 16.2 479 11,171

15 Choice of culture medium and cell source also affects EV outbut. StemPro is reconstituted by addition of a protein-rich supplement to a protein-free basal medium. When the amount of supplement was reduced by 100 times, cell growth significantly slowed, but EVs still accumulated to a similar extent (Figure 10). Specifically, cell-specific productivity was marginally decreased, but the EV output per population doubling increased by an order of magnitude (Table 1). Two other sources of primary MSCs showed faster cell growth but similar EV output as hTERT-MSCs (Figure 10, Table 1). Collectively, our data indicates the EV output per cell per population doubling (k,/kc) is positively correlated with doubling time (Figure 11), or equivalently, negatively correlated with rate of cell growth (kc).

Discussion

To the best of our knowledge, this study is the first time where EV number is tracked over time in situ. Other studies that examined the kinetics of EV output quantified EVs by protein content only after the EVs were isolated [56,601. Because efficiency varies significantly between isolation methods [731, measurement in situ is more accurate for estimating the true biological EV output from cultured cells. This, however, forbids the use of protein content for EV quantification as EVs are not separated from non-EV proteins in the conditioned medium. NTA objectively measures EV concentration without additional processing. Other objective methods of particle enumeration include tunable resistive pulse sensing which requires a priori knowledge of particle size and charge, and flow cytometry which requires fluorescent stains [74,75]. Our study is also the first attempt to quantify cell-specific productivity for EVs. We show that EV yields (i.e. normalized to cell number at a defined time point) can be time-dependent and cannot pinpoint whether an increase in EV output was due to increased cell growth, increased EV production, or a mixture of both. By considering all time points in the quantification of EV output, we accounted for both cell growth and EV output, and could observe differences in EV output per cell per population doubling. In an emerging field where much standardization is desired [761, our methodology offers a new standard for calculating and comparing EV output. A monophasic increase in EV output over time suggests that the cell population is at a steady state. From our estimates, each MSC cultured in 100% StemPro loses 0.2-0.8% of its protein mass to EVs between cell divisions, given that an EV contains about 0.1 fg of protein [771. Industrial manufacturing of antibodies, growth factors, and other soluble proteins typically operates in the order of 10 pg of protein output per cell per population doubling [611, about 3% of a cell's protein mass. Note that these steady-state estimates represent net output-the overall effect of production, degradation, re-uptake, etc. The large discrepancy between EV and protein output may not be surprising given that a much more complex machinery than is protein production regulates EV biogenesis and release [20,30,31]. However, our study reveals that varying cell source or culture medium can increase EV output, the latter of which could dramatically raise the output by an order of magnitude, to a point that matches protein output. Furthermore, we observed that EV output negatively correlates with cell growth, a phenomenon that has been observed in protein manufacturing and taken advantage of by inhibiting cell cycle progression to maximize protein production [78,79,801. Since the current paradigm of formulating culture media in industry aims to maximize cell growth, a new paradigm for EV manufacturing may be necessary to meet EV therapy demands.

Future Work

The generality of Equation 2 should be validated by repeating our protocol on cell types beyond MSCs, although mathematically equivalent behavior has been previously described for EVs released by bacteria [811. Boundary conditions, such as exponential cell growth, should be strictly imposed. Significant departure from monophasic and exponential accumulation of EVs will undermine the validity of Equation 2 even as a first- order approximation.

11"I IMIN MINN MIN IM MINN= 0 0 MMI Correlation between EV output and cell proliferation suggests that signaling pathways governing cell proliferation may also govern EV biogenesis or release. Conditions known to perturb MSC proliferation may be tested to modify EV output. Our new metric of EV output can also verify conditions previously claimed to modify EV output. The use of a 6 -well plate format in our protocol should facilitate parallel testing of multiple conditions while ensuring that sample volumes suffice for accurate measurement. Preliminary testing will be necessary to identify treatment durations where Equation 2 remains valid, since cells will require some time to reach a new steady state following treatment.

17 CHAPTER THREE

TANGENTIAL-FLOW FILTRATION FOR SCALE-UP OF EV ISOLATION

Summary . Compared to ultracentrifugation-the current standard for isolating EVs, tangential-flow filtration is more scalable and effective for depleting particles from cell culture media, which would otherwise contaminate EVs prepared from these media. . Although tangential-flow filtration may recover EVs from conditioned media, it may not sufficiently remove non-EV components to render the EV product pure.

Introduction Distinct from other components of the secretome, extracellular vesicles (EVs) are membranous structures that encapsulate and deliver small molecules, lipids, proteins, and nucleic acids. EVs typically range from 50 nm to 200 nm in diameter, but can be as small as 30 nm [68,74,821 and as large as 1 pm 1831. Purifying cell type-specific EVs is vital to determine the biological significance and therapeutic utility of EVs. First, cells should be cultured in media depleted of exogenous particles (e.g. serum EVs, protein aggregates) that could be co-isolated with EVs produced by the cultured cells. Second, the desired EVs should be separated from other secretory factors without affecting EV integrity. Whether for depletion or purification (Figure 12), the current standard for EV isolation is ultracentrifugation (UC)-a highly laborious method. Effective depletion of exogenous particles from serum- containing media requires at least 18 hours of UC [841. Throughput is limited: about 230 mL for bench-scale UC (e.g. Beckman Coulter Optima L-90K) and about 200 L for the largest commercially available UC (e.g. Alfa Wassermann KII). Moreover, optimization of UC protocol is not straightforward. Increasing spin time or rotor speed, while able to boost recovery or pelleting efficiency [851, risks irreversible clumping or even fusion of EVs [86,871. Hence, UC is generally considered difficult to scale up. As the EV industry grows, scalable methods of EV isolation are increasingly in demand. Other methods reported for isolating EVs from culture media include filtration [88,89], polymer- induced precipitation [90,91], chromatography [86,92], and acoustic separation [93], among which only precipitation and filtration are suitable for both depleting particles from serum-containing media and purifying EVs from conditioned media. Precipitation requires a polymer additive that is eventually removed by serial centrifugation, and thereby introduces complexity to scale-up. Conversely, in filtration, additives are not necessary, and EVs can readily undergo buffer exchange and volume reduction. However, most published protocols for EV filtration utilize dead-end filters, where EVs may be extruded or irretrievably lodged. In this study, we investigate the ability of tangential-flow filtration (TFF) to deplete particles from serum- containing media and purify EVs from conditioned media (Figure 12). Unlike dead-end filtration, TFF directs EVs along a porous membrane instead of across the membrane. Such a flow direction reduces the rate of membrane clogging and caking, and spares EVs from physical impact on or across the membrane. Compared to UC, TFF is commercially available at larger throughputs for both academic and industrial use: about 500 mL for bench-scale TFF (e.g. Pall Minimate), and 1000-2000 L for industrial TFF (e.g. Sartorius Sartoflow Beta, Pall Allegro MVP). We are essentially evaluating if TFF is a scalable alternative to UC for EV isolation.

18 A. DEPLETION 0; from serum-containing media a Remove particles 0 but retain nutrients for cell growth

A Protein TANGENTIAL-FLOW FILTRATION 0EV Fluid flow -*na .- Q) 'L A A A* A0 A. .0 OW AAA A A L.

A LA A* A ~q) A A A 0Porous membrane Cu 0~ AAAA A A A A A A A A A A A

B. PURIFICATION Recover EVs from conditioned media but remove non-EV factors

Figure 12. TFF has upstream and downstream applications in EV production. (A) Particles pre-existing in culture media such as serum-containing media should be removed before cell culture, but particle- depleted media should still provide sufficient nutrients to support cell growth. (B) To purify EVs, all secretory factors but EVs should be removed, so that only EVs are recovered.

Box 2. Glossary of terms.

Isolation: complete or incomplete separation of desired product from impurities Purification: complete separation of desired product from impurities Recovery: complete or incomplete retention of desired product with or without impurities

Buffer exchange: transfer of desired product from one solvent to other Volume reduction: removal of solvent resulting in increased concentration of desired product

Diafiltration: process of subjecting desired product (i.e. retentate) to a series of buffer exchanges and filtrations to increasingly dilute and remove impurities Discontinuous diafiltration: diafiltration where, in each cycle, the desired product is completely filtered before fresh buffer is added, resulting in alternating rises and declines in concentration of the desired product Continuous diafiltration: diafiltration where fresh buffer is continuously added to maintain the desired product at the initial concentration

19 Methods Depletion of particles from serum-containingmedia Serum-containing media was simulated by supplementing phosphate-buffered saline (PBS) with 10% fetal bovine serum (FBS; Atlanta Biologicals). Particles were depleted from PBS/FBS either by UC (Optima L-90K, Beckman Coulter) at 100,000g for 18 hours, or by TFF (Minimate, Pall) using a polyethersulfone membrane with a 300-kDa molecular weight cut-off (nominal pore size of 35 nm). Protein and particle content of undepleted PBS/FBS, UC supernatant, or TFF filtrate were respectively determined by Bradford assay (Sigma) and nanoparticle tracking analysis (NTA; NanoSight NS300, Malvern). Size distribution of intact proteinaceous particles in depleted media was analyzed by one-dimensional native polyacrylamide gel electrophoresis (Bio-Rad) under non-denaturing, non-reducing conditions, since the introduction of detergents to linearize proteins may disrupt EVs and protein aggregates and obscure their presence. To evaluate the impact of depleting FBS particles on cell growth, Dulbecco's modified Eagle's medium (DMEM; Gibco) supplemented with 10% FBS and penicillin/streptomycin (Gibco) was depleted of particles by UC and TFF as above. L cells (ATCC) were cultured in undepleted, UC-depleted, or TFF-depleted DMEM/FBS, and measured for their DNA content using CyQUANT cell proliferation assay kit (Invitrogen) 24 and 72 hours after plating. Cell morphology was examined by bright-field phase-contrast microscopy. Purificationof EVs from conditioned media Conditioned media was prepared from L cells cultured in DMEM/FBS depleted of particles by TFF at 30 psi. To compare UC with TFF, conditioned media was simulated using PBS/FBS. To investigate TFF- induced protein aggregation, we used StemPro serum-free medium (Gibco). Particles or EVs were purified either by UC at 100 ,000g for 70 min, or by TFF at 30 psi. For repeated rounds of UC, supernatants from the previous round were discarded while the pellets were resuspended in PBS and centrifuged again. For repeated cycles of TFF, filtrates from the previous cycle were discarded while the retentates were diluted with PBS and filtered again (i.e. discontinuous diafiltration). UC pellets and TFF retentates were analyzed for protein and particle content, using micro BCA protein assay kit (Thermo Scientific) and NTA respectively. Purity was indicated by protein-to-particle ratio [771.

Results TFF is more effective than UC at depleting particles from serum-containingmedia. Even after 18 hours of centrifugation at 100,000g-the longest so far reported [841, UC could deplete only about 50% of the particles initially detectable in serum-containing media (Figure 13). Almost all proteins remained in the supernatant, indicating that the contribution of particles to protein content was negligible. Gel electrophoresis supported that UC partially removed proteinaceous particles above 250 kDa (Figure 13) and spared the rest. In contrast, regardless of transmembrane pressure, TFF almost completely removed detectable particles from serum-containing media; most of these particles were proteinaceous and above 250 kDa (Figure 13). However, TFF also depleted proteins from serum-containing media, and increasingly did so with increasing transmembrane pressure. Since particles contributed minimally to protein content, proteins removed from the serum-containing media were likely not associated with detectable particles. Despite loss of both particles and protein, the serum-containing media were still able to support cell growth without significant compromise between 24 and 72 hours after seeding (Figure 14).

11111111111111110 111111111111 1 1 ''I'll I Efficiency of UC versus TFF at depleting ;; I I particles from EV-containing media 2 a 4 120% .9 m am CL DProtein IL IL L 100% m Particle

80% -25OkD -

60%

16 40% m.m ~13OkD

20% -IOOkD Y -r...m 0% -70k SI- UC TFF TFF TFF ~55kD 18h (10 psi) (20 psi) (30 psi)

Figure 13. TFF is more effective than UC at depleting particles from serum-containing media. (Left) TFF almost completely removed detectable particles from PBS/FBS whereas UC retained about 50% of the particles. (Right) Native gel electrophoresis showed that proteinaceous particles above 250 kD remained in UC supernatants but were undetectable in TFF filtrates. Each lane was loaded with the same total amount of protein.

Cell growth in undepleted versus Undepleted TFF (10 psi) EV-depleted media 6

<4 z .9 3 0)TFF (20 psi) C 2

0 Undepleted UC TFF TFF TFF 18h (10 psi) (20 psi) (30 psi)

Figure 14. Particle depletion does not compromise the ability of serum-containing media to support cell growth. (Left) Proliferation and (right) morphology were similar whether L cells were cultured in undepleted or depleted DMEM/FBS. Time points were taken at 24 and 72 hours after seeding.

21 TFF may recover but may not purify EVs from conditioned media. UC effectively purified particles from PBS/FBS: protein-to-particle ratio (PPR) decreased significantly after 1 spin and plateaued at around 1 fg/particle even after more spins (Figure 15). Consistent with our earlier observations (Figure 13), majority of proteins were removed after the first spin, and subsequent decreases in protein content corresponding to decreases in particle content were small compared to the total protein content of the initial media. UC therefore isolated particles including EVs from other proteins. However, TFF could not reduce particle or protein content even after repeated diafiltrations, and instead increased particle content (Figure 15). Additional particles may originate from the TFF equipment because of leaching or spalling, but we detected minimal particles when a protein-free buffer (i.e. PBS) was fed into the system (Figure 15), which could not account for the particle increase. When a protein-rich buffer (i.e. StemPro) was fed, we observed particle increase despite a concomitant loss of proteins (Figure 16). Likely, proteins in the retentate were concentrated during each TFF cycle and formed aggregates that were detectable as particles. Indeed, PPR for PBS/FBS plateaued above 100 fg/particle instead of the 1 fg/particle attained by UC (Figure 15), indicating presence of particles densely packed with proteins. When EVs were isolated by UC from L cell-conditioned DMEM/FBS (previously particle-depleted), PPR averaged between 1 and 10 fg/particle (data not shown). Although TFF could achieve a similar PPR, we were unable to conclude if the particles were EVs since particle content increased initially during TFF (Figure 16).

Discussion TFF is widely used and preferred over UC in the biotechnology industry for upstream (e.g. clarification of culture media) and downstream (e.g. purification, volume reduction, buffer exchange) processing of biologics including antibodies, viruses, and cells [94,95,96,971. To the best of our knowledge, there has been only one report of the use of TFF for EV applications [981. Specifically, Heinemann et al. developed a series of filtration steps involving dead-end, track-etched, and tangential-flow filters to isolate EVs. In our study, we investigated the use of only TFF for EV isolation. Both our studies are in agreement that TFF is superior to UC at depleting particles from serum-containing media, but our study additionally showed that the depleted media still supported cell growth. Although Heinemann et al. demonstrated enrichment in a putative EV marker when purifying EVs from conditioned media, they did not verify that non-EV proteins were sufficiently removed. In contrast, we paid particular attention to purity indicated by PPR. Because our data strongly suggested that TFF generates EV-sized particles by inducing protein aggregation, we can at best confirm that TFF can recover EVs, but we are unable to conclude if TFF can purify EVs. Meanwhile, our data supports that TFF is useful for concentrating and reducing the volume of conditioned media, during which buffer exchange can concurrently happen. The retentate, which would contain EVs and other particles including protein aggregates, can then be fractionated using chromatography where EVs may be separated from other particles based on size and charge. Since chromatography will dilute the EVs, TFF can be used again after chromatography to achieve an appropriate EV concentration for storage and administration. This sequence of steps resembles current downstream processes in the biotechnology industry [94,95,96,971, although optimization strategy will need to be catered for EV applications. Our study provides some parameters and concerns that may aid in optimization.

, " , 111 1 - 111111 NON I 1 11, 11 Particle retention after repeated Particle retention after repeated rounds of ultracentrifugation cycles of tangential-flow filtration 1E+13 1E+13

1E+12 1E+12

V -10% FBS -Protein-fre buffr 1E+11 1E+11 cc CL ft

09 12 1E+10 1E+10

1 E+9 1E+9 0 1 2 3 0 1 2 3 4 Number of spins at 100,000g for 70 min Number of cycles at 30 psi Protein retention after repeated Protein retention after repeated rounds of ultracentrifugation cycles of tangential-flow filtration 1 E+7 1 E+7

1E+6 1 E+6

1E+5 E 1E+5

C 1 E+4 8 1E+4

2 1 E+3 1E+3

1E+2 1 E+2

1E+1 1E+1 0 1 2 3 0 1 2 3 4 Number of spins at 100,000g for 70 min Number of cycles at 30 psi Purity of particles after repeated Purity of particles after repeated rounds of ultracentrifugation cycles of tangential-flow filtration 1 E+4 1E+4

1E+3 r 1E+3

1E+2 1E+2

1E+1 1E+1 E 0

1E.O 1 E+O 0 1 2 3 0 1 2 3 4 Number of spins at 100,000g for 70 min Number of cycles at 30 psi Figure 15. TFF is unable to purify particles from serum-containing media. (Left) UC significantly separated proteins from particles, resulting in a plateau in protein-to-particle ratio which indicates purified particles. (Right) TFF not only retained majority of proteins, but also generated additional particles in the retentate. Majority of these additional particles did not originate from the TFF equipment (e.g. leaching, spalling) since much less particles were detected when a protein-free buffer was fed into the system.

23 TFF retention of particles in TFF retention of proteins in Particle purity in retentate of unconditioned StemPro unconditioned StemPro unconditioned StemPro 5 8 1E+5

4 1E+4 4a M 6

o a, 5 1E+3 3c 8 a 4 )1E+2

2 3 -. L 1E+1 1 ~2 22 1 a 0 CL - 1E+O 0 0 1E-1 0 1 2 3 4 0 1 2 3 4 0 1 2 3 4 Number of cycles at 30 psi Number of cycles at 30 psi Number of cycles at 30 psi TFF retention of particles in TFF retention of proteins in Particle purity in retentate of L cell-conditioned DMEM L cell-conditioned DMEM L cell-conditioned DMEM 16 45 1E+5 U)4 40 .2 351E+4 M12E 3 - 0 1E+3 010 0 S25 1E+2 20 (D4 15 1LE+1 S100 S2 5 2C 0 0 1E-1 0 1 2 3 4 0 1 2 3 4 0 1 2 3 4 Number of cycles at 30 psi Number of cycles at 30 psi Number of cycles at 30 psi

Figure 16. TFF can induce protein aggregation in the retentate, and therefore may recover but may not purify EVs from conditioned media. (Top) An increase of particles with a concomitant loss of proteins in a protein-rich but particle-free buffer during TFF indicates protein aggregation. (Bottom) TFF also increased particles in conditioned media, but could restore particle content after diafiltration.

Future Work

A major technical limitation in our study is the difficulty in quantifying relative proportions of EVs, protein aggregates, and other particles detectable by NTA. Theoretically, EVs may be distinguished by their membranous structure, but our attempts at cryoelectron microscopy were unsuccessful because the EV preparations were too protein-dense to obtain clear images. An alternative is to quantify EV purity by multiple indicators in addition to PPR, such as lipid-to-particle ratio and RNA-to-particle ratio. A purer EV preparation would have lower PPR, higher lipid-to-particle ratio, and higher RNA-to-particle ratio, since other secretory factors are unlikely associated with lipid and RNA. These ratios obtained from UC-purified EVs can serve as benchmarks for TFF-purified EVs.

Three different types of complex biofluids were examined in our study of EV purification: PBS/FBS, StemPro, and L cell-conditioned DMEM/FBS. Since the same molecular weight cut-off was used in all TFF experiments, our data suggests that PBS/FBS contains mostly proteinaceous particles above 300 kDa since minimal proteins were removed from the retentate during TFF (Figure 15). Particles in both StemPro and L cell-conditioned DMEM appeared to be increasingly purified with more diafiltrations, since decreasing protein content and PPR suggest more complete separation of particles from proteins (Figure 16). However, with more diafiltrations, particle content appeared to increase for StemPro but not for L cell-conditioned DMEM. The new particles formed are hence likely of different nature in StemPro versus L cell-conditioned DMEM. Further analysis through gel electrophoresis, size-exclusion chromatography, or even mass spectrometry may explain the disparate behaviors of the biofluids during TFF.

24 During discontinuous diafiltration, the retentate is maximally concentrated before a buffer is added to dilute the retentate for the next diafiltration. In contrast, continuous diafiltration keeps the retentate at its initial concentration by continually adding buffer to replace volume lost as the filtrate, and thereby avoids excessive concentration of feed components that have yet to exit from the retentate into the filtrate. Continuous diafiltration may reduce protein aggregation if EVs are the only feed components above the molecular weight cut-off. Otherwise, analysis of protein size distribution in the feed by gel electrophoresis or size-exclusion chromatography will be necessary to select for the appropriate molecular weight cut-off. Because TFF membranes are provided at discrete increments of molecular weight cut-offs, any overlap in size between EVs and proteins would imply a need to choose between loss of smaller EVs or retention of larger proteins. We recommend that both scenarios should be investigated to rule out if TFF alone is truly unfeasible for EV purification and therefore requires other technologies such as chromatography to produce pure EVs.

25 CHAPTER FOUR

INTER-DONOR VARIABILITY IN EV POTENCY

Summary . EVs of mesenchymal stem cells from different donors elicit different effects on different cell types involved in inflammation: monocytes, macrophages, and endothelial cells. . Not a single assay but a suite of assays may be necessary to capture how EVs impact different cell types, and thereby predict the overall therapeutic outcome.

