The Perforin Pore Facilitates the Delivery of Cationic Cargos*

Background: Perforin is a pore-forming protein that delivers granzymes to eliminate compromised cells. Results: The perforin pore preferentially delivers cationic molecules in comparison to anionic or neutrally charged cargo, which are inefficiently delivered. Conclusion: Perforin delivers cationic cargos more efficiently than anionic or neutral cargos. Significance: This is the first report of charge-based discrimination by the perforin pore. Cytotoxic lymphocytes eliminate virally infected or neoplastic cells through the action of cytotoxic proteases (granzymes). The pore-forming protein perforin is essential for delivery of granzymes into the cytoplasm of target cells; however the mechanism of this delivery is incompletely understood. Perforin contains a membrane attack complex/perforin (MACPF) domain and oligomerizes to form an aqueous pore in the plasma membrane; therefore the simplest (and best supported) model suggests that granzymes passively diffuse through the perforin pore into the cytoplasm of the target cell. Here we demonstrate that perforin preferentially delivers cationic molecules while anionic and neutral cargoes are delivered inefficiently. Furthermore, another distantly related pore-forming MACPF protein, pleurotolysin (from the oyster mushroom), also favors the delivery of cationic molecules, and efficiently delivers human granzyme B. We propose that this facilitated diffusion is due to conserved features of oligomerized MACPF proteins, which may include an anionic lumen.


Cytotoxic lymphocytes eliminate virally infected or neoplastic cells through the action of cytotoxic proteases (granzymes).
The pore-forming protein perforin is essential for delivery of granzymes into the cytoplasm of target cells; however the mechanism of this delivery is incompletely understood. Perforin contains a membrane attack complex/perforin (MACPF) domain and oligomerizes to form an aqueous pore in the plasma membrane; therefore the simplest (and best supported) model suggests that granzymes passively diffuse through the perforin pore into the cytoplasm of the target cell. Here we demonstrate that perforin preferentially delivers cationic molecules while anionic and neutral cargoes are delivered inefficiently. Furthermore, another distantly related pore-forming MACPF protein, pleurotolysin (from the oyster mushroom), also favors the delivery of cationic molecules, and efficiently delivers human granzyme B. We propose that this facilitated diffusion is due to conserved features of oligomerized MACPF proteins, which may include an anionic lumen.
Cytotoxic lymphocytes (CLs) 2 eliminate infected or compromised target cells as part of the immune response. This elimination is achieved through exocytosis of lytic granules containing a pore-forming protein, perforin and the granzyme family of serine proteases (1). Granzymes induce apoptosis following delivery into the target cell cytoplasm by perforin, with human Granzyme B (hGrB) being the most potent cytotoxin of the family (2)(3)(4). Elimination of target cells by granule exocytosis is entirely dependent on perforin (5,6).
The mechanism through which perforin delivers granzymes into the target cell cytoplasm remains a topic of debate, reviewed in Ref. 7. One view is that perforin and granzymes are co-endocytosed by the target cell, and perforin disrupts an endocytic compartment allowing egress of granzymes into the cytoplasm (8 -10). However, there are multiple lines of evidence that suggest perforin forms a pore at the plasma membrane to allow access of granzymes into the target cell. These include images of pores on the target cell membrane after a killer cell attack (11); the finding that the calcium influx after treatment with perforin only lasts for 180 s (12); and perforinmediated delivery and release of various molecules including fluorescein isothiocyanate (FITC)-labeled dextrans, up to 20 kDa in size, into resealed human erythrocyte ghosts and giant unilamellar vesicles (13)(14)(15). It has recently been reported in models that examine physiologically relevant immune synapses that propidium iodide (PI) is very rapidly delivered across the target cell membrane by perforin, further supporting a diffusion model (12,16).
The requirements for granzyme delivery by perforin remain unclear, and the models do not provide much insight into whether the perforin pore only delivers granzymes, or if it is a simple aqueous channel, potentially allowing passive diffusion of many different cargos. During a natural killer cell attack the perforin pore on target cell membranes is only open very briefly before it is removed by wound healing, making efficient delivery of cargo essential (12). Electrostatic interaction is proposed to be important for efficient delivery of hGrB (17,18), as neutralization of specific cationic motifs on the surface of hGrB markedly reduce its cytotoxicity (18). These data suggest that perforin requires positive surface charge on cargo for effective delivery. Furthermore there are multiple reports that suggest perforin is unable to deliver various molecules, most of which are neutral in charge, into nucleated target cells. These include small dyes and proteins used as controls in various studies, such as 8-hydroxypyrene-1,3,6-trisulfonic acid (19); sytox green (10); FITC-labeled hemopoietic-specific protein HS1 and larger molecules including FITC-dextran (20 kDa) (20); green fluorescent protein (GFP); and FITC-glutathione S-transferase (GST) (21).
