Extracellular vesicles: Roles and applications in drug-induced liver injury

Abstract

Extracellular vesicles (EV) are defined as nanosized particles, with a lipid bilayer, that are unable to replicate. There has been an exponential increase of research investigating these particles in a wide array of diseases and deleterious states (inflammation, oxidative stress, drug-induced liver injury) in large part due to increasing recognition of the functional capacity of EVs. Cells can package lipids, proteins, miRNAs, DNA, and RNA into EVs and send these discrete packages of molecular information to distant, recipient cells to alter the physiological state of that cell. EVs are innately heterogeneous as a result of the diverse molecular pathways that are used to generate them. However, this innate heterogeneity of EVs is amplified due to the diversity in isolation techniques and lack of standardized nomenclature in the literature making it unclear if one scientist's “exosome” is another scientist's “microvesicle.” One goal of this chapter is to provide the contextual understanding of EV origin so one can discern between divergent nomenclature. Further, the chapter will explore the potential protective and harmful roles that EVs play in DILI, and the potential of EVs and their cargo as a biomarker. The use of EVs as a therapeutic as well as a vector for therapeutic delivery will be discussed.

1. Introduction

Extracellular vesicles (EV) consist of a heterogeneous population of nanosized particles, enclosed in a lipid bilayer that lack a functional nucleus [1]. There has been a surge of research into EVs ranging from their role in diseases [1–3], utility as biomarkers [4,5], and potential as a nanosized delivery system for therapeutics [6,7]. The heterogeneity of the cargo loaded inside EVs is now being fully appreciated as a variety of omic approaches: lipidomics, proteomics, and transcriptomics, including miRNA and non-coding RNA, are being applied to different cell types under varying experimental conditions. The magnitude of this challenge is further amplified because the same cell type may secrete EVs with differential cargo depending on the experimental condition or if the EV release is occurring on the basolateral or apical face of the cell [8,9]. This issue of cell polarity is relevant in studies of drug-induced liver injury (DILI), since the polarization of hepatocytes could strongly influence EV release, as the apical side faces the bile and the basolateral domain releases into the interstitial space and ultimately systemic circulation [2]. Many online databases have been created in an attempt to synthesize this information such as EV-TRACK (www.evtrack.org), ExoCarta (www.exocarta.org) and Vesiclepedia (www.microvesicles.org/) which serve as valuable resources for the nascent EV researcher.

In physiological and pathological conditions many different cell types secrete EV's as an integral mode of communication delivering a variety of bioactive cargo to recipient cells [10]. A diverse array of cargo ranging from proteins, lipids, carbohydrates, and nucleic acids can be packaged into an EV and can serve as a bolus dose of macromolecule constituents to the target cell [1]. Delivery of this package to the recipient cell has a functional consequence, mRNA cargo from an EV has been successfully transcribed into proteins by the recipient cell [11–13]. Moreover, EVs have been isolated from a variety of biological fluids, ranging from breast milk [14] and amniotic fluid [15] to blood [16] and bile [17]. Further, by enclosing the cargo in a lipid membrane the contents are shielded from the extracellular environment and exhibit greater stability [1] compared to soluble cellular mediators. These qualities of ubiquity and stability have made EVs a promising source for identifying novel biomarkers in DILI. One promising biomarker is miR-122 which is elevated in the serum and inside circulating EVs in models of alcoholic liver disease and acetaminophen (APAP)-induced liver injury [18]. However, EV miR-122 more closely correlated with liver injury than free miR-122 [19] and EV miRNAs such as miR-122 have been shown to be more stable than non-EV associated miRNAs [20]. The enhanced stability of miR-122 packaged into an EV in conjunction with the relatively non-invasive nature for blood collection from human patients makes this a promising target. Notably, any potential biomarkers require the isolation process to have high fidelity and reproducibility for it to have any clinical value. Research in this area is in its infancy and an aim of this review is to highlight the primary benefits and limitations of EV-associated biomarkers.

Many studies have investigated the role of EVs in disease, however, the importance for EVs in DILI has received increasing scrutiny. Drug-induced liver injury is the unanticipated damage by drugs or the resulting metabolites to hepatocytes and other non-parenchymal cells in the liver [21]. DILI is the most common cause of acute liver injury and APAP accounts for nearly 50% of acute liver failures in the United States [22]. The liver is the primary site for drug-induced injury due to its crucial role in concentrating and metabolizing xenobiotics. Susceptibility to DILI depends on an individual's genetics and the environment. For example, chronic alcohol consumption can decrease glutathione (GSH), the primary antioxidant defense system of the cell, increasing sensitivity to injury [23]. The role that EVs play in propagating and/or ameliorating injury has been investigated in a variety of hepatotoxic agents such as APAP, alcohol, and polycyclic aromatic hydrocarbons (PAHs). The primary purpose of these studies includes (1) characterization of cargo packaged into EVs (2) the functional effect of EVs on recipient cells (3) identification of novel biomarkers with superior efficacy (4) exploring the potential for EVs to facilitate recovery or regeneration after injury induction.

Idiosyncratic drug-induced liver injury (iDILI) has also become a focal area of research within academia and the pharmaceutical industry. Idiosyncratic drug reactions that lead to severe patient injuries or death in clinical trials or after FDA approval of the drug can cost companies billions of dollars in potential and lost revenue [21]. Evidence suggests that the primary modality of cell death in iDILI is due to dysregulated activation of the innate and adaptive immune system. This activation can lead to death receptor (DR)—mediated apoptosis [24] which can result in severe liver injury or death. This phenomenon represents an interesting intersectionality with ongoing research in EV biology because it has been demonstrated that EV-mediated delivery of death ligands to target cells result in a greater functional effect (e.g., induction of apoptosis), than that of the soluble death ligand alone [25,26]. Little research has been done in this area due to the nature of idiosyncratic injury and the lack of animal models.

Notably, the role of EVs in drug-induced injury targeting other organs is being actively explored with a focus on novel biomarker discovery [27,28] and the use of EVs as a therapeutic [29,30]. EVs have been studied in models of drug-induced kidney [28,31], muscle [28], intestine [32], and lung injury [29]. It has been demonstrated in models of acute lung injury that EVs attenuated injury [29,33], however, in a cadmium-induced nephrotoxicity model EVs contributed to injury [31]. While this may be attributable to the divergent models it also highlights the importance of defining the source of the EVs in one's model of interest. EVs are best thought of as a temporal snapshot of the cell, therefore, the function of an EV is a product of the physiological state of the cell and the parental cell source.

2. Extracellular vesicle nomenclature

EVs have been classically classified into three discrete categories depending on their size and origin pathway: exosomes, microvesicles, and apoptotic bodies (Table 1). As research and technology has expanded a new type of nanoparticle has been identified which has been given the moniker “exomere.” These exomeres are very small, approximately 35 nanometers (nm), cell particles that lack a lipid membrane which exhibit a distinct proteomic profile compared to larger EVs, however, more research is needed to uncover the origins of this cellular particle [45]. Nevertheless, the lion's share of scientific interest has been targeted toward exosomes which are derived from the endosomal network. The term “exosome” was first used in reference to any vesicles released from cultured cells that varied from 40 to 1000 nm [46], however, as the nomenclature has evolved it is now accepted that an exosome is an intraluminal vesicle (ILV) of multivesicular bodies (MVB) that is trafficked to the cell membrane and released into the extracellular environment (Fig. 1) [1]. Generally, the diameter of an exosome is 40–150 nm [10,47] and it has been speculated that the size of the exosome is likely constrained by their intraorganellular packaging in MVBs [48]. Microvesicles are much larger ranging from 100 to 1000 nm and are produced by the outward budding and fission of the plasma membrane [41]. Apoptotic bodies are a byproduct of apoptosis and differ from other EVs because they are not thought to be a regulated form of cell to cell communication. They also are larger ranging from 500 nm to 5 μm in diameter as they can consist of large cellular fragments that can be nearly equivalent to the size of a cell [34,49]. Studies interested in characterizing EV subpopulations usually do so based on size in the context of the isolation technique used, for example, large EVs pelleted at low speed (~ 2000 g), medium EVs (~ 10,000 g), and small EVs (sEVs) pelleted at high speed (~ 100,000 g), as opposed to an exosome or microvesicle, because discriminating between these EV subtypes can only be achieved if the genesis of the EV is witnessed through microscopy [50]. Proteomic analysis utilizing the aforementioned framework have revealed unique proteome profiles for these different EV subpopulations, although there is a large overlap in biochemical markers between categories [35]. This proteomic characterization revealed that many putative exosome markers such as the tetraspanins (CD63, CD9), flotillin and heat-shock proteins [34] were found in both large and sEV subpopulations demonstrating that there is not a single, “gold standard” biochemical marker that can be used to distinguish between exosomes and microvesicle subpopulations [35,50]. Moreover, they identified a subpopulation of sEVs (vesicles pelleted at 100,000 g) that were not derived from the endosome network, meaning they do not correspond with the canonical definition of an exosome [35]; which is in contrast to groups that have used the term “exosome” to refer to any vesicle pelleted at 100,000 g [36,51]. It is important to understand this to properly interpret the landscape for EV research, particularly for research involving DILI, as a primary goal is to delineate sensitive biomarkers. Nevertheless, the term extracellular vesicle will be used throughout this review to refer to both exosomes and microvesicles, as any further delineation beyond EV has to be carefully applied, as some authors have used it to mean an EV of a certain size, or others as means to denote de novo origin. However, short of catching the synthesis of the EV in the act it is extremely difficult to assign a biogenesis pathway [50]. While there is no currently agreed upon consensus for gold standard markers, many groups are working on this issue and evidence suggests that it will require a combination of markers to define an increasingly specific subcategory (e.g., sEVs CD63/CD81 negative, CD9 positive [35]). One can appreciate that determining the functional properties of different EV subpopulations is exponentially more difficult than the characterization and until these subpopulations can be rigorously defined, studies purporting a functional benefit and assigning it to be the result of exosomes or microvesicles need to be carefully interpreted. Therefore, throughout this review, tables summarizing the experimental design, the isolation methodology and key findings of a study are placed at the end of select sections.

Table 1.

Physical characteristics of EVs and commonly co-isolated contaminants.

	Diameter (nm)	Density (g/mL)	Shape	Sedimentation (g)	Classic markers	References [#]
Exosomes	40–150	1.13–1.19	Uniform spherea	100,000–200,000	TSG101, ALIX, tetraspanins (CD63)	[34–37]
Microvesicles	100–1000	1.02–1.20	Asymmetric/heterogeneous	10,000–12,000	Integrins, selectins, ARRDC1b	[38–40]
Apoptotic bodies	500–5000	1.16–1.28	Heterogeneous, dense	4000–12,000c	Phosphatidylserinehistones	[34,37,41]
LDL	24–28	1.01–1.06	Spherical	200,000	apoB-100	[42,43]
HDL	7–12	1.06–1.20	Spherical	200,000	A-II	[42,43]
Albumin	3.8	N/A	Ellipsoid	100,000–200,000	N/A	[44]

LDL and HDL measured by flotation in ultracentrifuge at 200,000 g for 1 h. Shape is determined by cryo-EM or NTA. Size is assessed by transmission electron microscopy (TEM). Density is determined from sucrose gradient. Sedimentation is determined by ultracentrifugation.

^aExosomes have been reported as having a “cup” shape, however, this is due to the exosome collapsing during the drying process.

^bARRDC1 is present on microvesicles collected from 120,000 g pellet.

^c100,000 g has isolated apoptotic bodies from UV-treated cells [37].

Fig. 1.

Extracellular vesicles: biogenesis, transport, and release. A progenitor structure is first formed from the inward budding of the plasma membrane termed an early endosome. As the early endosome matures, intralumenal vesicles (ILVs) form driven by both ESCRT-dependent and ESCRT-independent (tetraspanins/lipids) mechanisms giving rise to a multivesicular body (MVB). These mechanisms dictating the formation of ILVs are also responsible for sorting and loading cargo (proteins, lipids, RNA, miRNA). The MVB can go to the lysosome for degradation or it can migrate to the plasma membrane via Rab proteins. SNARES, Rab GTPases and a rise in intracellular calcium help facilitate docking and fusion of the MVB to the plasma membrane. Once the ILVs are released into the extracellular environment they are now termed exosomes. Tetraspanin enriched membrane domains (TEMs), indicated by a triangle, help facilitate vesicular uptake and are highly concentrated on both the surface of exosomes and target cells. Microvesicles (MVs) form from the outward budding of the plasma membrane and are controlled through similar mechanisms as exosome formation. Unique mechanisms involved in MV formation, such as AARDC1, have been described. The overlap in the systems governing exosome and microvesicle formation highlight the difficulty in determining ex vivo the biogenesis pathway of the EV.

