Aquaporin-3 regulates endosome-to-cytosol transfer via lipid peroxidation for cross presentation

Antigen cross presentation, whereby exogenous antigens are presented by MHC class I molecules to CD8+ T cells, is essential for generating adaptive immunity to pathogens and tumor cells. Following endocytosis, it is widely understood that protein antigens must be transferred from endosomes to the cytosol where they are subject to ubiquitination and proteasome degradation prior to being translocated into the endoplasmic reticulum (ER), or possibly endosomes, via the TAP1/TAP2 complex. Revealing how antigens egress from endocytic organelles (endosome-to-cytosol transfer, ECT), however, has proved vexing. Here, we used two independent screens to identify the hydrogen peroxide-transporting channel aquaporin-3 (AQP3) as a regulator of ECT. AQP3 overexpression increased ECT, whereas AQP3 knockout or knockdown decreased ECT. Mechanistically, AQP3 appears to be important for hydrogen peroxide entry into the endosomal lumen where it affects lipid peroxidation and subsequent antigen release. AQP3-mediated regulation of ECT was functionally significant, as AQP3 modulation had a direct impact on the efficiency of antigen cross presentation in vitro. Finally, AQP3-/- mice exhibited a reduced ability to mount an anti-viral response and cross present exogenous extended peptide. Together, these results indicate that the AQP3-mediated transport of hydrogen peroxide can regulate endosomal lipid peroxidation and suggest that compromised membrane integrity and coordinated release of endosomal cargo is a likely mechanism for ECT.


INTRODUCTION
The cross presentation of antigen is required in order to generate an effective cytotoxic CD8+ T cell (CTL) response to viruses, bacteria, parasites, and cancer (4). Whereas CTLs can be primed by direct presentation on MHC class I, cross presentation allows for the generation of adaptive immunity to exogenous antigens not expressed by antigen presenting cells (APCs).
Although some cross presentation can occur within the phagosome itself (5)(6)(7), it is far more common for cross presented antigen to escape from phagosomes to be processed in the cytosol by proteasomal proteolysis with the resulting peptides transported into the ER via the TAP1/TAP2 complex and loaded onto MHC class I molecules. Owing partly to phagosomes characterized by higher pH, reduced enzymatic activity, and antigen preservation (8)(9)(10)(11), DCs are particularly efficient at endosome to cytosol transfer (ECT) of antigen (12,13), yet how this process occurs remains a mystery. One view is that antigens cross the endosome membrane by a process analogous to the retrotranslocation of content from the ER lumen for degradation in the cytosol (ERAD) (14,15). However, exhaustive proteomic analysis of isolated endocytic organelles has failed to identify likely candidates (16,17) and definitive genetic evidence is lacking (18). In addition, recent work demonstrated that the reduced cross presentation observed when the ERAD transporter Sec61 was inhibited was not due to a decrease in ECT, as was originally suggested, but rather correlated with a decrease in the expression of the machinery necessary for MHC class I presentation in general (19).
Another mechanistic explanation for ECT could be a coordinated, but more generalized process of antigen escape. Indeed, evidence in support of this mechanism was provided by a recent study focused on the role of phagosomal NOX2 and the generation of ROS in the phagosomal lumen, resulting in lipid peroxidation and antigen release (20). Although this is a key observation, there are several important questions that remain: namely, is NOX2-dependent endosomal ROS the only source of free radicals necessary to induce lipid peroxidation and antigen release? And, as Cybb -/-(NOX2deficient) mice do not have cross presentation defects in vivo, what other mechanisms are important for ROS generation and endosomal lipid peroxidation? Here, we provide evidence that ECT is coupled to the pathogentriggered release of mitochondrial ROS (mROS), which may then mediate lipid peroxidation and subsequent membrane disruption following enhanced uptake of H 2 O 2 by the endosomal peroxide channel aquaporin-3.