Intriguingly, cryo-EM structures of the perforin pore reveal a structure around 16 nm in diameter, which is more than sufficient to allow passive diffusion of the molecules listed above (11,(22)(23)(24)(25). In comparison, a bacterial pore-forming protein, streptolysin O (SLO), is able to deliver a variety of molecules into mammalian target cells, up to 150 kDa without apparent restriction (26). SLO is a cholesterol-dependent cytolysin (CDC) for which the process of pore formation is well described (reviewed in Ref. 27). SLO binds to the membrane and then oligomerizes to form a pre-pore structure, which subsequently punctures the membrane to form an aqueous channel across the membrane, ϳ30 nm in diameter (28), through which molecules can passively diffuse (26,28,29).
Here we show that perforin is capable of delivering molecules and proteins other than granzymes into target cells. However perforin shows a preference for cationic cargo, as anionic and neutrally charged test molecules were inefficiently delivered into cells. By contrast, all test molecules and proteins were delivered by SLO, irrespective of charge. These observations were extended using the natural perforin cargo granzyme B, where moderating the surface charge of granzyme B reduced the efficiency of delivery by perforin. Interestingly, a similar preference for cationic cargo was observed for the distantly related MACPF family member pleurotolysin. We conclude that physiologically relevant perforin cargos have evolved to enter the cell via a process of facilitated diffusion through the perforin pore.
Recombinant Proteins and Fluorescent Molecules-Cytochrome c from equine heart was purchased from Sigma (cat no. 2506). Human Bid (N/C-Bid) plasmid, a gift from Prof. Newmeyer, was purified as previously described (30,31). Briefly, full-length bid fused to GST was expressed in BL21 (DE3) pLysS Escherichia coli. The soluble material was collected and purified over glutathione-Sepharose (HiTrap, GE Healthcare). This was then activated with thrombin using a cleavage site introduced in place of the caspase cleavage site in the full-length human Bid sequence. The activated (N/C-Bid) was then purified from the GST by glutathione-Sepharose chromatography (HiTrap, GE Healthcare) followed by nickel affinity chromatography (HiTrap, GE Healthcare). As this system is detergentfree, the pro-domain remains associated and hence produces a 21 kDa-activated human Bid (N/C-Bid), as described (31).
Human GrB (hGrB) was expressed and purified as previously described (32). The pro-protease form of mouse granzyme B (mGrB) was modified to include an N-terminal hexa-histidine (6-His) tag and bovine enterokinase cleavage site and cloned into the bacterial expression plasmid pEXP-His. The plasmid was transformed into BL21 arabinose-inducible (AI) E. coli and recombinant protein expressed in inclusion bodies was collected by sonication and subsequent centrifugation of the lysed bacterial pellet. Collected inclusion bodies were washed three times to remove non-protein impurities: Wash 1 (25 mM Tris pH 7.4, 2 mM EDTA, 20 mM DTT, 1% (v/v) Triton X-100); Wash 2 (25 mM Tris pH 7.4, 2 mM EDTA, 1 M NaCl); and Wash 3 (25 mM Tris pH 7.4, 2 mM EDTA). The insoluble protein pellet was collected by centrifugation and denatured by the addition of a solution of 7 M guanidine, 100 mM DTT, and 100 mM Tris, pH 8.3. The denatured protein was refolded by slow, dropwise addition into a Refolding Buffer (50 mM Tris pH 7.4, 1 mM EDTA, 1 mM L-cysteine, 3 mM L-cystine, 600 mM L-arginine, 500 mM NaCl, and 10% (v/v) glycerol). The solution containing soluble and correctly folded pro-granzyme was concentrated by Tangential Flow Filtration (TFF) before activation with bovine enterokinase as described previously (32). Mature mGrB was collected by applying the protein mixture to an SP Sepharose Fast Flow HiTrap FPLC column (GE Biosciences), washing and then eluting with an increasing NaCl gradient (according to the manufacturer's instructions). Methylation of mGrB surface lysine residues was achieved by following the protocol previously described by Walter et al. (33). Activity of methylated mGrB was assessed as described (2,34).
FITC-CM-dextrans and FITC-DEAE-dextrans were purchased from Tbd Consultancy and reconstituted at 1 mM in distilled water.