3. Extracellular vesicle biogenesis

Exosomes and microvesicles are a regulated, programmed form of intracellular communication [1,10,47], distinctly distinguishing them from apoptotic bodies. In support of this notion, exosomes and microvesicles have well-defined biogenesis pathways [10,47] while the formation of apoptotic bodies seem to be a necessary byproduct of apoptosis [34]. Correspondingly, the exosome and microvesicle synthesis pathways will be explored in depth as it will provide the contextual reasoning for the classic exosome markers (tetraspanins (CD63), ALIX, TSG101) or microvesicles (phosphatidylserine, integrins, selectins) and why these markers are limited in their efficacy.

3.1. Exosome biogenesis

Before exosome synthesis can be initiated, a progenitor structure, termed an early endosome, must first be formed [10]. The canonical role of the early endosome is to sort proteins and lipids and export them to other locations within the cell such as the trans-Golgi network [52]. The early endosome forms from the inward budding of the plasma membrane eventually pinching away forming an independent structure [10]. As the early endosome matures it can transform into what is termed a multivesicular body (MVB). This MVB is comprised of intralumenal vesicles (ILVs) that form from invaginations of the limiting membrane, lending it an internal topology of discrete vesicular structures [53]. These MVBs have several fates and the mechanism by which the MVB chooses a path is poorly understood [54,55]. Nevertheless, the MVB can fuse with the plasma membrane of the cell, releasing the ILVs into the extracellular environment at which time the ILVs are now termed exosomes [47,48,53,56,57]. Exosome biogenesis can be mechanistically binned into two pathways: ESCRT-dependent or ESCRT-independent pathways (Endosomal Sorting Complex Required for Transport) [53,58]. Lastly, studies done prior to ~ 2008 often used visualization techniques, such as electron microscopy or confocal microscopy in conjunction with fluorescent or Immuno-Gold labeling to study MVBs and the generation of ILVs irrespective of actual ILV release into the extracellular space (e.g., exosomes), studies done after ~ 2008 have a greater interest in exosomes (ILV after release into EC environment) and the role they may be playing in cell to cell communication. It is these latter studies that were interested explicitly in “exosomes” (EVs) where a concerted effort is used to discuss the “exosomes” in the context of the isolation methodology and characterization.

3.1.1. The ESCRT-dependent pathway of exosome formation

The ESCRT system is comprised predominately of five protein complexes, ESCRT-0, ESCRT-I, ESCRT-II, ESCRT-III, and the Vps4 (Vacuolar protein sorting 4) complex which facilitate MVB formation, cargo selection (e.g., proteins, lipids, RNA) and recycling of the ESCRT components (Table 2) [53,63]. ESCRT-0, I and II are responsible for cargo sequestration [54–56], ESCRT I and II induce bud formation, ESCRT-III orchestrates vesicle release, and the ATPase Vps4 dissociates ESCRT-III subunits back into its monomeric, soluble forms [53,62]. In general, proteins are tagged with a single ubiquitin modifier that allows for recognition by the ESCRTs allowing for proper sorting into budding ILVs [64], while the addition of multiple ubiquitin subunits can enhance efficiency [65], however, other ubiquitin-independent signals do exist [66]. Hepatocyte growth factor-regulated tyrosine kinase substrate (HRS) is an adaptor protein of the ESCRT-0 protein complex (Table 2) which is recruited to the membrane of the MVB-containing phosphatidylinositol 3-phosphate (PtdIns3P) and through a ubiquitin-binding motif recognizes and sorts cargo [67]. HRS then recruits TSG101 of the ESCRT-I complex, and either through ESCRT-II or ALIX, ESCRT-III is recruited, ultimately resulting in vesicle scission before dissociation and recycling by Vps4 [60].

Table 2.

ESCRT-dependent EV biogenesis.

Protein complex	Protein name	General function	EV marker	Biological impact when inhibited	References [#]
ESCRT-0		Cargo sequestration			[55]
	HRS	Binds to and sorts ubiquitinated cargo	ALIX/Flotillin CD63/MHCII/HSC70 or CD81	Reduced intraluminal transport of EGF	[59]
				Decrease in EVs between 50 and 200 nm	[60]
	STAM1	Cargo sorting	CD63/MHCII/HSC70 or CD81	Decrease in EVs less than 50 nm Impaired sorting of MHCII and CD63 in ILV	[60]
ESCRT-I		Bud formation, cargo sequestration			[53,55]
	TSG101	Membrane budding, recognizes ubiquitin	CD63/MHCII/HSC70 or CD81	MHCII and CD63 sorting in ILV is impaired	[60]   [59]
			ALIX/Flotillin	Reduced IL transport of EGF
ESCRT-II		Bud formation, cargo sequestration			[53,55]
	VPS22	Cargo sorting	Syntenin 1/CD63a	Reduced syndecan-CTF and syntenin in exosomes	[61]
ESCRT-III		Vesicle release			[62]
	CHMP4	Membrane scission	Syntenin 1/CD63a	Reduced syndecan-CTF, CD63 and syntenin in exosomes	[61]
Accessory proteins
	VPS4	Dissociates and recycles ESCRT machinery	Syntenin 1/CD63a ALIX/Flotillin	Reduced syntenin and CD63 in exosomes	[61]   [59]
No effect on PLP sorting
	ALIX	Promotes bud formation and abscission	Syntenin 1/CD63a CD63/MHCII/HSC70 or CD81	50% reduction in EVs ranging from 30 to 110 nm Decrease syndecan-CTF, CD63, HSP70 in exosomes MHCII sorting impaired	[61]   [60]
				Decrease syndecan-CTF, CD63, HSP70 in exosomes MHCII sorting impaired

EVs were isolated by differential ultracentrifugation with the 100,000 g pellet containing the EVs that were analyzed unless otherwise indicated.

^aEVs were collected from the 140,000 g pellet.

Early work demonstrated that small interfering RNA (siRNA) targeting HRS resulted in a decreased amount of ILVs [68] while more recent work using short hairpin RNA (shRNA) resulted in reduced EV release in a variety of cell lines [60,69,70]. This RNA-interference (RNAi) strategy has been employed to exhaustively assay many putative molecular constituents of exosome biogenesis, such as TSG101 and ALIX, which resulted in a decrease in EV number [53,61,71]. Specifically, Colombo and colleagues used an RNAi strategy targeting 23 different ESCRT proteins in HeLa-CIITA cells [60]. The EVs were collected via ultracentrifugation at 100,000 g and were classified as “exosomes” due to the presence of the tetraspanin CD63, MHC II and either HSC70 or tetraspanin CD81. The silencing of only 4 of the 23 ESCRT proteins assessed consistently resulted in perturbations in exosome size and release, these were HRS and STAM1 (ESCRT-0), TSG101 (ESCRT-1) and Vps4B. Interestingly, the depletion of HRS decreased the number of EVs between 50 and 200 nm compared to control cells without altering the number of EVs smaller than 50 nm, while the loss of STAM1 only decreased the number of EVs less than 50 nm in diameter. These results indicate that different sized vesicles may be generated through varying molecular mechanisms and supports the notion that EVs should be characterized based on their physical characteristics as opposed to their origin pathway. Notably, knockdown of TSG101 and STAM1 resulted in an increase in vesicles lacking both CD63 and MHC II, suggesting that these ESCRT proteins are required for sorting CD63 and MHC II into ILVs [60]. In support of this notion, it was shown that when four major components of the ESCRT system were silenced, epidermal growth factor receptor (EGFR) was not properly sorted through the MVB and EGFR degradation was strongly inhibited [58] which was also confirmed in a separate study [59]. Based in part on the aforementioned data, it has been suggested that ESCRT proteins are necessary for cargo sorting and maturation of the endosome while other ESCRT-independent mechanisms facilitate ILV formation [64]. Conversely, different cell types could have a unique molecular consortium of cargo loaded into MVBs and perhaps different MVBs have different compositions in lipid content resulting in altered mechanisms for EV biogenesis.

3.1.2. The ESCRT-independent pathway of exosome formation

Numerous studies demonstrated that silencing of the various ESCRT components could reduce the number of EVs, but not completely abrogate EV genesis. An early hypothesis was that this was due to either the incomplete knockdown of ESCRT proteins or due to inherent compensatory mechanisms (redundancy) built into the system. To address this, simultaneous siRNA knockdown of HRS (ESCRT-0), TSG101 (ESCRT-I), Vps22 (ESCRT-II), and Vps24 (ESCRT-III) in HEp-2 cells was employed and the resulting ILVs were assessed via electron microscopy. They found enlarged, CD63-positive ILVs and that these distorted ILVs colocalized with lyso-biphosphatidic acid (LBPA) [58] which is enriched in ILVs, but not on the limiting membrane of MVBs [72], suggesting a potential ESCRT-independent, lipid-driven pathway (Table 3). In congruence with this observation, in an oligodendroglial cell line, EV synthesis and secretion was not dependent on the ESCRT system, but instead relied on sphingomyelinase and the lipid ceramide [59]. This group isolated EVs using ultracentrifugation at 100,000 g, which resulted in a pellet enriched in sEVs with a diameter of approximately 50–100 nm that were Alix and flotillin positive and negative for ER, Golgi and early endosomal proteins. These sEVs lipid composition were assessed by nano-electrospray ionization tandem mass spectrometry which revealed they were enriched in cholesterol, sphingolipids, ceramides and had lower amount of phosphatidylcholine (PC) compared to the total cell membrane [59]. Ceramides have been consistently found in high concentrations in sEVs [75,76] and use of a neutral sphingomyelinase inhibitor (GW4869) to prevent the formation of ceramide has effectively reduced the amount of sEVs [59], while simultaneously increasing the amount of large EVs (pelleted at 14,000 g) [77]. Ceramide has favorable biophysical properties, a small headgroup promoting inward curvature (e.g., budding) of the limiting membrane of MVBs leading to ILVs [56], however, it fails to explain how cargo is sorted into EVs. Using GW4869 impaired the trafficking of proteolipid protein (PLP) through the endosomal system without impacting EGF trafficking [59]. Conversely, when the ESCRT machinery were silenced EGF trafficking to ILVs was prevented [58,59] without impacting intraendosomal transport of PLP [59]. Taken together, these results highlight that vesicle size and cargo selection may be differentially regulated by ESCRT-dependent and ESCRT-independent pathways.

Table 3.

ESCRT-independent mechanisms facilitating EV biogenesis.

ESCRT independent	Protein/lipid name	EV marker	Biological impact when inhibited	References [#]
Lipids
	Ceramide	ALIX/Flotillin	ILV biogenesis	[59]
			Cargo sorting of PLP
			No effect on EGF
	PLD2/PA	Syntenin 1/CD63	sorting ILV Biogenesis ARF6 exosome biogenesis	[73]
Tetraspanins
	CD63	CD63-positive ILVsa	ILV Biogenesis Cargo sorting of Pmel17	[74]

^aThis study used EM and fluorescent microscopy to study ILV formation, the necessary precursor to exosome synthesis.

It also has been demonstrated in pigment-producing mammalian cells that certain cargo (Pmel17) is trafficked independently of the ESCRT system [66] as well as ceramides [74]. Tetraspanins are highly enriched in EVs [78], particularly in sEVs (pelleted at 100,000 g) [35], and have been implicated in the formation of ILVs and sEVs [79]. The tetraspanin CD63 can regulate endosomal sorting of Pmel17 and can mediate ILV biogenesis such that Pmel17 is successfully delivered to the mature endosome, while ESCRT-dependent ILVs loaded with Pmel17 is degraded [74]. Therefore, while different size EVs have differing lipid compositions which can differentially regulate cargo sorting, we also have a tetraspanin-dependent mechanism (CD63) regulating the fate of cargo. Finally, other, ESCRT-independent mechanisms have been described such as phospholipase D2 (PLD2), which is enriched in sEVs (pelleted at 100,000 g) [80], and produces phosphatidic acid (PA) which has similar biophysical properties as ceramide, whereby its integration into limiting membrane can create a budding giving rise to ILVs [73]. These results emphasize that EVs or putative exosomes are not a single, uniform group but instead heterogeneous and that EVs with different lipid compositions, sizes, and cargos may be formed, shuttled, and released by diverging control mechanisms. Lastly, this is a multidirectional system where all the discrete variables are dynamically interacting giving rise to a final, coordinated product.