Introduction As agents of intercellular communication, extracellular vesicles (EVs) transfer biomolecules including proteins and RNA between cells. Significant molecular differences exist between EVs and their parent cells [13,99,100], indicating that not all cellular constituents are available for transport via EVs. Yet, for a variety of cell types, EVs do inherit sufficient molecular information to reproduce some characteristic functions of the parent cell [33,34,47]. This is especially so when the parent cells are therapeutic, such as dendritic cells [101], endothelial progenitor cells [102], and mesenchymal stem cells [103,104], whose EVs have been shown to elicit similar therapeutic effects. However, the link between molecular composition and biological function of EVs is not straightforward. Possible interactions between an EV and the *recipient cell range from receptor-ligand binding to fusion whereby membrane and cytosolic material is transferred from the EV to the cell [20,30]. Furthermore, as a submicron particle (typically 50 to 200 nm in diameter), an EV has a minuscule capacity and is unable to singly encapsulate the entire diversity of molecules detectable in an EV population. Stoichiometric analysis has revealed that the average number of copies of a given microRNA sequence in an EV is about 0.0083; even the most abundant microRNA sequence occurred at 1 copy per 9 EVs [261. Likewise, EVs can be fractionated into subsets each bearing a unique set of protein markers [22,23,24,25]. EVs harvested from a given cell population are therefore heterogeneous, and their biological functions are likely the synergistic outcomes of different EVs perturbing different cells via different modes of interaction. Consequently, unlike the case for other biologics such as proteins and viruses where a purified preparation is presumably homogeneous in composition, molecular profiling can at best suggest but cannot predict the biological function of EVs, at least until robust techniques are developed for single-vesicle analysis and tracking. Biological functions of therapeutics are quantified using potency assays. According to the United States Code of Federal Regulations Title 21 Part 600.3(s), potency refers to 'the specific ability or capacity of the product, as indicated by appropriate laboratory tests or by adequately controlled clinical data obtained through the administration of the product in the manner intended, to effect a given result'. Potency assays are instrumental for assessing lot-to-lot variability in product quality, and are required by regulatory authorities for lot release. A series of complementary potency assays in vitro informing and leading to animal testing and clinical studies are essential [1051. Importantly, when primary cells are used to produce therapeutics, which is the case for most EV therapies [33,34,47], intra- and inter-donor variability should be expected, demanding a first-line potency assay for donor selection [106]. In this study, we aim to develop in vitro cell-based assays that can detect inter-donor variability in EV potency. We focus on human mesenchymal stem cells (MSCs), since significant inter-donor variability has been reported for MSCs and their cytokine profiles [107,108,109 but not yet for MSC-EVs. Therapeutic efficacy of MSC-EVs has been demonstrated in multiple disease models including myocardial infarction [58,104,110], pulmonary hypertension [111,112,113], lung infection [114,115], kidney injury [103,116,117,118,119], liver injury [120,1211, brain injury [122,123], and transplant rejection [124,125,126].

26 Central to and common among these pathologies is inflammation, wherein lymphocytes, monocytes and their derivatives, and endothelial cells are heavily implicated. Indeed, MSC-EVs have been shown to promote anti-inflammatory and pro-angiogenic responses in vitro through interactions with peripheral blood mononuclear cells [127,128,129,130], primary and immortalized monocytes [124], as well as primary endothelial cells [54,131,1321, although results often vary between studies. We hypothesize that the variation between studies is due to inter-donor variability, and in this study, we adapt and modify these cell-based assays for donor screening.

Methods Purificationof EVs released from MSCs Primary MSCs from different donors were procured from Lonza and RoosterBio, while MSCs immortalized with human telomerase reverse transcriptase (hTERT) were procured from ATCC. All MSCs were cultured on CELLstart substrate (Gibco) in StemPro serum-free medium (Gibco) supplemented with penicillin/streptomycin (Gibco), and subcultured upon reaching 80% confluency. Primary MSCs, typically received at P2, were subcultured not more than 5 times (i.e. up to P7). No limit was imposed for subculturing immortalized MSCs, but they were kept below P20. Between subcultures, conditioned medium was discarded and fresh medium was added every 2-3 days, except when confluency was between 50% and 80%, during which the conditioned medium was instead kept for EV isolation. Conditioned medium was centrifuged once at 500g for 10 min to pellet dead cells and twice at 2,000g for 20 min to pellet cell debris including apoptotic bodies, at 4'C throughout [651. The supernatant was centrifuged twice at 100,000g for 70 min at 4*C to pellet EVs, rinsed once with ice-cold ultrapure calcium- and magnesium-free phosphate-buffered saline (PBS-'-) in between. Each EV pellet was resuspended in 100- 120 pL of ice-cold PBS- and frozen at -80*C. Fresh and thawed EVs were characterized using cryoelectron transmission microscopy (Swanson Biotechnology Center, Koch Institute for Integrative Cancer Research, MIT) for morphology, nanoparticle tracking analysis (NanoSight NS300, Malvern) for size distribution and particle concentration, and micro BCA protein assay kit (Thermo Scientific) for protein concentration. EV samples with protein-to-particle ratios below 10 fg were considered pure [771. Isolation, culture, and activation of peripheral blood mononuclear cells (PBMCs) Heparinized whole blood from different human donors was procured from Research Blood Components. On the same day of blood collection, PBMCs were extracted from the blood by Ficoll-Paque separation (GE Healthcare Life Sciences) per manufacturer's instructions. Following rinse with PBS-/, the PBMCs were treated with ACK lysing buffer (Gibco) to eliminate potential contamination by erythrocytes. Purified PBMCs were rinsed again with PBS-/- and frozen in CryoStor 5 (BioLife Solutions) which contains 5% dimethyl sulfoxide. For functional assays, PBMCs were thawed, resuspended in PBS- supplemented with 0.1% bovine serum albumin (BSA), and fluorescently stained with 1 pM CellTrace CFSE (Invitrogen) for 10 min at 37*C. X-VIVO 15 (Lonza), a chemically defined serum-free medium, was added to rinse and resuspend the stained PBMCs at a final concentration of x 106 cells/mL. To activate PBMCs, the cells were treated with lipopolysaccharide (LPS; Sigma) or anti-CD28/CD3 Dynabeads (DB; Gibco) at the reported concentrations, with or without MSC-EVs, and then incubated at 370 C for 1-4 days in 96-well plates. At each time point, treated PBMCs were compared with untreated but stained PBMCs using flow cytometry (Accuri C6, BD Biosciences). Successful activation leads to proliferation and is indicated by the appearance of cell populations with lower fluorescence than the control population.

27 Culture, differentiation, and activation of THP-1 monocytic cell line THP-1 cells (ATCC) were cultured in ATCC-formulated RPMI 1640 (ATCC) supplemented with 10% fetal bovine serum (Atlanta Biologicals), penicillin/streptomycin, and 50 pM molecular biology-grade 2- mercaptoethanol (Sigma). Culture medium was refreshed every 2-3 days and cell density was kept below Ix 106 cells/mL. For differentiation from a monocyte-like phenotype to a macrophage-like phenotype, THP-1 cells were treated with 25 nM phorbol 12-myristate 13-acetate (PMA; Sigma) for 48 hours. For activation, THP-1 monocytes or macrophages were treated with 100 ng/mL LPS for 24 hours with or without MSC-EVs in 96-well plates in ATCC-formulated RPMI 1640 supplemented with 0.1% BSA. Supernatants were collected and analyzed for their tumor necrosis factor a (TNFa) or interleukin-10 (IL-10) content using enzyme-linked immunosorbent assay (BioLegend). Culture and activation of primary human umbilical vein endothelial cells (HUVECs) HUVECs (Lonza) were cultured on 0.1% gelatin substrate (Sigma) in fully supplemented EGM-2 medium (Lonza) which contains 2% serum. Culture medium was refreshed every 2-3 days; HUVECs were subcultured upon reaching 90-95% confluency. For activation, HUVECs were transferred to gelatin-coated 96-well plates; incubated for 24 hours in complete medium; then in medium with reduced serum (0.5%) for 24 hours; and finally treated with TNFx (PeproTech) with or without MSC-EVs at the reported concentrations for 16 hours. To measure surface expression of intercellular adhesion molecule-1 (ICAM-1) or vascular cell adhesion molecule-1 (VCAM-1), HUVECs were fixed, stained with fluorescently labeled antibodies, and quantitatively imaged using high-content screening (Wyss Institute, Harvard University). Supernatants were collected and analyzed for their interleukin-8 (IL-8) content using enzyme-linked immunosorbent assay (BioLegend).

Results

MSCEVs from different donors are structurally similar but differently susceptible to freeze-thaw damage. Consistent with published findings [133], isolated MSC-EVs, whether fresh or previously frozen, were morphologically heterogeneous (Figure 17) and polydisperse (Figure 18). Differences between donors in these respects were negligible. After freezing and thawing, MSC-EVs decreased in particle concentration and increased in protein-to-particle ratio (Figure 19). Because bulk protein content remained unchanged (data not shown), some EVs likely became undetectable. Possible reasons include shrinkage of EVs below limit of detection, and degradation of EVs into particles below limit of detection. Fusion and aggregation were unlikely since size distribution did not change significantly. In subsequent experiments, particle and protein concentrations were determined after thawing frozen MSC-EVs, right before the EVs were subject to functional assays. MSCEVs from most donors do not inhibit DB-induced PBMC proliferation. To mimic inflammation, we attempted to induce PBMC proliferation, and could do so with DB but not LPS (Figure 20). Several factors may explain why LPS failed. First, LPS appears to be much stronger at stimulating cytokine release than proliferation [134,135]. Second, studies that demonstrated LPS-induced PBMC proliferation used serum-containing RPMI 1640 to culture PBMCs [134,135,1361, whereas we used serum-free X-VIVO 15 so that MSC-EVs can interact maximally with PBMCs and minimally with serum components. Serum contains LPS-binding protein which potentiates the binding of LPS to monocytes and macrophages [1371. Moreover, immune cells have been reported to behave differently in X-VIVO 15 versus RPMI 1640, even when both contain serum [138,1391. Using DB, we could induce majority of PBMCs to proliferate, suggesting that our PBMC samples were mostly T lymphocytes (CD28*/CD3*). This is consistent with our LPS results since LPS is mitogenic to NK cells but not T and B lymphocytes [134].

28 A. Unilarnellar EVs B. Multilamellar EVs C. Elongated EVs

Figure 17. MSC-EVs are morphologically heterogeneous. Whether fresh or previously frozen, (A) most MSC-EVs are unilamellar and spherical. (B) A distinct subpopulation are vesicles within spherical vesicles. (C) A rarer subpopulation consists of elongated vesicles.

Size distribution of MSC-EVs #24539 Size distribution of MSC-EVs #24655 140 200 -Fresh 180 -Fresh 120 -Frozen 2 160 -Frozen E E 100 e 140

120 E 80 100 60 r 80

40 60

40 0 20 20 0 0 0 100 200 300 400 500 600 0 100 200 300 40 0 500 600 Particle diameter (nm) Particle diameter (nm)

Figure 18. MSC-EVs are polydisperse. Whether fresh or previously frozen, MSC-EVs have a modal size between 100 and 200 nm that varies slightly between donors.

Particle concentration Protein-to-particle of MSC-EVs ratio of MSC-EVs 14 4.0

12 3.5 3.0 10 TM 2.5 P8 2.0 6 a. 1.5

1.0 02 C -+-24539 0.5 -+-24539 -+-24655 -+-24655 0 0.0 Fresh Frozen Fresh Frozen

Figure 19. MSC-EVs from different donors are differently susceptible to freeze-thaw damage. A decrease in particle concentration concomitant with an increase in protein-particle ratio is consistent with shrinkage or disintegration of EVs forming particles below limit of detection, or degradation of EVs. EVs from donor 24655 suffered significantly more damage than EVs from donor 24539. (n=3 per condition)

29 A. Treatment of PBMCs with lipopolysaccharide (LPS) Day 1 post-stimulation Day 2 post-stimulation Day 4 post-stimulation PBS LPS 10 ng/mL LPS 100 ng/mL - LPS 1 igImL

CFSE fluorescence CFSE fluorescence CFSE fluorescence

B. Treatment of PBMCs with anti-CD28/CD3 Dynabeads (DB) Day 1 post-stimulation Day 2 post-stimulation Day 4 post-stimulation PBS IDB:IOPBMC IDB:IPBMC

012 3 Mo4 w6 aO *7.2 a1 m2 ,3 ,04 J5 06 a72 a. j. 04 j 6 W7.2 CFSE fluorescence CFSE fluorescence CFSE fluorescence

Figure 20. Anti-CD28/CD3 Dynabeads (DB) but not lipopolysaccharide (LPS) induce PBMC proliferation. CFSE-stained PBMCs were plated at Ix 106 cells/mL in X-VIVO 15. (A) LPS up to I pg/mL did not induce PBMC proliferation, but (B) DB at Ix 10' or Ix 106 particles/mL did.

Inhibition of DB-induced PBMC proliferation by MSC-EVs from different donors 80% *No DB 70% DB only * DB+EV #22867 0 60% m 60% * DB+EV #24539 50% * DB+EV #24655 0 * DB+EV #24935 U40% 0% a DB+EV #03915 30%

S20%

C-&10% 0% 1DB:10PBMC 1DB:2PBMC Ratio of anti-CD28/CD3 Dynabeads (DB) to PBMCs Figure 21. MSC-EVs from only 1 out of 5 donors inhibit DB-induced PBMC proliferation. CFSE- stained PBMCs at Ix 106 cells/mL were treated with DB at the indicated ratios for 4 days, with or without MSC-EVs at Ix 108 particles/mL. (n=3 per condition)

30 We proceeded to use DB-induced PBMC proliferation as a cell-based assay to detect inter-donor variability in MSC-EV potency. A previous study using only one source of MSCs reported the inability of MSC-EVs to inhibit DB-induced PBMC proliferation [129]. We tested MSC-EVs from 5 donors, and one but not the rest could inhibit DB-induced PBMC proliferation (Figure 21). In conclusion, counteracting CD28/CD3 activation of T lymphocytes is unlikely a key mechanism of how MSC-EVs alleviate inflammation, and hence is not an accurate predictor of MSC-EV potency. THP-1 monocytes and macrophages collectively can measure the consistency of inflammatory action by MSC-EVs. We used THP-1 cells to simulate the monocyte-macrophage axis of the immune system. To resting THP- 1 monocytes, Ix 108 particles/mL EVs from hTERT-MSCs were as effective as 100 ng/mL LPS at inducing TNFa secretion (Figure 22). A milder pro-inflammatory action was observed for primary MSC-EVs from only 1 out of 3 donors. None of the MSC-EVs could reduce counteract TNFu secretion by LPS-activated THP-l monocytes. Our results disagree with a previous study which, utilizing only one source of MSCs, showed that MSC- EVs suppressed expression of pro-inflammatory cytokines and promoted expression of anti-inflammatory cytokines by modified THP-1 cells [124]. Although we measured cytokine release rather than transcriptional activity, we proceeded to investigate if the published observations apply to THP-1 macrophages instead of monocytes, in addition to testing MSC-EVs from multiple donors. Using PMA, we successfully differentiated non-adherent THP-1 monocytes into adherent macrophages (Figure 23).

TNFa release from THP-1 cell line due to MSC-EVs from different donors 2,000 " No EV 0. " EV #00043 1,500 EV #00055 EEV #00081 E .EV hTERT C 1,000 0 0 E500

0 ng/mL 100 ng/mL [Lipopolysaccharide] in initial media

Figure 22. MSC-EVs from some donors are pro-inflammatory to resting THP-1 monocytes but none are anti-inflammatory to LPS-activated THP-1 monocytes. In the absence of LPS, hTERT-MSC-EVs and primary MSC-EVs from 1 of 3 donors induced TNFO secretion, a pro-inflammatory response, within 24 hours. When LPS was present, none of the MSC-EVs could counteract its pro-inflammatory effect. MSC- EV concentration was kept at 1x 108 particles/mL. (n=3 per condition)

31 PMA induction of macrophage-like phenotype in THP-1 monocytes Day 2 post-treatment Day 2 post-treatment Bright-field micrograph of -PMA PMA-treated THP-1 cells -PMA +PMA

CD 11b expression CD14 expression Figure 23. THP-1 monocytes adopt a macrophage-like phenotype after PMA treatment. Within 48 hours of treatment with 25 nM PMA, THP-1 monocytes, which are non-adherent, began to adhere and upregulate their surface expression of CD11 b and CD 14-markers of macrophages.

TNFa release from PMA-treated THP-1 IL-10 release from PMA-treated THP-1 due to MSC-EVs from different donors due to MSC-EVs from different donors 900 80 2j mNo EV " No EV FL 800 mEV #00043 F70 " EV #00043 700 EV #00055 60 EV #00055 600 UEV #00081 0 *EV #00081 E EV hTERT E 50 * EV hTERT 9500V 40 0400 0 30 0 300 3 U 0 .S 200 . 20 L 100 - 10 nn 0 ng/mL 100 ng/mL 0 ng/mL 100 ng/mL [Lipopolysaccharide] In initial media [Lipopolysaccharide] In Initial media

Figure 24. MSC-EVs are anti-inflammatory to LPS-activated THP-1 macrophages. To resting THP-1 macrophages, all MSC-EVs did not induce secretion of TNFa or IL-10, except for hTERT-MSC-EVs. In the presence of LPS, however, all MSC-EVs suppressed TNFo secretion; and EVs from hTERT-MSCs but not primary MSCs induced IL-10 secretion. MSC-EV concentration was kept at Ix 10' particles/mL. (n=3 per condition)

In contrast to THP-1 monocytes, resting THP-1 macrophages did not secrete TNFax in response to MSC- EVs from any donor, and all MSC-EVs could suppress TNFo secretion by LPS-activated THP-1 macrophages, indicating a strong anti-inflammatory action (Figure 24). We further examined the secretion of anti- inflammatory cytokines, using IL-10 as a marker. EVs from hTERT-MSCs but not primary MSCs could induce IL-10 secretion in both resting and LPS-activated THP-1 macrophages. These results are in better agreement with the previous study. Collectively, the use of both THP-1 monocytes and macrophages can indicate whether MSC-EVs from a given source is consistently pro- or anti-inflammatory (Table 2).

TNFaactivated HUVECs can distinguish anti-inflammatory actions of MSC-EVs from different donors.

The endothelium responds to inflammation by upregulating surface adhesion molecules (e.g. ICAM-1, VCAM-1) and secreting pro-inflammatory cytokines (e.g. IL-8) to facilitate the recruitment of circulating immune cells to inflamed tissues. To minimize the consumption of EVs, we used high-content screening to quantitatively image the expression of ICAM-1 and VCAM-1 by HUVECs in response to TNFa treatment. Because VCAM-1 expression showed a larger dynamic range and hence higher sensitivity (Figure 25), we used VCAM-l expression to compare MSC-EV effects between different donors.

32 A. ICAM-1 expression by HUVECs after TNFa treatment Fold change in median fluorescence of TNFa-treated versus untreated cells 15 -

-2 -1 0 1 2 4 log[TNFa (ng/mL)]

B. VCAM-1 expression by HUVECs after TNFa treatment Fold change in median fluorescence - versus untreated cells of TNFa-treateda

.2 -1 0 1 2 3 4 log[TNFa (ng/mL)]

Figure 25. VCAM-1 expression is more sensitive than ICAM-1 expression in response to TNFa treatment. High-content screening was used to measure expression of (A) ICAM-I and (B) VCAM-1 as an inflammatory response of HUVECs to TNFu treatment.

VCAM-1 expression by HUVECs IL-8 release by HUVECs due to MSC-EVs from different donors due to MSC-EVs from different donors 4,000 I 9,000 a No EV "No EV 2 8,000 . 3,500 nEV#00043 E "EV#00043 7000 S 3,000 EV #00055 SEV #00055 *EV #00081 o 6,000 *EV #00081 u2,500 I. . EV hTERT 2 EVhTERT r 5,000 th 2,000 0 E 4,000 0 1,500 _ 3,000

1,000 2,000 500 1,000

0 0 0 ng/mL II1 ng/mL 0 ng/mL 1 nyimL 10 rig/mL [TNFaI in initial media [TNFa] in initial media

Figure 26. TNFa-activated HUVECs can distinguish anti-inflammatory actions of MSC-EVs from different donors. Following 16 hours of TNFx treatment, HUVECs were imaged by high-content screening to quantify VCAM-1 expression, and supernatants were analyzed for IL-8 content. MSC-EV 8 concentration was kept at Ix 10 particles/mL.

33 Primary MSC-EVs from 2 of 3 donors, and not hTERT-MSC-EVs, could significantly suppress VCAM- 1 expression by TNFcfactivated HUVECs (Figure 26). We further examined the cytokine profile of HUVECs, and found IL-8 secretion to correlate positively with ICAM-1 expression in response to MSC-EVs. None of the MSC-EVs influenced resting HUVECs.

Discussion The availability of potency assays is essential to ensure reproducible performance of manufactured biologics. Variability can originate in starting material or from process perturbations, and be propagated during scale-up [140,141]. In this study, we show that inter-donor variability in MSC-EV potency exists. Previously, inter-donor variability has been reported in only one study, which investigated variability in yield of dendritic cell-derived EVs from individual patients for autologous therapy [1421. Furthermore, to the best of our knowledge, our study is the first to employ a suite of functional assays to indicate inter-donor variability in EV potency. Although we acknowledge that disease-specific potency assays would be more translational, our potency assays interrogate the impact of EVs on inflammation from multiple angles. Consistent with the compositional and structural heterogeneity of EVs, we demonstrated that EVs even from the same donor can influence different cell types differently, supporting complex mechanisms of action in EV therapy. To predict therapeutic efficacy in vivo, collective interpretation of functional data from a suite of potency assays may be necessary to make decisions for lot release. Based on our observations in this study, we envision a potential scoring system (Box 3) to indicate the overall inflammatory action of MSC-EVs (Table 2). The overall scores further validate inter-donor variability.

Box 3. Potential scoring system to determine overall inflammatory action from a suite of potency data.

1. Given no stimulus (e.g. resting cells), induction of a pro-inflammatory ('pro') phenotype indicates 'pro' action, while induction of an anti-inflammatory ('anti') phenotype indicates 'anti' action; otherwise data is inconclusive. 2. Given a 'pro' stimulus (e.g. activated cells), suppression of a 'pro' phenotype or induction of an 'anti' phenotype indicates 'anti' action; otherwise data is inconclusive. 3. Conclusive events are scored 1 while inconclusive events are scored 0. Overall therapeutic score is the number of 'anti' events in excess of 'pro' events.

Table 2. Example of scoring overall inflammatory action of MSC-EVs for lot release.