Enhanced green fluorescent protein (eGFP) was modified to include a 6-His tag at the N terminus, for the construction of eGFP-heparin binding site fusion (eGFP/HS) the heparin biding site identified in interferon-␥ (AAKTGKRKRSQML-FRGRRASQ) (35) was added to the C terminus. These were cloned into the pET28b plasmid and eGFPϮHS were produced in BL21 AI E. coli and purified by nickel affinity chromatography (HiTrap, GE Healthcare) followed by phenyl-Sepharose chromatography (HiTrap, GE Healthcare).
Supercharged GFP expression plasmids were a gift from Prof. Liu and were expressed and purified as previously described (36), with additional purification conducted with appropriate affinity chromatography (SP, Q, or phenyl-Sepharose).
The SLO expression plasmid, a gift from Prof. Bhakdi, was produced and purified as previously described (37). Recombinant mouse perforin was produced as previously described (38).
Full-length PlyA and PlyB genes were chemically synthesized (TopGene Technology) with a 5Ј NdeI and a 3Ј BglII restriction enzyme recognition sites outside the coding region. Full-length PlyA was cloned into NdeI and BamHI sites of pET3a (Agilent Technologies). The vector was transformed into BL21 (DE3) pLysS (Novagen).
PlyA culture was grown at 310 K until A 600 of 0.6 when protein expression was induced with 1 mM isopropyl ␤-D-1-thiogalactopyranoside. The cell culture was then allowed to cool on the bench for 20 min before being grown for a further 20 h at 289 K, after which they were harvested by centrifugation. The pellet was resuspended in 10 mM Tris-HCl pH 8.4, 10 mM EDTA, and 1 mg ml-1 lysozyme. The cells were then sonicated and insoluble material pelleted. Soluble PlyA was precipitated with 1.4 M (NH 4 ) 2 SO 4 and subsequently collected by centrifugation (15,000 ϫ g, 30 min). The precipitant was dissolved in 10 mM Tris-HCl, pH 8.4 then dialyzed against 10 mM Tris-HCl, pH 8.4 buffer for 24 h with 3 buffer changes to remove remaining (NH 4 ) 2 SO 4 . PlyA was then purified over a Q FF column (HiTrap, GE Healthcare), phenyl-Sepharose FF column (HiTrap, GE Healthcare), and finally size-exclusion chromatography (Superdex 75 16/60, GE Healthcare). The mature form of PlyB (145-1569 bp) was amplified and cloned into NdeI/BamHI sites of pET3a expression vector (Agilent Technologies). This vector was then transformed into E. coli expression cells BL21 (DE3) pLysS (Novagen). For protein production cells were grown to A 600 of 0.4 and expression induced with 1 mM isopropyl ␤-D-1-thiogalactopyranoside for 4 h. Cells were harvested by centrifugation (3,000 ϫ g, 20 min) and then resuspended in 50 mM Tris-HCl, pH 8.0 buffer containing 1% (w/v) Triton X-100, 1% (w/v) sodium deoxycholate, 100 mM NaCl, 5 mM MgCl 2 , 0.1 mg ml Ϫ1 DNaseI, and 1 mg ml Ϫ1 lysozyme. The suspension was incubated at room temperature for 30 min before the cells were sonicated (4 ϫ 30 s). Inclusion body pellet was collected by centrifugation (15,000 ϫ g, 277 K, 30 min) and resuspended in 50 mM Tris-HCl, pH 8.0 buffer containing 0.5% (w/v) Triton X-100, 100 mM NaCl, 1 mM EDTA, and centrifuged (15,000 ϫ g, 15 min). This was repeated until inclusion body appeared white (typically 1-3 times). Inclusion bodies were then washed with 50 mM Tris-HCl, pH 8.0 buffer containing 1 mM EDTA. The inclusion body was dissolved in 8 M urea buffer containing 50 mM Tris-HCl, pH 8.5, 10 mM ␤-mercaptoethanol and incubated at room temperature for 30 min. A 1/250 (v/v) dilution of dissolved inclusion body, containing PlyB, was added dropwise into a 50 mM Tris-HCl, pH 8.5 refold buffer containing 150 mM NaCl, 0.1 mM EDTA, 5% (v/v) glycerol, leaving to stir for 16 h. For purification of PlyB, adapted from (39), ammonium sulfate was slowly added to the refolded protein at a final concentration of 1.2 M and left for 30 min before the solution was bound to the phenyl-Sepharose resin (GE Healthcare) then further purified using size-exclusion chromatography (Superdex 200 10/30, GE Healthcare).