3.2. MVB migration to the plasma membrane and exosome release

MVBs have several options as they mature, they can sort the cargo to the recycling endosome, the trans-Golgi network, the lysosome or they can fuse with the plasma membrane [54]. The general mechanisms facilitating mobilization, fusion and ultimately exosome release requires multiple molecular systems such as actin and microtubules in conjunction with kinesins and myosins, small Rab GTPases and the fusion machinery such as SNARES and tethering factors [10]. The Rab GTPases are critical for coordinating this array of molecular machinery and producing a successful symphony of intracellular vesicular trafficking [81]. Early proteomic analysis of sEVs (pelleted at 110,000 g) found Rab7 and Rab11 enriched in dendritic cells [37] and shortly after it was demonstrated that Rab11 was necessary for the release of HSC70-positive EVs as well as regulating sorting of the transferrin receptor (TfR) [82]. Rab11 has been shown to promote docking and fusion of MVBs to the cell membrane, however, the fusion event was inhibited by administering a calcium chelator, suggesting that Rab11-positive MVBs require an intracellular rise in calcium for successful fusion with the cell membrane [83]. Hsu et al. [84] performed a proteomic screen for Rab GTPases in sEVs (pelleted at 100 k) collected from Oli-neu cells (immortalized cell-line derived from murine oligodendrocytes) and identified multiple Rab GTPases, including Rab35 which was enriched compared to the other Rab GTPases. They showed that a constitutively active Rab35 mutant (Q67A) had increased localization at the plasma membrane and led to a twofold increase in PLP-positive sEVs compared to an inactive Rab35 mutant. In contrast, when Rab35 was depleted there was an increase in the total number of intracellular vesicles specifically in late endosomes/lysosome (MVB derivatives) which suggests that Rab35 is not critical in the biogenesis of ILVs but is mediating a downstream process. This group used total internal reflection fluorescence microscopy (TIRFM) to show that Rab35^Q67A increased vesicle immobilization at the plasma membrane suggesting that it is involved in vesicular tethering/docking [84]. (PLP is the primary component of myelin in the central nervous system and is highly expressed in Oli-neu derived sEVs) [59].

It was first demonstrated in the melanocyte that Rab27a forms a complex with an adaptor protein, melanophilin, which acts as a Rab27a effector in turn recruiting the molecular motor myosin-Va (actin-based vesicle motor) [85] which then can shuttle these vesicles to the cell periphery [86,87]. In support of this finding, a RNAi screen in human HeLa cells revealed five novel Rab GTPases that promoted the secretion of sEVs (pelleted at 100,000 g), two of which, Rab27a and Rab27b when silenced led to a reduction in CD63-positive MVBs at the plasma membrane [88]. Notably, silencing of Rab27a and Rab27b also reduced the number of sEVs without influencing the protein cargo. This suggests again that a Rab family member is mediating a process occurring after cargo loading, such as MVB transport to the periphery or subsequent fusion. Using TIRFM it was demonstrated that knockdown of Rab27a and Rab27b corresponded with a reduction in time of vesicle immobility [88]. Taken together, these results suggest that cargo sorting and ILV formation is mediated upstream through ESCRT-dependent and independent processes, and then Rab27, perhaps through its interaction with myosin-Va, shuttles and docks the MVBs at the cell periphery. Importantly, Rab27 is working in concert with effector proteins such as synaptotagmin-like protein 4 (Slp4) and Slac2 (exophilin 5), in HeLa cells [88], and synaptotagmin-like protein 2 (Slp2) in the melanocytes [86] to facilitate this intraorganellar trafficking.

The next ingredient for exosome release into the extracellular milieu is fusion of MVBs to the plasma membrane. The auspicious facilitator of the MVB-membrane fusion event are the SNARE proteins [89,90] which are controlled by Rab GTPases [81]. SNAREs have predominately been studied in mediating docking/fusion of synaptic vesicles with the presynaptic membrane of neurons and the general principle is that v-SNARES (on the vesicles) interact with t-SNARES (target compartment, in the context of exosomes, the plasma membrane) [89]. Rab27a and Rab3a putatively interact with SNARE synaptosomal-associated protein 25 (SNAP25) to mediate docking of vesicles to the plasma membrane. Intracellular calcium has been demonstrated to be critical for Rab11 mediated fusion of MVBs to the membrane [83] and synaptotagmin, a SNARE component, binds calcium [91] and phospholipids facilitating vesicle fusion [92]. There is a diversity in the docking and fusion systems which could be partially explained by the varying expression pattern for Rab proteins in endosomes at different stages of maturation. For example, Rab11 [82] and Rab35 [84] localize within recycling and sorting endosomes while Rab27a and Rab27b [88] are found within late endosomes. While it is often said that EVs/exosomes are heterogeneous it stands to reason that this heterogeneity is mediated far earlier than release at the plasma membrane. Finally, selective silencing of ESCRT proteins, ceramides, tetraspanins or Rab GTPases only partially prevents EV release indicating that this is a multi-factorial process with many complementary systems involved in the successful release of EVs.

3.3. Microvesicle biogenesis

Microvesicles (MVs) are formed by the outward budding of the plasma membrane giving rise to a 100–1000 nm diameter vesicle delimited by a lipid bilayer [41]. There is a dearth of research on microvesicles compared to exosomes that seems to be attributable to two factors: the first is that cargo loading into the MV has been purportedly thought of as a stochastic process, and secondly using the classic differential ultracentrifugation protocol (~ 3000, 14,000, 100,000 g) the pellet formed from the 14,000 g centrifugation is the putative MVs, which is far more likely to be contaminated with non-EV constituents (e.g., cell debris) [34], making it technically more feasible to study the 100,000 g pellet (e.g., sEVs or what many call “exosomes”). MVs cargo is similar to the cargo found in exosomes, ranging from lipids, proteins, and RNA/microRNA [41], with the recognition that successful separation of an MV and an exosome is extremely difficult to achieve as many isolation techniques do so based on biophysical properties of the EV (e.g., size and density), therefore a large exosome (~ 150 nm) easily could co-precipitate with a small MV. In HEK293T cells, the established microvesicle marker, arrestin domain-containing protein 1(ARRDC1) [38,93] failed to be purified from CD9 or CD81-positive EVs (suggestive of endosomal origin) when using sucrose gradient ultracentrifugation [93]. Notably, the microvesicles were collected from the 120,000 g ultracentrifugation pellet. Moreover, the larger the EV, the greater propensity for indiscriminately selected molecules to be loaded [50].

As a testament to evolutionary conservation many of the molecular systems involved in exosome biogenesis, transport, and secretion are also intimately involved in MV genesis. TSG101 (ESCRT-I) [38,93], multiple ESCRT-II and ESCRT-III proteins, ATPase Vps4 [93], and phospholipase D (PLD) [94] have been implicated in the biogenesis of plasma-derived vesicles (microvesicles) suggesting significant overlap between these systems. Nevertheless, the microvesicle origin pathway does have unique features, it was demonstrated in ARF6-positive microvesicles (ADP-ribosylation factor 6; pelleted at 10 k) that microvesicle release depends on PLD activity [94], perhaps by modulating the lipid environment promoting membrane budding [73], as well as starting a phosphorylation cascade resulting in ERK, myosin light chain kinase (MLCK), and MLC activation ultimately leading to actin and myosin facilitated contraction resulting in release of the microvesicle [94]. Nabhan et al. [38] demonstrated in HEK293T cells that TSG101 was recruited from the early endosome to the plasma membrane by ARRDC1 and with the ATPase Vps4 led to ARRDC1-positive MV (ARMMs) formation. Further, silencing of TSG101 greatly reduced the amount of AARDC1 inside the MV (pelleted at 120,000 g), alternatively, silencing of AARDC1 reduced the amount of TSG101 within the MV [38]. This suggests that cargo sorting within MVs are a controlled process and highlights an intersectionality between exosome biogenesis and MV biogenesis. Proteomic analysis of ARMMs (essentially, small microvesicles) identified many ESCRT proteins to be enriched as well as ADAM10 and NOTCH2. It was then demonstrated that NOTCH2-containing ARMMs could be delivered to a recipient cell and activated through sequential cleavage by ADAM10 and γ-secretase, resulting in increased gene expression of NOTCH2 target genes in the recipient cell [93]. This demonstrates a functional role for MVs and a potential form of long-range cell communication.

The lipid composition in the plasma membrane can promote budding [95] and it has been postulated that vertical trafficking of cargo occurs to align with the site of MV biogenesis [96]. The cargo proteins themselves can promote the necessary membrane curvature to form an MV, glycosylphosphorylinositol (GPI) GPI-anchored proteins can bend the membrane through its molecular mass [97]. Additionally asymmetric lipids (e.g., lipids that distribute asymmetrically in the inner and outer leaflet of the cell membrane), such as ceramides [98], which promote inward curvature [56], and phosphatidylethanolamine which consist of small, polar heads in proportion to large, hydrophobic tails, and gangliosides can result in membrane curvature [99]. It also had been speculated that recruitment of lipases and flippases to distinct regions of the plasma membrane can also contribute to MV formation [100]. PLD2, a lipase, has been implicated in both MV [94] and ILV biogenesis [73], where it locally increases PA, and in the former it was suggested that PA may also recruit components of the ERK pathway such as Raf to the plasma membrane.

4. EV targeting and uptake

Understanding the mechanisms that facilitate the homing of EVs to a target cell and how the cargo is delivered is critical for ensuring efficacy for EV-based delivery of therapeutics. Moreover, the role of EVs in physiological and pathological processes is contingent upon successful distribution of cargo to the recipient cell. EV cargo delivery can occur as a result of fusion with the membrane of the target cell or the entire EV can enter the cell where it can fuse with an endosomal compartment [101]. Numerous fluorescent, lipophilic dyes have been used to visualize EV delivery [102], however, it has been shown that use of these dyes can artificially increase the size of the EV [103] which may influence EV uptake. To circumvent these limitations, EV-associated proteins such as the tetraspanins, CD9 and CD63 have been GFP-tagged and have been used to successfully demonstrate EV sequestration in a variety of models [104,105]. Additionally, EVs have molecules exposed on the surface that can act in a paracrine manner to trigger signaling cascades within the recipient cell. It was demonstrated that EVs (pelleted at 100,000 g) derived from natural killer (NK) cells exposed FasL, a cytotoxic molecule, that could activate the Fas receptor on the target cell and promote cytotoxicity [106]. The heterogeneity in EVs results in a similar heterogeneity in EV uptake mechanisms by the cell ranging from clathrin-mediated endocytosis (CME) [102], micropinocytosis [107], phagocytosis [108], and endosomal membrane fusion [105]. Cell type seems to be one determinant for mode of cell uptake, for example, ovarian cancer cells sequestered EVs through CME [109], melanoma cells used a lipid raft-mediated endocytosis mechanism [110], while other innate immune cells, such as macrophages and microglia have been shown to use multiple mechanisms such endocytosis [111], phagocytosis [111,112] and micropinocytosis [107,112].

Early evidence for EV targeting specificity was demonstrated when EVs collected from human intestinal epithelial cells favored interaction with dendritic cells instead of T or B lymphocytes [113]. Moreover, when EVs from red blood cells were collected and injected into a rat the majority of EVs were taken up by liver Kupffer cells (KC) [114], conversely, EVs collected from melanoma cells that were intravenously injected into mice were mainly sequestered by the lungs and spleen [115]. This suggests that the biodistribution of EVs is source dependent; however, it does not provide evidence of the molecular mediators involved in EV internalization. The importance of EV-surface proteins was demonstrated when EVs collected from SKOV3 cells were treated with Proteinase K which reduced their uptake by ovarian cancer cells [109]. The primary proteins believed to be involved in mediating uptake are the tetraspanins and integrins, however, a diverse range of protein-protein interactions have been validated [1,10,102]. Tetraspanins interact with themselves and other transmembrane and cytosolic proteins to form microdomains, coined TEM domains (tetraspanin-enriched membrane domains) [116]. These domains are prevalent on the surface of EVs [117] as well as recipient cells and are facilitators of EV uptake [102]. Generally, the role for a specific protein in mediating EV sequestration is established using a specific inhibitor or antibody that can prevent the protein-protein interaction. One such study used dendritic cells treated them with antibodies against CD81 or CD9 which reduced uptake of EVs [118]. In a rat adenocarcinoma model that overexpressed the tetraspanin, Tspan8, it was demonstrated that Tspan8 recruited VCAM-1 (CD106) and CD49d (integrin alpha-4) into sEVs (pelleted at 150,000 g in sucrose gradient, Hsp70/TSG101 positive) [119] and when Tspan8 and CD49d complexed on the EV surface there was enhanced uptake by endothelial and pancreatic cells [117]. Furthermore, when the integrins CD51 and CD61 were blocked there was a reduction in EV internalization by dendritic cells [118]. These studies highlight the integral role that tetraspanins and integrins play independently as well as in association with one another for mediating EV uptake.