Assay Outcome #00043 #00055 #00081 hTERT

Resting monocytes Induce TNFa Not pro Pro Not pro Pro

Activated monocytes Suppress TNFa Not anti Not anti Not anti Not anti

Resting macrophages Induce TNFQ Not pro Not pro Not pro Not pro Induce IL-10 Not anti Not anti Not anti Anti

Activated macrophages Suppress TNFu Anti Anti Anti Anti Induce IL-10 Not anti Not anti Not anti Anti

Resting endothelium Induce VCAM-1 Not pro Not pro Not pro Not pro Induce IL-8 Not pro Not pro Not pro Not pro

Activated endothelium Suppress VCAM-1 Anti Not anti Anti Not anti Suppress IL-8 Anti Not anti Not anti Not anti

Pro-inflammatory incidence (P): 0 1 0 1 Anti-inflammatory incidence (A): 3 1 2 3 Therapeutic score (A-P): 3 0 2 2

34 An ideal design for potency assays is low sample consumption. Particularly for EVs where productivity is much lower than other classes of biologics (Chapter 2), limiting product consumption for quality testing is imperative to minimize cost and maximize production. Therefore, we restricted assay formats to 96-well plates, and employed surrogate analytical techniques such as high-content screening in lieu of flow cytometry. Consequently, our assays are amenable to measuring dose response and high-throughput testing of EVs derived from different donors, culture conditions, isolation techniques, and investigative methods that perturb EV potency.

Future Work To validate our potency assays, dose-response curves indicating half-maximal effective and inhibitory concentrations should be generated. Not only will more EV concentrations be tested, more suitable positive controls may need to be established as potency standards. Examples include concanavalin A or phytohaemagglutinin for PBMC proliferation, and IL-I or interferon y for activating monocytes, macrophages, or endothelial cells, up to saturating concentrations. Control EVs may be tricky to use until EV standards of consistent functional quality are developed, especially given the variability between EV sources demonstrated by our study. However, conditioned media from the parent cells may be useful to indicate if the EVs are derived from truly therapeutic cells. Profiling of RNA and protein content may reveal correlations between certain EV components with EV potency. Further comparison of molecular composition and potency between EVs and their parent cells may uncover how parent cells impact the potency of their EVs.

35 CHAPTER FivE

MODULATING SURFACE COMPOSITION OF VESICLES

Summary " A ligand construct bearing distinct intracellular and extracellular domains can be genetically incorporated into vesicles and remain functionally intact.

" Purely mechanical methods such as homogenization and fractionation can reduce molecular diversity of the vesicle surface by selectively enriching for some membrane proteins. " Reducing endogenous proteins on the surface may improve the performance of exogenous ligands.

Introduction

Recognized as 'Nature's drug delivery system' 150,52,143], extracellular vesicles (EVs) are known to transfer membrane and cytosolic components between distant cells, including genetic material [13,144,145] and surface molecules [42,146] that impart new functions or restore lost functions to the recipient cell. Upon binding to a cell surface, an EV can trigger receptor-mediated signaling, or even fuse with the cell [20,30]. Because of their ability to co-deliver multiple cellular signals, EVs can dramatically shift cellular phenotypes, sometimes leading to stable epigenetic changes that last for at least several weeks [41,42,43,147,148]. As EVs naturally prevail in vivo and transmit information across physiological barriers, interest in using EVs to deliver exogenous biomaterial is rapidly emerging [20,50,51,52]. Key to how an EV functions is its heterogeneous surface, a protein-laden phospholipid bilayer. Proteins that commonly found on the EV surface include major histocompatibility complexes (antigen presentation), integrins (adhesion), flotillins (lipid rafts), and tetraspanins (membrane fusion), all of which are multimeric transmembrane assemblies with distinct functions [31,143]. This is a unique advantage of EVs over synthetic drug carriers, since artificially incorporating such assemblies in the native polarity while preserving function and fluidity is highly challenging [149,150,1511. However, surface heterogeneity complicates prediction and control of EV fate and function. Addition of targeting ligands onto the EV surface, while able to increase retention at target sites, is unable to reduce off-target binding [152,153,154,155, since exogenous ligands must compete with endogenous surface proteins to determine the overall target. Likewise, exogenous ligands may interfere with signaling and fusion events that require functional endogenous surface proteins. The demand for more control over the drug delivery performance of EVs has given rise to a spectrum of artificial cell-derived vesicles (CDVs), ranging from vesicles formulated entirely from cell membranes by chemically or mechanically disrupting cells, to hybrid vesicles that comprise both cellular and synthetic materials [143,149,151,156]. CDVs have been successfully utilized to deliver small molecules [157,158,159], peptides [1601, mRNA [1611, siRNA [1621, and plasmid DNA [1631, indicating that CDVs may replicate the ability of EVs to fuse with the recipient cell. Even without incorporating exogenous cargo, CDVs can elicit complex biological effects including proliferation [164] and differentiation [165]. Importantly, surface molecules presented by CDVs can be controlled by selecting for cell surface fragments from which to formulate the CDVs [166,167]. A key logistical benefit of using CDVs compared to EVs is significantly higher yield from parent cells [157], which facilitates the study of surface molecules in their native environment [168]. Meanwhile, adding exogenous ligands to the vesicle surface may interfere with endogenous surface proteins. Direct attachment to endogenous surface proteins via click chemistry [169], noncovalent binding [155,170], or recombinant technology [152,153,170,171] risks loss of protein activity or availability that may be vital to vesicle function, or relies on 'markers' that are now found to be specific to only a subset of vesicles [22,23,24,25]. In a recent study, cells from which vesicles would be derived were functionalized by fusion with

36 ligand-bearing liposomes [1721. While this method spares endogenous surface proteins from modification, it is limited to ligands that are amenable to bioorthogonal chemistry. In this study, our goal is to modulate the surface content of CDVs so that the exogenous ligand can dictate the overall fate and function of the CDV. We first assess how the method of formulating CDVs may differently select for or even reduce endogenous surface proteins. Next, we investigate the feasibility of genetically inserting a transmembrane protein construct that is fully exogenous (i.e. not conjugated to any endogenous protein) and whose extracellular and intracellular domains can bear functional proteins. Combining this two approaches may increase the effectiveness of the exogenous ligand.

Methods

Preparationof raftosomes and nanoghosts HEK293 cells (ATCC) were cultured in Dulbecco's modified Eagle's medium (Gibco) supplemented with 10% fetal bovine serum (Atlanta Biologicals) and penicillin/streptomycin (Gibco), and subcultured upon reaching 90% confluency. To prepare CDVs, the cells were rinsed twiced in calcium- and magnesium- free phosphate-buffered saline (PBS-/-), and then detached by mild scraping in 500 mM sodium bicarbonate solution containing cOmplete Protease Inhibitor Cocktail (Roche). Enzymatic detachment was avoided to preserve surface proteins. Thereafter, the detached cells underwent one of two treatments. In one treatment, detached cells were homogenized by 150 strokes in a Dounce tissue grinder on ice. After being loaded on top of a discontinuous linear gradient from 20% to 15% to 10% of OptiPrep density gradient medium (Sigma), the suspension was centrifuged at 5 2 ,000g for 90 min at 4*C. Fractions were analyzed by Western blot; those enriched in membrane proteins were resuspended in PBS-/-, extruded through a 0.2-pm membrane, and rinsed by centrifugation at 100,000g for 2 hours at 4*C. This method is adapted from published protocols for purifying lipid rafts and caveolae [173,174,1751; the resultant vesicles are termed raftosomes. In the other treatment, detached cells were enucleated by mild homogenization (<20 strokes) in a Dounce tissue grinder on ice to weaken the plasma membrane, and then by sucrose cushion centrifugation at 4*C to remove the denser nucleus from the rest of the cell. Removal of nuclear material was confirmed by co-staining cell membranes with DiI (Molecular Probes) and nuclei with Hoechst 33342 (Molecular Probes) followed by inspection under a fluorescence microscope. Material above the sucrose cushion was resuspended in PBS-/-, extruded through a 0.2-pm membrane, and rinsed by centrifugation at 100,000g for 2 hours at 4'C. This method is adapted from published protocols for preserving the majority of surface proteins when artificially reducing cells to vesicles [160,163]; the resultant vesicles are termed nanoghosts. Samples of raftosomes and nanoghosts was submitted to Swanson Biotechnology Center at Koch Institute for Integrative Cancer Research at MIT for transmission cryoelectron microscopy (cryoTEM) to examine morphology. Size distribution was measured by dynamic light scattering. To analyze membrane proteins, samples were submitted in triplicate to Swanson Biotechnology Center for mass spectrometry. From the peptide sequences, membrane proteins and their subcellular locations were identified using UniProt. Insertion of ligand construct into vesicle surface Plasmids were a gift from Alan S. Kopin, M.D. of Tufts University. The construct consisted of an intracellular domain bearing the green fluorescent protein (GFP), a transmembrane domain derived from the human herpes simplex virus type 1 glycoprotein C, and an extracellular domain bearing a Myc tag and terminating with exendin-4 (EXE4) [1761. EXE4 is a well-established agonist of the glucagon-like peptide 1 receptor (GLP1R).

37 Cells were transfected using Lipofectamine (Invitrogen) according to manufacturer's instructions. Expression and subcellular localization of GFP was verified by confocal microscopy. After 48 hours, transfected cells were converted to raftosomes. Co-localization of GFP and myc-tag with membrane proteins in the raftosomes was verified by Western blot. To verify that EXE4 remains functional on the raftosomes, reporter cells that transduce GLP1R activity to luciferase expression [1771-also a gift from Alan S. Kopin-were treated with EXE4-raftosomes and, as a control, soluble EXE4 (American Peptide Company). The specificity of EXE4-raftosomes or soluble EXE4 at activating GLP 1 R was investigated by adding EXE9-39 (AnaSpec), a truncated form of EXE4 that antagonizes GLP1R, in the presence of EXE4-raftosomes or soluble EXE4.

Results Raftosomes and nanoghosts are vesicles differently enriched in membrane proteins. Density gradient centrifugation divided homogenized cells into 10 fractions with increasing density from top to bottom. The top 5 fractions were enriched in flotillin-1, a membrane protein, but devoid of protein kinase C alpha (PKCct), a cytosolic protein (Figure 27A); hence membrane material was sequestered into these fractions. After extrusion and rinsing, raftosomes were collected from these fractions, and appeared as vesicular structures under cryoTEM (Figure 27B) with a mean size of 191.6 nm under dynamic light scattering. To prepare nanoghosts, cells were first enucleated to form ghosts. Mild homogenization followed by sucrose cushion centrifugation separated nuclear material from the rest of the cell (Figure 27C). The resultant ghosts were extruded and rinsed to form nanoghosts which appeared as vesicular structures under cryoTEM (Figure 27D) with a similar mean size of 210.7 nm under dynamic light scattering. Proteomic analysis through mass spectrometry revealed that <5% of detected peptide fragments in both raftosomes and nanoghosts were associated with secreted proteins, indicating that the samples contained minimal non-vesicular material. Presence of intracellular proteins indicates intact outer membranes. Peptide fragments associated with membrane proteins were more abundant in raftosomes, while peptide fragments associated with non-membrane proteins were more abundant in nanoghosts (Figure 28). This correlates with the difference in preparation, since raftosome preparation subjected cells to more vigorous disruption and finer fractionation which could better separate membrane and non-membrane components. In particular, raftosome preparation selectively enriched for membrane proteins from mitochondria and the nucleus, and depleted membrane proteins from endolysosomes, peroxisomes, and the Golgi apparatus (Figure 29). Although we could not directly determine which membrane proteins were presented on the surface of the vesicles, majority of the vesicles (Figure 27B,D) were unilamellar, suggesting that most membrane proteins were localized on the surface. Hence, the degree of homogenization and fractionation suffices to modulate the surface content of CDVs. Ligand construct colocalizes with membrane proteins and remains functional on raftosomes. Since nanoghosts were not fractionated and were expected to preserve the majority of membrane proteins [160,1631, for the rest of the study we focused on raftosomes which are selectively enriched in some membrane proteins (Figure 28). Specifically, we examine if a genetically inserted, fully exogenous, transmembrane ligand construct can still be presented on raftosomes despite the selective enrichment. In transfected cells, green fluorescence was largely localized at the plasma membrane (Figure 30A). This indicates that GFP as the intracellular domain of the construct was functional, and that the transmembrane domain was properly recognized. After homogenization and fractionation, both intracellular (GFP) and extracellular (Myc) domains co-localized in fractions where membrane proteins were sequestered and from which raftosomes were derived (Figure 30B). This is in spite of the fact that Myc is naturally a non-membrane protein in the nuclear compartment.

38 A. Fractions containing raftosomes Fractions from top 1 2 3 4 5 6 7 8 9 10 kDa PKCa 9 55 Flotillin-1 43 i1* .

C. Enucleation of cells to form ghosts

Figure 27. Raftosomes and nanoghosts are mostly unilamellar vesicles. (A) Fractionation of homogenized cells separated membrane proteins (e.g. flotillin-1) from cytosolic proteins (e.g. PKCx). After extrusion and rinsing, fractions containing membrane proteins were harvested as raftosomes, which (B) appeared mostly as uhilamellar vesicles under cryoTEM. (C) Mild homogenization followed by sucrose cushion centrifugation removed nuclear material (blue) from cell membranes (red), forming ghosts, which were subsequently extruded to form nanoghosts. (D) Nanoghosts also appeared as unilamnllar vesicles.

Peptides associated with membrane versus non-membrane proteins Membrane Non-membrane I

0.8

U .. U0 = 0.4 0 0) (a 0 W0.2.-

0 -

o 0 0 *00 -0.2

00-0.2 L

-0. 0. -0.6 to 0 (A -0.8. 0t

-1 - Peptide identified by mass spectrometry, sorted by abundance

Figure 28. Raftosomes and nanoghosts are differently enriched in membrane and non-membrane proteins. Peptides with unique sequences detected under mass spectrometry were identified as fragments of membrane or non-membrane proteins. For each peptide, the ratio of its abundance in raftosomes to nanoghosts was calculated. Since raftosomes contained more membrane proteins than nanoghosts, differences between their preparations indicate methods that can modulate the surface content of vesicles.

39 Membrane proteins associated with Membrane proteins associated with plasma membrane endoplasmic reticulum 08 0.4-

06 0.3 004 10.2

02 0.1 Le 0 -02

-0.1

-0.2 e 0 -0.3 I ~04 Membrane proteins associated with Membrane proteins associated with endolysosomes/peroxisomes Golgi apparatus 04

02 j0 2 0 L ON W 0 K S 04

-06 C -08

-0.8 I I -08 11 Membrane proteins associated with Membrane proteins associated with mitochondria nucleus L1 08 0.8 - 06 1 08

8 g 04 0.4 9-02 jIg 02 0,8 02 1.0 08 -0.6 .g0

-0.8 -02

-0 4

Figure 29. Membrane proteins in raftosomes and nanoghosts are differentially associated with subcellular organelles. Raftosomes were enriched in components of mitochondria and the nucleus, while nanoghosts were enriched in components of endolysosomes, peroxisomes, and the Golgi apparatus.

40 B. Fractions containing construct Fractions from top 1 2 3 4 5 6 7 8 9 10 kDa

Myc 43 6,0 IA U.n

GFP 27

Figure 30. Ligand construct remains structurally intact in raftosomes. (A) Green fluorescence, afforded by functional GFP in the intracellular domain of the ligand construct, was largely sequestered to the plasma membrane. (B) Following homogenization and fractionation, both intracellular (GFP) and extracellular domains of the construct co-localized in fractions from which raftosomes were derived.

[EXE4-raftosomes] [soluble EXE4] A (ng proteinlmL) B (ng proteinmL) S120- ,-i I 0 . 7 120 0.001 0.012 0.123 1.23 12.3 0.0002 0.002 0.02 0.2 2.0 i-100 0 EC : -250 FU/mL EC5: -5.0 pM -

40- ~40 'X0(

0J No 0 1 2 3 4 No -14 -3 -12 -11 40 ligand [EXE4-raftosomea] ligand [soluble EXE4] log(FU/mL) log(M)

[EXE4-raftosomsj] [soluble EXE4] C Q 2.7 log(FUlmL) D -11 log(M) 120- 120- -b0 1004 100, 80

LIT * o* .I 1Z 40- IC50: ~-0 nM IC50: -47 nM 20 20- 0 0- -i2 i o 4i2 -10 4 4 [soluble EXE,.3 ] [soluble EXE .39] log(M) 109(M) Figure 31. Extracellular domain of ligand construct is functional and properly oriented. (A) EXE4- raftosomes, measured in fluorescence units (FU), activated GLPiR in luciferase reporter cells and achieved up to 80% of the maximum activity afforded by (B) soluble EXE4. In the presence of (C) EXE4- raftosomes or (D) soluble EXE4 at a fixed, saturating concentration, the antagonist EXE9 3 9 inhibited

EXE4 activity in a concentration-dependent manner. The half-maximal effective concentrations (EC5o) of EXE4-raftosomes and soluble EXE4 were 250 FU/mL and 5.0 pM respectively, while the half-maximal inhibitory concentration (IC5o) of EXE 93 9 was 60 nM against EXE4-raftosomes and 47 nM against soluble EXE4.

41 To further validate that the construct exhibited the intended structure, polarity, and localization, we probed for EXE4 function in the raftosomes. EXE4-raftosomes were able to activate GLP1R and achieve up to 80% of the maximum activity afforded by soluble EXE4 (Figure 31A-B). At saturating concentrations, both EXE4-raftosomes and soluble EXE4 were inhibited by similar concentrations of the GLPIR antagonist EXE9. 39 (Figure 3 1C-D), strongly indicating that EXE4-raftosomes and soluble EXE4 interacted with the same GLP 1Rs via the same binding mechanism. Hence, the weaker action (Figure 31A-B) of EXE4-raftosomes is unlikely due to the incomplete ability of vesicle-bound EXE4 to fit snugly into the active sites of GLPIR. Rather, an interaction between a GLP1R and a vesicle-bound EXE4 may obscure neighboring GLP1Rs from other vesicle-bound EXE4, whereas soluble EXE4 can better access GLP1Rs.

Discussion In this study, we restricted our methodology to purely mechanical methods to generate CDVs without the addition of synthetic agents. This preserves not only the structure but also the function of endogenous proteins, as shown in previous studies [157,159,160,161,162,163,164,165,168]. Chemical methods for formulating CDVs (e.g. use of synthetic amphiphiles) risk protein denaturation; the possibility of preserving the binding ability of endogenous proteins by careful titration and incorporation of phospholipids and cholesterol was only very recently reported [167]. Preserving the quality of transmembrane proteins is particularly essential since the delivery of transmembrane proteins is a unique advantage of CDVs (and EVs) that is challenging for other drug carriers to replicate. Modulation of the vesicle surface is a tricky balance between retention of endogenous molecules with desired functions and addition of exogenous molecules at sufficient quantities that can perturb the overall fate and function of the vesicle. Towards this end, we show that retention of endogenous proteins in CDVs may be tuned by varying the degree of homogenization and fractionation. We also show that a ligand construct bearing functional intracellular and extracellular domains can be genetically incorporated onto the raftosome surface; whether a similar effect can be seen in nanoghosts remains to be determined. Constructs of the same design have been previously demonstrated to be tunable: increasing plasmid concentration during transfection will increase the number and activity of ligands on the cell surface [176,178,1791. Furthermore, constructs bearing different ligands can be co-expressed in the same cell, and even interact cooperatively to function like multimeric proteins [180]. While there may be ways to independently tune endogenous and exogenous molecules on the vesicle surface, several technical limitations impede the assessment of overall vesicle function. Given that both raftosomes and nanoghosts are unilamellar vesicles and yet harbor membrane proteins associated with intracellular organelles, these CDVs must have originated from different organelles and are likely to differ significantly in surface content even between two vesicles from the same preparation. Developing a single pharmacological assay that measures the endogenous function of all vesicles is therefore highly challenging. Likewise, whether the exogenous ligand is presented on all vesicles is difficult to determine, requiring single- vesicle analysis such as specialized flow cytometry [74,751. Hence, new tools will be necessary to investigate the competition between exogenous and endogenous molecules on the same vesicle surface. This should be a prerequisite for testing vesicles in vivo, because without first understanding the surface heterogeneity of the vesicles, predicting and controlling the fate of the vesicles in the body-where enormous possibilities of surface interactions abound-will be near impossible. To the best of our knowledge, only one group has applied a similar genetic construct to modify the surface of EVs for targeted delivery, but whether ligand density on the EV surface can be tuned remains unknown [181]. Their ligand bore an extracellular domain that binds the epidermal growth factor receptor, and a transmembrane domain derived from platelet-derived growth factor receptor. An intracellular domain was omitted, which could have been beneficial for tracking EVs and fusion events. Alike our ligand design,

42 no rational attempt was made to direct their construct specifically to EVs, and yet their ligand was still expressed on EVs. Hence, we postulate that our ligand technology will be applicable to EVs.

Future Work In this work, CDVs were derived using only 2 conditions of fractionation and homogenization. More conditions should be tested, with or without sucrose cushion or density gradient centrifugation, by varying the number of strokes with the Dounce tissue grinder to reveal the degree of control over the surface content of CDVs. A digital tissue homogenizer (e.g. Omni THQ) may provide finer control, but agitation parameters would have to be re-optimized because the mechanism of homogenization is different. Following transfection with the ligand construct, expression and function of the ligand construct should be examined on EVs in addition to the CDVs. To obtain EVs purely from transfected cells, the cells should be cultured in particle-free media (see Chapter 3), and persistence of particulate transfection reagents in the final EV or CDV preparation should be examined. Kinetics and stability of ligand expression on EVs or CDVs must be studied to determine if differences in ligand density between EVs and CDVs are time- dependent. Plasmid load during transfection may be varied to tune ligand density, which can be estimated by normalizing total ligand content (indicated by a reporter intracellular domain such as GFP) measured with immunosorbent assays to particle content measured with nanoparticle tracking analysis. These measurements may also explain the discrepancy in pharmacological activity between EXE-4 raftosomes and soluble EXE4 (Figure 3 lA-B). Once our ligand technology is shown to apply to both CDVs or EVs, its utility in targeting CDVs or EVs by masking endogenous proteins may be investigated. The most abundant endogenous protein on the surface may first be identified with mass spectrometry. Then a ligand bearing an extracellular domain that neutralizes the most abundant protein can be designed and functionally tested. Masking of a receptor may reduce binding of the parent cells on a substrate coated with the counter-ligand [182,183,184,185,186]. Such a substrate may also be able to detect a decrease in the number of substrate-bound CDVs or EVs following modification with a masking ligand, but nanoparticle imaging technologies such as hyperspectral imaging [187,188] or surface plasmon resonance imaging [189,190,191] may be necessary to detect CDVs or EVs. If masking is effective in vitro, in vivo biodistribution of unmodified versus modified CDVs or EVs following systemic administration may be examined. Reduction in off-target accumulation due to endogenous surface proteins will be crucial for control of targeting using exogenous ligands.