Heparin-Sepharose Pull-down Assay-Heparin-Sepharose (100 l of 50% (v/v) slurry) (GE Healthcare) was washed into buffer containing 20 mM Tris-HCl, pH 7.5, 10 mM NaCl then incubated at room temperature for 30 min with 15 g of eGFP or eGFP/HS in the same buffer. The resin was centrifuged and washed twice in the above buffer. Bound protein was eluted three times with 200 l of buffer containing 20 mM Tris, pH 7.5, 2 M NaCl. Samples were mixed with an equal volume of nonreducing Laemmli Sample Buffer (62.5 mM Tris, pH 6.8, 2% (w/v) sodium dodecyl sulfate (SDS), 10% (v/v) glycerol, 0.1% (w/v) bromphenol blue), and analyzed by 12% SDS-polyacrylamide gel electrophoresis (PAGE) followed by analysis on a Typhoon Trio (GE Healthcare) scanner with excitation at 488 nm and emission detected at 520 nm. Images were collected using 500 volts and 200 microns.
Cytotoxicity Assay-Cell death assays with serial dilutions of the required cytotoxic protein in the presence of SLO or perforin were carried out as previously described (18,40). Briefly perforin and SLO were titerd on the same day of each experiment to establish a baseline of 10 -30% lysis in the cell population (sub-lytic concentration) to ensure that the concentration of lytic agent was not a limiting factor. The protein was serially diluted in perforin or SLO, and cells were then added at 4 ϫ 10 6 per ml and incubated for 30 -60 min at 37°C. Cells were then washed in complete medium and plated out in triplicate, allowed to recover overnight, and analyzed for survival.
Fluorescent Molecule Delivery Assay-Fluorescent molecules (5 M final) were incubated with K562 cells (4 ϫ 10 6 cells/ml) in the presence or absence of a sub-lytic concentration of delivery agent (SLO, perforin or Ply) for 30 min at 37°C. Complete medium was added and cells were either washed into ice cold flow cytometry (FACS) buffer (phosphate-buffered saline containing 0.5% (v/v) heat-inactivated fetal calf serum, and 2 mM EDTA) or placed on slides for imaging. Immediately before cells were analyzed by FACS. PI was added to a final concentration of 4 g/ml. Bromphenol blue was also added immediately before analysis by FACS to a final concentration of 1 mM to quench any fluorescence of surface-bound material (41). For FACS data analysis, all PI-positive cells were excluded and viable cells were then analyzed for delivery of the fluorescent molecule above the basal level of uptake via endocytosis. For imaging, cells were placed on Cell Tak (BD Biosciences) treated slides for at least 10 min at 37°C, washed in PBS and treated with PBS with DAPI at 0.1 g/ml for 10 min at room temperature. Cells were imaged using an upright Nikon epifluorescent microscope.
Native Isoelectric Focusing-5 g of total protein was loaded onto a native isoelectric focusing (IEF) gel (Novex) and analyzed by Coomassie Brilliant Blue staining. The gel was run as per the manufacturer's instructions.
Size-exclusion Chromatography (SEC)-A minimum of 100 g of protein was loaded over an S200 Superdex gel filtration column. The column was run in 20 mM Hepes pH 7.5, 150 mM NaCl or the same buffer containing 1 M NaCl. Protein standards were used to calibrate the column in the same buffer, these were a mixture of ovalbumim (45 kDa), ribonuclease A (13.7 kDa), and a blue dextran (2 kDa).

Perforin Preferentially Delivers Cationic
Cargo-To investigate whether perforin has a requirement for cationic cargo we used a panel of positive and negatively charged FITC-dextrans ranging from 4 to 70 kDa. This approach allowed us to compare molecules, which are the same size and structure but have opposite charges, detecting delivery as fluorescence in the cytoplasm of target cells treated with perforin or SLO. As cationic molecules are electrostatically attracted to the plasma membrane and non-specifically endocytosed, we saw a higher basal level of fluorescence (in the absence of pore-forming proteins) with the positive FITC-dextrans than with the negative counterparts. Therefore we present the data from FACS analysis as a ratio between the mean fluorescence intensity (MFI) in the presence of the delivery agent (delivery) and the MFI in its absence (basal endocytosis), hence a value of 1 represents no delivery. Perforin delivered the positive FITC-dextrans regardless of molecular weight, but excluded all of the negative FITCdextrans (Fig. 1, A and B). Epifluorescence images are consistent with the cationic fluorescent dextrans reaching the cell cytoplasm when delivered, rather than being trapped in another intracellular compartment (Fig. 1A). By contrast, when SLO was used both positive and negative FITC-dextrans were delivered into K562 cells with a similar efficiency (Fig. 1, A and  B). These data suggest that perforin preferentially delivers positively charged molecules and provides visual evidence that this cargo, when delivered, reaches the cytoplasm of target cells.