Two independent functional genomics screens identify aquaporin-3 (AQP3) as a regulator of ECT
To identify genes that might be involved in transferring internalized protein antigens from endosomes to the cytosol, we adapted a wellcharacterized -lactamase-CCF4 assay to assess ECT as a function of gene expression level (12,13,15). Bone marrow-derived dendritic cells (BMDCs) were loaded with a cytosolic fluorescent substrate that is cleaved by lactamase following its escape from endosomes (Fig. 1A). Live BMDCs were then sorted based on high levels of cleaved substrate ("ECT+") and no cleaved substrate ("ECT-") ( Fig. 1B). Differential expression analysis following RNA-Seq of both populations revealed a small fraction (~0.1%) of genes that were statistically significantly enriched or diminished (Table S1). A gene that was highly enriched in the most efficient ECT+ population was AQP3 (false discovery rate adjusted p-value<0.001, Fig. 1C). In parallel, we developed a second, independent screen using ECT reporter assay in the context of gene expression knockdown (Fig. 1D, E). In this approach, cells expressing the GAL4-UAS reporter element were incubated in the presence of recombinant GAL4 transactivating domain protein (GAL4-TA). For reporter expression to occur, GAL4-TA has to undergo ECT, translocate to the nucleus, and bind to its cognate upstream activating sequence. Using the GAL4-UAS ECT assay, we performed a siRNA knockdown screen in HEK293 cells to identify candidates involved in ECT. AQP3 was once again identified as a potential positive regulator of ECT. Knockdown of AQP3 and the resulting reduction in GAL4-UAS ECT was later confirmed in BMDCs, along with a second hit, ATP7a (a AAA-ATPase copper transporter) (Fig. 1F).
Since two independent screens identified AQP3 as a positive regulator of ECT, we investigated its function in greater detail.