5. Drug-induced liver injury

5.1. Acetaminophen-induced liver injury

APAP is the leading cause of DILI in the United States [120] as well as a critical experimental model for studying DILI. While APAP at therapeutic doses is safely metabolized by UDP-glucuronosyltransferases (UGT1A1 and 1A6) and sulfotransferases (SULT1A1), at high doses these pathways become overwhelmed and APAP is metabolized by CYP2E1 into a reactive electrophile N-acetyl-p-benzoquinone imine (NAPQI) [121]. The formation of excess NAPQI will deplete GSH, whereby NAPQI will then readily react with sulfhydryl groups of proteins [122]. This protein binding event is associated with mitochondrial stress leading to an initial formation of reactive oxygen species (ROS) which triggers activation of c-Jun-N-terminal kinase (JNK) [123]. Activated JNK will translocate to the mitochondria increasing oxidative stress and exacerbating mitochondrial injury eventually facilitating cell death [124,125].

Many studies have shown that APAP administration will increase the amount of sEVs (pelleted at 100,000 g) in vivo and in vitro in both mouse and rat models (Fig. 2). Untreated hepatocytes will secrete approximately 10 million EVs per one million cells while APAP-treated hepatocytes will secrete 65 million EVs per one million cells [126]. These APAP-derived sEVs are uniform in shape and tend to be 50–150 nm in size according to nanoparticle tracking analysis (NTA) [12,127]. Similarly, APAP overdose patients have increased plasma EVs with a predominant increase in sEVs ranging between 100 and 150 nm, however, EV number was independent of circulating ALT values [127]. Correspondingly, mice treated with 300 mg/kg APAP have increased plasma sEVs at 2 h before any change in ALT, and the amount of plasma sEVs decreased at 4 and 6 h while ALT values continued to increase [127].

Fig. 2.

Impact of xenobiotics on EV dynamics and the functional consequences. Xenobiotic exposure induces cellular stress resulting in a nearly ubiquitous response by increasing the release of EVs which are metabolically active and enriched in cytochrome P450 enzymes. These xenobiotic-EVs may deliver arginase to endothelial cells resulting in decreased intracellular arginine driving endothelial dysfunction. Extensive work has demonstrated in models of alcohol-induced injury that EVs can elicit a pro-inflammatory immune response by delivering specific molecular cargo. Additionally, drug-protein adducts are present in EVs which can be internalized by dendritic cells and may contribute to T cell proliferation. Ultimately, the cargo loaded into the EVs is xenobiotic-dependent which will dictate the downstream functional consequence.

Early work investigating both mouse progenitor hepatocytes and primary rat hepatocytes (PRHs) demonstrated that released EVs have a unique consortium of RNA transcripts packaged in them. PRHs secrete three subpopulations of EVs that were identified based on their densities: the 1.13 g/mL fraction was CD81-positive and contained CYP2E1 transcripts, the 1.19 g/mL fraction was CD81/Flotillin-positive and was enriched in albumin (generally used as a negative control for contamination [50]) and GAPDH, the 1.23 g/mL was CD81/TSG101-positive which contained CYP2E1 [12]. Furthermore, CYP2E1 mRNA and protein have been shown to be enriched in human plasma sEVs and metabolically active [128]. CYP2E1 is the primary enzyme responsible for the bioactivation of APAP to its reactive metabolite NAPQI [121]. The presence of sEVs containing metabolically active CYP2E1 begets the question if these EVs could potentially propagate APAP-induced liver injury. Royo et al. [12] isolated sEVs (pellet 100,000 g) from PRH and added them to a stellate-like cell line (8b cells) which led to the production of the activation marker nitric oxide synthase. Further, when the PRH-sEVs were pretreated with an RNAse and Triton X-100 and then added to the 8b cells nitric oxide synthase was not produced [12]. These data suggest that sEVs are delivering active mRNA transcripts that are altering the phenotype of the recipient cell. A similar mechanism for EV-mediated transfer for deleterious or protective transcripts could be occurring during APAP overdose, such as the transfer of CYP2E1. In one study mice were treated with 300 mg/kg APAP and 24 h later EVs were collected from the plasma (APAP-EVs) and then added to untreated primary mouse hepatocytes (PMHs) for 24 h. These APAP-EVs increased cell death in PMHs and increased CYP2E1 protein levels [129]. However, the APAP-EVs were not assessed for CYP2E1 protein expression, which has been shown to be packaged into EVs [12], therefore, the elevated protein levels in the PMH could be the result of APAP-EVs containing CYP2E1. Alternatively, CYP2E1 transcripts could have been delivered to recipient cells and induced CYP2E1 expression, as the functional transfer of RNA transcripts from EVs has been demonstrated [93].

One study did report that human plasma EVs containing CYP2E1 added to alcohol-treated HepaRG cells increased cell death compared to alcohol treatment alone. Then it was shown that the use of CYP2E1 siRNA mitigated alcohol-induced toxicity but did not prevent the EV plus alcohol-induced toxicity [130]. Nevertheless, while APAP-EVs increase toxicity to PMHs and in vitro data suggests a potential role for EV-associated CYP2E1 to promote injury, there is no data demonstrating a functional transfer of CYP2E1 to neighboring hepatocytes and subsequent propagation of liver injury in vivo. Lastly, it has been shown that other CYP P450s are present in circulating rat sEVs after APAP and are metabolically active [131]. Although CYP2E1 is the canonical P450 associated with APAP bioactivation, it has been suggested that other P450s may be important in NAPQI formation at increasing APAP concentrations [121] and therefore may promote toxicity.

During APAP overdose rapid depletion of the GSH is a hallmark characteristic of APAP metabolism as the reactive metabolite, NAPQI, will rapidly conjugate with GSH [121]. APAP-sEVs (derived from PRHs treated with 10 mmol/L APAP) are enriched in glutathione transferases (GST) [131], which can conjugate GSH to NAPQI, although it has been shown that NAPQI/GSH reaction will occur spontaneously [132]. GST is also enriched in hepatic-sEVs collected from mouse plasma after dosing with 300 mg/kg APAP, suggesting a potential systemic delivery to other tissues [133]. However, EV-mediated transfer of GST is not useful if the substrate, GSH, is depleted. Interestingly, it was demonstrated that hepatic-sEVs (untreated) incubated with rat serum leads to a four-fold increase in 5-l-glutamyl-l-alanine a substrate of gamma-glutamyl transferase (GGT) [134], an enzyme located on the exterior of the plasma membrane responsible for the breakdown of extracellular GSH and GGT is also found to be packaged into sEVs [135]. Notably, GGT can produce cysteine, the rate-limiting substrate for de novo intracellular GSH production [136]. Moreover, sEVs (10 mmol/L APAP) contained increased amounts of enzymes associated with cysteine synthesis, such as cystathionine-beta-synthase, which produces cystathionine, a cysteine precursor [131]. These data hint at the possibility for APAP-EVs to deliver protective enzymes and substrates to stressed hepatocytes. While direct evidence is currently lacking for this phenomenon in APAP-induced liver injury, in a model of ischemia/reperfusion it was shown that key enzymes, ceramidase and sphingosine kinase 2 (SK2), and the necessary substrate, ceramide, was delivered to hepatocytes by hepatic-EVs and promoted proliferation/regeneration [137].

Additional evidence supporting a potential beneficial role for hepatic-sEVs has been suggested by analyzing other metabolites that are increased when serum is incubated with APAP-sEVs (PRH—10 mmol/L APAP). One study utilizing this paradigm found 27 metabolites were differentially expressed after the serum was incubated with APAP-sEVs compared to untreated-sEVs [126]. Two notable metabolites were alpha-linoleic acid, which was protective in rats that had arsenite-induced DNA damage [138], and 9,10-dihydroxy-octadecanoic (9,10-DHOME) which is a peroxisome proliferator-activated receptor gamma (PPARγ) agonist [139]. The latter metabolite is particularly interesting as PPARγ activation promotes mitochondrial biogenesis [140] which has been shown to limit APAP hepatotoxicity [141]. Mitochondrial biogenesis has also been suggested to be critical for ensuring hepatocyte recovery and regeneration after liver injury [142]. Hepatic-sEVs (untreated) incubated with serum have also been shown to increase deoxyinosine, a purine nucleoside that can maintain ATP levels [134], even when the mitochondria are depolarized [143], and increase methyl-histidine and anserine [126], both of which are involved in the same metabolic pathway for antioxidant defense. Taken together, these data seem to suggest that hepatic-sEVs/APAP-sEVs may provide the necessary energy and detoxification substrates to mitigate APAP hepatotoxicity and perhaps even promote recovery.

However, not all metabolic effects by hepatic-sEVs or APAP-EVs in serum are beneficial as there is a decrease in the antioxidant ascorbic acid (vitamin C) [134] as well as tetracosanoic acid [126]. Mice fed a vitamin-E deficient diet have increased susceptibility to APAP-induced liver injury [144] and a decrease in tetracosanoic acid has been associated with cardioembolic stroke [145]. Additional functional evidence has shown that APAP-sEVs have high levels of arginase activity and greatly reduce arginine in the serum [134]. Arginine is the primary substrate of endothelial nitric oxide synthase (eNOS) and elevated arginase activity has been associated with dysregulated eNOS which leads to endothelial dysfunction [146]. When APAP-sEVs were incubated with isolated rat pulmonary arteries the pulmonary endothelial had an attenuated response to acetylcholine-induced relaxation, suggestive of impaired endothelial function [134]. Furthermore, confocal microscopy revealed that the fluorescently labeled sEVs colocalized with the endothelium hinting at the possibility that hepatic-sEVs could deliver arginase to endothelial cells and diminish intracellular arginine [134]. Sinusoidal endothelial cells are susceptible to APAP-induced injury and while it is thought to occur prior to parenchymal toxicity [147], one can envision a situation where there is bidirectional EV-mediated communication between endothelial cells and hepatocytes causing an amplification of this potentially deleterious process.

Cho et al. [129] showed that the addition of APAP-sEVs to PMHs resulted in the phosphorylation of c-Jun N-terminal Kinase (JNK), which is a critical step in APAP hepatotoxicity [148]. Next, APAP-sEVs were administered via i.p. injection to healthy mice which resulted in an increase in TNF-α and IL-φ at 4 h [129]. However, there was no change in alanine aminotransferase (ALT) suggesting a lack of liver injury and pJNK was detected in both the control and experimental mice. Importantly, any study assessing the functional effects of EVs must consider that soluble mediators will co-precipitate with the sEVs after ultracentrifugation [50]. A host of damage-associated molecular patterns (DAMPs) and cell death promoting molecules such as calpain [149] will also be a constituent of this dose of sEVs, therefore, it is not possible to distinguish if the effects characterized were mediated by the EVs and their recipient cargo or soluble mediators. Lastly, the authors pretreated PMH with two putative EV inhibitors, one being GW4869 (a neutral sphingomyelinase inhibitor) and demonstrated a reduction in TNF-α and cell death [129]. These data suggest that inhibition of EV secretion attenuates APAP-induced hepatotoxicity. However, the data cannot be interpreted because both EV inhibitors are dissolved in DMSO, a potent inhibitor of CYP2E1 [150], which prevents the metabolic activation of APAP and all downstream mediated toxicity.

In summary, APAP-sEVs have unique transcriptomic profiles compared to untreated hepatic-sEVs [131]. Further, APAP-sEVs are metabolically active and will produce a multitude of metabolites that are associated with both protective and harmful processes [126]. While some evidence suggests that plasma-derived APAP-sEVs are toxic to hepatocytes [129], more work is needed to confirm these findings. It is important to understand that plasma-derived EVs represent a complex mixture of EVs derived from various cell types, and the limited data available suggests that the majority of circulating EVs come from immune cells [151], therefore exogenous administration of foreign plasma EVs may have a negative effect independent of APAPs influence on EV composition. Currently, interpreting the functional role for EVs in APAP-induced injury is difficult due to the use of different isolation protocols, the EVs being collected from APAP-treated animals at differing times, and the use of various model systems (Table 4). It is difficult to delineate the function of an EV if different subpopulations of EVs are being compared. This is further confounded when trying to compare the functional consequence of these APAP-sEVs between mice and rats as the latter are highly resistant to APAP-induced injury [152]. It seems plausible that EVs derived from APAP-treated animals may be loaded with different cargo throughout the APAP toxicity time course. Therefore, the functional impact of these EVs may differ over time perhaps contributing to the propagation of injury early on while later serving as facilitators of recovery.