43 CHAPTER SIx

ECONOMIC OPTIMIZATION OF EV BIOPROCESSING

Summary . Modeling costs of EV manufacturings reveals ultrafiltration as the cheapest option for harvesting EVs.

. Biological parameters governing cell proliferation and EV accumulation are the strongest cost drivers, followed by labor rate and price of consumables for EV harvest.

Introduction Extracellular vesicles (EVs) are increasingly recognized as a ubiquitous mode of intercellular signaling and as a mechanism of action in some cell therapies [47,481. Although the notion of using EVs for therapy started to gain traction only in the mid-2000s, EV therapy has already been tested in !8 human studies [33,341 and 3 companies are currently developing EV therapeutics [461, indicating a rapidly emerging industry. Key to successful commercialization is the cost-effective scale-up of production. While much research and development initiate in the laboratory, bench-scale technologies are rarely adequate for meeting industrial demands. To scale up, manufacturers embark on a complex and iterative process of identifying and testing new technologies, which is especially costly for producing cell-derived therapeutics [192,1931. Furthermore, since dosage, market size, and hence product demand can vary considerably between disease applications, the most cost-effective solution is case-specific: a bioprocess may be optimal for one case but not another. The practice of using decision-support tools that optimize process, quality, and costs in silico has been remarkably powerful at mitigating scale-up challenges in other industries [194,1951 and is beginning to be adopted for producing cell-derived therapeutics [193,1961. While industry experts can provide overall cost estimates [197], systematic and modular modeling of the bioprocess offers deeper insight into cost structure and allows consideration of case-specific needs and constraints [198,199]. In particular, correlations through sensitivity analysis can pinpoint specific aspects where costs can be dramatically reduced [1981. By accelerating the iterative process of scale-up, decision-support tools can shorten time to market and facilitate earlier patient access to new therapies. The utility of bioprocess modeling has been demonstrated for manufacturing cells and antibodies [193] but not EVs. Providing decision-support tools now for the EV industry may be particularly timely as more companies are developing EV products, and will lay a foundation for benchmarking future innovations. In this study, we aim to build a modular framework which can identify combinations of upstream (cell expansion) and downstream (EV harvest) technologies that minimize costs of goods (COG).

Methods EV demand is parametrized by lot size (EVs/lot) and annual demand (lots/year). One lot is defined as one stage of cell expansion followed by EV harvest. Such a modular framework allows the user to build multi- stage bioprocesses and consider parallel processing without needing to alter the computational algorithm. Our protocol modifies a previously published model for cell manufacturing [198,1991 to accommodate modular modeling; other published models do not specify technologies [2001 or costs [2011. Figure 32 summarizes our approach. The user inputs lot size (Not), annual demand (Nj'J), and biological parameters that characterize cell proliferation (k,) and EV accumulation (k,) (see Chapter 2). We begin by using percent recovery of each EV harvest technology (TH) to determine the number of EVs needed at the end of cell expansion before EV harvest (NE). For a given TH, we compute the number of units (UE) of each cell expansion technology (TE) and the corresponding COG (ZE). By matching the volume output of

_11M 11VWRRMfl each TE to the volume input of the given TH, we compute the number of TH units (UH), COG for EV harvest (ZH), and the total COG (z). Each entry in Tables 4 and 5 defines one unit of cell expansion or EV harvest technology. Recognizing space and time constraints, we impose upper limits to UE and UH WE, UH) for each technology, and a minimum lag of 24 hours between cell seeding and EV harvest. Iterative comparisons of all TE continue for each TH until z is minimized and the cheapest pair of TE and TH is identified. Box 4 lists our assumptions and limitations; Table 3 to Table 5 list the numerical values used.

To compute N: Divide lot size by the percent recovery (yj) of a given TH.

NE. (3) vj Yj

To compute uE: First determine cell number (Nc) using cell density (d,) and culture surface area. While a planar TE has a fixed surface area (ain), a microcarrier-based single-use bioreactor (SUB) has a surface area proportional to the surface area (amc) and volume density (dmc) of the microcarriers, as well as the SUB volume (Visub).

N + amcdmcVisub (4) ub t erofoplna TEan apin azeofor Us The same equation can apply to all TE by setting Visub at zero for planar TE and a at zero SUBs Next, determine the maximum number of EVs (Ngax) that one unit of TE can produce, which is when cells grow from the seeding density (d'n) to the maximum allowable density (dgax) (see Chapter 2).

Nvgjax = ti - Nfmax) = L! (drnin - dimax)(apln + amcdmcViub) (5) VLk (Nc" C C ki m m

Finally, compute the number of units each TE needs to meet the lot size, rounded up to the nearest integer (square brackets without lower horizontal bars denote the ceiling function).

UEi INVx]NI (6)

To compute uH :

Divide the total media consumption of the TE by the maximum sample volume (V) of one TH unit, and round up to the nearest integer. Media consumption is normalized to surface area for planar TE (V7"n) but not for SUBs.

In a pn+V-sub UH,j = UEJ 'i Vs (7)

To compute zE: Three types of costs are considered: consumables, labor, and equipment. The cost of consumables for one lot depends on the price of vessel (pess), media (Pied), and microcarriers (pmc).

C'oflS = UE,[ + pmed a i sub) + pmcdmc isub(8) To determine labor cost per lot, first compute the number of operators (mE,i) needed to handle UEi, taking into account that each operator can manage up to U' units.

mEJ -=I (9)

45 USER For each EV For each cell INPUTS START harvest tech expansion tech

Nv'OfT; = (W) Compute Nj El3o Compute ur.

Lastr TT;C r.

To noCltrtm

Lowest z 2 24h ?

yes

Compute UV.J

no~U1. 5 UH~j ?

yes

no Z, + Zi < Z* ?

yes

Z; = TZ,

Figure 32. Two optimization loops compute costs and identify the cheapest combination of cell expansion and EV harvest technologies for a given demand and lot size. The inner loop optimizes for the cheapest combination for a given EV harvest technology, while the outer loop ensures that every EV harvest technology is considered. See text for explanation of symbols.

46 Box 4. Assumptions and limitations imposed.

1. Cells are not harvested at the end of each lot. Users intending to harvest cells in addition to EVs can refer to studies where various cell harvest technologies are considered [199]. 2. EV harvest occurs only once per lot. In each lot, culture media introduced during seeding needs not be changed, and no new media will be added. EV harvest is the only instance when media is removed. 3. There are no EVs in the culture media when cells are initially seeded. 4. Equations and parameters governing cell proliferation and EV accumulation apply to all technologies equally. 5. To exactly match EV demand, cell culture is terminated when just enough EVs have accumulated. This is equivalent to allowing the cells to reach the maximum allowable density and discarding extra EVs. Hence, a non- integral uE value can be rounded up to the nearest integer. 6. Likewise, to exactly match volumes between cell expansion and EV harvest technologies, EV-free buffer will be added such that a non-integral number of EV harvest units can be rounded up to the nearest integer. Cost of this. EV-free buffer is assumed to be negligible. 7. This study also does not consider costs of pipettes, tips, common buffers, and other consumables or equipment not mentioned herein, as well as COG associated with storage, packaging, and shipping. We perform sensitivity analysis to account for uncertainty in our estimations. 8. All EV harvest technologies generate EVs of sufficient purity and usable concentration for the end application. 9. Wherever possible, disposable or single-use technologies are considered to reflect the shift in industry preference away from hardpiped, steel-based equipment 1202,2031.

Table 3. Key process and cost parameters.

Process parameter Cost parameter

t 2 Seeding density (d" in) 3000 cells/cm Cell culture media (Pmed) $150/L 2 Maximum allowable density (d'ax) 25,000 cells/cm Microcarriers (Pmc) $5/g 2 Surface area of a microcarrier (am) 2930 cm /g Labor rate (Plab) $200/h Microcarrier density (dmc) 6.3 g/L Labor multiplier (fl) 2.2 Biosafety cabinet, BSC (Pbsc) $17,000 BSC capacity (Ubsc) 1 operator/BSC Depreciation period (tdep) 10 years

47 Table 4. Process and cost parameters for cell expansion technologies. Media usage Labor (h)

Cell expansion Surface area Vessel price Planar SUB Operator Seed Collect Biosafety Incubator Incubator Ancillary Ancillary Maximum number per lot technology (cm2) ($) (ml./cm) (mL) capacity time time cabinet? capacity price ($) capacity price ($) of units vess seed qcol Uanc anc a In Vypn vsub Um. U Inc pnc Pi UE,

T-flasks T-175 175 9 0.25 N.A. 10 0.38 0.38 1 100 17,835 N.A. N.A. 80 T-225 225 10 0.25 N.A. 10 0.38 0.38 1 100 17,835 N.A. N.A. 80 T-500 500 15 0.40 N.A. 10 0.38 0.38 1 100 17,835 N.A. N.A. 80 Multi-layers L-1 636 60 0.25 N.A. 1 0.15 0.15 1 60 17,835 N.A. N.A. 80 L-2 1,272 73 0.25 N.A. 1 0.15 0.15 1 60 17,835 N.A. N.A. 80 L-5 3,180 241 0.25 N.A. 1 0.20 0.20 1 24 17,835 N.A. N.A. 80 L-10 6,360 507 0.25 N.A. 1 0.25 0.25 0 12 17,835 N.A. N.A. 80 L-40 25,440 1,265 0.25 N.A. 4 0.08 0.08 0 16 30,000 16 425,000 80 Compact flasks cT 1,720 19 0.33 N.A. 10 0.38 0.38 1 100 17,835 N.A. N.A. 80 Compact multi-layers cL-12 6,000 575 0.22 N.A. 1 0.20 0.20 0 24 17,835 N.A. N.A. 80 cL-36 18,000 1,050 0.22 N.A. 1 0.25 0.25 0 12 17,835 N.A. N.A. 80 cL-120 60,000 3,000 0.20 N.A. 4 0.08 0.08 0 16 30,000 16 425,000 80 Multi-layer bioreactors bL-10 6,360 2,506 0.27 N.A. 1 0.75 0.25 0 6 17,835 1 56,000 80 bL-50 31,800 5,586 0.19 N.A. 1 0.75 0.25 0 4 17,835 1 56,000 80 bL-180 114,480 13,986 0.17 N.A. 1 0.75 0.25 0 2 17,835 1 56,000 80 Hollow-fiber bioreactors HF 21,000 12,000 0.37 N.A. 1 0.20 0.20 0 N.A. N.A. 1 150,000 80 Microcarrier-based single-use bioreactors 20L N.A. 2,000 N.A. 15 4 0.08 0.08 0 N.A. N.A. 1 185,000 8 200L N.A. 4,500 N.A. 150 4 0.08 0.08 0 N.A. N.A. 1 215,000 8 500L N.A. 5,850 N.A. 375 4 0.08 0.08 0 N.A. N.A. 1 320,000 8 1000L N.A. 8,850 N.A. 750 4 0.08 0.08 0 N.A. N.A. 1 425,000 8 2000L N.A. 10,500 N.A. 1,500 4 0.08 0.08 0 N.A. N.A. 1 575,000 8

48 Table 5. Process and cost parameters for EV harvest technologies.

EV harvest technology Percent Unit sample Consumables Operator Labor Ancillary Ancillary Maximum number recovery volume (mL) price ($) capacity time (h) capacity price ($) of units per lot cons U"m anc y3 Vi pryoc Uanc UHi Ultracentrifugation UC Remove cells and cell debris 0.50 240 51 2 0.83 1 86,000 8 1. 6 PS tubes (10 min) -+ 3 ,000g in BC Pellet and rinse EVs 2. 6 PA (15 min) -* 100,000g in UC 3. 6 PA (15 min) -* 100,000g in UC 4. Collect (10 min) Ancillary equipment: 1 BC, 1 UC Polymer-induced precipitation PPT Remove cells and cell debris 0.80 10 46.50 12 1.75 12 18,000 96 1. 1 PS tube (10 min) -+ 3 ,000g in BC Precipitate EVs 2. 1 PS tube + 2 mL EQ-TC (10 min) - 1,500g in BC 3. Aspirate (5 min) -+ 1,500g in BC Remove polymer from EVs 4. 1 microtube + 1 spin column (60 min) -+ 800g in MC 5. 1 microtube (10 min) -> 800g in MC 6. Collect (10 min) Ancillary equipment: 1 BC, 1 MC Size-exclusion chromatography SEC1 Remove cells and cell debris 0.90 0.5 13.50 30 0.42 30 7,000 240 1. 1 microtube (5 min) -> 3 ,000g in MC Elute EVs 2. 1 microtube + 1 qEV column (10 min) 3. Collect (10 min) Ancillary equipment: 1 MC SEC2 Remove cells and cell debris 0.90 240 212 2 0.83 1 21,000 16 1. 6 PS tubes (10 min) -> 3 ,000g in BC Concentrate and rinse EVs 2. 16 DEF1 (20 min) -> 3,500g in BC Elute EVs 3. 1 PS tube + 1 HiPrep column (10 min) -> SEC pump 4. Collect (10 min) Ancillary equipment: 1 BC, 1 SEC pump

49 Table 5 (continued). Process and cost parameters for EV harvest technologies. EV harvest technology Percent Unit sample Consumables Operator Labor Ancillary Ancillary Maximum number recovery volume (mL) price ($) capacity time (h) capacity price ($) of units per lot cons anc tProc Uanc UHi yj V p1 UHmi i Pi Ultrafiltration UFI Remove cells and cell debris 0.50 150 235 4 0.67 1 15,000 16 1. 1 DEF2 (10 min) Concentrate and purify EVs 2. 1 TFF (20 min) -> TFF pump 3. Collect (10 min) Ancillary equipment: 1 TFF pump UF2 Remove cells and cell debris 0.50 3,000 297 4 1.00 1 15,000 16 1. 3 DEF3 (20 min) Concentrateand purify EVs 2. 1 TFF (20 min) -> TFF pump 3. Collect (20 min) Ancillary equipment: 1 TFF pump

PS tubes: polystyrene tubes, $0.50 ea DEF1: Amicon 15-mL centrifugal filters, $11 ea UC: Beckman Coulter L90k ultracentrifuge with 1 rotor, $75,000 ea PA tubes: polyallomer tubes, $4 ea DEF2: EMD Millipore Stericup 250-mL filters, $10 ea BC: Eppendorf 5804R benchtop centrifuge with 1 rotor, $11,000 ea Microtubes, $0.25 ea DEF3: EMD Millipore Stericup 1-L filters, $24 ea MC: Eppendorf 5424R microcentrifuge with 1 rotor, $7,000 ea EQ-TC: ExoQuick for Tissue Culture, $19/mL TFF: Spectrum Labs MidiKros filters, $225 ea SEC pump: GE Healthcare AKTA system, $10,000 ea Spin column, $7 ea TFF pump: Spectrum Labs KrosFlo with pressure monitor, $15,000 ea qEV column, $65 ea, reusable up to 5 times HiPrep column, $650 ea, reusable up to 20 times

50 Then compute the total wages from the hourly rate (Plab) and the time taken to seed cells (teed) and collect conditioned media (tf 11). A multiplier fl accounts for labor costs beyond that of an operator (e.g. supervisors and management).

Cab = m (teed + col )(i + f) (10)

Equipment for cell expansion is split into three categories: incubators where cells undergo expansion, biosafety cabinets (BSCs) where cells are handled under aseptic conditions, and ancillary equipment such as those for automation. The price of incubators (pinc) or ancillary equipment (pLflc) depends on the specific TE and each incubator or ancillary equipment can process up to Ugnc or Uganc units simultaneously. The price of a BSC (Pbsc) is the same for any expansion technology and a BSC can be used by up to Ubsc operators at a time. A delta function (S) indicates if a particular TE requires a BSC. Add the three costs to compute equipment costs per lot.

[UEi l rmE"'l an Uafl C Pi =pifnC +Pbsc UbSc +pafinc

Use NYrlot to computempttha the annual COG. Since equipment can be shared between lots, an additional lot in the same year does not increase equipment cost, but equipment can depreciate over time (taep). Annual COG of each TE is stored as zi, while the global minimum (i.e. annual COG of cheapest combination of TE and TH) among all values of z1 is stored as ZE-

z = N (Ccons + Cab) +(1 lot L - +dep

To compute zfj: Likewise, annual COG for EV harvest is broken down into consumables, labor, and equipment. One TH unit represents a multi-step process of purification, with an overall price of consumables (pqOns). Consumables include single-use tubes, chromatography columns, and polymers for precipitation.

Ccons = H -ons (13)

Similarly compute labor cost per lot for EV harvest units, but instead using the total labor time (tproc for the entire multi-step process. Waiting between steps is not considered labor.

mH, = M (14) mHH~j

C~iab = mH,J ' Piab * C(] + f) (15) Equipment for EV harvest involves only BSCs and ancillary equipment such as benchtop centrifuges. Because BSCs may also used for cell expansion, they are considered in computing ZH only if additional BSCs are required for EV harvest. Annual COG of each TH is stored as zj; the global minimum is stored as ZH. anc un mH' iE if [MH,j mEij (p UHnc Pbsc U IbIscI Ubsc I Ubsc Ubsc C. - (16) P anc [UH,J 1 mHj _E,i j Uanc Ubsc Ubsc

= N (Ccons + Clab) + (17) tdep

51 Results Cost estimates and cheapest technologies are similar whether a one-stage or multi-stage bioprocess is modeled. To validate the accuracy of our model, we first attempted to reproduce cost estimates previously published for cell expansion bioprocesses [198,1991, specifically the Simaria model. Unlike our model which focuses on one stage of cell expansion, the Simaria model assumes a four-stage bioprocess (i.e. four passages from P1 to P4) regardless of the demand for cells. Each stage would utilize a different cell expansion technology, and optimization would seek to minimize COG incurred throughout all four stages. The algorithm starts with P4, uses P4 parameters to pick the cheapest technology for P3, and so on until P1. Simaria et al. provided absolute numbers only in a case study comparing three bioprocesses at 50 doses/lot versus 1,000 doses/lot at a dose size of 10' cells/dose and annual demand of 10,000 doses/year [198]; hence we used the same conditions in the case study to validate our estimates (Figure 33). The Simaria model indicates that the cheapest bioprocesses to produce 50 and 1,000 doses/lot would utilize L-40 (Figure 33A) and cL-120 (Figure 33D) respectively at the last stage (i.e. P4). Because annual demand is the same, a difference in lot size reflects a difference in the number of lots per year. At 50 doses/lot, the same bioprocess would be repeated 20 times more in a year than it would at 1,000 doses/lot. Hence, quality control, which incurs the same cost per lot regardless of lot size, would cost more per year with a smaller lot size. Meanwhile, costs of labor and consumables are sensitive to lot size. Generally, for larger-scale technologies, consumables cost more because of higher media consumption, but labor costs less because of automation. When lot size is small, although larger-scale technologies would save labor, they would 'waste' media as cells need not be cultured to maximum density to meet the lot size; smaller-scale technologies therefore become more favorable despite the higher labor cost. Equipment does not cost much annually since they can be reused between lots and depreciate over time, even though the start-up cost of ancillary equipment (e.g. $425,000 for cL-120) can be significant. A key concern when modeling a four-stage bioprocess is the need to consider compatibility of candidate technologies between stages, which the Simaria model attempts to address by imposing 'priority rules' when optimizing for the cheapest technology in an earlier stage. Although the 'priority rules' were not specified, the rest of the optimization algorithm was well documented. Strictly following the algorithm, we found that cost estimates become only slightly lower without the 'priority rules', and ranking between candidate technologies for P4 remains unchanged (Figure 33B,E). Deeper analysis into cheapest technologies for individual stages showed that the 'priority rules' may apply additional constraints on the number of units or type of technology, at the expense of cost (Table 6).

Table 6. Cheapest bioprocesses for 1,000 doses/lot. Simaria model Removal of 'priority rules'

P4 cL-120 (2 units) cL-120 (2 units) P3 cL-36 (4 units) cT (34 units) P2 cT (6 units) L-10 (2 units) P1 T-500 (3 units) L-5 (1 unit)

52 P4 only A Cost structure by Simaria model B Removal of 'priority rules' C P4 only

sEquipment mConsumables oLabor *Qualitycontroi *Equipment -Consumables oLabor *Qualitycontrol *Equipment -Consumables aLabor nQualitycontrol $4.5M $4.5M $4.5M $4.OM $4.OM $4.M 53.5M $3.5M $3.5M $ $3.OM I$3.OM I M 0 $2.5M $2.5M $2.5M 10 810M $2.OM $2.OM c $1.5M $1.5M 4 $1.OM $1.0M

$0.5M $0.5M $0.5M $O.5M Si OM $0.M - - $S.OM -"'"" - 50.DM -- - - L-10 L-40 cL-120 L-10 L-40 cL-120 L-10 L-40 cL-120 50 doses/lot 50 doses/lot 50 doses/lot

D Cost structure by Simara model E Removal of 'priority rules' F P4 only UEquipment xConsumables sLabor *Qualitycontrol aEquipment gConsumables sLabor nQualitycontrol EEquipment Consumables ELabor EQualitycontrol

$1.6M $1.6M $1.6M

SlAM $AM $1.4M

$$1.2M $1.2M $S1.2M 0 0 0 a 0 $1.0M $1.0M 0 o $0.8M a $0.8M o $0.8MI- t$0.6M $0.6M C 4C $0.4M $0.4M $0.4M

$0.2M $0.2M $0.2M

SO.OM $S.OM $.OM- L-10 L-40 cL-120 L-10 L40 cL-120 L-10 L40 cL-120 1000 doses/lot 1000 doses/lot 1000 doses/lot

Figure 33. Cost estimates for cell expansion change only slightly when 'priority rules' are removed, and are dominated by the last passage. (A) The Simaria model indicates that a four-stage bioprocess ending with L40 is the most economical for a lot size of 50 doses/lot. (B) Removing 'priority rules' from the Simaria model slightly lowers cost estimates but preserves the cheapest technology for P4. (C) Reducing the cell expansion bioprocess from 4 stages to 1 stage further lowers cost, and still yields the same cheapest technology. (D-F) The same observations are made for a lot size of 1,000 doses/lot.

53 4

A Contribution by each stage B Contribution by each stage OP1 UP2 EP3 EP4 nP1 GP2 uP3 EP4 100% 100%

90% 90%

0 80% 0 80%

70% 70% C C C 60% 60%

50% 0 0 *, 40% *40%3 0%

30% C 30%

20%

10% 10%

0% 0% L-10 L-40 cL-120 L-10 L-40 cL-120 50 doses/lot 1000 dosesilot

Figure 34. The largest contribution to the annual cost of a four-stage cell expansion bioprocess originates from the final stage. Costs of consumables, labor, and equipment are broken down by stage of cell expansion for lot sizes of (A) 50 doses/lot and (B) 1,000 doses/lot.