To further assess the apparent enhanced delivery of cationic molecules by perforin, we used mouse granzyme B (mGrB) on which surface lysine residues had been artificially methylated, thus moderating its overall surface charge. The predicted isoelectric point (pI) of mGrB is 9.72, and when evaluated by native IEF we found that mGrB does not enter the gel due to its high pI. By contrast, the methylated mGrB in comparison does just enter the gel, suggesting that the pI has been lowered (to ϳ8.5) but it remains cationic (Fig. 1C). Importantly, methylation did not affect the enzymatic function of mGrB at all (data not shown). As expected, when delivered by SLO there was no difference in the killing ability of mGrB when methylated (Fig.  1D). However when delivered by perforin a 4-fold increase in the EC 50 of the methylated mGrB was observed, indicating that the decrease in positive charge had lowered its delivery efficiency (Fig. 1, E and F). Consistent with the dextran results, this suggests that the more positive charge a cargo carries the more successfully it is delivered by perforin. Furthermore it supports the notion that cationic cargo are delivered more efficiently than neutral or negative molecules in the immune synapse.
Perforin Facilitates Delivery of Cationic Cytochrome c but Does Not Deliver Neutral-activated Human Bid-Dextrans are polysaccharides with an elongated, branched chain-like structure, and therefore may not be effective models for the delivery of (globular) proteins (42). Next we sought to assess whether positive charge is sufficient for the delivery of natural proteins small enough to fit through the perforin pore. We used cytochrome c (cytc) from equine heart and cleaved (activated) human Bid (N/C-Bid) (described previously, 31). These proteins are both involved in the induction of apoptosis in cells, hence cytotoxicity can be used as a measure of delivery into the target cell cytoplasm. Activated Bid acts to initiate the release of cytc from the mitochondria through the action of Bax and Bak (43). Cytc initiates formation of the apoptosome and subsequent activation of the caspase cascade (44).
Cytc and N/C-Bid, 14 kDa and 21 kDa, respectively ( Fig. 2A), are both easily small enough to passively diffuse through the lumen of the full-sized, 16 nm, perforin pore (25). Cytc has been shown experimentally to have a pI of 10.5 (45) however, the pI for Bid has not been reported. We found that while the theoretical pI of N/C-Bid was 9.7 (46), the actual pI as determined by native IEF was far lower, 6.7 (Fig. 2B). This indicated that N/C-Bid's net surface charge is neutral. Cytc on the other hand (pI of 10.5) did not enter the isoelectric focusing gel, as expected for proteins with a pI higher than 8.3. Size exclusion chromatography further demonstrated that both cytc and N/C-Bid are monomeric and ϳ14 kDa and 21 kDa respectively (Fig. 2, C and  D). Therefore if perforin forms a simple pore to allow the nonselective entry of granzymes and potentially other cargo by passive diffusion, both proteins should pass through the pore and initiate apoptosis in the target cells. Furthermore SLO should deliver these small cytotoxic proteins by passive diffusion as shown previously for numerous molecules smaller than 150 kDa (26,47).
As previously reported, when delivering hGrB there is no difference in the EC 50 when using SLO compared with perforin ( Fig. 2E) (18). As predicted cytc and N/C-Bid were delivered by SLO and induced cell death (Fig. 2, F and G). The average EC 50 for cytc and N/C-Bid was 3.8 M and 4.1 M respectively. In marked comparison, perforin was only capable of delivering cytc, with an EC 50 of 24.2 M (Fig. 2F). There was no cell death induced when N/C-Bid was incubated with perforin, suggesting that perforin is unable to deliver N/C-Bid into the cytoplasm of target cells (Fig. 1G). To ensure that N/C-Bid does not interfere with perforin pore formation we performed a hGrB cytotoxicity assay in the presence of 32 M N/C-Bid (the highest concentration used in the previous assay). This showed that perforin is able to efficiently deliver hGrB in the presence of N/C-Bid and thus perforin's core function was not impaired (Fig. 2H). Together, these results suggest that perforin is able to deliver cationic proteins such as hGrB and cytc, but is unable to efficiently deliver neutrally charged proteins such as N/C-Bid that are small enough to diffuse through the perforin pore.