AQP3 localizes to phagosomes and transports hydrogen peroxide
In order to examine the attributes of AQP3 important for ECT, we expressed other aquaporin family members and AQP3 mutants in HEK293 cells and performed the -lactamase ECT assay. Although WT AQP3 increased ECT compared to empty vector, an AQP3 channel mutant (A213W) (21) did not, demonstrating that channel function was required ( Fig. 2A). We next examined AQP2, an aquaporin family member that has been proposed to localize to early endosomes (22) but that strictly transports water (23), whereas AQP3 also transports glycerol and hydrogen peroxide (H 2 O 2 ) (24,25).
AQP2 overexpression did not increase ECT ( Fig. 2A), suggesting that aquaporin-mediated transport of a substrate other than water was important.
To further test this hypothesis, we generated an AQP3 mutant that was rendered water-specific by changing the glycine at position 203 to a histidine (AQP3 G203H). Aquaporins that transport only water have a histidine at position 203 while aquaporins that have broader substrate specificities exhibit small, uncharged amino acids at position 203 (glycine or alanine) (26). Indeed, expression of the water-selective AQP3 G203H mutant failed to increase ECT ( Fig. 2A). AQP3 and AQP9 are the only aquaporin family members expressed by BMDCs, with both thought to transport similar substrates (water, glycerol, (24). However, in the RNA-Seq ECT screen, AQP9 was not differentially expressed (Table S1) and its expression in HEK293 cells did not increase ECT ( Fig. 2A). Unlike AQP3, AQP9 is thought to be localized primarily at the plasma membrane. In fact, by fluorescence microscopy using RFP-tagged constructs, the distributions of the two aquaporins were distinct, with AQP9 being found at the plasma membrane with only a small amount of intracellular staining in the perinuclear region. By comparison, AQP3 was localized almost entirely to intracellular compartments (that were not AQP9-positive) (Fig. 2B) and could also be readily detected by immuno-electron microscopy on the phagosomal membrane in HEK293 cell transfectants allowed to internalized latex beads ( Fig. 2C). Immunofluorescence staining of endogenous AQP3 in BMDCs corroborates the pattern observed in transfected HEK293s, with AQP3 localized on or near the plasma membrane and dotting the interior of the cell ( Fig. 2D). Since these observations suggested that endolysosomal localization was an important factor in AQP3's ability to facilitate ECT, we generated a "mislocalization mutant" by mutating the two cytoplasmic domain dileucine motifs (AQP3 2xLLmut) previously shown to be important for endolysosomal targeting (16,27). By both cell fractionation (Fig. 2E) and fluorescence microscopy ( Fig. 2F), the AQP3 2xLLmut was found not to reach latex phagosomes and also failed to increase ECT relative to WT AQP3 ( Fig. 2A). In summary, the phagolysosomal localization and broader substrate specificity for H 2 0 2 and glycerol of AQP3 are key to its ability to promote ECT.
Reactive oxygen species (ROS) have been shown to play a role in cross presentation as a regulator of intraphagosomal pH or cysteine protease activity (28,29). We next asked if H 2 O 2 was also important for ECT. WT BMDCs were incubated in the presence of exogenously applied H 2 O 2 and found to significantly increase -lactamase release into the cytosol (Fig. 3A). To determine if endogenous ROS/H 2 O 2 could similarly increase ECT, we loaded BMDCs with the ROS sensor CM-H 2 DCFDA and treated with viral or bacterial components, using dextran as a negative control. Interestingly, -lactamase, which is bacterial in origin and a rich source of endotoxin, proved to be a potent stimulator of ROS production, superior to peptidoglycans (PGN) or lipopolysaccharide (LPS) (Fig. 3B). Based on staining with the reporter MitoSOX red, this increase appeared largely due to enhanced mitochondrial ROS (mROS) (Fig. 3C). In addition, the -lactamase effect on cellular ROS was significantly diminished in the presence of the mitochondrial electron transport chain inhibitor 2-thenoyltrifluoroacetone (TTFA) (Fig. 3B). TTFA also reduced the egress of -lactamase to the cytosol to a similar extent (Fig. 3D), consistent with the possibility that mROS was associated with ECT. In contrast, the phagosome-associated NOX2 system seemed to have little if any role in modulating ECT. BMDCs from Cybb -/-(NOX2-deficient) mice showed no significant decrease in -lactamase-induced ROS or in its release to the cytosol by ECT (Fig. S1).   These results implicated viral or bacterial products as potent stimuli of mROS in DCs, perhaps not surprising given a recent study demonstrating that pathogen sensing of internalized material was important for mROS production in macrophages (30). Interestingly, induction of mROS in macrophages was associated with the recruitment of mitochondria to phagosomes and a TRAF6-dependent assembly of the electron transport chain (30). As TRIF is upstream of TRAF6 in a pathogen-sensing signaling cascade, we investigated whether TRIF deficiency would alter ROS production and ECT.
Indeed, BMDCs from TRIF -/mice were deficient at ECT (Fig. 3E), consistent with a previous report (15), and produced less ROS in response to -lactamase ( Fig. 3B). Also consistent with the macrophage data, by electron microscopy we often found mitochondria in proximity to latex bead phagosomes (Fig. 2C), although this occurred regardless of whether TLR agonists were present.
Since -lactamase and other microbial agonists appear capable of generating mROS in DCs, we next asked if AQP3 might have a role in the transport of H 2 O 2 into the lumen of endocytic organelles. Although H 2 O 2 is often thought of as being membrane-permeable, passive diffusion across membranes is inefficient and is greatly facilitated by the presence of an appropriate aquaporin (31). We tested this possibility using recombinant HyPer (rHyPer) protein, a highly sensitive and specific H 2 O 2 sensor (32).
rHyPer was added to the media of HEK293 cells transfected with WT AQP3 or the channel mutant AQP3 A213W. When H 2 O 2 was added to the media, AQP3expressing cells had elevated phagosomal H 2 O 2 as compared to AQP3 channel mutants (Fig. 3F). rHyPer can be sensitive to pH changes in addition to H 2 O 2 , therefore, we tested the possibility of a difference in endo-lysosomal pH between HEK293 cells that expressed wild-type AQP3 or AQP3 channel mutants. We observed no difference in endo-lysosomal pH between cells expressing either construct (Fig. S2A)

