Perforin-deficient NOD mice are protected from diabetes, suggesting that cytotoxic granule contents of CD8+ T-cells have a significant role in killing β-cells. Despite this, cytotoxic granule effects on human or mouse pancreatic islets have not been reported. We tested the susceptibility of human and mouse islet cells to purified recombinant perforin and granzyme B and measured apoptotic death using a number of assays. Perforin and granzyme B impaired insulin secretion from islet cells, and this was accompanied by cytochrome c release, caspase activation, and DNA fragmentation. Granzyme B–mediated apoptotic changes only occurred in the presence of perforin. When compared with hemopoietic cells, traditionally used as targets to measure cytotoxic T-cell function in vitro, islet cells were relatively resistant to perforin and granzyme B. Inhibition of caspases prevented DNA fragmentation but not cytochrome c release, indicating that mitochondrial disruption due to granzyme B is independent of caspase activation. Consistent with this, islet cells from mice deficient in the BH3-only protein Bid were resistant to cytochrome c release and were protected from apoptosis after exposure to perforin/granzyme B. Our data suggest that Bid cleavage by granzyme B precedes mitochondrial disruption and apoptosis in pancreatic islets.
Type 1 diabetes is an autoimmune disease mediated by T-cells specific for antigens found in pancreatic β-cells. The pancreatic islets of people with type 1 diabetes have a lymphoid infiltrate termed “insulitis” (1). Insulitis is also observed in animal models of autoimmune diabetes. The infiltrate consists of CD8+ and CD4+ T-cells, B-cells, and macrophages (2). These cells play critical roles in the pathogenesis of type 1 diabetes, and at least in animal models, all are required for efficient progression of β-cell destruction and diabetes (3–6). CD8+ T-cells appear to be the most directly damaging cell type. β-Cell death occurs by several molecular pathways leading to apoptosis. Pathways triggered by the contents of the cytolytic granules of CD8+ T-cells, cell death receptors such as Fas, and proinflammatory cytokines, including tumor necrosis factor, interleukin-1, and interferon-γ (7), have been implicated.
The pore-forming protein perforin and the granzyme family of serine proteases are key constituents of cytolytic granules. After conjugate formation, perforin and granzymes are released toward the target cell membrane where they synergize to cause apoptotic cell death (8). Data from the NOD mouse model suggest that cytotoxic granules from CD8+ T-cells are important mediators of β-cell death. Perforin-deficient NOD mice have been shown to develop insulitis but have a markedly reduced incidence of diabetes, with ∼16% of female mice becoming diabetic compared with 77% of wild-type NOD mice (9,10). Despite this, insulitis remains a histopathological feature in perforin-deficient NOD mice. Granzyme B is indispensable for an effective immune response to some viruses (11). Evidence for granzyme B having a role in human islet destruction in an allogeneic setting comes from a study of diabetic recipients of pancreatic islet grafts who had increased expression of granzyme B in their peripheral blood preceding graft rejection (12). Additionally, we have shown that granzyme B has a critical role in β-cell death induced by alloreactive cytotoxic T-cells (CTLs) (13). Granzyme B is a serine protease with a strong preference for cleaving proteins after specific aspartate residues (14,15). It can enter cells alone but requires perforin to induce cell death (16,17). Multiple caspase proteins are potential substrates for granzyme B, including effector caspases 3, 6, and 7, but direct activation of caspases to a lethal threshold by granzyme B within intact cells is an inefficient process, and cell death generally requires the release of proapoptotic mitochondrial factors (18–20).
The Bcl-2 family of homologous proteins regulates mitochondrial outer membrane integrity (21). Proapoptotic members include Bak, Bax, and the BH3-only proteins Bid, Bim, NOXA, and PUMA. Each proapoptotic member has a critical role in transmitting death signals from different stimuli (21). Granzyme B has been shown to cleave Bid specifically after Asp75 within 5 min of entering the cell (18), and this results in the release of mitochondrial factors, including Smac/Diablo and cytochrome c. This is followed by full-blown caspase activation and cell death (19,22). Embryonic fibroblasts and primary hemopoietic cells from Bid-deficient mice are 30- to 50-fold resistant to recombinant perforin and granzyme B compared with Bid-sufficient cells (20).
Many studies have analyzed the interaction between β-cells and recombinant effector molecules such as cytokines and Fas ligand, but there have been no studies to date analyzing the effect of perforin and granzymes on β-cells. In this study, we have directly tested the susceptibility of human and mouse β-cells to perforin and granzyme B. We demonstrate that pancreatic β-cells undergo apoptosis that correlates with a loss of islet insulin secretion capacity. Compared with hemopoietic cells, islet cells were found to be less sensitive to the combined effects of perforin and granzyme B in vitro. Finally, we show that the BH3-only protein Bid plays an important role in the interaction between caspase and mitochondrial cell death pathways activated by perforin and granzyme B in islets. A detailed understanding of the effects of perforin and granzyme B on β-cells should ultimately aid our efforts to design ways of inhibiting β-cell death.