Perforin Excludes eGFP, and the Addition of a Cationic Heparin Binding Site Is Not Sufficient for Delivery-Although our experiments with small cytotoxins and FITC-dextrans show perforin selectively delivers cationic cargo, these molecules are not geometrically similar to hGrB, which is a larger globular molecule (48,49). Next we sought to see if the requirement for cationic charge could be met by adding the heparin binding site from interferon-␥ to a protein normally excluded by perforin, enhanced (e)GFP. As reported previously, hGrB binds heparin through two cationic sites, which are also important for efficient delivery by perforin and subsequent cytotoxicity (18). These sites are in close proximity in the hGrB tertiary structure but are separated by 123 residues in the primary sequence and therefore do not form an appropriate tag sequence for adding to eGFP (18). Instead the heparin binding site from interferon-␥ was used, which comprises two adjacent clusters of basic residues similar to that of hGrB (35). eGFP is a 27-kDa globular molecule with similar dimensions to hGrB. It has an experimentally determined neutral pI (and therefore net neutral surface charge) and is excluded from cells that have been permeabilized by perforin (21,50). We added the interferon-␥ heparin binding site (HS) to the C terminus of eGFP to determine whether this is sufficient to allow its delivery by perforin. Recombinant eGFPϮHS were purified and analyzed by SEC and native IEF (Fig. 3, A-C). This showed that the preparation was pure and monomeric. When resolved by native IEF the pI for eGFP alone was 5.2 and with the addition of the heparin binding site the pI increased to ϳ6.1 (Fig. 3C). The functionality of the HS tag was confirmed by a heparin-Sepharose pull-down assay (Fig. 3D).
Both proteins were delivered by SLO, assessed by microscopy as fluorescence in the cytoplasm, consistent with previous data (47), and a large shift in MFI above that of basal endocytosis was observed by FACS (Fig. 3, E and G). However when incubated in the presence of perforin, both microscopy and FACS analysis showed no increase in cytoplasmic fluorescence above basal endocytosis for either eGFP or eGFP/HS (Fig. 3, F and G). This was not cell-type specific as the results were reproduced in P815 cells (data not shown). To exclude the possibility that eGFP inhibited perforin's function, hGrB was delivered by perforin in the presence of eGFP and again, similar to N/C-Bid, this showed perforin's activity was not compromised under these assay conditions (Fig. 3H). These results show that the addition of a structured cationic heparin binding site, similar to that on hGrB, is not sufficient to allow delivery by perforin, and/or suggest that a higher overall surface charge is required by perforin cargo.
Overall Increase in Positive Charge Increases Efficiency of Delivery by Perforin-The majority of the surface of hGrB is positively charged whereas eGFP is mostly neutral. Therefore we reasoned that a GFP with an increased, evenly distributed positive surface may be delivered by perforin. We utilized previously characterized supercharged GFPs, where residues at the surface were mutated to impart a net negative or positive charge (36). We obtained the Ϫ30 GFP, stGFP (not mutated standard), ϩ15 GFP, and ϩ36 GFP mutants generated by Lawrence et al. The supercharged GFP variants were purified and analyzed by native IEF. This showed that non-mutant stGFP had a pI of ϳ6.0, Ϫ30 GFP had a pI of 4.3 while ϩ15 and ϩ36 GFP both had pIs greater than 8.3 as they did not run into the gel (Fig. 4, A and B), consistent with previous results (36). In addition, all mutants were shown to be monomeric by size exclusion chromatography and hence were theoretically small enough to pass through the perforin pore (Fig. 4, C and D) (25). When delivered by SLO all of the supercharged GFPs arrived in the cytoplasm (Fig. 4E). When incubated with perforin stGFP and Ϫ30 GFP were inefficiently delivered over basal endocytosis as expected (Fig. 4F), while ϩ15 and ϩ36 GFP were delivered, but less efficiently than with SLO.