Table 4.

Functional role for EVs in APAP toxicity.

Model/EV source	Dose (APAP)	Isolation method	EV marker	Result	References [#]
PRH Cell media	10 mmol/L Collected 36 h later	100,000 g UC	None reported	Metabolically active APAP-sEVs produce unique metabolites in serum	[126]
C57BL/6J Plasma	300 mg/kg Collected plasma 2, 4, 6 h PT	100,000 g UC	ALIX/CD63	Increased EVs at 2 h prior to change in ALT. No APAP-CYS adducts present in EV fraction. Increased plasma-EVs in APAP overdose patients	[127]
PRH Cell media	10 mmol/L Collected 36 h later	Size-exclusion chromatography	CD63	Cyp2d1 is present in EVs and metabolically active	[131]
PRH Cell media	10 mmol/L Collected 36 h later	100,000 g UC	Hsp70/Hsp90/Rab8	Increased Arginine synthase, Arginase 1, GST, cysteine synthesis	[131]
C57BL/6 Ja Plasma	300 mg/kg Collected from plasma 24 h later	Exoquick (commercial)	CD63/TSG101	In vitro: Increased cell death in vivo: Increased TNF-α, IL-1β, JNK activation, no change in ALT	[129]
BALB/c Plasma	300 mg/kg Collected from plasma 1 and 3 h post treatment	Modified Exoquick	Hsp90/Hsp70/CD63	GST packaged in plasma APAP-EVs	[133]
PRH Cell media	10 mmol/L Collected 36 h later	100,000 g UC, SG 1.18–1.23 g/mL	ALIX/CD81/CD63/Hsp70/RAB8	Hepatic-EV can increase deoxyinosine, APAP-EVs can deplete arginine and interfere with endothelial function	[134]
Human Plasma	–	Plasma Exo Kit (commerical)	CD63	Cyp2e1 is metabolically active and packaged into plasma EVs	[128]
PRH Cell media	–	100,000 g UC	CD81/TSG101/Flotillin 1/AIP1	Defined three subpopulations of EVs from PRH. Evidence EVs can transfer mRNA to stellate-like cells and activate them	[12]
Human Plasma	–	Exoquick (commercial)	CD63/CD81	Human plasma contains functional Cyp2e1	[130]
Winstar Rats Urine	–	100,000 g UC, SG	CD63/CD81/Flotillin/TSG101 Albumin negative	GGT present in urinary vesicles	[135]

^aStudy treated mice with 300 mg/kg APAP, collected EVs from plasma, then took these APAP-EVs and treated to primary mouse hepatocytes or i.p. injection to mice.

5.2. Alcohol-induced liver injury

Alcohol is predominately metabolized in the liver into acetaldehyde by three enzymes: alcohol dehydrogenase (ADH), catalase, and CYP2E1 [153]. Acetaldehyde can form DNA or proteins adducts which can lead to impaired protein function, activation of the immune system and/or promote the risk for liver cancer [154,155]. The K_m of ADH for alcohol is 1 mmol/L and is easily saturated at relatively low alcohol concentrations [153]. As the alcohol concentration increases in a system, CYP2E1 is thought to play greater role in alcohol metabolism and can produce superoxide anion and hydrogen peroxide. However, the contribution of CYP2E1 to early phase alcohol-induced liver injury (AILI) has received significant scrutiny [156]. Nevertheless, data has indicated that excessive ROS can lead to lipid peroxidation causing significant liver injury [157]. The situation is further exacerbated as alcohol is a potent inducer of CYP2E1. The biological role of EVs in alcohol-induced liver injury (AILI) has been an area of great interest, particularly, for EVs role in cell to cell communication [158].

The number of plasma EVs (CD63 positive) are greatly increased in alcohol-treated rats and mice [159] as well as in alcoholic human patients [160]. Furthermore, a variety of CYP enzymes, CYP2E1, CYP2A6, CYP1A1/2, CYP4B are increased in EVs from alcoholic patients compared to controls which also was recapitulated in female Fischer F344 wild-type rats treated with three oral doses of alcohol (6 g/kg/dose). In CYP2E1-null mice or mice treated with a CYP2E1 inhibitor (chlormethiazole), there was a reduction in total EV number and CYP enzymes in the alcohol-EVs [160]. Further evidence using HepG2 cells overexpressing either ADH or CYP2E1 revealed that 100 mmol/L alcohol only increased EV number in the HepG2^CYP2E1 cells and this effect could be abrogated through CYP2E1 inhibition [161]. This suggests that EV release and enrichment of EV-CYP enzymes corresponds to the magnitude of injury and CYP2E1 bioactivation of alcohol. The selective enrichment of CYP2E1 in alcohol-EVs raises the question of its potential role in propagating liver injury. PMH were treated with plasma-derived alcohol-EVs which resulted in JNK activation and a reduction in cell viability [160]. A similar toxicity profile occurred in a monocytic cell line (U1) that were incubated with alcohol and alcohol-EVs [130]. Collectively, these studies seem to suggest an unbiased deleterious role for alcohol-EVs perhaps mediated by CYP2E1. However, the limitations present in a similar study investigating APAP once again apply here. JNK is activated in the control PMH and there is no appropriate control for the potential of soluble mediators present in the EV fraction [160]. Using an antibody-targeted capture, such as CD63, to collect CD63-presenting EVs is a potential approach to addressing this problem. Additionally, HepaRG cells treated with CYP2E1 siRNA prevented alcohol-mediated toxicity, but failed to prevent cell toxicity when alcohol and alcohol-EVs-containing CYP2E1 were co-administered [130]. This suggests that a CYP2E1 independent mechanism is contributing to the EV-induced toxicity. More generally, when mice were fed an alcohol diet and given a Rho kinase (ROCK1) inhibitor (fasudil: water soluble), a putative EV inhibitor there was a reduction in ALT values and inflammation compared to pair fed controls [161]. This simple experiment suggests that EVs are harmful contributors to the injury process, however, global ROCK1 inhibition certainly will affect cellular processes independent of EV biogenesis. Currently, due to confounding variables it is difficult to properly interpret the role of these alcohol-EVs in the promotion of hepatotoxicity.

Another mechanism implicated in regulating EV release occurs via a caspase-mediated signaling pathway. Alcohol exposure in a dose-dependent manner increased caspase-3 cleavage and EV production and these effects were inhibited by using a caspase inhibitor or caspase-3 shRNA [161]. Caspase-3 activation is an important step in apoptosis and cell death potentially contributes to artificially inflating EV number as cellular debris can be difficult to distinguish from EVs [50]. Therefore, is the caspase activation mediating alcohol-induced EV release or is it contamination from cells undergoing apoptosis? While unclear, this concern is potentially unwarranted as the authors demonstrated via TUNEL assay that the highest concentration of alcohol utilized did not significantly induce apoptosis compared to controls [161]. Lastly, the authors also used a ROCK1 inhibitor which also reduced alcohol-EV release. These results suggest that while molecules such as CYP2E1 and caspase-3 may play a novel role in controlling alcohol-EVs release, that “classical” EV biogenesis molecules like ROCK1 are integral in the production of alcohol-EVs.

Activation of the immune system and chronic inflammation is believed to play a major role in AILI [162]. Proteomic analysis of alcohol-EVs revealed an enrichment in proteins involved in immune activation, inflammatory response, cellular infiltration as well as a high abundance of ADH and fatty acid synthase (FAS) [159]. Investigators have demonstrated a potential role for alcohol-EVs to activate macrophages and contribute to a pro-inflammatory environment [130,159,161]. Various EV-associated molecules have been implicated in mediating the activation of the innate immune system such as CD40L [161], Hsp90 [159], and miRNA-122 [163]. Initial work demonstrated that miRNA-122 was selectively enriched in alcohol-EVs (100 mmol/L) derived from both Huh 7.5 cells (hepatocyte-derived carcinoma cell) and primary human hepatocytes while the miRNA29b was downregulated [163]. The authors fluorescently labeled these alcohol-EVs and incubated them with THP human monocytes demonstrating that these EVs are internalized. THP cells have very low basal levels of miRNA-122 but the addition of alcohol-EVs leads to a fivefold increase in miRNA-122 [163]. However, internalization of an EV within a cell and gross measurement for RNA does not indicate successful delivery. The EV once internalized in the recipient cell needs to release its contents and the cargo needs to be partitioned to the correct intracellular compartments. To address this, heme oxygenase-1 (HO-1), a target of miRNA-122 silencing, was measured and found to be decreased with a corresponding increase in a host of LPS-induced inflammatory cytokines that are negatively regulated by HO-1 [163]. Taken together, these data suggest a selective process is controlling miRNA packaging into alcohol-EVs and miRNA-122 can affect the polarization state of the recipient cell. These findings were further extended when it was shown that alcohol-EVs derived from monocytes contained high amounts of miRNA-27a and when incubated with naïve monocytes could induce M2 polarization [164]. One can envision in vivo a scenario where alcohol-induced stress could activate this potential communication system between hepatic cells and immune cells. Moreover, these immune cells once activated could amplify and sustain their own activation state through immune-derived EV transmission.

Another putative molecule involved in alcohol-EV mediated macrophage activation is CD40L. A cytokine/chemokine array revealed that alcohol-EVs (110,000 g UC, Tsg101/CD63 positive) from HepG2 cells had high expression levels of CD40L and that it was located on the exterior surface of the alcohol-EVs [161]. Addition of these alcohol-EVs containing CD40L to THP-1 cells (monocytes) induced production of pro-inflammatory cytokines and the use of CD40L-antibody attenuated this effect. This idea was further explored in alcohol-fed mice treated with a pan-caspase inhibitor or CD40 KO mice (CD40^−/−). Both groups of mice had reduced hepatic infiltration of macrophages, reduced levels of pro-inflammatory cytokines, and decreased ALT values compared to WT mice [161]. While this investigation provides compelling mechanistic evidence that CD40L-EVs can activate macrophages in vitro and that CD40^−/− mice are protected from alcohol-induced injury, direct evidence for CD40L-EVs activating macrophages and promoting AILI is lacking. As one example, it is highly likely that there are other sources of CD40L and CD40^−/− mice would have been as equally unresponsive to those signals. Therefore, the outstanding question remained: does this process of EV cargo delivery between the parenchymal and non-parenchymal tissues occur within a biological system and what are the molecular mediators responsible? Saha et al. [159] fed mice a Lieber-DeCarli alcohol diet for 5 weeks which induces features of alcoholic liver disease (ALD) such as steatosis, inflammation, and moderate fibrosis. Plasma EVs (CD63 positive) were collected from these alcohol-fed mice (ALD-EVs) and then injected into naïve mice [159]. It was observed that both control-EVs and ALD-EVs were sequestered by hepatocytes and KC, however, the ALD-EV recipient mice had increased total monocyte infiltration consisting of both pro- and anti-inflammatory monocytes, while the KC were induced to a pro-inflammatory phenotype. Proteomic analysis of these ALD-EVs revealed a high abundance in Hsp90 and the addition of either ALD-EVs (containing Hsp90) or recombinant Hsp90 to RAW macrophages induced an inflammatory phenotype that was prevented by an Hsp90 inhibitor (17-DMAG) [159]. These data suggest another potential EV-associated molecular mediator that could facilitate intercellular communication between hepatocytes and the immune system. Moreover, the evidence suggests that these ALD-EVs are contributing to a chronic inflammatory state in ALD mice. Notably, this study extends previous work by demonstrating that in vivo ALD-EVs are sequestered in hepatocytes and monocytes of naïve mice and promote monocyte infiltration and activation. Nevertheless, the exact molecular mediator(s) of this intercellular communication in vivo remains elusive, although strong evidence has been presented for the role of CD40L [161] and Hsp90 [159].

The current landscape strongly supports a role for alcohol-EVs to facilitate an immune response, however, the contribution of EV-associated CYP2E1 to facilitating liver injury requires further validation. The alcohol-EVs are metabolically active [130,160], deliver active transcripts to recipient cells [163,164], and can influence the state of the recipient cell [159,161,163,164] (Table 5). The use of different model systems and various isolation protocols makes it difficult to compare the function of alcohol-EVs across different studies. Importantly, while many groups have shown that DILI will increase the amount of EVs in circulation, little effort has been given to determining the cellular origin of these plasma EVs. One study investigating EVs from a murine NASH model demonstrated that hepatic-EVs account for less than 1% of the circulating EVs and that the majority of plasma EVs come from platelets and immune cells [151]. Therefore, the functional effect of plasma alcohol-EVs cannot simply be assumed to be the result of a hepatic-EVs. The study of AILI is complex due to its chronic nature, and different AILI models aim to recapitulate different features of the pathology. Correspondingly, alcohol-EVs from one study may contain a different assortment of cargo compared to alcohol-EVs from another study. This could accurately represent what occurs clinically: as a patient progresses through AILI the EV and its associated cargo represent the current state of the cell, which varies throughout the disease pathology.