RESEARCH DESIGN AND METHODS
Mice and cell lines.
Four- to 8-week-old male NOD and C57BL/6 mice were obtained from the Walter and Eliza Hall Institute specific pathogen-free animal breeding facility (Melbourne, Australia). Bid-deficient C57BL/6 mice were provided by Dr. S. Korsmeyer (Howard Hughs Medical Institute, Boston, MA) and Dr. A. Strasser (Walter and Eliza Hall Institute, Melbourne, Australia). All mouse experiments were carried out with the approval of the institutional animal ethics committee.
The mastocytoma cell line P815 and the Eppstein Bar virus–transformed human B-cell line JY-EBV (23) were cultured at 37°C, 5% CO2 in RPMI (Life Technologies, Gaithersburg, MD) supplemented with 10% FCS, antibiotics, and nonessential amino acids.
Perforin, granzyme B, cytokines, and other reagents.
Recombinant perforin was expressed and purified as previously described (24). Recombinant human granzyme B was purified as previously described (25). Mega Fas Ligand (used at 100 ng/ml) was a gift from Dr. J. Tschopp (University of Lausanne, Lausanne, Switzerland). Z-Val-Ala-Asp-fluoromethylketone (zVAD.fmk) and Z-Phe-Ala-fluoromethylketone (zFA.fmk) (Enzyme Systems Products, Livermore, CA) were used at 100 μmol/l.
Human pancreata were obtained, with informed consent, from heart-beating, brain-dead donors. Pancreas procurement and experimental procedures were approved by the human ethics committee at St. Vincent’s Hospital (Melbourne, Australia). Islets were isolated as previously described (26) by intraductal perfusion and digestion of the pancreas with collagenase followed by purification using ficoll density gradients.
Mouse islets were isolated as previously described (27). Purified islets were washed, hand picked, and cultured at 37°C, 5% CO2 in CMRL-1066 (Life Technologies) supplemented with 10% FCS, antibiotics, and glutamine (CMRL).
Treatment of islet cells with perforin and granzyme B.
Islets were washed in PBS and resuspended in trypsin solution (0.125 mg/ml trypsin and 3 mmol/l EDTA in PBS) for 12 min at 37°C. Islets were mechanically dispersed using a pipette. Cell suspensions were then recovered for 45 min in complete medium at 37°C. The cysteine protease inhibitors (zVAD.fmk or zFA.fmk) were added at this stage if required. Cell suspensions or intact islets were washed in PBS, suspended in perforin buffer (Hanks’ balanced salt solution, 0.4% BSA, 25 mmol/l HEPES, and 2 mmol/l CaCl2), prewarmed to 37°C, transferred to 48-well plates, and incubated for 2–4 h at 37°C, 5% CO2 with purified perforin and granzyme B. Control islets were incubated in the same buffer without addition of perforin and granzyme B.
Intact or dispersed islets were suspended in 300 μl CMRL containing 200 μCi [51Cr]sodium chromate (Amersham Pharmacia Biotech, Piscataway, NJ) and incubated for 90 min, resuspending every 30 min. Islets/cells were washed three times in medium and transferred to 96-well plates with 10 islets/well or 1 × 104 cells in 100 μl buffer. Recombinant perforin and/or granzyme B was added to make the reaction volume 200 μl. Buffer (100 μl) or 100 μl 2% Triton X-100 was added to wells to determine spontaneous and maximum 51Cr release, respectively. All data points were performed in triplicate. After incubation for 4–20 h, plates were centrifuged for 5 min at 1,000 rpm. Supernatant (100 μl) was removed and analyzed on a gamma counter (Perkin-Elmer). Specific lysis was calculated using the following formula:
Nuclear fragmentation was quantified as previously described (28). Cells were resuspended in 200 μl hypotonic buffer (50 μg/ml propidium iodide [Sigma, St. Louis, MO], 0.1% sodium citrate, and 0.1% Triton X-100), which stains nuclear DNA. The cells were then analyzed on a FACSCalibur (Becton Dickinson, Mountain View, CA) using the FL-3 channel. Apoptotic cells were identified by their subdiploid DNA content.
Caspase 3 activity.