Pleurotolysin Shows a Preference for Cationic Cargo-Ply is a MACPF protein identified in the oyster mushroom Pleurotus ostreatus. It is composed of two proteins, PlyA and PlyB (39,51). PlyA binds to the membrane in a sphingomyelin-dependent manner, PlyB then binds to PlyA and inserts into the membrane (39,51). This two-component system forms small pores in the order of 3-5 nm in diameter (51). Ply is evolutionarily distant from perforin and therefore makes an interesting com- . eGFP is not delivered by perforin and the addition of a cationic heparin binding site is not sufficient for delivery. A, 15% SDS-PAGE loaded with 5 g of purified eGFP and eGFP/HS stained with Coomassie Brilliant Blue. B, native isoelectric focusing gel with a gradient of pH3-10 loaded with 5 g of eGFP and eGFP/HS and stained with Coomassie Brilliant Blue. The pI of the folded eGFP was found to be 5.3, and the addition of the heparin binding site increased this to 6.3. C, representative SEC trace for eGFP. This confirmed that eGFP was monomeric and eluted at 16.5 ml, corresponding to a size of ϳ30 kDa. D, heparin-Sepharose pull-down assay with eGFP versus eGFP/HS. Samples were run non-reduced and non-boiled on 12% SDS-PAGE, which was then scanned for green fluorescence. E, K562 cells were incubated with 5 M eGFPϮHS in the presence of SLO, and representative images at 40ϫ are shown. F, representative images of delivery of eGFPϮHS as per E but using perforin in place of SLO. G, delivery of eGFPϮHS with SLO (n ϭ 2) and perforin (n ϭ 3) represented as the MFI ratio as previously described in Fig. 1B. H, cytotoxicity of hGrB when delivered into P815 cells with perforin in the presence or absence of 5 M eGFP. MARCH 28, 2014 • VOLUME 289 • NUMBER 13 parator to assess whether the apparent selectivity of the pore is a feature of this family. First we examined the ability of Ply to deliver hGrB. Ply delivered hGrB with a mean EC 50 of 12 nM, which is slightly higher than when delivered by either of SLO or perforin (typically 1-5 nM for Jurkat cells) (Fig. 5A). Next we assessed delivery of cytc and N/C-Bid, much smaller proteins that should be more easily delivered by Ply. Indeed cytc was delivered with an EC 50 of 41 M, higher than when delivered by either SLO or perforin (Fig. 5B). On the other hand, N/C-Bid and eGFPϮHS were excluded by Ply (Fig. 5, C and D).

Facilitated Diffusion through the Perforin Pore
Overall, these results suggest that the Ply pore is similar to perforin and preferentially delivers cationic molecules. It is formally possible that concentrations of N/C-Bid applied (17 M) were not high enough (cytc was delivered ϳ10 times worse than with SLO, which if also true for N/C-Bid might mean an expected EC 50 as high as 44 M). On the other hand if the Ply pore is non-selective, eGFP should have been delivered similarly to hGrB as they are very similar in size and dimensions.

DISCUSSION
In this study we have conclusively shown that perforin is able to deliver molecules other than granzymes into target cells. However perforin does not deliver all cargo equally, and preferentially delivers cationic molecules compared with anionic or neutral molecules, an observation not previously reported. What is the explanation for this charge-based facilitated diffusion? We originally hypothesized that granzyme delivery by perforin involves electrostatic interaction at the plasma mem-  brane where the positively charged granzyme reversibly binds to negatively charged moieties in the membrane (18). This would concentrate granzymes in closer proximity to the perforin pore, allowing more efficient delivery by diffusion. The data presented here, however, suggest that the mechanism of granzyme delivery by perforin is more complicated.
Previously there has been little direct evidence to suggest that perforin is able to deliver positively charged molecules other than granzymes into nucleated cells, for example there is conflicting data on whether perforin is capable of delivering the 8 kDa cationic protein, azurin (20,21). Indeed, the only evidence of a foreign molecule, other than PI, being delivered by perforin into target cells is from Thiery et al. who demonstrated the delivery of a cationic 10 kDa Texas red labeled dextran (TRdextran) (8). They attribute this to TR-dextran's ability to be readily endocytosed (8), however we consider endocytosis an unlikely explanation because these same studies showed that hGrB-mediated apoptosis proceeded in the absence of endosome formation, and another study has shown that hGrB is delivered just as efficiently when endocytosis is blocked (12).
We have demonstrated here that even though it has a much smaller pore size perforin is able to deliver positively charged FITC-dextrans as efficiently as SLO. Perforin delivers the 20 -70 kDa FITC-DEAE-dextrans as well or better than SLO, however the 4 kDa FITC-DEAE-dextran is delivered less efficiently than the larger counterparts by perforin. The inverse is true for SLO with this smaller dextran being delivered more efficiently. This difference may be reconciled by considering the difference in total positive charge, as the smaller the dextran the less charge it carries. The preference of perforin for charged dextrans is inconsistent with previous studies that show delivery of neutral dextrans into liposomes (13), however the latter observations are likely explained by the fact that there is no wound healing response in an artificial membrane system and therefore pores remain on the surface for an extensive period of time (up to 45 min). In mammalian cells the pores are only available for seconds before they are removed due to wound healing (12).