We further examined if similar observations can be drawn by modeling one stage of cell expansion instead of four. Importantly, P4 among all stages contributes the most to annual cost (Figure 34). Because P4 utilizes the largest-scale technology and incurs the least labor costs, excluding P1 to P3 affects cost structure, but because P4 dominates overall costs, ranking between candidate technologies is preserved (Figure 33C,F). For a large range of annual demands and lot sizes, both the Simaria model and our model identify the same cheapest technologies, with a few exceptions (Figure 35). In these exceptions, the cost discrepancies between the cheapest P4 technologies identified by the Simaria model, versus the cheapest technologies identified by our one-stage model, are subtle (Figure 36).

We concluded that our one-stage model is comparable to the Simaria model in its ability to make recommendations as a decision-support tool. Additionally, our one-stage model is amenable to modular modeling, as shown by how a 4-stage bioprocess (Figure 33B,E) can be assembled from a series of one-stage bioprocesses (Figure 33C,F) under user-defined conditions.

According to industry experts, typically L-10 is used in industrial settings [204], and about 2,500 doses at 108 cells/dose are produced each year, the COG of which totals between $3.1 1M and $3.74M annually [197]. To meet this demand, our one-stage model estimates that the use of L-10 would cost $5.27M, $3.64M, $3.07M, and $2.83M annually at lot sizes of 10, 25, 50, and 100 doses/lot respectively (Figure 37). These increase by 7-11% to $5.62M, $4.05M, $3.42M, and $3.14M if a 2-stage bioprocess is allowed; by 9-14% to $5.75M, $4.14M, $3.48M, and $3.20M if a 3-stage bioprocess is allowed; and by 11-15% to $5.86M, $4.19M, $3.53M, and $3.23M if a 4-stage bioprocess is allowed. Given that commercial lot sizes would be at least 100 doses/lot at 108 cells/dose [205], our estimates capture actual COG in industrial settings with reasonable accuracy. Moreover, our model shows that L-10, while currently preferred by industry, may not be the most economical option, and recommends larger-scale planar vessels such as LAO and cL-120, or even SUBs (Figure 37), as industry experts have recently proposed [204].

Therefore, for the rest of the study, we proceeded with only our one-stage model to analyze cell expansion and EV harvest bioprocesses.

54 A Lot size (cellsilot) 1x1012 2.5x1012 5x1013 5x1O 1x108 5x108 1x109 5x109 1x1010 5x1010 1x1011 5x10

1x109 L-5 L-10

' 5x1Q9 L-5 L-10 L-10 1x101 L-5 L-10 L-40 L-40 A 5x10 L-40 L-40 cL-120 1x101 L-40 L-40 cL-120 cL-120 "0 5x1011 cL-120 cL-120 cL-120 1 4) M 1X10125x1Q cL-120 cL-120 cL-120 cL-120 E x1012 cL-120 cL-120 500L

1x1013 cL-120 cL-120 500L 500L C 5x1013 500L 500L 2000L 1x1014 500L 500L 2000L 14 x1I 2000L

Lot size (cells/lot)

5x1011 1x1012 2.5x1012 1x1013 5x107 1x108 5x108 1x109 5x1O9 1x1010 5x1010 1x101 1x109 L-5 cL-12 B 5x10 L-5 cL-12 cT 1x1010 L-5 cL-12 cT cT 5x1010 L-40 L-40 cL-120 1x101 L-40 L-40 cL-120 cL-120 0 5x1011 cL-120 cL-120 cL-120

V 1x1012 cL-120 cL-120 cL-120 CL-120 12 5x10 CL-120 cL-120 2000L CU C 1x1013 CL-120 cL-120 2000L 500L 0 5x101-5x 13 2000L 500L 2000L .- - 2000L 500L 2000L 1x1014 2000L 5x1014

cheapest technology for a large range of annual demands and lot sizes. Figure 35. Reducing a four-stage model to one-stage still yields the same space indicates conditions that no technology can meet Solution space that is empty either falls below 10 lots/year or exceeds 200 lots/year. Shaded (A) P4 in the Simaria model and (B) our one-stage model within the limitations imposed (e.g. maximum number of units). Solutions that differ between within space constraints. are highlighted in blue. SUBs are considered only when no planar vessel can meet the conditions

55 A 5Exi0 cells/lot, 10 lots/year B 5x1i01 cells/lot, 200 lots/year UEquipment Consumables a Labor 0 Quality control a Equipment Consumables .Labor *Quality control 5O5M $120M

VI 5O.4M IM$80m V 0 5O.3M VI 0 U $60M 5O.2M $40M

4 S.1AM 4 $01M $20M .m $OM

5x10 cells/lot, 20 lots/year Ix1O" cellsllot, 100 lots/year aEquipment Consumables .Labor uQualitycontrol *Equipment Consumables *Labor n Quality control $0.9M $90M SOAM $80M 50.7M $70M SO.6M $60M $0.5M $50M $04M $40M 50.3M $30M $0.2M $20M 50.1 M $10m $0.0m $OM

5x10I cells/lot, 100 lots/year 5x10" cells/lot, 20 lots/year

*Equipment rConsumables .Labor a Quality control sEquipment Cnmbes Labor w~ualitycontrol $4.'M $7M $3.5M SiGm56M q $3.0M $5M $2.5M $4M S2.OM $1.5m $3M 2M

SO 5M SGGM 5GM c 1L

1 5x10S cells/lot, 200 lots/year 1M cells/lot, 10 lots/year a Equipment Consumables a Labor @ Quality control a Equipment Consumables aLabor Qualityontrol $8.OM $7M

VI S6M $6.0m 58GM $5M 0 $4M S4.'M - $3M

C S2.OM C $2M 4 $1.0M IIIII~i1Ii__ SiM 57GM .; 'Vt 0

Figure 36. Lot size, more so than annual demand, drives a switch in cheapest technology. (A) For a fixed lot size, varying annual demand does not drastically change the cheapest technology. (B) For a fixed annual demand, varying lot size quickly drives a change in scale and hence a change in cheapest technology. Excluded technologies either violate space constraints, or already exceed lot size upon seeding.

Ml - ' 10 doses/lot, 250 lotslyear 25 doses/lot, 100 lots/year -Consrnables sLabor .Quahtycwntrol 4 Equipment -Consumabies sLabor nQualityontroI mEquipment $14M $10m $9M $12M s8M $10m $7M O $6M I $8M a $4M c $3M $4M _' Ii C $2M 52M SiM 5GM

60 doses/lot, 50 lots/year 100 doses/lot, 25 lots/year a Equipment Consumables mLabor wQuaflycontrol mEquipment -Consumabnles Labor eQuaitycontrol $10M $9M $9M $aM $8M $7M $7M I $6M I $6M $5M $5m CM $4M $4M 0 $3M C $2M. $2M $imI $OM

Figure 37. Technologies (e.g. L-10) currently preferred by industry may not be the most economical to meet market demands (2,500 doses/year). Use of larger-scale planar vessels (e.g. L-40, cL-120) or SUBs (e.g. 20L, 200L) may further reduce cost.

Lot size, rather than annual demand, determines economy of scale. In general, for a given lot size, cost structure remains largely unchanged with increasing annual demand (Figure 36A). Costs of consumables, labor, and quality control scale up almost linearly as the number of lots per year increases, whereas equipment costs are fixed per year. Hence, the contribution to total costs remains largely unchanged for consumables, labor, and quality control, but decreases for equipment. Essentially, the number of cell expansion units remains the same as long as lot size is fixed; comparison is made within the same set of candidate technologies since no new technologies would violate space constraints (Figure 32). With increasing annual demand, the same lot is simply repeated more times with no additional constraints unless consecutive lots begin to overlap (e.g. >200 lots/year). Consequently, the cheapest technology generally remains unchanged when lot size is fixed (Figure 35). When annual demand is fixed, increasing lot size shifts the set of valid candidate technologies from smaller-scale to larger-scale technologies (Figure 36B,37) and the cheapest technology changes (Figure 35). Smaller-scale technologies are excluded when lot size is large enough for them to violate space constraints, while larger-scale technologies are excluded when lot size is small enough such that the number of cells required for seeding already exceeds the lot size (Figure 32). Using larger lot sizes to meet a given annual demand requires less labor and quality control, maximizes the capacity of culture vessels, operators, and equipment, and thereby incurs less cost. Ultrafiltrationis generally most cost-effective for harvesting EVs. Having validated the accuracy of our estimates for cell expansion, we next modeled whole bioprocesses incorporating both cell expansion and EV harvest. We selected EV harvest technologies (Table 5) among published protocols specifically for isolating EVs from cell culture supernatants. UC represents

57 ultracentrifugation, the current standard for EV isolation [46,73,76,881. PPT and SECI represent two commercial kits, namely ExoQuick from System Biosciences and qEV from IZON, which are the first to be marketed for isolating intact EVs by polymer-induced precipitation and size-exclusion chromatography respectively-emerging methods that are rapidly gaining interest [46,89]. Since SECI is relatively small-scale, taking only 0.5 mL of sample per column, we added SEC2, a size-exclusion chromatography protocol published by an academic group that allows up to 240 mL of sample per unit [861. Yet another emerging method is ultrafiltration. While commercial kits dedicated to EV filtration are not yet available, several academic groups including us (see Chapter 3) have adopted commercially available filtration devices that are traditionally used for isolating other forms of biological particles. UP represents a published protocol employing a series of dead-end and tangential-flow filters [981. Recognizing that the tangential-flow filters can take 20 times more volume than what was published, we added UF2, a slightly modified version of UF1 to match the same tangential-flow filter with larger-volume dead-end filters of the same brand, without changing any other parameters in the protocol (Table 5). In general, larger-scale versions of each method may be possible with continuous-flow ultracentrifuges (e.g. 8 L) [206], continuous chromatography (e.g. 16 L) [2071, and other industrial equipment, but feasibility of their use for EVs has not been reported. Figure 38 shows the cheapest technologies for a range of EV demands. Costs due to quality control were disregarded since release criteria for EV products are still under debate [45,76], but this would not affect ranking between candidate technologies because the same cost of quality control would apply to all (Figure 36,37). Cheapest cell expansion technologies span both planar vessels and SUBs, indicating that the range of lot sizes and annual demands considered was sufficiently extensive (Figure 38). Although cell numbers required to generate EVs are within the range of cell numbers in bioprocesses solely for cell expansion (Figure 35), solutions differ because cell expansion technologies for EV harvest are additionally limited by volume constraints: candidate technologies producing the same number of cells or EVs can significantly differ in media consumption, and require dramatically different numbers of EV harvest units. Likewise, volume constraints favor EV harvest technologies that accept larger volumes, since EV-free buffer can be added to conditioned media to meet the minimum volume of any EV harvest technology, increasing the range of lot sizes that larger-volume technologies can tolerate without violating space constraints. Smaller-volume technologies, namely PPT and SEC1, require a large number of units, which quickly drives up costs of consumables and labor (Figure 39,40). Meanwhile, despite differing in percent recovery, UC, SEC2, and UF2 yield comparable COG in most conditions. Even when percent recovery is set at 100% for all technologies, solutions remain largely the same (Figure 41). This is because a smaller percent recovery needs not translate to requiring more units; a longer culture time in the same number of vessels will suffice to produce more EVs, which accumulate exponentially. Among all methods of EV isolation considered in this study, UF1 and UF2 collectively remain valid for the widest range of lot sizes. While they may not be cheapest for all conditions considered, their COG stay close to that of the cheapest technology. Moreover, UF2 is essentially equivalent to UF1, the only exception being that larger-volume dead-end filters are used prior to the common step of tangential-flow filtration. Until more protocols utilizing larger-volume technologies for EV isolation are published, we conclude that ultrafiltration is currently the most versatile and cost-effective for scale-up. Harvest costs dominate COG, but can be reduced at large lot sizes. In general, EV harvest accounts for more than 50% of annual COG. However, when lot size is sufficiently large, the need to use larger-scale cell expansion technologies lowers the contribution of EV harvest to overall COG. Figure 42 shows an example using UF2. L-10 is most widely used in the current industry [204], while L-40, cT, cL-120, and 20L have been recommended for cost-effectiveness. When coupled to these cell expansion technologies, UF2 can contribute as low as 20% to the annual COG. Meanwhile, HF and 20L are emerging cell expansion technologies being investigated for scaling up EV production [56,2081.

58 A Lot size (EVs/Iot) 1x101' 5x1013 1x1014 5x1014 1x101 5x101I 1x101 5x1011 1x1012 5x1012 1x1011 T-175 5x1011 T-175 L-1 (U 1 X1012 T-175 L-1 L-1 "0U) > 5x1012 L-1 L-1 cT 1x1013 L-1 L-1 CT CT 13 5x10 cT CT CT E 1x1014 cT cT cT 20L 14 5x10 cT 20L 20L 20L 1x1015 cT 20L 20L 15 5x10 20L 20L 20L 1 X1016 20L B Lot size (EVs/lot) V 1x1014 5x1014 1x1010 5x1010 1x10l 5x1011 1x1012 5x1012 1x1013 5x1013

1x101l UF1 (U 5x1011 UF1 SEC2 1x1012 UC SEC2 SEC2 5x1012 UC SEC2 UF2 1x101, UC SEC2 UF2 UF2 5x1013 UF2 UF2 UF2 E 1x1014 UF2 UF2 UF2 UF2 CD 5x1014 UF2 UF2 UF2

1x1015 UF2 UF2 UF2 UF2 5x1015 UF2 UF2

1x1016 UF2 UF2 considered. Optimization is based on the total COG Figure 38. Ultrafiltration dominates as the cheapest EV harvest technology in most conditions indicate the cheapest pair of (A) of every possible combination of cell expansion and EV harvest technologies. Corresponding solutions in both tables is empty either falls below 10 lots/year or cell expansion and (B) EV harvest technologies for a given lot size and annual demand. Solution space that imposed. exceeds 200 lots/year. Shaded space indicates conditions where no technology can meet within the limitations

59 1 x 1010 EVs/lot, 10 lots/year Cell expansion + EV harvest % due to EV harvest Harvest costs for using T-175

.UC *PPT .SEC2 .UF NSECI aLIF2 mUC *PPT uSECI uSEC2 -UF1 .*F2 *Equtpment Consumables ELa;of $45K 90% $30K S40K 80% I $25K $35K C70% U $30K 60% $20K C 0 0 $25K 50% C C SIOKS15K 0 $20K 40% 0

$15K C 30% SOK -C SIOK 20% ~1 S5K 10% C 4 50K 0% $0K T-175 1-225 T-500 *i1, 1-225 T-500 UC III!PPT SEC1 SEC2 UF1 UF2 Cell expansion technology Cell expansion technology EV harvest technology

1 x1010 EVs/lot, 50 lots/year Cell expansion + EV harvest % due to EV harvest Harvest costs for using T-175

.UC .PPT .SECd .SEC2 'IUFI .F21 mUC .PPT uSECI -SEC2 -UF1 2 eEquopment -Conswmabes *LAbor S10K 90% $120K S160K 80% I $100K $140K 8 70% U

S120K 60% $80K C 0 0 5100K 50% C 580K $80K I 40% 0 S60K 30% S40K C $40K 20% *1 $20K S20K 10% C C oil 4 SOK 01- S0K Ii T1751-225 T-500 T-175 T-225 T-500 Uc PPT SEC1 SEC2 UFI U 2 Cell expansion technology Cell expansion technology EV harvest technology

1 x 1010 EVs/lot, 100 lots/year Cell expansion + EV harvest % due to EV harvest Harvest costs for using T-175

|u .PPT uSEC1 9SEC2 -UFI sUF2 SUC *PPT MSECd eSEC2 UF1 eLF2 UEquipmernt Conamables *L01o

S350K 90% $250K

S300K 80% 70% 5200K $250K I 60% V 5200K S150K I 50% 8 *0 $150K .2 40% 1C 5100K 30% SI00K I 1I 20% I 10% ! S50K $50K 0% 175 T-500 Cell e nio t T-500 UC PPT SECI SEC2 UFI UF2 Cell expansion technology Cell expansion technology EV harvest technology

Figure 39. At bench scale, most EV harvest protocols are comparable in annual COG. EV harvest generally costs more than cell expansion in a given bioprocess. For a fixed lot size, varying annual demand does not drastically change the cost structure of EV harvest technologies except in the case of UC, thereby mostly preserving the ranking between technology combinations. Excluded technologies violate the limitations imposed.

60 5x1010 EVs/lot, 100 lots/year Cell expansion + EV harvest % due to EV harvest Harvest costs for using L-1

MUC APPT uSEC1 *SEC2 UFI *UF2 IUC .PPT mSEC1 wSEC2 UFI gUF2 I Equipment COnsunbaeS wLabof

$500K 100% $350K 90% $300K 80% LI 70% 5250K $500K 60% 0U 5200K S300K 50% 'C S150K C 40% S200K 0U 0 30% $100K C 20% 5100K $50K 10%

50K 0% SK UC PPT SEC1 SEC2 UFI UF2 Cell expansion technology Cell expansion technology EV harvest technology

1 x 1011 EVs/lot, 50 lots/year Cell expansion + EV harvest % due to EV harvest Harvest costs for using L-1

NUC m PPT MSEC1 wSEC2 UFI *LW2 GUC .PPT uSEd1 aSEC2 UFI 1F2 F me im nsumaes mLatow

$700K 100% 5160K 0 90% 5160K $600K 0 80% $140K $500K 70% S120K 0 60% 0 $400K S100K 50% S80K 5300K 40% C 30% C S200K 0 $40K 20% 5100K C 10% S20K 50K U-- $0K LJ IL- 0% "1* :P,4>, S7> 4 UC PPT SECI SEC2 UF1 UF2 Cell expansion technology Cell expansion technology EV harvest technology

5x1011 EVs/lot, 10 lots/year Cell expansion + EV harvest % due to EV harvest Harvest costs for using L-1 UFI UF2 UL.bo, | muC .PPT uSEG1 aSEC2 UFI mUF21 mUC .PPT uSECi NSEC2 *EqumeIm , Consumables 5180K 100% 5120K 90% IO$ 100K $140K 20% C70% $120K $80K C 60% $100K 50% $60K I 40% 580K C S40K 30% 4 S40K 20% 3 $20K $20K 10% 0% $0K $OK idlikitiLi UC PPT SEC1 SEC2 UFI UF2 Cell expansion technology Cell expansion technology EV harvest technology

Figure 40. Ultrafiltration remains valid over a large range of lot sizes. As lot size increases, larger-scale cell expansion technologies become valid, for which only larger-scale EV harvest technologies can match.

61 A Lot size (EVs/Iot) J1x10O 5x1010 1x10" 5x1011 1x1012 5x1012 1x1013 5x101 1x101 5x10" 1x1011 T-175 I.. 5x1011 T-175 0 1 x1012 T-175 5x1012 L-1 L-1 L-5 1x1013 L-1 L-1 L-5 cT 0 5x1013 -5 cT cT 1x1014 -5 cT CT 20L 5x1014 cT 20L 20L 1x1015 cT 20L 20L 20L 5x1015 20L 20L 1x 1016 20L 20L B Lot size (EVs/Iot)

10 1 2 1x10 5x1010 1x1011 5x10 ' 1x101 5x10 1x1013 5x1013 1x1014 5x1014 Ix1011 UF1 5x1011 UC SEC2 1 1x1012 UC SEC2 SEC2 1 5x10 UC UC UF2 1XJ130 UC UC UF2 UF2 0 5xl 13 UF2 UF2 UF2

1x101 UF2 UF2 UF2 UF2 5x1014 UF2 UF2 UF2 115" UF2 UF2 UF2 UF2 5x101 UF2 UF2 1x1016 UF2 UF2

Figure 41. Cheapest technologies remain largely the same when all EV harvest technologies are assumed to recover 100% of EVs. The key outcome of this additional assumption is that a given EV demand and lot size will now require the same number of units of cell expansion technology regardless of the EV harvest technology; hence, comparison is chiefly made between EV harvest technologies, rather than whole bioprocesses. Solutions that differ from those in Figure 38 are highlighted in blue.

62 % due to EV harvest using UF2 60% n 5el1 - 1e12 50% n5e12 0 1e13 40% * e 13

C 30% C

.2 20%

10% 0

0% T-175 L-10 L-40 cT cL-120 HF 20L Cell expansion technology

Figure 42. When lot size is sufficiently large, EV harvest can become cheaper than cell expansion. The legend shows different lot sizes in EVs/lot; all conditions are normalized at 100 lots/year.

Because HF is relatively costly among cell expansion technologies (Figure 36,37), adding EV harvest to a HF bioprocess will consume only about 10% extra COG.

Trends observed in cell expansion bioprocesses (Figure 36,37) when varying lot size and annual demand remain largely applicable after EV harvest is incorporated (Figure 39,40). However, since the EV harvest technologies considered have not been demonstrated to be automatable, we cannot attribute economy of scale to automation. Indeed, labor dominates in most harvest costs. Given that automation of purifying biologics is possible and being developed [209,210], harvest costs will likely decrease further in the future.

Biological parameters are the strongest cost driver.

To identify key cost drivers as well as account for uncertainty in our estimates, we perform sensitivity analysis by varying cost or process parameters and examining their individual impact on annual COG of selected technology combinations (Figure 43). Unsurprisingly, since harvest costs dominate overall COG in most conditions (Figure 39,40), the price of cell culture media-which concerns only cell expansion costs- hardly affects annual COG. On the contrary, labor rate and the price of consumables for EV harvest can influence annual COG almost proportionally to their degree of change. Labor rate, being involved in both cell expansion and EV harvest, is a particularly strong cost driver, regardless of the technology combination, although its impact on our solutions is mild (Figure 44), even when lowered by 10 times (Figure 45).