After 2 h of treatment, caspase 3 activity was measured as previously described (29). Cells were lysed in 5 mmol/l Tris-Cl, pH 7.5, 5 mmol/l EDTA, and 0.5% nonidet P-40 on ice for 30 min. Equivalent amounts of protein (15 ng) from each extract were added to caspase buffer (50 mmol/l HEPES, pH 7.4, 10% sucrose, 0.1% CHAPS, and 10 mmol/l dithiothreitol) with 2.5 μmol/l fluorogenic peptide substrate Ac-Asp-Glu-Val-Asp-AMC (Bachem, Bubendorf, Switzerland). Fluorescent emission was measured using a POLARstar OPTIMA fluorescence microplate reader (BMG Labtechnologies, Offenburg, Germany) for 1 h with readings made every 5 min (excitation filter 405 nm; emission filter 510 nm).
Cytochrome c release.
After 2 h of treatment, cytochrome c release was measured using the method previously described (30). Cells were incubated for 2 min on ice in 100 μl digitonin buffer (80 mmol/l KCl, 50 ng/ml digitonin, and 1 mmol/l EDTA in PBS). The permeabilized cells were fixed with 4% paraformaldehyde in PBS, washed in PBS, and incubated for 1 h in blocking buffer (3% BSA and 0.05% Saponin in PBS). Cells were then stained with mouse anti–cytochrome c antibody (BD Pharmingen) for 24 h at 4°C followed by phycoerythrin–anti-mouse IgG (BD Pharmingen) for 30 min at room temperature. Phycoerythrin fluorescence was then detected by flow cytometry.
Glucose-stimulated insulin secretion.
Glucose-stimulated insulin secretion assays was assessed as previously described (31). After treatment, islets were washed in KRB buffer containing 0.1% BSA and 3 mmol/l glucose. Groups of 20 islets were then incubated for 30 min in either 3 or 20 mmol/l glucose. Insulin was assayed by radioimmunoassay (LINCO Research, St. Charles, MO). Each treatment was performed in quadruplicate.
Fluorescent labeling and confocal microscopy.
An Alexa 488 labeling kit (Molecular Probes, Eugene, OR) was used to label recombinant human granzyme B. Islet and P815 cells were incubated with granzyme B–Alexa 488 for 1 h, washed in PBS, and then analyzed by flow cytometry. For confocal microscopy, cells were centrifuged onto microscope slides, fixed with 4% paraformaldehyde, and stained with guinea pig anti-insulin polyclonal antibody (Dako, Glostrup, Denmark) followed by anti–guinea pig Alexa 568 (Molecular Probes). Slides were viewed with a BIORAD MRC 1024 confocal imaging system (Bio-Rad, Richmond, CA). Whole islets treated with 500 ng/ml perforin, 0.5% Triton X-100, or normal saline were transferred onto a concave chamber slide in PBS containing 1 μg/ml propidium iodide. Slides were viewed immediately by confocal microscopy.
Analyses of data were performed using GraphPad Prism (GraphPad Prism Software, San Diego, CA). Groups were analyzed by one-way ANOVA with Bonferroni’s post test for comparison of multiple columns. Data are represented as means ± SE of multiple independent experiments.
Dispersed islet cells show similar sensitivity to perforin as hemopoietic cells, but intact islets are relatively resistant.
Polymerization of perforin damages the plasma membranes of many cell types, but the sensitivity to perforin can vary (15). Cell lysis induced by high-dose perforin was studied by loading intact islets, dispersed islet cells, and hemopoietic cells with 51Cr and exposing them to increasing concentrations of perforin for 4 h (Fig. 1A). Perforin induced a similar degree of 51Cr release from dispersed human and mouse islet cells as it did from mouse mastocytoma (P815) and human B-cell (JY-EBV) cell lines. In contrast, intact islets were resistant to perforin-induced lysis. Even concentrations of perforin resulting in 50% lysis of dispersed cells caused minimal 51Cr release from intact islets (Fig. 1A). This difference could not be put down to inefficient labeling of intact islets, because both cell preparations released a similar amount of 51Cr when lysed with detergent (not shown).
Perforin is highly lipophylic, and the outer layers of the intact islet may prevent perforin penetration to cells deeper within the islet. To examine the penetration of perforin into intact islets, we analyzed membrane integrity of cells in whole islets after exposure to high concentrations of perforin. This was done by staining perforin-treated islets with propidium iodide and examining unfixed islets using confocal microscopy (Fig. 1B). We observed that only those cells around the periphery were permeable to the propidium iodide, whereas islets treated with 0.5% Triton X-100 had propidium iodide uptake into all cells, even those at the core of the islet. These data suggest that recombinant perforin is unable to adequately penetrate whole islets in vitro. This could be due to structural properties of intact islets and/or functional properties of perforin.