The preference of perforin for cationic cargos was supported using perforin's natural cargo, mGrB. When methylated, the surface charge of mGrB was slightly reduced and the EC 50 was increased 4 fold, demonstrating that positive charge contributes to delivery by perforin, and is not important for the nonselective delivery agent SLO. In our previous study we identified 2 clusters of basic residues on hGrB that are important for delivery by perforin but not SLO. Here we found that a similar heparin binding site added to eGFP was functional (in binding to heparin) but it was not sufficient for delivery by perforin. The addition of a heparin binding sequence as a C-terminal tag on eGFP either did not provide enough positive charge and/or failed to recapitulate a specific three-dimensional surface of hGrB. The supercharged positive GFPs were more efficiently delivered than their negative and neutrally charged counterparts as expected, suggesting that surface positive charge is an important requirement for delivery through the perforin pore. It is, however, interesting that while the supercharged positive GFPs are similar to the FITC-DEAE-dextrans in that they bind to cell membranes and are endocytosed extremely well (as pre-viously reported (52)), they are not as efficiently delivered by perforin as the FITC-DEAE-dextrans. This difference suggests that positive charge alone is not the only requirement for delivery by perforin. This is also the first report of perforin delivering proteins unrelated to the granzymes and not found in the CL granule (thus would not be cargo for perforin in vivo). Cytc was delivered by both SLO and perforin, however there was a 6-fold decrease in efficiency when delivered by perforin. This may simply be explained by the differences in perforin and SLO pore diameter (16 nm versus 30 nm) even though both pores are much larger than the diameter of cytc (3-3.4 nm) (53). Yet when perforin and SLO deliver hGrB or mGrB there is no difference in the EC 50 (2,18), which is unexpected given the difference between perforin and SLO when delivering the smaller equally cationic cytc (or supercharged GFPs). This indicates that granzymes have evolved specific features beyond simple positive charge that maximize efficient engagement with perforin, ameliorating the restrictive effects of its smaller pore size (compared with SLO). For instance, positive charge may be presented to perforin in a structured way (for example, as part of an element reminiscent of a heparin binding site), or granzymes may possess an additional feature besides positive charge that promotes engagement with the pore. These are not mutually exclusive possibilities. Interestingly there is indirect evidence that granulysin, a component of the CL granule, is delivered by perforin into target cells (54). This is consistent with data presented here as granulysin is known to be cationic (55). Therefore it is conceivable that perforin not only delivers granzymes but other cationic granular contents as well.
How the perforin pore might facilitate the diffusion of positively charged molecules is not clear. It is conceivable that negatively charged features on the rim or in the lumen of the pore promote the diffusion of cationic molecules. This may be, in principle only, similar to the case for anthrax protective antigen, which is selective for protonated cargo (56). A high-resolution model of the perforin pore would allow examination of the charge-distribution of the pore lumen, however the current cryo-EM data do not permit a comprehensive model to be constructed. The finding that Ply also appears to favor positively charged cargo is very interesting as perforin and Ply are different in terms of pore assembly mechanism and are evolutionarily distant from one another. These data suggest that the preference for positively charged cargo could be a common feature of the pores of MACPF family proteins. The major similarity between perforin and Ply is the MACPF domain and fold, hence it would be interesting to see if these pores have a negatively charged lumen or surface feature. It is instructive that delivery of GrB by Ply is 2-3-fold worse than delivery by SLO; whereas delivery of (the smaller and highly cationic) cytc by Ply is Ͼ10 fold worse than delivery by SLO. It is possible that the secret of cargo delivery lies in such a conserved feature of the oligomerized MACPF proteins that granzymes have evolved to exploit.
With the insights provided recently by Voskoboinik et al. (12,57), our data support facilitated diffusion of granzymes through a plasma membrane perforin pore. However we cannot exclude an alternative hypothesis that perforin need not form a classical pore to deliver cargo. It has been proposed that incomplete perforin oligomers form "arcs" or proteo-lipidic pores that might be capable of granzyme delivery (reviewed in Ref. 58), although evidence for formation and function of such structures in vivo has not yet been presented. In this scenario a lipidlined segment might contribute to an electrostatic mode of delivery (16).
There is evidence the composition of the membrane is important for perforin function and appears to affect the type of pore formation, where some lipid combinations appear to favor proteo-lipidic pores over barrel stave (16,59). In addition to this, it has been reported that perforin affects the plasma membrane in a similar way to Equinatoxin II, which is predicted to form proteo-lipidic pores (16,60). Hence it is possible that perforin is forming both traditional pore structures as well as proteo-lipidic arcs, which can deliver cargo into the target cell.
In conclusion, we have shown that perforin (and the distantly related Ply) is able to deliver cargos other than granzymes, however cationic cargo is preferentially delivered. We therefore suggest that granzyme entry into the target cell cytoplasm is achieved through a facilitated diffusion mechanism through a plasma membrane perforin pore that promotes their rapid delivery during a killer attack. It is conceivable that there are conserved anionic elements on the perforin pore, and that granzymes have evolved surface properties that exploit features of an oligomerised MACPF domain. This work provides further insight to the perforin-granzyme synergy in target cell destruction by cytotoxic lymphocytes.