Thus far, cost computations have been based on biological parameters empirically obtained from immortalized cells (Table 1; see Chapter 2). Cultured in the same medium and density, primary cells from two out of three donors behaved similarly and yielded almost identical COG, but cells from the third donor, which produced about 3 times as much EVs per population doubling, can reduce COG by more than 60% (Figure 46). Even more impressively, when cells are cultured under conditions that boost EV output per population doubling by an order of magnitude, COG can be lowered by approximately 6 times (Figure 46). These comparisons were based on cheapest technology combinations, which could differ between biological conditions for the same lot size and annual demand. If comparison between biological conditions was made for the same technology combination at the same lot size and annual demand, cost reduction would be more dramatic. This is because an increased EV output per population doubling means that a culture vessel produces more EVs with the same initial and final cell densities, such that EVs would be more concentrated in the conditioned media, leading to less cell expansion units, less volume, and ultimately less EV harvest units. Biological parameters are therefore the strongest cost driver.

63 T-175 at 5x10 EVs/lot L-10 at 5x10" EVs/lot CT at 1x1O" EVs1lot

UF2 UF? t Ml195 I I~I-UF2 UF1 Price of cell I C SEC2 culture media I UF1 SEC2 Base: SECI $150/L I PPT PPT SEC UIC V I -40% .20% 0% 20% 40% "402% 0% 20% 40% -40% -M n a C Change in annual COG Change in annual COG Change In annual COG

T-175 at 5x10 EVs/lot L-10 at x10" EVs/lot cT at 1xIO" EVs/lot

IU UF 2 UF2 I" ri'2f LPF U Labor I SEC2 I rate I UF1 .i$2eo SEC .n 2WMs .ES14M Base: SECI I I PP $200/h T PPT SEC2 U U I I I I i -40% -20% 0% "% 40% -40% .20% 0% 20% 40% -40% -20% 0% 20% 40% Change In annual COG Change In annual COG Change In annual COG

T-175 at 5x10O EVsflot L-10 at 5x10" EVsIlot cT at 1x1i0" EVs/lot

UF 2 |=+30% UF2 M -30% | U,2 1W-30%

Price of I fASEC consumables I UF1 SEC2 for EV I SECT C harvest I SPPT PP. SEC2 UC UC I I i -40% -20% 0% 20% 40% -4 -20% 0% 20% 40% -40% -20% 20% 40% Change In annual COG Change in annual COG Change In annual COG

Figure 43. Labor rate and price of consumables for EV harvest, but not price of cell culture media, are key cost drivers. Each cost parameter is varied by 30%, and changes in annual COG at 100 lots/year are computed for different technology combinations at different lot sizes.

64 A Lot size (EVs/Iot) 4 14 10 11 11 12 12 13 1x10 10 5x0 1x10 5x10 x10 5x10 x10 5x10 1x101 5x10

1x1011 T-175 5x1011 T-1 75 L-1 1x0 12 T-1 75 L-1 L-1 5x1012 L-1 L-1 cT 1x10 13 L-1 L-1 cT cT 3 cT 5x101 cT CT 1x1014 cT cT cT 20L C 5x1 014 cL-1 20 20L 20L 1x1015 cL-1 20 20L 20L 20L (U 15 5x10 20L 20L 6 1xlO 20L 20L Lot size (EVs/Iot)

1 4 1 4 10 1 12 13 3 5x10 1x10 5x10' 1x10 5x10 1x10 ' 5x10 1x10 5x101 1x10

1x1011 UF1 5x1011 UF1 SEC2 C 2 1x10 UC SEC2 SEC2 5x101 2 UC SEC2 UF2 1x1013 UC SEC2 UF2 UF2 5x101 UF2 UF2 UF2

1x1014 UF2 UF2 UF2 UF2 5x10 14 UF2 UF2 UF2 UF2 1x1015 UF2 UF2 UF2 5x1 015 F2 F2 1x101, UF2 UF2

by 30%. Solutions that differ from those in Figure 38 are Figure 44. Cheapest technologies remain largely the same when labor rate is increased those in Figure 38, except that UFI is replaced by UC for highlighted in blue. When labor rate is decreased by 30%, solutions remain identical to Ix 1010 EVs/lot at 5x10 " EVs/year.

65 A Lot size (EVs/Iot) 10 1 1 12 3 3 4 4 1x1'O 5x10 1x10" 5x " 1x10 5x10 1x10 5x10 1x10 5x10 1x10" T-175 5x1011 T-175 T-225 T-1 75 T-225 LT-225 5x1 012 T-225 T-225 cT T-225 T-225 cT cT (U 5x1013 cT cT cT lx1014 cT cT cT 20L 5x1 014 cT 20L 20L 1x101, cT 20L 20L 20L cc 5x10015 20L 20L 1x10"6 20L wCU 20L E B> Lot size (EVs/Iot) 1 11 12 1 13 4 1x10 5x101 1x10 5x10 1x10 5x10" 1x10 5x10 1x10 5x1014 1x1011 UF1 5x10 UC SEC2 1x1012 UC SEC2 SEC2 B 5x12 UC UC UF2 1x1013 UC UC UF2 UF2 5x13 UF2 UF2 UF2 lx1014 UF2 UF2 UF2 UF2 5x10 14 UF2 UF2 UF2 1x10 UF2 UF2 UF2 UF2 5x15 UF2 UF2 1x10 UF2 UF2

Figure 45. Cheapest technologies change more significantly for bench-scale lot sizes when labor rate is reduced to $20/h, which is typical for a paid research student. Hence, from an economic perspective, academic laboratories have little incentive to switch to a different cell expansion or EV harvest technology. Solutions that differ from those in Figure 38 are highlighted in blue.

66 Impact of culture conditions on COG $1.6M -ATCC #ATERT @100% SP $1.4M -- Lonz #2 @100% SP -- Lonza #3 @ 100% SP $1.2M -- RoosterBio #1 @ 100% SP $1.-M RoosterBio #1 @1% SP N E $0.8M 0 $0.6M 0 O $0.4M

$0.2M

$0.0M IE+10 1E+11 1E+12 IE+13 1E+14 IE+15 Lot size (EVs/lot) Figure 46. Changes in cell source or culture conditions can significantly influence COG. Biological parameters were selected from Table 1; SP represents StemPro serum-free medium which was used to empirically determine the biological parameters (see Chapter 2). Lot size was varied at 100 lots/year.

Discussion

New methods continually emerge to tackle technical and logistical challenges in EV isolation. Generally, biological effects of EV preparations have been observed regardless of the isolation method, so any method could potentially be used to manufacture EV products. Due to the heterogeneity of EVs, each method typically enriches for a different EV subset with varying quality (e.g. purity, aggregation, structural integrity). Active substances and their mechanisms of action likely differ between EV products even if the products originate from the same cell source [451. Process and product become tightly intertwined; the bioprocess should be kept consistent as much as possible during scale-up to generate EV fractions with reproducible performance. Such is the case already seen in the manufacturing of cell therapies [211,212] and glycosylated proteins [213] including antibodies [214], whereby small adjustments in bioprocessing can propagate to large variability in product quality. Bioprocess design should therefore be established as early as possible, even before empirical iterations, emphasizing the importance of decision-support tools like our model. By accommodating technological variations, our model can additionally evaluate new or hypothetical methods (e.g. UF2), particularly the extent of their economic advantage over existing methods, and identify key cost drivers that may inspire and guide future innovations.

Critically, our model reveals the relative scalability between EV harvest technologies, at least from an economic perspective. Scalability, historically derived from the concept of expansion flexibility, refers to the ability of a system to accommodate capacity changes without requiring significant new designs [2151. Ultracentrifugation is generally considered not scalable. Conventional ultracentrifuges (e.g. Beckman Coulter Optima L-90K) employ swing-bucket or fixed-angle rotors whose rotational axes are outside the sample chamber; to ensure gravitational balance between the symmetrically positioned but individually and manually filled chambers, the rotors are designed to be substantially heavier than the samples. Consequently, majority of electrical power is 'wasted' on accelerating the rotor instead of the sample. At industrial scale, such a design will warrant unrealistically massive rotors and exorbitant power consumption. On the contrary, continuous- flow ultracentrifuges (e.g. Alfa Wassermann KII) are designed so that the rotational axis is inside the sample chamber. A fluid sample will naturally balance itself symmetrically around the axis, eliminating the need to add weight to the rotor. Despite being more amenable for large-scale processing, continuous-flow

67 ultracentrifuges are relatively new and have not yet been widely adopted by academic groups. Meanwhile, compared to ultracentrifugation, size-exclusion chromatography (e.g. SEC1, SEC2) and ultrafitration (e.g. UF1, UF2) are more scalable: although larger columns and filters may be necessary for industrial-scale manufacturing, system operation and design remain the same. Perhaps the least scalable of all is polymer- induced precipitation (e.g. PPT). The need to remove the polymer additive increases the number of steps and reliance on other purification technologies including centrifugation and chromatography. Solutions from our model, which incorporates both cost and bioprocess parameters in identifying cheapest technologies, correlate with the relative scalability of these isolation methods. Currently, because the majority of clinical trials investigating EV therapy are in phase I [33,34], the effective dose range of EVs is unclear. We also recognize that dose will likely vary between disease applications. Hence, lot sizes and annual demand were expressed in EVs instead of doses, to allow users to define their own dose. Nevertheless, to the best of our knowledge, one phase II trial has been completed and reported, in which 22 patients were intradermally administered a median of 247 pg (by protein mass) of EVs per dose over a median of 7 doses per patient [216]. Given that pure EVs contain about 0.1 fg of protein each [771, a median dose would consume about 2.5x10" EVs. In this study, we explored annual demands up to 200 lots/year but the maximum lot size we could explore was 1x 10" EVs/lot, beyond which no combination of the cell expansion and EV harvest technologies in consideration is feasible within the boundary conditions. This maximum limit corresponds to about 4,000 doses per year, treating about 570 patients a year, based on the Phase II trial. This already requires 3 units of 20L and 15 units of UF2, or 60,000 units of UC if the current standard for EV isolation was utilized. However, we note that primary endpoints were not met in this phase I trial. Since preclinical doses that showed efficacy in animal studies are on the order of 10" EVs/kg, perhaps the effective clinical dose would be two orders of magnitude higher than what was tested in the phase II trial. If this is true, none of the technologies investigated in this study would be feasible to meet the demand. Larger-scale cell expansion technologies are already being investigated for scaling up EV production [12,108]; we recommend that future studies should focus on larger-scale EV harvest technologies capable of meeting clinical demands. Who would be in the best position to conduct such studies? Given that the bioprocess would produce both cells and EVs, the most economical strategy may be to develop both cells and EVs as products. Our model predicts that adding EV harvest to an existing cell expansion process can cost as low as 10% of annual COG, especially if lot size is large (e.g. 100 lots/year). Therefore, current cell manufacturers may face the least barriers by converting conditioned media, which is otherwise regarded as waste, to commercializable products [461. EVs purified by PPT are currently sold by Systems Biosciences as 'standards'. However, since PPT will unlikely be the method of choice for clinical manufacturing of EVs, 'standards' generated by scalable means might be more appropriate. A potential disadvantage for cell manufacturers is the constraint in choice of cell culture media. If their existing bioprocesses employ EV-containing media (e.g. serum), they would not be able to harvest EVs purely generated by the cultured cells. Moreover, the optimal media composition for maximizing quality and quantity of cellular products may not be optimal for producing EVs. Production of both cell and EV therapies may therefore be challenging; commercializing one product for clinical use while reserving the other for research use may be more realistic. By commercializing EV products generated from clinical-grade cells, cell manufacturers would not only advance large-scale methods for EV isolation, but also provide reliable and relevant 'standards' which the academic community has been advocating for [761.

Future Work Besides relaxing the assumptions and limitations imposed in this study (Box 4), several changes to our model may further improve its applicability. First, space constraints could be accounted for by tracking the space consumption of each technology, rather than setting a limit to the number of units. This may be accompanied by an option for the user to define the available space for cell expansion versus EV harvest.

68 Detailed accounting of space use can facilitate adoption of 'lean manufacturing' to improve production efficiency [2171. Second, volume reduction steps could be incorporated to normalize final EV concentration. This would add a third iterative loop to the algorithm, and require evaluation of technologies that have been shown to concentrate EVs. Furthermore, the saturation point of EVs would have to be empirically determined, as EVs could aggregate when overly concentrated [86]. Third, variation in biological parameters due to a change in cell expansion technology could be studied. While this may not be feasible in a single academic setting, empirical data could be collected from research groups exploring a variety of cell expansion technologies for EV production. This might also uncover more differences between cell types and culture conditions in their impact on EV output.

69 CHAPTER SEVEN

TOWARDS CLINICAL TRANSLATION OF EV THERAPY

Passage from discovery to clinical adoption more often than not comprises the 'valley of death', where a developing technology, while deemed promising, is too new to attract sufficient funding for continued development and validation of its commercial potential [2181. Translation of EV therapy will be no exception. While minimizing costs of goods (Chapter 6) is one way to ease financial pressure, securing a strong portfolio of intellectual property will strengthen commercial advantage. Currently, EV-related patents appear to be relatively scattered with higher activity in technology areas associated with immunology and stem cells (Figure 47). Majority of the patents concern novel compositions. Manufacturing methods, however, are no less important, as demonstrated by nanomedicine patents over the last two decades [2191. A potentially intensive area of innovation is the proprietary formulation of cell culture media that maximize EV output for clinical-grade cells (Chapter 2). Also patentable as application-specific technologies are potency assays that predict clinical outcomes and assure consistency in product quality (Chapter 4), as well as methods that enhance product quality (Chapter 5). Although scalable isolation methods may appear to be broadly applicable, their technical feasibility depends on the complexity of the specific cell culture supernatants (Chapter 3). Since EV manufacturing innovations likely need to be application-specific for reduction to practice, there is ample room in the near future for securing patents that will aid in crossing of the 'valley of death'. Cost-effective bioprocess design and an aggressive patent strategy will likely accelerate the clinical translation on EV therapy.

-V

g Peept muholipid sm sEce 40M 0..'dNd 4micrtm -I.~ 4 EP _1~ 1

P Ib . wl0 SW

A* Q

Bio o~cai mpie

a single patent. Terminologically related patents are clustered together within contour lines that indicate strength of correlation. 'Snow-capped' peaks have the highest patent activity. (Source: Thomson Reuters)

70 RE FERENCES x 1 Singer SJ, Nicolson GL. The fluid mosaic model of the structure of cell membranes. Science. 1972;175(4023):720-731.

2 Lodish H, Berk A, Kaiser CA, Krieger M, Scott MP, Bretscher A, Ploegh H, Matsudaira P. Molecular Cell Biology. 6th ed. New York: W. H. Freeman; 2007. 3 Mellman I, Emr SD. A Nobel prize for membrane traffic: vesicles find their journey's end. J Cell Biol. 2013;203(4):559-561. 4 Wells WA. The discovery of synaptic vesicles. J Cell Biol. 2005; 168(1):12. 5 Kerr JFR, Wyllie AH, Currie AR. Apoptosis: a basic biological phenomenon with wide-ranging implications in tissue kinetics. Br J Cancer. 1972;26(4):239-257. 6 Lockshin RA, Zakeri Z. Programmed cell death and apoptosis: origins of the theory. Nat Rev Mol Cell Biol. 2001;2(7):545- 550. 7 Harding C, Heuser J, Stahl P. Receptor-mediated endocytosis of transferrin and recycling of the transferrin receptor in rat reticulocytes. J Cell Biol. 1983;97(2):329-339. 8 Harding C, Heuser J, Stahl P. Endocytosis and intracellular processing of transferrin and colloidal gold-transferrin in rat reticulocytes: demonstration of a pathway for receptor shedding. Eur J Cell Biol. 1984;35(2):256-263.

9 Pan BT, Teng K, Wu C, Adam M, Johnstone RM. Electron microscopic evidence for externalization of the transferrin receptor in vesicular form in sheep reticulocytes. J Cell Biol. 1985; 101(3):942-948. 10 Johnstone RM, Adam M, Hammond JR, Orr L, Turbide C. Vesicle formation during reticulocyte maturation: association of plasma membrane activities with released vesicles (exosomes). J Biol Chem. 1987;262(19):9412-9420. 11 Johnstone RM, Bianchini A, Teng K. Reticulocyte maturation and exosome release: transferrin receptor containing exosomes shows multiple plasma membrane functions. Blood. 1989;74(5):1844-1851.

12 Raposo G, Nijman HW, Stoorvogel W, Liejendekker R, Harding CV, Melief CJ, Geuze HJ. B lymphocytes secrete antigen- presenting vesicles. J Exp Med. 1996; 183(3):1161-1172. 13 Valadi H, Ekstr6m K, Bossios A, Sj6strand M, Lee JJ, L6tvall JO. Exosome-mediated transfer of mRNAs and microRNAs is a novel mechanism of genetic exchange between cells. Nat Cell Biol. 2007;9(6):654-659. 14 Johnstone RM, Mathew A, Mason AB, Teng K. Exosome formation during maturation of mammalian and avian reticulocytes: evidence that exosome release is a major route for externalization of obsolete membrane proteins. J Cell Physiol. 1991;147(1):27-36.

15 Warren BA, Vales 0. The release of vesicles from platelets following adhesion to vessel walls in vitro. Br J Exp Pathol. 1972;53(2):206-215. 16 Nicolson GL, Smith JR, Poste G. Effects of local anesthetics on cell morphology and membrane-associated cytoskeletal organization in BALB/3T3 cells. J Cell Biol. 1976;68(2):395-402. 17 Scott RE, Perkins PG, Zschunke MA, Hoerl BJ, Maercklein PB. Plasma membrane vesiculation in 3T3 and SV373 cells: morphological and biochemical characterization. J Cell Sci. 1979;35:229-243.

18 Black BL, Yoneyama Y, Moog F. Microvillous membrane vesicle accumulation in media during culture of intestine of chick embryo. Biochim Biophys Acta. 1980;601(2):343-348.

19 Gould SJ, Raposo G. As we wait: coping with an imperfect nomenclature for extracellular vesicles. J Extracell Vesicles. 2013;2:20389.

20 El Andaloussi S, Mager I, Breakefield XO, Wood MJA. Extracellular vesicles: biology and emerging therapeutic opportunities. Nat Rev Drug Discov. 2013;12(5):347-357. 21 Lbtvall J, Hill AF, Hochberg F, Buzas El, Di Vizio D, Gardiner C, Gho YS, Kurochkin IV, Mathivanan S, Quesenberry P, et al. Minimal experimental requirements for definition of extracellular vesicles and their functions: a position statement from the International Society for Extracellular Vesicles. J Extracell Vesicles. 2014;3:26913.

71 22 Bobrie A, Colombo M, Krumeich S, Raposo G, Thery C. Diverse subpopulations of vesicles secreted by different intracellular mechanisms are present in exosome preparations obtained by differential ultracentrifugation. J Extracell Vesicles. 2012;1:18397. 23 Jeppesen DK, Hvam ML, Primdahl-Bengtson B, Boysen AT, Whitehead B, Dyrskjot L, Orntoft TF, Howard KA, Ostenfeld MS. Comparative analysis of discrete exosome fractions obtained by differential centrifugation. J Extracell Vesicles. 2014;3:25011. 24 Willms E, Johansson HJ, Mager I, Lee Y, Blomberg KEM, Sadik M, Alaarg A, Smith CIE, Lehti6 J, El Andaloussi S, et al. Cells release subpopulations of exosomes with distinct molecular and biological properties. Sci Rep. 2016;6:22519. 25 Kowal J, Arras G, Colombo M, Jouve M, Morath JP, Primdal-Bengtson B, Dingli F, Loew D, Tkach M, Thery C. Proteomic comparison defines novel markers to characterize heterogeneous populations of extracellular vesicle subtypes. Proc Natl Acad Sci USA. 2016;113(8):E968-E977. 26 Chevillet JR, Kang Q, Ruf IK, Briggs HA, Vojtech LN, Hughes SM, Cheng HH, Arroyo JD, Meredith EK, Gallichotte EN, et al. Quantitative and stoichiometric analysis of the microRNA content of exosomes. Proc Natl Acad Sci USA. 2014;111(41):14888-14893. 27 Hurley JH, Meyer T. Subcellular targeting by membrane lipids. Curr Opin Cell Biol. 2001; 13(2):146-152. 28 Mellman I,Nelson WJ. Coordinated protein sorting, targeting and distribution in polarized cells. Nat Rev Mol Cell Biol. 2008;9(11):833-845. 29 Longatti A, Tooze SA. Vesicular trafficking and autophagosome formation. Cell Death Differ. 2009; 16(7):956-965. 30 Colombo M, Raposo G, Thery C. Biogenesis, secretion, and intercellular interactions of exosomes and other extracellular vesicles. Annu Rev Cell Dev Biol. 2014;30:255-289. 31 Cicero AL, Stahl PD, Raposo G. Extracellular vesicles shuffling intercellular messages: for good or for bad. Curr Opin Cell Biol. 2015;35:69-77. 32 Vader P, Breakefield XO, Wood MJA. Extracellular vesicles: emerging targets for cancer therapy. Trends Mol Med. 2014;20(7):385-393. 33 Gy6rgy B, Hung ME, Breakefield XO, Leonard JN. Therapeutic applications of extracellular vesicles: clinical promise and open questions. Annu Rev Pharmacol Toxicol. 2015;55:439-464. 34 Fais S, O'Driscoll L, Borras FE, Buzas E, Camussi G, Cappello F, Carvalho J, da Silva AC, Del Portillo H, El Andaloussi S, et al. Evidence-based clinical use of nanoscale extracellular vesicles in nanomedicine. ACS Nano. 2016; 10(4):3886-3899. 35 Nolte-'t Hoen E, Cremer T, Gallo RC, Margolis LB. Extracellular vesicles and viruses: are they close relatives? Proc Natl Acad Sci USA. 2016;1 13(33):9155-9161. 36 Capulli AK, Tian K, Mehandru N, Bukhta A, Choudhury SF, Suchyta M, Parker KK. Approaching the in vitro clinical trial: engineering organs on chips. Lab Chip. 2014; 14(17):3181-3186. 37 Benam KH, Dauth S, Hassell B, Herland A, Jain A, Jang KJ, Karalis K, Kim HJ, MacQueen L, Mahmoodian R, et al. Engineered in vitro disease models. Annu Rev Pathol. 2015;10:195-262. 38 Fatehullah A, Tan SH, Barker N. Organoids as an in vitro model of human development and disease. Nat Cell Biol. 2016; 18(3):246-254. 39 Hentze H, Graichen R, Colman A. Cell therapy and the safety of embryonic stem cell-derived grafts. Trends Biotechnol. 2007;25(1):24-32. 40 Trounson A, McDonald C. Stem cell therapies in clinical trials: progress and challenges. Cell Stem Cell. 2015; 17(1):11-22. 41 Aliotta JM, Pereira M, Li M, Amaral A, Sorokina A, Dooner MS, Sears EH, Brilliant K, Ramratnam B, Hixson DC, et al. Stable cell fate changes in marrow cells induced by lung-derived microvesicles. J Extracell Vesicles. 2012;1:18163. 42 Liu S, Liu D, Chen C, Hamamura K, Moshaverinia A, Yang R, Liu Y, Jin Y, Shi S. MSC transplantation improves osteopenia via epigenetic regulation of Notch signaling in lupus. Cell Metab. 2015;22(4):606-618. 43 Ng KS, Kuncewicz TM, Karp JM. Beyond hit-and-run: stem cells leave a lasting memory. Cell Metab. 2015;22(4):541-543. 44 Quesenberry PJ, Aliotta J, Deregibus MC, Camussi G. Role of extracellular RNA-carrying vesicles in cell differentiation and reprogramming. Stem Cell Res Ther. 2015;6:153.