AQP3 regulates ECT via endosomal lipid peroxidation
One possible mechanism whereby increased phagosomal H 2 O 2 might increase ECT is its well-known ability to cause lipid peroxidation and membrane damage (33,34). To assess directly whether AQP3 enhanced the extent of lipid peroxidation of phagosomal membranes, we exposed BMDCs to OVA-coated beads and a fluorescence-based lipid peroxidation indicator C11-bodipy. Cells were then homogenized and the beads were analyzed by flow cytometry. BMDCs from AQP3 -/mice had decreased phagosomal lipid peroxidation, but only when lipid peroxidation was induced with a lactamase bacterial stimulus during the bead/C11-bodipy incubation period ( Fig. 4A).  (8,11), and second, a signature of oxidative stress that is associated with enhanced ROS (Fig. 4B). Interestingly, the expression profile also exhibited the diminished expression of several iron-sequestering enzymes, which might favor iron-based Fenton reactions that are needed to produce membrane-damaging hydroxyl radicals from H 2 O 2 (35). The increased expression of the copper transporter ATP7A (Fig. 1F) is also intriguing, given the well-known role of copper (Cu-II) in catalyzing lipid peroxidation (36,37). These considerations are consistent with a model where antigen is released from phagosomes via compromised membrane integrity from lipid peroxidation damage.
We therefore asked if another approach to compromising the integrity of endocytic compartments might similarly yield an increase in ECT or cross presentation. ISCOMATRIX™ adjuvant is a formulation of saponin, phospholipids, and cholesterol that has demonstrated immune-stimulating properties, including the ability to increase CD8+ T cell priming in vitro and in vivo in a MyD88-dependent manner (38). Although the precise mechanism of action of ISCOMATRIX™ adjuvant is unknown, saponin is an active component suggesting a role for membrane permeabilization. We found that ISCOMATRIX™ adjuvant increased ECT (Fig. 4C) and cross presentation (Fig.   4D) in WT BMDCs. Importantly, the effect on cross presentation was sensitive to the proteasome inhibitor epoxomicin, indicating that the ISCOMATRIX™ adjuvant-induced pathway utilized the canonical pathway of antigen cross presentation (Fig. 4D). Thus, direct membrane disruption by ISCOMATRIX™ adjuvant enhanced both ECT and antigen presentation, consistent with the possibility that ROS-mediated phagosome disruption would perform similarly. Previous models of cross presentation have proposed that a transloconlike membrane channel such as Sec61 would serve as the route for antigen release and ECT (14,15). Although the identity of this putative channel has remained elusive, we devised an experiment to test this mechanism. Since translocation through an ERAD channel absolutely requires protein unfolding (39), we generated a fusion protein of -lactamase and mammalian dihydrofolate reductase (DHFR). DHFR fusion proteins have been used to evaluate the role of protein unfolding in various settings because DHFR forms a nearly irreversible high affinity complex with methotrexate (MTX) that stabilizes a folded conformation (40). Even though the -lactamase-DHFR fusion protein was sensitive to MTX inhibition indicating that at least the DHFR moiety was in a folded state (Fig. 4E), the -lactamase fusion protein underwent ECT at the same efficiency as PBS-treated protein (Fig. 4F), suggesting that its escape from endocytic compartments in DCs reflected a more non-specific process, such as coordinated leak, secondary to a loss of membrane integrity.