Table 5.

Functional role for EVs in alcohol-induced liver injury.

Model/EV source	Dose (EtOH)	Isolation method	EV marker	Result	References [#]
HepG2ADH HepG2CYP2E1 Cell media	100 mmol/L Collected 24 h later	110,000 g UC	TSG101/CD63/LAMP1/RAB5	Increased EVs in HepG2CYP2E1 but not HepG2ADH. CYP2E1 inhibitor prevents increase in EV release	[161]
Fischer F344 Plasma	Three oral doses of 6 g/kg	Exoquick (commercial)	CD63	Increased EVs and CYP enzymes (CYP2E1) packaged into Alc-EVs. Suggests Alc-EVs mediate injury through CYP2E1
Alcoholic Patients Plasma	–				[160]
Fischer F344a Plasmab	Three oral doses of 6 g/kg	Exoquick (commercial)	CD63	In vitro: Alc-EVs added to PMH decrease cell viability, increase pJNK, and increase cleaved caspase 3	[160]
HepRGb Monocytic cells (U1) Cell media	Alc-EVsc and 50 mmol/L EtOH	Exoquick (commercial)	CD63/CD81	CYP2E1 knockdown prevents Alc toxicity but not Alc plus Alc-EV toxicity	[130]
Huh 7.5 PHH Cell media	100 mmol/L (24, 48, 72 h)	Exoquick-TC (commercial)	CD63	miRNA122 enriched in Alc-EVs. miRNA122 can be transferred from Alc-EVs to monocyte and modulate gene expression	[163]
THP-1 Monocytes Cell Media	100 mmol/L (24, 48, 72 h)	Exoquick-TC (commercial)	CD63	miRNA-27A enriched in Alc-EVs and can induce M2 polarization in naïve monocytes	[164]
HepG2 Cell Media	100 mmol/L Collected 24 h later	110,000 g UC	TSG101/CD63/LAMP1/RAB5	Alc-EVs enriched in CD40L and located on exterior surface. In vitro and in vivo evidence for EV-CD40L macrophage activation	[161]
C57BL/6 Plasma	Liber DeCarli Diet (5% v/v) for 5 weeks	Exoquick-TC (commercial)	CD63 GRP-78 negative	Alc-EVs enriched in immune/inflammation pathways. Alc-EVs taken up by hepatocytes and Kupffer cells in naïve mice. Increased monocytic infiltration	[159]
RAW Macrophages Cell Media	Alcohol-EVs (50 μL at 50 μg/mL)		CD63	Alc-EVs contain Hsp90 and activate macrophages. Hsp90 inhibitors prevent activation. Suggest Hsp90 mediates increased monocytic infiltration in vivo	[159]

^aStudy treated mice with three oral doses of 6 g/kg alcohol, collected EVs from plasma, and then took these alcohol-EVs and added to primary hepatocytes.

^bDoes not mention concentration of EVs added.

^cAlc-EVs were collected from plasma of C57BL/6 mice that had been treated with one dose of Alc (5 g/kg).

5.3. Other drugs

Environmental exposure to compounds such as polycyclic aromatic hydrocarbons (PAHs) have been associated with liver injury, however, the role of EVs in contributing to injury has been relatively unexplored. One group aimed to remedy this by investigating three common PAHs: benzo[a]pyrene (BP), dibenzo[a,h]anthracene (DBA), and pyrene (PYR) [165]. These three molecules are commonly found in food and share an affinity for the aryl hydrocarbon receptor (AhR). AhR is a cytosolic ligand-activated transcription factor that is referred to as a “xenobiotic sensor” because of its affinity for an assortment of xenobiotics, particularly aromatic hydrocarbons, whereby AhR activation leads to the transcription of drug metabolizing enzymes (DMEs) such as CYP1A1 [166]. All three PAHs increased EV release from PRH and WIF-B9 hepatocytes, particularly EVs with a diameter between 90 and 200 nm (100,000 g). Moreover, chronic treatment of PAHs to rats led to a significant increase in EVs in plasma [165]. This finding is in concordance with previous work investigating other drugs (APAP, alcohol etc.) and collectively seems to suggest an overarching principle that drugs at toxic doses result in increased EV synthesis. PAH treatment caused an increase in apoptosis and caspase 3/7 activity in both hepatocyte cultures which was prevented by using Z-VAD, a specific inhibitor of caspase 3/7 [165]. Previous work using an AILI model suggested a role for caspases in modulating EV release [161], however, in the current work Z-VAD attenuated cell death without interfering with EV release. Moreover, HMEC-1 (human endothelial cells) treated with PAHs led to an increase in EVs without causing an increase in apoptosis. Interestingly, in this model an AhR inhibitor prevented the release of sEVs but did not influence the amount of large EVs released [167]. This highlights that xenobiotics can differentially modulate the distinct EV biogenesis pathways presumably by influencing different molecular mechanisms. In support of this, it was demonstrated that BP- and DBA-induced EV production required AhR activation while PYR EV production was independent of AhR. However, it was demonstrated that PYR EV production was dependent on another common xenobiotic sensor, constitutive androstane receptor (CAR). Collectively, this suggests a requirement for the induction of DMEs for the facilitation of enhanced EV release. Nevertheless, TCDD, an AhR agonist, failed to increase EV release in the hepatocytes while still increasing apoptosis. Therefore, while the activation of xenobiotic sensors is a necessary upstream event for aromatic hydrocarbon induced EV production, it is a downstream consequence that dictates the enhanced EV release. Many early studies investigating basic EV biology have underscored the prominent role of lipids and membrane composition in EV biogenesis (as explored in Section 3.1.2). While BP and DBA increased the fluidity of the plasma membrane, which was necessary for an increase in apoptosis and EV release [165], TCDD has been shown to decrease fluidity of the membrane [168]. Conversely, PYR mediated cell death and EV release occurred independently of alterations in membrane fluidity [165]. Taken together, the data hint that molecules which primarily activate AhR (BP, DBA, and TCDD) require increased fluidity of the plasma membrane to induce increased EV release, whereas molecules that activate CAR (PYR) facilitate apoptosis and EV release independent of membrane composition (Table 6).

Table 6.

Functional role for EVs in drug-induced liver injury.

Model/EV source	Dose	Isolation method	EV marker	Result	References [#]
RPMI 8866 B lymphocytes Cell media	0.5 mg/mL Biotinylated amoxicillin/amoxicillin	200,000 g UC	None reported	AX-EVs contain haptenated HSP70, EF-2, actin, α-enolase. AX-EVs were internalized by B lymphocytes	[169]
Long Evans Plasma	0.8 mg/kg PAH via oral gavage 3/wk. for 90 days	100,000 g UC	ALIX/TSG101/CD63/CD81/Flotillin1	Chronic PAH treatment leads to increased EVs in plasma	[165]
PRH WIF-B9	100 nmol/L BP, DBA, or PYRa	100,000 g UC	ALIX/TSG101/CD63/CD81/Flotillin1	BP and DBA require AhR activation to increase EV release. PYR functions through CAR. BP and DBA increase plasma membrane fluidity. Increase in EV release is independent of apoptosis	[165]
HMEC-1 Cell media	B[a]P 1–10,000 nmol/L	10,000 g UC (large EVs)   100,000 g UC (sEVs)	TSG101, Caveolin-1, Flotillin-1, Hsc70   CD63, TSG101, Caveolin-1, Flotillin-1, Hsc70	AhR inhibitor (10 μmol/L) prevents B[a]P induced sEV release without effecting large EVs Suggests that EV subpopulations are differentially impacted by B[a]P	[167]
PHH	AX 0.05 mmol/L FX 0.05 mmol/L Isoniazid 0.03 mmol/L SMX-NO 0.01 mmol/L Collected 24 h later	Exoquick-TC (commercial)	CD63b	AX-, FX-, and SMX-NO-EVs contain haptenated proteins. Dendritic cells sequester drug-EVs via phagocytosis AX-EVs fail to activate dendritic cells	[170]

^aPRH were treated for 2–18 h, WIF-B9 hepatocytes were treated 5–72 h.

^bDrug-EVs were CD9, CD63, CD81, Hsp70 positive via mass spectrometry, western blot only detected CD63. HMEC1 is a human endothelial cell line. AX, amoxicillin, FX, flucloxacillin, SMX-NO, nitroso-sulfamethoxazole.

A prominent mechanism for DILI is the formation of drug-protein adducts which can trigger an immune response, at which point the drug-protein adduct is considered a hapten. It is thought that many idiosyncratic DILIs (iDILI) are caused by the adaptive immune system recognizing a hapten and launching a response [21]. Amoxicillin (AX) is a commonly prescribed antibiotic and is the most commonly associated drug that causes iDILI. Many xenobiotics can elicit an immune response but the exact mechanism in which haptens are presented to T cells is not well understood. Gomez et al. [169] showed that AX covalently binds to endogenous proteins and is packaged into circulating EVs. Further, these EVs were sequestered by endothelial cells, suggesting delivery of these AX-haptenated proteins and demonstrating that EVs may have the capacity to transfer haptens to antigen-presenting cells [169]. In a tour de force the proteome of EVs isolated from PHH treated with subtoxic concentrations of either flucloxacillin, amoxicillin, isoniazid, and nitroso-sulfamethoxazole (SMX-NO) were analyzed and 2109 proteins were identified. Of these aforementioned drugs only AX, flucloxacillin (FX), and SMX-NO led to the covalent modifications of proteins packaged into EVs [170]. The FX-EVs and SMX-NO-EVs were added to human monocyte-derived dendritic cells where these EVs were internalized via phagocytosis and endocytosis. Yet, these drug derived EVs were unable to activate the dendritic cells. Nevertheless, an amoxicillin-modified peptide was constructed that had a high probability of binding to an HLA class 1 allele (associated with amoxicillin-clavulanate liver injury). This amoxicillin-modified peptide was loaded to dendritic cells and then co-cultured with naïve T cells leading to proliferation of the T cells [170]. These results suggest that the covalent modification of endogenous proteins by amoxicillin can trigger an immunogenic response. However, extensive work needs to be carried out to demonstrate how this process could occur in vivo by EV delivery. Notably, the current work suggests that this process fails to occur, as the dendritic cells failed to respond to drug derived EVs. Although an EV containing haptenated cargo can be sequestered by a dendritic cell, the process in which a hapten is excised from the EV and presented on the surface of an APC remains to be elucidated.

6. Extracellular vesicle biomarkers in DILI

The most commonly used markers for liver injury come from plasma or serum measurements of hepatic enzymes such as ALT, AST, or alkaline phosphatase (ALP). Increases in ALT/AST reflect a hepatocellular pattern of DILI while ALP increases indicate damage to the biliary epithelial cells [21]. While these diagnostic biomarkers are considered the gold standard, none can distinguish between the different types of DILI nor are they completely liver specific [171]. For example, ALT and AST are expressed in skeletal muscle and are measurable in circulation after extreme exercise [172]. Therefore, the identification of novel biomarkers in DILI with much greater sensitivity and specificity has been an area of great interest.

EVs are attractive candidates due to their ubiquity across biofluids and biophysical properties (e.g., can enclose soluble molecules and shield them in an extracellular environment). A variety of proteins [133,135,173,174], miRNAs/mRNAs [18,19,28,175], and DMEs [128,130,131] are enriched in drug derived EVs, however, many of these molecules have not undergone any further validation for potential efficacy as a biomarker. Furthermore, many of these EV-associated molecules suffer the same drawbacks as the current generation of biomarkers: they lack liver specificity and cannot distinguish between the different types of liver injury. However, miR-122 is a promising biomarker that was increased in the plasma/serum and EV fraction in various models of liver injury [18,176,177].