72 45 Lener T, Gimona M, Aigner L, Barger V, Buzas E, Camussi G, Chaput N, Chatterjee D, Court FA, del Portillo HA, et al. Applying extracellular vesicles based therapeutics in clinical trials-an ISEV position paper. J Extracell Vesicles. 2015;4:30087.

46 Smith JA, Ng KS, Mead BE, Dopson S, Reeve B, Edwards J, Wood MJA, Carr AJ, Bure K, Karp JM, et al. Extracellular vesicles: commercial potential as byproducts of cell manufacturing for research and therapeutic use. Bioprocess Int. 2015; 13(4s):20,2 2 , 2 4-2 6.

47 Lamichhane TN, Sokic S, Schardt JS, Raiker RS, Lin JW, Jay SM. Emerging roles for extracellular vesicles in tissue engineering and regenerative medicine. Tissue Eng Part B Rev. 2015;21(1):45-54.

48 Kourembanas S. Exosomes: vehicles of intercellular signaling, biomarkers, and vectors of cell therapy. Annu Rev Physiol. 2015;77:13-27.

49 Robbins PD, Morelli AE. Regulation of immune responses by extracellular vesicles. Nat Rev Immunol. 2014; 14(3):195-208. 50 van der Meel R, Fens MHAM, Vader P, van Solinge WW, Eniola-Adefeso 0, Schiffelers RM. Extracellular vesicles as drug delivery systems: lessons from the liposome field. J Control Release. 2014;195:72-85. 51 Hall J, Prabhakar S, Balaj L, Lai CP, Cerione RA, Breakefield XO. Delivery of therapeutic proteins via extracellular vesicles: review and potential treatments for Parkinson's disease, glioma, and schwannoma. Cell Mol Neurobiol. 2016;36(3):417-427.

52 Vader P, Mol EA, Pasterkamp G, Schiffelers RM. Extracellular vesicles for drug delivery. Adv Drug Deliv Rev. 2016; 106(A):148-156.

53 Pizzirani C, Ferrari D, Chiozzi P, Adinolfi E, SandonA D, Savaglio E, Di Virgilio F. Stimulation of P2 receptors causes release of IL-1 P-loaded microvesicles from human dendritic cells. Blood. 2007; 109(9):3856-3864. 54 Lopatina T, Bruno S, Tetta C, Kalinina N, Porta M, Camussi G. Platelet-derived growth factor regulates the secretion of extracellular vesicles by adipose mesenchymal stem cells and enhances their angiogenic potential. Cell Commun Signal. 2014; 12:26. 55 Glebov K, Lachner M, Jabs R, Lau T, Merkel 0, Schloss P, Steinhauser C, Walter J. Serotonin stimulates secretion of exosomes from microglia cells. Glia. 2015;63(4):626-634. 56 Mitchell JP, Court J, Mason MD, Tabi Z, Clayton A. Increased exosome production from tumour cell cultures using the Integra CELLine Culture System. J Immunol Methods. 2008;335(1-2):98-105.

57 King HW, Michael MZ, Gleadle JM. Hypoxic enhancement of exosome release by breast cancer cells. BMC Cancer. 2012; 12:42 1.

58 Bian S, Zhang L, Duan L, Wang X, Min Y, Yu H. Extracellular vesicles derived from human bone marrow mesenchymal stem cells promote angiogenesis in a rat myocardial infarction model. J Mol Med. 2014;92(4):387-397.

59 Yeo RWY, Lai RC, Zhang B, Tan SS, Yin Y, Teh BJ, Lim SK. Mesenchymal stem cell: an efficient mass producer of exosomes for drug delivery. Adv Drug Deliv Rev. 2013;65(3):336-341.

60 Lamparski HG, Metha-Damani A, Yao JY, Patel S, Hsu DH, Ruegg C, Le Pecq JB. Production and characterization of clinical grade exosomes derived from dendritic cells. J Immunol Methods. 2002;270(2):211-226.

61 Kempken R, Berthold W, Thomse K. Significance of specific productivity for effectiveness and consistency of mammalian cell cultivation. In: Beuvery EC, Griffiths JB, Zeijlemaker WP, editors. Animal Cell Technology: Developments Towards the 21st Century. 1st ed. Netherlands: Springer; 1995. p. 561-565.

62 Clarke C, Doolan P, Barron N, Meleady P, O'Sullivan F, Gammell P, Melville M, Leonard M, Clynes M. Predicting cell- specific productivity from CHO gene expression. J Biotechnol. 2011;151(2):159-165.

63 Rani S, Ryan AE, Griffin MD, Ritter T. Mesenchymal stem cell-derived extracellular vesicles: toward cell-free therapeutic applications. Mol Ther. 2015;23(5):812-823.

64 van der Pol E, Coumans FAW, Sturk A, Nieuwland R, van Leeuwen TG. Refractive index determination of nanoparticles in suspension using nanoparticle tracking analysis. Nano Lett. 2014;14(11):6195-6201.

65 Crescitelli R, Lasser C, Szab6 TG, Kittel A, Eldh M, Dianzani I, Buzas EI, L6tvall J. Distinct RNA profiles in subpopulations of extracellular vesicles: apoptotic bodies, microvesicles and exosomes. J Extracell Vesicles. 2013;2:20677. 66 Sokolova V, Ludwig AK, Hornung S, Rotan 0, Horn PA, Epple M, Giebel B. Characterisation of exosomes derived from human cells by nanoparticle tracking analysis and scanning electron microscopy. Colloids Surf B Biointerfaces. 2011;87(1):146-150.

73 67 Sitar S, Kejiar A, Pahovnik D, Kogej K, Tusek-Znidaric M, Lenassi M, 2 agar E. Size characterization and quantification of exosomes by asymmetrical-flow field-flow fractionation. Anal Chem. 2015;87(18):9225-9233. 68 Chernyshev VS, Rachamadugu R, Tseng YH, Belnap DM, Jia Y, Branch KJ, Butterfield AE, Pease LF,3rd, Bernard PS, Skliar M. Size and shape characterization of hydrated and dessicated exosomes. Anal Bioanal Chem. 2015;407(12):3285-3301. 69 Chase LG, Lakshmipathy U, Solchaga LA, Rao MS, Vemuri MC. A novel serum-free medium for the expansion of human mesenchymal stem cells. Stem Cell Res Ther. 2010;1(1):8. 70 Chase LG, Yang S, Zachar V, Yang Z, Lakshmipathy U, Bradford J, Boucher SE, Vemuri MC. Development and characterization of a clinically compliant xeno-free culture medium in Good Manufacturing Practice for human multipotent mesenchymal stem cells. Stem Cells Trans Med. 2012;1(10):750-758.

71 Patrikoski M, Juntunen M, Boucher S, Campbell A, Vemuri MC, Mannerstr6m B, Miettinen S. Development of fully defined xeno-free culture system for the preparation and propagation of cell therapy-compliant human adipose stem cells. Stem Cell Res Ther. 2013;4(2):27.

72 Neuhuber B, Swanger SA, Howard L, Mackay A, Fischer I. Effects of plating density and culture time on bone marrow stromal cell characteristics. Exp Hematol. 2008;36(9):1176-1185.

73 Gudbergsson JM, Johnsen KB, Skov MN, Duroux M. Systematic review of factors influencing extracellular vesicle yield from cell cultures. Cytotechnology. 2016;68(4):579-592.

74 van der Pol E, Coumans FAW, Grootemaat AE, Gardiner C, Sargent IL, Harrison P, Sturk A, van Leeuwen TG, Nieuwland R. Particle size distribution of exosomes and microvesicles determined by transmission electron microscopy, flow cytometry, nanoparticle tracking analysis, and resistive pulse sensing. J Thromb Haemost. 2014; 12(7):1182-1192.

75 Maas SLN, de Vrij J, van der Vlist EJ, Geragousian B, van Bloois L, Mastrobattista E, Schiffelers RM, Wauben MHM, Broekman MLD, Nolte-'t Hoen ENM. J Control Release. Possibilities and limitations of current technologies for quantification of biological extracellular vesicles and synthetic mimics. 2015;200:87-96.

76 Witwer KW, Buzis El, Bemis LT, Bora A, Lisser C, L6tvall J, Nolte-'t Hoen EN, Piper MG, Sivaraman S, Skog J, et al. Standardization of sample collection, isolation and analysis methods in extracellular vesicle research. J Extracell Vesicles. 2013;2:20360.

77 Webber J, Clayton A. How pure are your vesicles? J Extracell Vesicles. 2013;2:19861. 78 Ibarra N, Watanabe S, Bi JX, Shuttleworth J, Al-Rubeai M. Modulation of cell cycle for enhancement of antibody productivity in perfusion culture of NSO cells. Biotechnol Prog. 2003;19(l):224-228.

79 Du Z, Treiber D, McCarter JD, Fomina-Yadlin D, Saleem RA, McCoy RE, Zhang Y, Tharmalingam T, Leith M, Follstad BD, et al. Use of a small molecule cell cycle inhibitor to control cell growth and improve specific productivity and product quality of recombinant proteins in CHO cell cultures. Biotechnol Bioeng. 2015;112(1):141-155.

80 Park JH, Noh SM, Woo JR, Kim JW, Lee GM. Valeric acid induces cell cycle arrest at G1 phase in CHO cell cultures and improves recombinant antibody productivity. Biotechnol J. 2016; 11(4):487-496.

81 Biller SJ, Schubotz F, Roggensack SE, Thompson AW, Summons RE, Chisholm SW. Bacterial vesicles in marine ecosystems. Science. 2014;343(6167):183-186.

82 van der Pol E, Hoekstra AG, Sturk A, Otto C, van Leeuwen TG, Nieuwland R. Optical and non-optical methods for detection and characterization of microparticles and exosomes. J Thromb Haemost. 2010;8(12):2596-2607.

83 Minciacchi VR, You S, Spinelli C, Morley S, Zandian M, Aspuria PJ, Cavallini L, Ciardiello C, Sobreiro MR, Morello M, et al. Large oncosomes contain distinct protein cargo and represent a separate functional class of tumor-derived extracellular vesicles. Oncotarget. 2015;6(13):11327-11341.

84 Shelke GV, Lasser C, Gho YS, L6tvall J. Importance of exosome depletion protocols to eliminate functional and RNA- containing extracellular vesicles from fetal bovine serum. J Extracell Vesicles. 2014;3:24783. 85 Cvjetkovic A, L6tvall J, Lasser C. The influence of rotor type and centrifugation time on the yield and purity of extracellular vesicles. J Extracell Vesicles. 2014;3:23111.

86 Nordin JZ, Lee Y, Vader P, Mager 1, Johansson HJ, Heusermann W, Wiklander OPB, Hallbrink M, Seow Y, Bultema JJ, et al. Ultrafiltration with size-exclusion liquid chromatography for high yield isolation of extracellular vesicles preserving intact biophysical and functional properties. Nanomedicine. 2015; 11(4):879-883.

74 87 Yamashita T, Takahashi Y, Nishikawa M, Takakura Y. Effect of exosome isolation methods on physicochemical properties of exosomes and clearance of exosomes from the blood circulation. Eur J Pharm Biopharm. 2016;98:1-8. 88 Thery C, Amigorena S, Raposo 0, Clayton A. Isolation and characterization of exosomes from cell culture supernatants and biological fluids. Curr Protoc Cell Biol. 2006;30:3.22.1-3.22.29.

89 Lobb RJ, Becker M, Wen SW, Wong CSF, Wiegmans AP, Leimgruber A, Miller A. Optimized exosome isolation protocol for cell culture supernatant and human plasma. J Extracell Vesicles. 2015;4:27031. 90 Shin H, Han C, Labuz JM, Kim J, Kim J, Cho S, Gho YS, Takayama S, Park J. High-yield isolation of extracellular vesicles using aqueous two-phase system. Sci Rep. 2015;5:13103.

91 Rider MA, Hurwitz SN, Meckes DG,Jr. ExtraPEG: a polyethylene glycol-based method for enrichment of extracellular vesicles. Sci Rep. 2016;6:23978.

92 Welton JL, Webber JP, Botos LA, Jones M, Clayton A. Ready-made chromatography columns for extracellular vesicle isolation from plasma. J Extracell Vesicles. 2015;4:27269.

93 Lee K, Shao H, Weissleder R, Lee H. Acoustic purification of extracellular microvesicles. ACS Nano. 2015;9(3):2321-2327.

94 Kalyanpur M. Downstream processing in the biotechnology industry. Mol Biotechnol. 2002;22(1):87-98.

95 Gronemeyer P, Ditz R, Strube J. Trends in upstream and downstream process development for antibody manufacturing. Bioengineering. 2014;1(4):188-212.

96 Nestola P, Peixoto C, Silva RRJS, Alves PM, Mota JPB, Carrondo MJT. Improved virus purification processes for vaccines and gene therapy. Biotechnol Bioeng. 2015;112(5):843-857.

97 Pattasseril J, Varadaraju H, Lock LT, Rowley JA. Downstream technology landscape for large-scale therapeutic cell processing. Bioprocess Int. 2013; 1 1(3s):38,40-42,44-47. 98 Heinemann ML, Ilmer M, Silva LP, Hawke DH, Recio A, Vorontsova MA, Alt E, Vykoukal J. Benchtop isolation and characterization of functional exosomes by sequential filtration. J Chromatogr A. 2014;1371:125-135.

99 Nolte-'t Hoen ENM, Buermans HPJ, Waasdorp M, Stoorvogel W, Wauben MHM, 't Hoen PAC. Deep sequencing of RNA from immune cell-derived vesicles uncovers the selective incorporation of small non-coding RNA biotypes with potential regulatory functions. Nucleic Acids Res. 2012;40(18):9272-9285.

100 Baglio SR, Rooijers K, Koppers-Lalic D, Verweij FJ, Lanz6n MP, Zini N, Naaijkens B, Perut F, Niessen HWM, Baldini N, et al. Human bone marrow- and adipose-mesenchymal stem cells secrete exosomes enriched in distinctive miRNA and tRNA species. Stem Cell Res Ther. 2015;6:127.

101 Zitvogel L, Regnault A, Lozier A, Wolfers J, Flament C, Tenza D, Ricciardi-Castagnoli P, Raposo G, Amigorena S. Eradication of established murine tumors using a novel cell-free vaccine: dendritic cell-derived exosomes. Nat Med. 1998;4(5):594-600.

102 Deregibus MC, Cantaluppi V, Calogero R, Iacono ML, Tetta C, Biancone L, Bruno S, Bussolati B, Camussi G. Endothelial progenitor cell-derived microvesicles activate an angiogenic program in endothelial cells by a horizontal transfer of mRNA. Blood. 2007; 110(7):2440-2448.

103 Bruno S, Grange C, Deregibus MC, Calogero RA, Saviozzi S, Collino F, Morando L, Busca A, Falda M, Bussolati B, et al. Mesenchymal stem cell-derived microvesicles protect against acute tubular injury. J Am Soc Nephrol. 2009;20(5):1053-1067. 104 Lai RC, Arslan F, Lee MM, Sze NSK, Choo A, Chen TS, Salto-Tellez M, Timmers L, Lee CN, El Oakley RM, et al. Exosome secreted by MSC reduces myocardial ischemia/reperfusion injury. Stem Cell Res. 2010;4(3):214-222.

105 Mire-Sluis AR. Progress in the use of biological assays during the development of biotechnology products. Pharm Res. 2001;18(9):1239-1246.

106 Ahrlund-Richter L, De Luca M, Marshak DR, Munsie M, Veiga A, Rao M. Isolation and production of cells suitable for human therapy: challenges ahead. Cell Stem Cell. 2009;4(1):20-26.

107 Zhukareva V, Obrocka M, Houle JD, Fischer I, Neuhuber B. Secretion profile of human bone marrow stromal cells: donor variability and response to inflammatory stimuli. Cytokine. 2010;50(3):317-321.

108 Ranganath SH, Levy 0, Inamdar MS, Karp JM. Harnessing the mesenchymal stem cell secretome for the treatment of cardiovascular disease. Cell Stem Cell. 2012;10(3):244-258.

75 109 Mendicino M, Bailey AM, Wonnacott K, Puri RK, Bauer SR. MSC-based product characterization for clinical trials: an FDA perspective. Cell Stem Cell. 2014;14(2):141-145. 110 Teng X, Chen L, Chen W, Yang J, Yang Z, Shen Z. Mesenchymal stem cell-derived exosomes improve the microenvironment of infarcted myocardium contributing to angiogenesis and anti-inflammation. Cell Physiol Biochem. 2015;37(6):2415-2424. 111 Lee C, Mitsialis SA, Aslam M, Vitali SH, Vergadi E, Konstantinou G, Sdrimas K, Fernandez-Gonzalez A, Kourembanas S. Exosomes mediate the cytoprotective action of mesenchymal stromal cells on hypoxia-induced pulmonary hypertension. Circulation. 2012;126(22):2601-2611. 112 Chen JY, An R, Liu ZJ, Wang JJ, Chen SZ, Hong MM, Liu JH, Xiao MY, Chen YF. Therapeutic effects of mesenchymal stem cell-derived microvesicles on pulmonary arterial hypertension in rats. Acta Pharmacol Sin. 2014;35(9):1121-1128. 113 Aliotta JM, Pereira M, Wen S, Dooner MS, Del Tatto M, Papa E, Goldberg LR, Baird GL, Ventetuolo CE, Quesenberry PJ, et al. Exosomes induce and reverse monocrotaline-induced pulmonary hypertension in mice. Cardiovasc Res. 2016;110(3):3 19- 330. 114 Zhu YG, Feng XM, Abbott J, Fang XH, Hao Q, Monsel A, Qu JM, Matthay MA, Lee JW. Human mesenchymal stem cell microvesicles for treatment of Escherichia coli endotoxin-induced acute lung injury in mice. Stem Cells. 2014;32(1):116-125. 115 Monsel A, Zhu YG, Gennai S, Hao Q, Hu S, Rouby JJ, Rosenzwajg M, Matthay MA, Lee JW. Therapeutic effects of human mesenchymal stem cell-derived microvesicles in severe pneumonia in mice. Am J Respir Crit Care Med. 2015;192(3):324-336. 116 Gatti S, Bruno S, Deregibus MC, Sordi A, Cantaluppi V, Tetta C, Camussi G. Microvesicles derived from human adult mesenchymal stem cells protect against ischaemia-reperfusion-induced acute and chronic kidney injury. Nephrol Dial Transplant. 2011;26(5):1474-1483. 117 Bruno S, Grange C, Collino F, Deregibus MC, Cantaluppi V, Biancone L, Tetta C, Camussi G. Microvesicles derived from mesenchymal stem cells enhance survival in a lethal model of acute kidney injury. PLoS One. 2012;7(3):e33115. 118 He H, Wang Y, Sun S, Yu M, Wang C, Pei X, Zhu B, Wu J, Zhao W. Bone marrow stem cells-derived microvesicles protect against renal injury in the mouse remnant kidney model. Nephrology. 2012;17(5):493-500. 119 Zhou Y, Xu H, Xu W, Wang B, Wu H, Tao Y, Zhang B, Wang M, Mao F, Yan Y, et al. Exosomes released by human umbilical cord mesenchymal stem cells protect against cisplatin-induced renal oxidative stress and apoptosis in vivo and in vitro. Stem Cell Res Ther. 2013;4(2):34. 120 Li T, Yan Y, Wang B, Qian H, Zhang X, Shen L, Wang M, Zhou Y, Zhu W, Li W, et al. Exosome derived from human umbilical cord mesenchymal stem cells alleviate liver fibrosis. Stem Cells Dev. 2013;22(6):845-854. 121 Tan CY, Lai RC, Wong W, Dan YY, Lim SK, Ho HK. Mesenchymal stem cell-derived exosomes promote hepatic regeneration in drug-induced liver injury models. Stem Cell Res Ther. 2014;5(3):76. 122 Xin H, Li Y, Cui Y, Yang JJ, Zhang ZG, Chopp M. Systemic administration of exosomes released from mesenchymal stromal cells promote functional recovery and neurovascular plasticity after stroke in rats. J Cereb Blood Flow Metab. 2013;33(11):1711-1715. 123 Zhang Y, Chopp M, Meng Y, Katakowski M, Xin H, Mahmood A, Xiong Y. Effect of exosomes derived from multipluripotent mesenchymal stromal cells on functional recovery and neurovascular plasticity in rats after traumatic brain injury. J Neurosurg. 2015;122(4):856-867. 124 Zhang B, Yin Y, Lai RC, Tan SS, Choo ABH, Lim SK. Mesenchymal stem cells secrete immunologically active exosomes. Stem Cells Dev. 2014;23(11):1233-1244. 125 Gennai S, Monsel A, Hao Q, Park S, Matthay MA, Lee JW. Microvesicles derived from human mesenchymal stem cells restore alveolar fluid clearance in human lungs rejected for transplantation. Am J Transplant. 2015;15(9):2404-2412. 126 Koch M, Lemke A, Lange C. Extracellular vesicles from MSC modulate the immune response to renal allografts in a MHC disparate rat model. Stem Cells Int. 2015;2015:486141. 127 Budoni M, Fierabracci A, Luciano R, Petrini S, Di Ciommo V, Muraca M. The immunosuppressive effect of mesenchymal stromal cells on B lymphocytes is mediated by membrane vesicles. Cell Transplant. 2013;22(2):369-379. 128 Conforti A, Scarsella M, Starc N, Giorda E, Biagini S, Proia A, Carsetti R, Locatelli F, Bernado ME. Microvesicles derived from mesenchymal stromal cells are not as effective as their cellular counterpart in the ability to modulate immune responses in vitro. Stem Cells Dev. 2014;23(21):2591-2599.