AQP3 facilitates antigen cross presentation in vitro and in vivo
We next evaluated the functional effects of AQP3 modulation in dendritic cells. Of the resident CD11c+ DC populations in mouse, the CD8+ and XCR1+ subsets are considered specialized for cross presentation (41), with recent data implicating XCR1+ DCs as the most efficient cross presenting cells in vivo (42)(43)(44). Previous studies have suggested that a significant portion of cross presentation efficiency in certain DC subsets results from limited degradation and increased ECT of internalized antigen (10,45). Therefore, we hypothesized that the more efficient the cross presenting cell, the more sensitive it would be to ECT perturbations induced by deleting AQP3. As shown in Figure 5A-C, AQP3 -/-BMDCs, as well as both CD8+ and XCR1+ splenic DCs, exhibited reduced ECT compared to WT controls. However, the magnitude of the difference between WT and AQP3 -/was greatest in XCR1+ DCs (Fig. 5C). The decrease was not explained by reduced -lactamase uptake by the mutant DCs (Fig. S2B-C). These results demonstrate a reliance on AQP3 for efficient ECT in all DCs tested, with the largest difference observed in the XCR1+ subset that is specialized for cross presentation. harvested and the generation of antigen-specific CD8+ T cell clones was evaluated using a HER-2/neu tetramer. *P<0.05, **P<0.01, two-tailed t-test or Mann-Whitney U test.
As AQP3 may play a partial role in the controlled disruption of endosomal compartments and thus cytosolic release of internalized antigen, we next asked if it was involved in antigen cross presentation in functionally relevant settings. We first overexpressed AQP3 in BMDCs by viral transduction and performed cross presentation assays. AQP3 overexpression in BMDCs increased cross presentation to both soluble and antibody-conjugated antigen (Fig. 5D, E). Importantly, presentation of pre-processed peptide was not affected in AQP3-overexpressing cells (Fig. 5D); antigen uptake ( Fig. S3A, B) and MHC class II antigen presentation were also unchanged (Fig. 5F). In contrast, AQP3 shRNA knockdown in WT BMDCs decreased ECT (Fig. 5G) and reduced cross presentation (Fig. 5H), with no measurable difference in antigen uptake (Fig. S3C, D).
To explore the role of AQP3 modulation in vivo, we evaluated the ability of AQP3 -/mice to control an infection with lymphocytic choriomeningitis virus (LCMV), a process that relies on efficient cross presentation of viral antigens (46)(47)(48). AQP3 -/mice were more susceptible to LCMV challenge as indicated by higher viral titer in the kidney 8 days post infection (Fig. 5I). In addition, AQP3 -/mice displayed a partial but significant reduction in the generation of LCMV antigen-specific IFN CD8+ T cells (Fig. 5J), again consistent with impaired cross presentation.
In order to follow up on the viral challenge results, we next asked whether AQP3 deletion altered the ability to prime naïve CD8+ T cells in response to exogenous antigen captured and cross presented by APCs. In these studies, WT or AQP3 -/mice were immunized intraperitoneally with recombinant HER-2/neu antigen and subsequent analysis of splenocytes 1 week later revealed a significant reduction in newly generated HER-2/neuspecific tetramer-positive CD8+ T cells in AQP3 -/mice (Fig. 5K). In contrast, the CD4+ T cell-dependent antigen-specific antibody response was similar in AQP3 -/mice compared to controls (Fig. S3E). Taken together, these data demonstrate that AQP3 can affect cross presentation in vivo, with no observable effect on CD4+ T cell-dependent/MHC class II-based antigen presentation.

DISCUSSION
This study provides new insight into the long-standing question as to how internalized material escapes from endocytic compartments. Although ECT is not wholly dependent on AQP3, expression of functional AQP3 capable of transporting H 2 O 2 into endocytic compartments increased the efficiency of ECT, while AQP3 deletion decreased ECT. Our data are inconsistent with a role for specific, unknown protein channels; however they are not inconsistent with the possibility that the ER itself has a role in these events, as has been suggested many times. Indeed, in yeast and other non-immune cells, there is increasing evidence of direct contact between ER and endosomal elements, which are important for endosomal sorting functions (49).
Additionally, we found that a variety of pathogenic stimuli enhanced production of mROS, thereby providing additional substrate for AQP3. This process was partially dependent on TRIF, which sits upstream of TRAF6, a molecule critically involved in ROS signaling (30,50). These results provide further support for a link between pathogen sensing and increased antigen presentation efficiency, and suggest that ECT is another step in the cross presentation pathway that to some degree is modulated by pathogenic stimuli.
Although we did observe a decreased anti-viral response and reduced ability to cross present extended peptide in the AQP3 -/mice, we no longer saw a consistent cross presentation defect in AQP3 -/once the mice were crossed to the C57BL/6 background and immunized with OVA-expressing necroptotic cells (51). In addition, given that the results presented here support an ECT model in which cargo egress occurs as a result of compromised membrane integrity from endosomes that have experienced lipid peroxidation and ostensibly is not antigen specific, we also tested if AQP3 deletion had any effect on STING pathway activation in macrophages (52). In these series of experiments, irradiated cells transfected with STING agonists were incubated with bone marrow-derived macrophages and indicators of intracellular STING pathway activation were measured, with no difference observed between AQP3 -/and WT control cells. The reason behind the lack of an effect of AQP3 deletion in these experiments is unclear, although the contribution of AQP3 to ECT may be most pronounced when assessed in the context of soluble protein antigen, as both functional genomics screens that identified AQP3 relied on this antigen delivery method.
Finally, while AQP3 is important in regulating ECT efficiency, it should be remembered that it is one of a number of specializations that together confer DCs with an enhanced capacity for cross presentation. Like AQP3, none of the specializations such as reduced levels of lysosomal proteases to preserve antigen, activation of mROS in response to TLR agonists, or expression of IL-12 to prime CD8+ T cells are DC specific, suggesting that ECT is a process that can occur in most cell types, albeit with reduced efficiency.
Indeed, a variety of internalized substances gain access to cytosolic compartments to facilitate surveillance by the innate immune system (16,53).
Viewed in this light, it is possible ECT is an evolutionarily conserved mechanism to enable all cells to sample pathogenic endosomal material and engage cytosolic sensors while maintaining a relatively safe topological barrier between the pathogen and the interior of the cell. Perhaps DCs and other antigen presenting cells have co-opted this process for cross presentation by rendering it more efficient and subject to some level of regulation by linking it to mROS generation.