Early work using an APAP overdose mouse model identified miR-122 and miR-192 as potential biomarkers due to dose-dependent responsiveness and an earlier increase than ALT [176]. These findings were then extended when it was shown that miR-122 and miR-192 were elevated in the serum of human APAP overdose patients while these microRNAs were only slightly increased in chronic kidney disease patients suggesting a degree of liver specificity [178]. miR-122 has been shown to be more sensitive than ALT in the clinic as well as being specific to liver injury, as a model for drug-induced kidney injury failed to influence miR-122 levels [179]. Mice treated with 500 mg/kg APAP had a dose-dependent increase in plasma EV-associated miR-122 [18]. Further, EV-miR-122 is more stable and resistant from RNAses than “free” miR-122 in circulation conferring a potential clinical advantage for EV-miR-122 as a biomarker [20]. Ultimately, if EV-miR-122 is more efficacious then it needs to have the same properties as “free” miR-122 (e.g., liver specificity and greater sensitivity than ALT/AST). In this direction, Cho et al. [28] showed that APAP-induced liver injury increased levels of miR-122 and miR-192 in EVs without increasing the kidney specific miR-146a or muscle specific miR-206 suggesting that EV-miR-122/192 is liver specific. However, multiple groups have demonstrated that there are much greater amounts of miR-122 in the protein rich fraction (estimated to be ~ 90%) compared to EV-miR-122 (~ 13%) [177] during the DILI time course [18,175]. Additionally, when rats were treated with a 1400 mg/kg of APAP the percentage of EV-miR-122 relative to the total miR-122 in the plasma (EV-associated + free) drastically decreased over time [175], interestingly, the opposite result has been reported where the proportion of EV-miR-122 increased over time [177], although due to high variability it failed to reach statistical significance. It was suggested that severe acute liver injury may cause “leaking” of miR-122 into circulation as the hepatocytes undergo necrosis which could account for the percent decrease in EV-miR-122 compared to the total miR-122 in circulation. In support of this, in a drastically less severe model of liver injury the majority of miR-122 was associated with the EV fraction instead of the protein rich fraction [18]. Nevertheless, it is unclear if EV-miR-122 confers any significant advantage over free miR-122 as a biomarker. One drawback for free miR-122 is its high degree of variability in human patients, therefore if EV-miR-122 has lower variability it may be more advantageous, although this also seems unlikely as EV-miR-122 has been shown to be highly variable in rodents [177]. Many other molecules are enriched in DILI-derived EVs, however, the advantage as biomarkers for these molecules over their soluble/free counterparts remains to be elucidated and have been explored exhaustively elsewhere [180].

While EVs represent an exciting area for biomarker research due to their ability to shield molecules from the extracellular milieu, the available data highlights the lack of fundamental understanding in the basic function of EVs. ALT/AST measurements increase during liver injury because cells lose the integrity of their membranes and cytosolic proteins leak into circulation. Are EVs being secreted during DILI randomly or is it driven by some coordinated process? Is the packaging of miR-122 into EVs an intentional attempt by the stressed cell to signal to other cells? As discussed in Section 5.2, miR-122 was packaged into EVs and shown to communicate with immune cells leading to their activation [163]. If this EV signaling is an intentional process, then their use as a biomarker would be the quantification of a coordinated response, not the stochastic leaking of cytosolic enzymes which begets the question, why should EV-contained molecules correlate with the magnitude of injury in a system? One answer is that EVs are being released from the stressed cells with the intention of facilitating a protective response, which if that was the case one would expect that system of EV-mediated signaling to fail if the stressor was too great (e.g., initial increase in the putative EV-associated biomarker followed by a plateauing, as seems to be the case with EV-associated miR-122 [175]). Additionally, if the adaptive response effectively fended away the toxic insult the EV-associated biomarker should also then return to baseline. Therefore, if EVs are intentionally released to facilitate an effect, the half-life of an EV-associated biomarker may be shorter than its free partner. Unfortunately, data measuring the half-life of endogenous EVs in vivo is scarce [181], however, pharmacokinetic modeling has estimated that plasma sEV secretion rate is 18 μg/min [182]. Exogenously administered plasma EVs have a distribution half-life that can be measured in minutes [115,183] and an elimination half-life of less than 4 h [115]. One group reported that ablation of macrophages increased the amount of circulating exogenously administered EVs, suggesting a shift in the balance of secretion and clearance [182]. Perhaps endogenous EVs have a longer half-life as they can be recognized as “self’ and escape targeting by macrophages. Lastly, if EVs are being used to shed waste products or hazardous materials then that EV-associated biomarker will continue to increase in relation to injury/necrosis. While the functional importance of EVs in many chronic conditions is evident [184], a dual role for EVs as a system for waste disposal and cell signaling during bouts of extreme acute stress is plausible. A recent, comprehensive report summarized the conclusions of 18 articles that compared the utility of non-EV miRNA compared to EV-miRNA as a biomarker and concluded that 75% of the papers suggested that EV miRNA had superior efficacy [185]. Alas, this highlights the complexity of EV biology as the aforementioned papers were not stratified according to model or condition, therefore, the function of EVs and the associated miRNA of interest could represent fundamentally different cellular processes. Nevertheless, a combination of different miRNAs associated and/or non-associated with EVs may be needed to understand the magnitude and progression of injury to make an informed decision in the clinic.

7. Extracellular vesicles: Potential applications in DILI

The clinical value for EVs can be conceptualized into three categories: (i) EVs as a drug target (ii) EVs as a therapeutic (iii) EVs as vectors for drug delivery. EVs are an attractive therapeutic option as they can be patient derived with a long shelf life and have innate qualities such as low or absent immunogenicity (dependent on the parental cell source), ability to cross the blood-brain barrier, and lack of replicability [186,187]. Both hepatic-EVs and mesenchymal stem cell-EVs (MSC-EVs) have been demonstrated to be effective at promoting regenerative responses following liver injury [137,188–190]. Neutral ceramidase and SK2 were identified as the critical molecules housed in hepatic-EVs that promoted proliferation in target hepatocytes by increasing local production of sphingosine-1-phosphate (S1P) [137]. Glutathione peroxidase 1 (GPX1) was enriched in MSC-EVs and when GPX1 was knocked down the protective effects of MSC-EVs in a model of DILI was attenuated [191]. Furthermore, various strategies have been employed to enhance the potency of MSC-EVs, such as culturing MSCs in hypoxic conditions [192] or using differential UC to collect key fractions of the MSC culture media [189]. Many promising phase 1 and phase 2 clinical trials are underway using EV-based therapies with early results indicating safety and efficacy [30]. These results are encouraging because there is a need for novel therapies that can enhance regeneration following DILI [125].

7.1. Extracellular vesicles as a drug target: Potential and limitations

Many studies have shown that EVs have the potential to propagate injury, thereby designating EVs as a drug target. The early work utilized to understand the mechanisms by which EVs are formed and how cargo loading is controlled have supplied a plethora of drugs that can inhibit specific EV subpopulations from being secreted. However, this sledgehammer approach of decreasing total EV synthesis/release/uptake has limited therapeutic value, as EVs may play a crucial role in fundamental physiological processes [186]. In the various models of DILI where the role for EVs have been explored it is not clear if EVs are harmful. Moreover, EVs may be important facilitators of recovery after injury [137,190] and therefore therapeutic targeting of EVs would have to consider the time of injury induction. Clinically, this is very difficult to determine as patients with drug overdoses/acute liver injury are not reliable sources of information.

Ultimately, it is the cargo associated with the EV that is believed to be promoting injury. In this framework, a targeted approach to control what populations of cargo is being loaded into the EV may be more advantageous. However, this has multiple layers of difficulty, as the fundamental mechanisms governing cargo sorting are still poorly understood and it has been suggested that while certain types of molecules (miRNA) may be selectively packaged, the packaging of cytosolic proteins may simply be stochastic [170,184]. Therefore, it would be a matter of identifying the deleterious molecule(s) of interest in one's experimental model at which point specific targeting of that molecule may be more advantageous and feasible. Nevertheless, it has already been demonstrated that loading of specific proteins of interest can be controlled by exposure to blue light in bioengineered EVs [193]. Many creative approaches are being developed in this burgeoning field and are explored in depth elsewhere [30]. While the active research in this area is primarily focused on overexpressing key proteins/RNA that have therapeutic efficacy into bioengineered EVs, these studies also provide evidence that this cargo loading process can be targeted and exploited to prevent loading of harmful molecules.

7.2. Extracellular vesicles as a therapeutic

Few studies have been conducted investigating the therapeutic value for EVs in DILI. The earliest forays evaluating the therapeutic value for EVs in liver biology utilized a model of partial hepatectomy (PH), where approximately 70% of the liver is surgically removed. From this early work it was revealed that human liver stem cell-EVs (HLSC-EVs) could accelerate liver regrowth which was mediated by the horizontal transfer of mRNAs [194]. This regenerative promoting property is not exclusive to stem cell-derived EVs as hepatic-EVs have also been shown to increase liver mass after PH as well as induce increased hepatocyte proliferation following ischemia-reperfusion (I/R) [137]. Notably, this proliferative effect was specific to hepatic-EVs as EVs derived from KC or liver sinusoidal cells (LSC) reduced proliferation or had no effect. These hepatic-EVs contained SK2, which was absent from the other liver derived EVs, which was delivered to recipient hepatocytes and enabled production of S1P [137], a critical sphingolipid signaling metabolite that regulates cell growth [195] and suppresses apoptosis [196]. Additionally, a direct comparison of fibroblast-EVs (FB-EVs) to MSC-EVs also revealed a parental cell specificity, only the MSC-EVs successfully suppressed the induction of pro-inflammatory cytokines and reduced injury after hepatic I/R [190].

The majority of work using in vivo models have used carbon-tetrachloride (CCl₄) to induce liver injury. Initial work suggested that MSC-EVs could attenuate injury through activation of proliferative and regenerative responses [188], while more recent work indicated that delivery of GPX1 by MSC-EVs to hepatocytes reduced oxidative stress and prevented cell death [191]. MSC-EVs also have been shown to mitigate oxidative stress by delivering peroxiredoxins and GST [197]. Taken together, these MSC-EVs are essentially serving as a vector enabling efficient transfer of a host of antioxidants to target cells. One group did report in vitro that hepatocytes treated with APAP (2 mmol/L) and MSC-EVs were protected although the hepatocytes failed to upregulate antioxidant genes [188]. This suggests that MSC-EVs have the capacity to mitigate oxidative insult during DILI by direct transfer of the antioxidant machinery and not by induction of the antioxidant repertoire in the recipient cell.

In vitro models using APAP have revealed that MSCs and MSC-EVs can prevent cell death [188,189] with the MSC-EVs displaying greater protective capabilities [189]. The precise mechanism has yet to be defined, however, initial findings suggested that this protective effect was through upregulation of anti-apoptotic Bcl-xL and decreased caspase 3/7 [188] or perhaps by reducing the amount of ROS [189]. We have previously discussed why metrics of apoptosis have limited relevance during APAP-induced injury as the primary mode of cell death is necrosis not apoptosis [198]. Temnov et al. [192] demonstrated that MSCs and soluble factors cultured under mildly hypoxic conditions had a greater ability to attenuate liver injury following APAP (350 mg/kg) than MSCs and soluble factors under normal oxygen levels. While an exact mechanism was not described, the hypoxic-MSCs and media was enriched in proteins involved in intracellular transport and cellular respiration [192]. These findings were later extended when they collected MSCs and the conditioned media (under hypoxic conditions) and fractionated it into different protein fractions based on mass, > 50, 50–30, 30–10, and 10–3 kDa. The 10–30 kDa fractionation had the greatest potency successfully reducing various measures of liver injury as well as stimulating a regenerative response [199]. Ultimately, extensive work is needed to delineate the mechanisms in which MSC-EVs are conferring protection following APAP-induced liver injury.

Collectively, these findings demonstrate the therapeutic potential for MSC-EVs and hepatic-EVs (Table 7) as well as highlighting the lack of information in this area. Do hepatic-EVs have potential as a treatment in DILI? Are the mechanisms in which MSC-EVs and hepatic-EVs conferring protection distinct, the same, or overlapping? Are MSC-EVs activating proliferative and regenerative responses or are they merely reducing the initial injury? One important caveat to consider is the timing of the EV administration relative to when the hepatotoxic drug was dosed. If the EVs are to be assessed for their ability to promote regeneration, then they need to be administered following the metabolism and injury inducing phase of the drug. For example, if the EVs are co-administered with the hepatotoxicant it becomes impossible to determine if the therapeutic value is due to a reduction in injury or the promotion of the regenerative potential of the cell. The present studies utilizing CCl₄ had dosing schemes involving co-administration [188] as well as 24 h post CCl₄ [189] suggesting that MSC-EVs are beneficial during the injury phase as well as during recovery. Conversely, the studies investigating APAP predominately used a co-administration strategy [188,189,192], except for one study where it was administered at 4 h post APAP [199]. Therefore, it is unclear if MSC-EVs can increase the regenerative and proliferative response of the cell during APAP toxicity, however, based on models of PH it seems plausible. As the role for MSC-EVs and EVs are explored in other models of DILI, it is important to recognize that the timing of EV administration will directly affect the conclusions one can draw from their data.