76 129 Del Fattore A, Luciano R, Pascucci L, Goffredo BM, Giorda E, Scapaticci M, Fierabracci A, Muraca M. Immunoregulatory effects of mesenchymal stem cell-derived extracellular vesicles on T lymphocytes. Cell Transplant. 2015;24(12):2615-2627.

130 Mongui6-Tortajada M, Roura S, Gdtvez-Mont6n C, Pujal JM, Aran G, Sanjurjo L, Franquesa M, Sarrias MR, Bayes-Genis A, Borras FE. Nanosized UCMSC-derived extracellular vesicles but not conditioned medium exclusively inhibit the inflammatory response of stimulated T cells: implications for nanomedicine. Theranostics. 2017;7(2):270-284.

131 Zhang HC, Liu XB, Huang S, Bi XY, Wang HX, Xie LX, Wang YQ, Cao XF, Lv J, Xiao FJ, et al. Microvesicles derived from human umbilical cord mesenchymal stem cells stimulated by hypoxia promote angiogenesis both in vitro and in vivo. Stem Cells Dev. 2012;21(18):3289-3297.

132 Shabbir A, Cox A, Rodriguez-Menocal L, Salgado M, van Badiavas E. Mesenchymat stem cell exosomes induce proliferation and migration of normal and chronic wound fibroblasts, and enhance angiogenesis in vitro. Stem Cells Dev. 2015;24(14):1635-1647.

133 Issman L, Brenner B, Talmon Y, Aharon A. Cryogenic transmission electron microscopy nanostructural study of shed microparticles. PLoS One. 2013;8(12):e83680.

134 Haller D, Blum S, Bode C, Hammes WP, Schiffrin EJ. Activation of human peripheral blood mononuclear cells by nonpathogenic bacteria in vitro: evidence of NK cells as primary targets. Infect Immun. 2000;68(2):752-759. 135 Jansk L, Reymanova P, Kopeck9 J. Dynamics of cytokine production in human peripheral blood mononuclear cells stimulated by LPS or infected by Borrelia. Physiol Res. 2003;52(6):593-598.

136 Yeap SK, Alitheen NB, Ali AM, Omar AR, Raha AR, Suraini AA, Muhajir AH. Effect of Rhaphidophora korthalsii methanol extract on human peripheral blood mononuclear cell (PBMC) proliferation and cytolytic activity toward HepG2. J Ethnopharmacol. 2007; 114(3):406-411.

137 Meszaros K, Aberle S, White M, Parent JB. Immunoreactivity and bioactivity of lipopolysaccharide-binding protein in normal and heat-inactivated sera. Infect Immun. 1995;63(1):363-365.

138 Jonuleit H, Kfihn U, Mfller G, Steinbrink K, Paragnik L, Schmitt E, Knop J, Enk AH. Pro-inflammatory cytokines and prostaglandins induce maturation of potent immunostimulatory dendritic cells under fetal calf serum-free conditions. Eur J Immunol. 1997;27(12):3135-3142.

139 Tkachenko N, Wojas K, Tabarkiewicz J, Rolinski J. Generation of dendritic cells from human peripheral blood monocytes- comparison of different culture media. Folia Histochem Cytobiol. 2005;43(1):25-30.

140 Gottschalk U, Brorson K, Shukla AA. The need for innovation in biomanufacturing. Nat Biotechnol. 2012;30(6):489-492.

141 Brindley DA, Bure KE. Automation of cell therapy biomanufacturing. Bioprocess Int. 2013;11(3s):18,20-22,24-25.

142 Patel S, Mehta-Damani A, Shu H, Le Pecq JB. An analysis of variability in the manufacturing of dexosomes: implications for development of an autologous therapy. Biotechnol Bioeng. 2005;92(2):238-249.

143 van Dommelen SM, Vader P, Lakhal S, Kooijmans SAA, van Solinge WW, Wood MJA, Schiffelers RM. Microvesicles and exosomes: opportunities for cell-derived membrane vesicles in drug delivery. j Control Release. 2012;161(2):635-644.

144 Mittelbrun M, Gutierrez-Vizquez C, Villarroya-Beltri C, GonzAlez S, Sanchez-Cabo F, Gonzilez MA, Bernad A, SAnchez- Madrid F. Unidirectional transfer of microRNA-loaded exosomes from T cells to antigen-presenting cells. Nat Commun. 2011;2:282.

145 Tomasoni S, Longaretti L, Rota C, Morigi M, Conti S, Gotti E, Capelli C, Introna M, Remuzzi G, Benigni A. Transfer of growth factor receptor mRNA via exosomes unravels the regenerative effect of mesenchymal stem cells. Stem Cells Dev. 2013;22(5):772-780.

146 Cossetti C, Iraci N, Mercer TR, Leonardi T, Alpi E, Drago D, Alfaro-Cervello C, Saini HK, Davis MP, Schaeffer J, et al. Extracellular vesicles from neural stem cells transfer IFN-y via IfngrI to activate Start1 signaling in target cells. Mol Cell. 2014;56(2):193-204.

147 Hao S, Ye Z, Li F, Meng Q, Qureshi M, Yang J, Xiang J. Epigenetic transfer of metastatic activity by uptake of highly metastatic B16 melanoma cell-released exosomes. Exp Oncol. 2006;28(2):126-131.

148 Peinado H, Aleekovi6 M, Lavotshkin S, Matei I, Costa-Silva B, Moreno-Bueno G, Hergueta-Redondo M, Williams C, Garcia- Santos G, Ghajar C, et al. Melanoma exosomes educate bone marrow progenitor cells toward a pro-metastatic phenotype through MET. Nat Med. 2012;18(6):883-891.

77 149 Yoo JW, Irvine DJ, Discher DE, Mitragotri S. Bio-inspired, bioengineered and biomimetic drug delivery carriers. Nat Rev Drug Discov. 2011;10(7):521-535. 150 Weingart J, Vabbilisetty P, Sun XL. Membrane mimetic surface functionalization of nanoparticles: methods and applications. Adv Colloid Interface Sci. 2013; 197-198:68-84. 151 Meyer RA, Sunshine JC, Green JJ. Biomimetic particles as therapeutics. Trends Biotechnol. 2015;33(9):514-524. 152 Alvarez-Erviti L, Seow Y, Yin H, Betts C, Lakhal S, Wood MJA. Delivery of siRNA to the mouse brain by systemic injection of targeted exosomes. Nat Biotechnol. 2011;29(4):341-345. 153 Tian Y, Li S, Song J, Ji T, Zhu M, Anderson GJ, Wei J, Nie G. A doxorubicin delivery platform using engineered natural membrane vesicle exosomes for targeted tumor therapy. Biomaterials. 2014;35(7):2383-2390. 154 Wiklander OPB, Nordin JZ, O'Loughlin A, Gustafsson Y, Corso G, Mager I, Vader P, Lee Y, Sork H, Seow Y, et al. Extracellular vesicle in vivo biodistribution is determined by cell source, route of administration and targeting. J Extracell Vesicles. 2015;4:26316. 155 Qi H, Liu C, Long L, Ren Y, Zhang S, Chang X, Qian X, Jia H, Zhao J, Sun J, et al. Blood exosomes endowed with magnetic and targeting properties for cancer therapy. ACS Nano. 2016;10(3):3323-3333. 156 Kooijmans SAA, Vader P, van Dommelen SM, van Solinge WW, Schiffelers RM. Exosome mimetics: a novel class of drug delivery systems. Int J Nanomedicine. 2012;7:1525-1541. 157 Jang SC, Kim OY, Yoon CM, Choi DS, Roh TY, Park J, Nilsson J, LJ6tvall J, Kim YK, Gho YS. Bioinspired exosome-mimetic nanovesicles for targeted delivery of chemotherapeutics to malignant tumors. ACS Nano. 2013;7(9):7698-77 10. 158 Peng LH, Zhang YH, Han J, Zhang CZ, Wu JH, Wang XR, Gao JQ, Mao ZW. Cell membrane capsules for encapsulation of chemotherapeutic and cancer cell targeting in vivo. ACS Appl Mater Interfaces. 2015;7(33):18628-18637. 159 Gao J, Chu D, Wang Z. Cell membrane-formed nanovesicles for disease-targeted delivery. J Control Release. 2016;224:208- 216. 160 Toledano Furman NE, Lupu-Haber Y, Bronshtein T, Kaneti L, Letko N, Weinstein E, Baruch L, Machluf M. Reconstructed stem cell nanoghosts: a natural tumor targeting platform. Nano Lett. 2013;13(7):3248-3255. 161 Jo W, Jeong D, Kim J, Cho S, Jang SC, Han C, Kang JY, Gho YS, Park J. Microfluidic fabrication of cell-derived nanovesicles as endogenous RNA carriers. Lab Chip. 2014; 14(7):1261-1269. 162 Lunavat TR, Jang SC, Nilsson L, Park HT, Repiska G, hasser C, Nilsson JA, Gho YS, L6tvall J. RNAi delivery by exosome- mimetic nanovesicles-implications for targeting c-Myc in cancer. Biomaterials. 2016;102:231-238. 163 Kaneti L, Bronshtein T, Dayan NM, Kovregina I, Khait NL, Lupu-Haber Y, Fliman M, Schoen BW, Kaneti G, Machluf M. Nanoghosts as a novel natural nonviral gene delivery platform safely targeting multiple cancers. Nano Lett. 2016;16(3):1574- 1582. 164 Jeong D, Jo W, Yoon J, Kim J, Gianchandani S, Gho YS, Park J. Nanovesicles engineered from ES cells for enhanced cell proliferation. Biomaterials. 2014;35(34):9302-9310. 165 Oh K, Kim SR, Kim DK, Seo MW, Lee C, Lee HM, Oh JE, Cho EY, Lee DS, Gho YS, et al. In vivo differentiation of therapeutic insulin-producing cells from bone marrow cells via extracellular vesicle-mimetic nanovesicles. ACS Nano. 2015;9(12):11718-11727. 166 Ding Q, Chen J, Wei X, Sun W, Mai J, Yang Y, Xu Y. RAFTsomes containing epitope-MHC-II complexes mediated CD4+ T cell activation and antigen-specific immune responses. Pharm Res. 2013;30(1):60-69. 167 Molinaro R, Corbo C, Martinez JO, Taraballi F, Evangelopoulos M, Minardi S, Yazdi IK, Zhao P, de Rosa E, Sherman ME, et al. Biomimetic proteolipid vesicles for targeting inflamed tissues. Nat Mater. 2016; 15(9):1037-1046. 168 Moonschi FH, Effinger AK, Zhang X, Martin WE, Fox AM, Heidary DK, DeRouchey JE, Richards CI. Cell-derived vesicles for single-molecule imaging of membrane proteins. Angew Chem Int Ed Engl. 2015;54(2):481-484. 169 Smyth T, Petrova K, Payton NM, Persaud I, Redzic JS, Graner MW, Smith-Jones P, Anchordoquy TJ. Surface functionalization of exosomes using click chemistry. Bioconjug Chem. 2014;25(10):1777-1784. 170 Morishita M, Takahashi Y, Matsumoto A, Nishikawa M, Takakura Y. Exosome-based tumor antigens-adjuvant co-delivery utilizing genetically engineered tumor cell-derived exosomes with immunostimulatory CpG DNA. Biomaterials. 2016; 111:55- 65.

78 171 Delcayre A, Estelles A, Sperinde J, Roulon T, Paz P, Aguilar B, Villanueva J, Khine S, Le Pecq JB. Exosome display technology: applications to the development of new diagnostics and therapeutics. Blood Cells Mol Dis. 2005;35(2):158-168.

172 Lee J, Lee H, Goh U, Kim J, Jeong M, Lee J, Park JH. Cellular engineering with membrane fusogenic liposomes to produce functionalized extracellular vesicles. ACS Appl Mater Interfaces. 2016;8(11):6790-6795.

173 Smart EJ, Ying YS, Mineo C, Anderson RGW. A detergent-free method for purifying caveolae membrane from tissue culture cells. Proc Natl Acad Sci USA. 1995;92(22):10104-10108.

174 Macdonald JL, Pike UJ. A simplified method for the preparation of detergent-free lipid rafts. J Lipid Res. 2005;46(5):1061- 1067.

175 Yao Y, Hong S, Zhou H, Yuan T, Zeng R, Liao K. The differential protein and lipid compositions of noncaveolar lipid microdomains and caveolae. Cell Res. 2009; 19(4):497-506.

176 Fortin JP, Zhu Y, Choi C, Beinborn M, Nitabach MN, Kopin AS. Membrane-tethered ligands are effective probes for exploring class BI G protein-coupled receptor function. Proc Natl Acad Sci USA. 2009; 106(19):8049-8054.

177 Fortin JP, Schroeder JC, Zhu Y, Beinborn M, Kopin AS. Pharmacological characterization of human incretin receptor missense variants. J Pharmacol Exp Ther. 2010;332(1):274-280.

178 Choi C, Fortin JP, McCarthy E, Oksman L, Kopin AS, Nitabach MN. Cellular dissection of circadian peptide signals with genetically encoded membrane-tethered ligands. Curr Biol. 2009;19(14):1167-1175.

179 Fortin JP, Chinnapen D, Beinborn M, Lencer W, Kopin AS. Discovery of dual-action membrane-anchored modulators of incretin receptors. PLoS One. 2011;6(9):e24693.

180 Harwood BN, Fortin JP, Gao K, Chen C, Beinborn M, Kopin AS. Membrane tethered bursicon constructs as heterodimeric modulators of the Drosophila G protein-coupled receptor rickets. Mol Pharmacol. 2013;83(4):814-821.

181 Ohno S, Takanashi M, Sudo K, Ueda S, Ishikawa A, Matsuyama N, Fujita K, Mizutani T, Ohgi T, Ochiya T, et al. Systemically injected exosomes targeted to EGFR deliver antitumor microRNA to breast cancer cells. Mol Ther. 2013;21(1):185-191.

182 Sarkar D, Vemula PK, Teo GSL, Spelke D, Karnik R, Wee LY, Karp JM. Chemical engineering of mesenchymal stem cells to induce a cell rolling response. Bioconjugate Chem. 2008;19(11):2105-2109.

183 Sarkar D, Vemula PK, Zhao W, Gupta A, Karnik R, Karp JM. Engineered mesenchymal stem cells with self-assembled vesicles for systemic cell targeting. Biomaterials. 2010;31(19):5266-5274.

184 Zhao W, Loh W, Droujinine IA, Teo W, Kumar N, Schafer S, Cui CH, Zhang L, Sarkar D, Karnik R, et al. Mimicking the inflammatory cell adhesion cascade by nucleic acid aptamer programmed cell-cell interactions. FASEB J. 2011;25(9):3045- 3056.

185 Sarkar D, Spencer JA, Phillips JA, Zhao W, Schafer S, Spelke DP, Mortensen LJ, Ruiz JP, Vemula PK, Sridharan R, et al. Engineered cell homing. Blood. 2011;1 18(25):e184-e191.

186 Levy 0, Zhao W, Mortensen LJ, LeBlanc S, Tsang K, Fu M, Phillips JA, Sagar V, Anandakumaran P, Ngai J, et al. mRNA- engineered mesenchymal stem cells for targeted delivery of interleukin-10 to sites of inflammation. Blood. 2013;122(14):e23- e32.

187 Xiong J, Miller VM, Hunter LW, Li Y. Leukocyte- and platelet-derived microvesicle interactions following in vitro and in vivo activation of Toll-like receptor 4 by lipopolysaccharide. PLoS One. 201 1;6(9):e25504.

188 Moore T, Sorokulova I, Pustovyy 0, Globa L, Vodyanoy V. Microscopic evaluation of vesicles shed by rat erythrocytes at elevated temperatures. J Therm Biol. 2013;38(8):487-492.

189 Im H, Shao H, Park YI, Peterson VM, Castro CM, Weissleder R, Lee H. Label-free detection and molecular profiling of exosomes with a nano-plasmonic sensor. Nat Biotechnol. 2014;32(5):490-495.

190 Zhu L, Wang K, Cui J, Liu H, Bu X, Ma H, Wang W, Gong H, Lausted C, Hood L, et al. Label-free quantitative detection of tumor-derived exosomes through surface plasmon resonance imaging. Anal Chem. 2014;86(17):8857-8864.

191 Sina AAI, Vaidyanathan R, Dey S, Carrascosa LG, Shiddiky MJA, Trau M. Real time and label free profiling of clinically relevant exosomes. Sci Rep. 2016;6:30460.

192 Kirouac DC, Zandstra PW. The systematic production of cells for cell therapies. Cell Stem Cell. 2008;3(4):369-38 1.

79 193 Rekhi R, Smith JA, Arshad Z, Roberts M, Bountra C, Bingham I, Bure K, Brindley DA. Decision-support tools for monoclonal antibody and cell therapy bioprocessing: current landscape and development opportunities. Bioprocess Int. 2015;13(11):30,32-39. 194 Kodiyalam S, Yang RJ, Gu L, Tho CH. Multidisciplinary design optimization of a vehicle system in a scalable, high performance computing environment. Struct Multidiscipl Optim. 2004;26(3):256-263. 195 Schmidt FR. Optimization and scale up of industrial fermentation processes. Appl Microbiol Biotechnol. 2005;68(4):425-435. 196 Tan TE, Peh GSL, George BL, Cajucom-Uy HY, Dong D, Finkelstein EA, Mehta JS. A cost-minimization analysis of tissue- engineered constructs for corneal endothelial transplantation. PLoS One. 2014;9(6):e100563. 197 Malik N. Allogeneic versus autologous stem-cell therapy. Biopharm Int. 2012;25(7). 198 Simaria AS, Hassan S, Varadaraju H, Rowley J, Warren K, Vanek P, Farid SS. Allogeneic cell therapy bioprocess economics and optimization: single-use cell expansion technologies. Biotechnol Bioeng. 2014;11 1(1):69-83. 199 Hassan S, Simaria AS, Varadaraju H, Gupta S, Warren K, Farid SS. Allogeneic cell therapy bioprocess economics and optimization: downstream processing decisions. Regen Med. 2015;10(5):591-609. 200 McCall MJ, Williams DJ. Developing cell therapies: enabling cost prediction by value systems modeling to manage developmental risk. J Commer Biotechnol. 2013;19(2):27-39. 201 Ungrin MD, Clarke G, Yin T, Niebrugge S, Nostro MC, Sarangi F, Wood G, Keller G, Zandstra PW. Rational bioprocess design for human pluripotent stem cell expansion and endoderm differentiation based on cellular dynamics. Biotechnol Bioeng. 2012;109(4):853-866. 202 Langer ES, Rader RA. Single-use technologies in biopharmaceutical manufacturing: a 10-year review of trends and the future. Eng Life Sci. 2014;14(3):238-243. 203 Shukla AA, Gottschalk U. Single-use disposable technologies for biopharmaceutical manufacturing. Trends Biotechnol. 2013;31(3):147-154. 204 Rowley J, Abraham E, Campbell A, Brandwein H, Oh S. Meeting lot-size challenges of manufacturing adherent cells for therapy. Bioprocess Int. 2012;10(3):16,18-22. 205 Brandenberger R, Burger S, Campbell A, Fong T, Lapinskas E, Rowley JA. Cell therapy bioprocessing: integrating process and product development for the next generation of biotherapeutics. Bioprocess Int. 201 1;9(ls):3 0-3 2 ,3 4,36-3 7. 206 Hahn TJ, Courbron D, Hamer M, Masoud M, Wong J, Taylor K, Hatch J, Sowers M, Shane E, Nathan M, et al. Rapid manufacture and release of a GMP batch of avian influenza A (H7N9) virus-like particle vaccine made using recombinant baculovirus-Sf9 insect cell culture technology. Bioprocessing J. 2013; 12(2): 1-10. 207 Bisschops M, Frick L, Levison P. Continuous chromatography is now possible for clinical manufacturing. Bioprocess Int. 2016;14(6s): 26-2 9. 208 Watson DC, Bayik D, Srivatsan A, Bergamaschi C, Valentin A, Niu G, Bear J, Monninger M, Sun M, Morales-Kastresana A, et al. Efficient production and enhanced tumor delivery of engineered extracellular vesicles. Biomaterials. 2016; 105:195-205. 209 Godawat R, Konstantinov K, Rohani M, Warikoo V. End-to-end integrated fully continuous production of recombinant monoclonal antibodies. J Biotechnnol. 2015;213:13-19. 210 Dong J, Migliore N, Mehrman SJ, Cunningham J, Lewis Mj, Hu P. High-throughput, automated Protein A purification platform with multiattribute LC-MS analysis for advanced cell culture process monitoring. Anal Chem. 2016;88(17):8673- 8679. 211 Davie NL, Brindley DA. Streamlining cell therapy manufacture. Bioprocess Int. 2012; 10(3s): 24-26, 2 8 ,49. 212 Campbell A, Brieva T, Raviv L, Rowley J, Niss K, Brandwein H, Oh S, Karnieli 0. Process development considerations for cell therapy. Stem Cells Transl Med. 2015;4(10):1155-1163. 213 Rosenl6cher J, Sandig G, Kannicht C, Blanchard V, Reinke SO, Hinderlich S. Recombinant glycoproteins: the impact of cell lines and culture conditions on the generation of protein species. J Proteomics. 2016;134:85-92. 214 Ivarsson M, Villiger TK, Morbidelli M, Soos M. Evaluating the impact of cell culture process parameters on monoclonal antibody N-glycosylation. J Biotechnol. 2014; 188:88-96. 215 Spice P, Yip-Hoi D, Koren Y. Scalable reconfigurable equipment design principles. Int J Prod Res. 2005;43(22):4839-4852.

80 216 Besse B, Charrier M, Lapierre V, Dansin E, Lantz 0, Planchard D, Le Chevalier T, Livartoski A, Barlesi F, Laplanche A, et al. Dendritic cell-derived exosomes as maintenance immunotherapy after first time chemotherapy in NSCLC. Oncoimmunology. 2016;5(4):e1071008.

217 Chowdary BV, George D. Improvement of manufacturing operations at a pharmaceutical company: a lean manufacturing approach. J Manuf Technol Manag. 2012;23(1):56-75. 218 Moran N. Public sector seeks to bridge 'valley of death'. Nat Biotechnol. 2007;25(3):266.

219 Neuman D, Chandhok JN. Patent watch: nanomedicine patents highlight importance of production methods. Nat Rev Drug Discov. 2016;15(7):448-449. x