RNA-Seq
RNA from flow-cytometry sorted BMDCs was isolated using the RNeasy kit (Qiagen), per the manufacturer's instructions. PolyA RNA-Seq was performed by Otogenetics (Norcross, GA) using an Illumina HiSeq sequencer. Cufflinks was used to carry out differential gene expression analysis.

ECT assays
For the CCF4 ECT system, cells were loaded with 2 M CCF4-AM (Life Technologies) substrate for one hour at room temperature. After 3 washes,

Measurement of DHFR inhibition by MTX
DHFR activity of the recombinant -lactamase-DHFR fusion protein was evaluated using the DHFR Assay Kit (Sigma-Aldrich), following the manufacturer's instructions. In the cell-free DHFR assay and the CCF4 ECT experiments, -lactamase-DHFR was exposed to 500 nM MTX (Sigma-Aldrich) for 5 minutes at room temperature prior to initiating analysis.

Phagocytosis measurements
Alexa Fluor (AF)-488-labeled or AF-647-labeled -lactamase or OVA were incubated for 30 minutes at the indicated doses. 1 g/ml AF-647-labeled DEC205 antibody was loaded at 4°C for 20 minutes, washed, and then incubated for an additional 30 minutes at 37°C.

BMDC viral transduction
293T cells were used to generate retrovirus. Lin-hematopoietic progenitors were isolated from bone marrow from WT mice using a lineage depletion kit (Miltenyi). Lin-progenitors were transduced with retroviral supernatants by spinfection followed by incubation for 3.5 hours at 32°C. Transduction media was removed, replaced with media containing 50 ng/ml SCF (Peprotech), 10 ng/ml GM-CSF, and 5 ng/ml IL-4, and cells were incubated for 2 days at 37°C.
On day 2, transduced progenitors were sorted by flow cytometry and replated in GM-CSF/IL-4. Cultures received fresh GM-CSF/IL-4 media on day 5 and were analyzed on day 6. AQP3 knockdown efficiency was 93%, as determined by real-time PCR.

In vivo experiments
For LCMV, LCMV Armstrong stocks were prepared and quantified as previously described (56). Mice (CD-1 background) were infected intravenously with 2x10 6 plaque-forming units (PFU). 8 days after infection, mice were euthanized and tissue was analyzed. For LCMV antigen-specific IFN producing cells, splenocytes were isolated, stimulated with 1 g/ml LCMV NP 118-126 peptide for 1 hour and then in the presence of GolgiPlug (BD Biosciences) and peptide for an additional 4 hours, followed by staining with rat anti-mouse IFN (eBiosciences) using eBiosciences intracellular staining reagents. For viral titer, monolayers of MC57 cells were infected with serially diluted tissue homogenates. 72 hours after infection, the cells were fixed with 4% paraformaldehyde and permeabilized with 0.5% Triton-X. Viral plaques were stained with anti-LCMV nucleoprotein (anti-LCMV NP, clone VL4) and