Table 7.

Role for EVs as a therapeutic.

Model	EV Source	Dose	EV Marker	Result	References [#]
Sprague-Dawley Partial Hepatectomy (PH)	Human liver stem cells	EVs (i.v.) 30 μg/mL immediately following PH	None reported	HLSC-EVs promote hepatocyte proliferation and reduce apoptosis. Effect is abrogated when HLSC-EVs are treated with RNAse   HLSC-EVs promote functional and morphological recovery of the liver through an RNA-dependent mechanism	[194]
C57BL/6J PH I/R	Mouse hepatocytes, Kupffer cells, liver sinusoidal endothelial cells	EVsa (i.v.) after surgery and again at 24 h EVs 24 and 48 h after I/R	TSG101/CD81/CD63 Positive EAA-1 and GRP78 Negative	Hepatocyte derived—EVs, but not other liver EVs, induce hepatocyte proliferation in vitro and in vivob   Hepatocyte-EVs deliver SK2 to target cells allowing local production of S1P, stimulating proliferation and reducing apoptosis	[137]
C57BL/6c	Human umbilical cord	CCl4 0.05 mL/kg (i.p.) followed by 0.4 μg MSC-EVs by intrasplenic injection	None reported	Increased PCNA and Cyclin D1, suggests increased proliferation	[188]
Wistar RatHepG2	Bone marrow derived from Wistar Rats	CCl4 20% (v/v), 5 mL/kg (i.p.). 24 h later HPV 50 μg MSC-EVs   In vitro: APAP 8 mmol/L cotreatment with 0.5 μg/mL MSC-EVs	CD9/CD63/CD81	In vivo: demonstrates MSC-EVs have greater potency than MSCs to mitigate liver injury   >In vitro: Increased cell viability, and decreased ROS in MSC-EVs + APAP compared to APAP alone	[189]
Balb/c—nu/nu Female	Human umbilical cord	0.15–0.35 mL/kg CCl4 at 10% conc. in mineral oil (i.p.) 24 h later 8, 16, or 32 mg/kg via tail vein or oral gavage	CD9/CD63	Dose-dependent protective effect for MSC-EVs, oral gavage compared to tail vein injection of MSC-EVs results in similar efficacy profiles   MSC-EVs deliver GPX1 to target hepatocytes and reduce oxidative stress and apoptosis	[191]
CD-1 micec	Femur bone marrow of CD-1	350 mg/kg APAP (i.p.) followed by 5 mg of MSC + soluble factors	–	Compares ncMSCs to hcMSCS, hcMSCs more effective at reducing liver damage, hcMSCs enriched in proteins involved in intracellular transport and cellular respiration	[192]
BalB miced Femur bone marrow of CD-1	Femur bone marrow of CD-1	270 mg/kg APAP (i.p.) 4 h later injected with 1 mL ncMSC or hcMSC (0.5 mL conditioned media to 0.5 mL lysate, protein conc. 10 mg/mL)	–	10–30 kDa fraction from hypoxic cells reduce liver injury, fraction enriched in 10 proteins	[199]

Unless otherwise reported, EVs were collected via ultracentrifugation at 100,000 g.

^aEV dose not reported.

^bNot directly reported, references two papers that characterize multiple methods.

^cTFF, Tangential flow filtration, followed by HPLC, further concentrated by using 100 kDa MWCO filter.

^dCells cultured in hypoxic conditions (5% O₂) = hcMSCs, cells cultured in normal oxygen conditions (21% O₂) = ncMSCs.

7.3. Extracellular vesicles as vectors for drug delivery

A promising area of research is the use of EVs to deliver small molecules and drugs. EVs can shield its drug cargo from the extracellular environment and can readily cross biological barriers. While EVs can have therapeutic value, their potency can be enhanced by increasing packaging of therapeutic cargo [186]. The process of drug loading into EVs can either be done exogenously or endogenously [30]. Exogenous loading requires isolation of EVs from the source cell where then the cargo of interest is incorporated into the EVs. This has been done via passive loading (coincubation of the drug with EVs) [200], or with transfection techniques such as electroporation or lipofection [201]. Preclinical evidence demonstrated that curcumin loaded into EVs resulted in greater bioavailability and a stronger anti-inflammatory effect than curcumin alone [200]. Correspondingly, a clinical trial is being conducted with curcumin-loaded EVs (NCT01294072) as well as a separate clinical trial utilizing chemotherapeutic drug-loaded EVs (NCT01854866). Endogenous loading is the genetic manipulation of the parent cell to autonomously package the cargo of interest and exploits the EV biogenesis process. Elegant work demonstrated efficacy for this system of endogenous loading where the CRIPSR/Cas system was transfected into mammalian cells which then was endogenously packaged into EVs which the authors termed GEDEX (genome editing with designed extracellular vesicles) [202]. Next, in a proof of principle experiment, alpha-napthylisothiocyanate (ANIT) was used to induce liver injury in mice which were then treated with GEDEX targeting Hepatocyte Growth Factor (Hgf) gene. They demonstrated an increase in HGF in treated animals as well as a reduction in liver injury [202]. GEDEX represents a powerful system as a research tool and as an EV-based delivery system for therapeutics. Other such systems exist such as the MSC-iExosome [203] which currently is being evaluated in a clinical trial for metastatic pancreatic cancer (NCT03608631). Notably, unmodified MSC-EVs also have been used as a therapy in human patients with chronic kidney disease (CKD) which resulted in improved kidney function and decreased inflammation [204].

While EVs as therapeutics and drug delivery systems represent exciting and promising solutions to challenging biomedical problems there is still much to learn about the basic biology of EVs. This represents a problem because exogenous administration of high doses of EVs could interfere with endogenous EV function as intercellular mediators or the exogenous EVs may have a host of off-target effects. These therapeutic EVs would be loaded with a wide array of molecular cargo both in its lumen as well on the vesicle surface leading to many potential functional consequences. Therefore, the use of EVs as a drug delivery system has to be considered in the context of what advantages it has over other synthetic vectors (e.g., liposomes or nanoparticles). The two oft-cited advantages for EVs as a vector are reduced immunogenicity [205,206] and ability to bypass biological barriers [186,201]. However, both of these properties are nuanced, as an example the maturation state of dendritic cells (DC) can impact the immunomodulatory effects DC-EVs, whereby immature DCs produce EVs that are immunosuppressive while mature DCs produce EVs that stimulate the immune system [207]. This underscores the rich diversity in EVs, as EVs function is dependent not only on the parental cell source but also on the temporal, physiological state of the cell. In essence, EVs are a photograph capturing the complexity of the cell at a current point in time under a set of environmental conditions. This feature captures both the potential for EVs as a vector as well as its inherent limitations.

8. Conclusion and future perspectives

While great advances have been made in understanding the basic biology of EVs, it has led to a multitude of deeper questions. Many articles talk about the “heterogeneity” of EVs as an innate quality, however, this heterogeneity exists far earlier than the release of ILVs at the plasma membrane into the extracellular environment. As the early endosome forms within the cell and matures into a MVB, the lipid composition of the burgeoning MVB and associated proteins involved in cargo sorting (e.g., ESCRT/tetraspanins) are potentially influencing the composition of molecular cargo and the functional effect of EVs that will be released by that MVB. Moreover, what are the mechanisms that dictate the fate of the MVB, as it can either go to the cell periphery or to the lysosome for degradation. How under pathological conditions, such as DILI, is the balance between ESCRT-dependent and ESCRT-independent sorting affected, as these mechanisms control if therapeutic or deleterious cargo is loaded into the released EVs conferring the protective or harmful effect of EVs.

Research has increasingly evolved away from understanding the cellular origin pathways of EV biogenesis and moved toward a question of “what can EVs do…” in a certain model system, in a certain disease state etc. What EVs can functionally do is a direct result of the molecules that are loaded into them, therefore enhancing our understanding of the pathways governing cargo selection and loading under periods of cellular stress is paramount. This lack of foundational understanding and rampant misuse of terminology makes it difficult to interpret the current landscape of EV biology in DILI, especially in regard to the potential functional and therapeutic applications of EVs. Variations in isolation protocols and lack of quality control in studies makes sifting through the literature problematic as it is unclear as to what is being compared. While a fairly ubiquitous response to DILI, irrespective of the drug of interest, is increased release of EVs, whether this contributes to the recovery or detriment of the cell remains to be elucidated. The data suggests that the role of EVs is drug-dependent and that the functional impact of EVs may differ over time.

A critical, unanswered question is the distribution, uptake, and half-life kinetics for endogenous EVs. Nearly all the kinetic studies have collected EVs and then administered these exogenous EVs tracing them over time. These exogenous EVs may or may not mimic endogenous EVs, especially considering macrophages contribute significantly to degradation of EVs in the plasma [182]. However, the work investigating the use of EVs as vectors for therapeutic delivery, involving the repurposing of the endogenous EV biosynthesis system, can be potentially exploited to study endogenous EV kinetics. This understanding is necessary to validate the efficacy of EV-associated biomarkers.

The primary goals of EV research in DILI are to uncover novel biomarkers with greater utility and unlock the therapeutic potential of EVs engineered or endogenous. To effectively translate the data from preclinical models to the clinic requires rigorous quality control and specifically defining the EV subpopulation of interest (e.g., sEVs CD63/CD81 negative, CD9 positive) to ensure adequate comparisons across studies. Furthermore, as increasing evidence supports the role of EVs as a therapeutic, the necessary toxicologic, pharmacokinetic, and pharmacodynamic profiles need to be defined in preclinical models to help guide dosing regimens in human patients.

Abbreviation list

9, 10-DHOME

9, 10-dihydroxy-octadecanoic

ADH

alcohol dehydrogenase

AF4

asymmetric-flow-field-flow fractionation

AhR

aryl hydrocarbon receptor

AILI

alcohol-induced liver injury

ALIX

apoptosis-linked Gene 2-Interacting Protein X

ALP

alkaline phosphatase

ALT

alanine aminotransferase

ANIT

alpha-naphthyl isothiocyanate

APAP

acetaminophen

ARF6

ADP-ribosylation factor 6

ARMM

ARRDC1-positive MV

ARRDC1

arrestin domain-containing protein 1

BP

benzo[a]pyrene

CAR

constitutive androstane receptor

CKD

chronic kidney disease

DAMP

damage-associated molecular patterns

DBA

dibenzo[a,h]anthracene

DILI

drug-induced liver injury

DME

drug metabolizing enzymes

EGFR

epidermal growth factor receptor

eNOS

endothelial nitric oxide synthase

ESCRT

Endosomal Sorting Complex Required for Transport

EV

extracellular vesicle

FAS

fatty acid synthase

FX

flucloxacillin

GEDEX

genome editing with designed extracellular vesicles

GGT

gamma-glutamyl transferase

GPI

glycosylphosphorylinositol

GPX1

glutathione peroxidase 1

GSH

glutathione

GST

glutathione transferase

Hgf

hepatocyte growth factor gene

HO-1

heme oxygenase-1

HRS

hepatocyte growth factor-regulated tyrosine kinase substrate

HSLC-EVs

human liver stem cell-EVs

I/R

ischemia/repurfusion

IDILI

idiosyncratic drug-induced liver injury

ILV

intralumenal vesicle

JNK

c-jun N-terminal kinase

KC

Kupffer cells

LBPA

lyso-biphosphatidic acid

MLCK

myosin light chain kinase

MV

microvesicle

MVB

multivesicular bodies

NAPQI

N-acetyl-p-benzoquinone imine

NK

natural killer

NTA

nanoparticle tracking analysis

PA

phosphatidic acid

PAH

polyaromatic hydrocarbon

PAH

polycyclic aromatic hydrocarbons

PH

partial hepatectomy

PLD2

phospholipase D2

PLP

proteolipid protein

PMH

primary mouse hepatocyte

PPARγ

peroxisome proliferator-activated-gamma

PRH

primary rat hepatocyte

PtdIns3P

phosphatidylinositol 3-phosphate

PYR

pyrene

RNAi

RNA-interference

ROCK1

rho kinase

S1P

sphingosine-1-phosphate

sEV

small extracellular vesicle

shRNA

short hairpin RNA

siRNA

small interfering RNA

SK2

sphingosine kinase 2

Slp

synaptotagmin-like protein

SMX-NO

nitroso-sulfamethoxazole

SNAP25

synaptosomal-associated protein

STAM1

signal transducing adaptor molecule

TEM

tetraspanin-enriched membrane domains

TfR

transferrin receptor

TSG101

tumor susceptibility gene 101

Vps4

vacuolar protein sorting 4