Entry of granzyme B into intact islets and dispersed islet cells.
Granzyme B has been reported to enter cells via receptor and/or fluid phase endocytosis (32–34). To demonstrate entry of granzyme B into islet cells directly, we incubated human islets with Alexa 488–labeled granzyme B for 1 h and then analyzed the cells by flow cytometry (Fig. 1C) or confocal microscopy (Fig. 1D). Intact islets or islet cells analyzed by flow cytometry demonstrated an increase in fluorescence consistent with granzyme B entry into all cells similar to that of P815 cells (Fig. 1C). Analysis of dispersed islet cells by confocal microscopy demonstrated insulin-positive β-cells (Fig. 1D, red), with punctate granzyme B staining (Fig. 1D, green), consistent with granzyme B entry into endocytic vesicles distinct from those containing insulin. The staining pattern was similar to that observed in HeLa cells exposed to labeled granzyme B (34).
Insulin secretion is reduced in human islets exposed to perforin and granzyme B in vitro.
The effect of perforin and granzyme B on β-cell function was measured by assaying glucose-stimulated insulin secretion of intact islets after exposure to perforin and granzyme B for 4 h. The stimulation of insulin secretion from low (3 mmol/l) to high (20 mmol/l) glucose was significantly less in islets treated with a sublytic dose of perforin and 120 nmol/l granzyme B compared with untreated islets or islets treated with perforin alone (Fig. 2). Insulin secretion did not recover significantly when islets were cultured for up to 4 days in fresh medium after treatment with the cytotoxic molecules (not shown). This effect on insulin secretion was surprising because we were not able to detect any significant difference in DNA fragmentation between untreated whole islets and islets exposed to perforin/granzyme B (not shown). These data provide evidence that soluble perforin and granzyme B can be directly toxic to human β-cells and significantly impair their function, despite the limited diffusion capacity of perforin into intact islets. Granzyme B is able to cleave extracellular matrix proteins (35), and this may offer a possible explanation for the inhibition of insulin secretion by perforin and granzyme B in whole islets.
Human and mouse islet cells are less susceptible to purified perforin/granzyme B than hemopoietic cells.
Primary and transformed hemopoietic cells are generally very sensitive to purified cytotoxic granule constituents. To compare the susceptibility of pancreatic islet cells with hemopoietic cells in vitro, dispersed mouse islet cells and P815 cells were treated with perforin and granzyme B, and DNA fragmentation was measured. More than 50% of P815 cells had fragmented DNA after 8 h of treatment with perforin and granzyme B (Fig. 3A). In contrast, little granzyme B–dependent DNA damage in islet cells was evident until between 8 and 16 h after treatment (Fig. 3B). At 16 h, there was a twofold difference between DNA fragmentation in mouse (67% DNA fragmentation) and human (30% DNA fragmentation) islet cells (Fig. 3B–D). Human islet cells are generally more resistant than mouse islets to toxic stimuli, including inflammatory cytokines and streptozotocin. An increase in antioxidant expression has been observed in human islets compared with mice, offering one explanation for this difference (36).
As the concentration of granzyme B was increased, an increase in DNA fragmentation was observed, but even at concentrations as high as 240 nmol/l, the percentage of DNA fragmentation was one-half that seen in the P815 cell line (Fig. 3E). These data demonstrate that islet cells exposed to purified perforin and granzyme B undergo DNA fragmentation and apoptotic cell death. However, apoptotic changes appear to proceed at a slower rate requiring higher concentrations of granzyme B compared with hemopoietic cells. It is possible that β-cells are particularly more sensitive than other islet cells to perforin/granzyme B, influencing the overall susceptibility of all islet cells in this in vitro assay. This is unlikely because the NIT-1 insulinoma cell line also required high concentrations of granzyme B and a longer time for apoptotic changes to appear (not shown), suggesting that β-cells themselves are relatively resistant to perforin/granzyme B.
Caspase 3 activity is induced by perforin/granzyme B and is essential for DNA fragmentation in pancreatic islet cells in vitro.
Activation of the caspase protease cascade has been observed in some cell types soon after exposure to perforin and granzyme B and is an indication of progression toward apoptosis. Caspase 3 activity was measured in cytosolic lysates from mouse islet cells and P815 cells treated with perforin and granzyme B for 2 h using a fluorogenic caspase 3 substrate. Perforin and granzyme B caused an increase in caspase 3 activity in islet cells that was dependent on the dose of granzyme B (Fig. 4A). Islet cells treated with 240 nmol/l granzyme B in the presence of perforin had less caspase 3 activity than P815 cells treated with 120 nmol/l granzyme B and perforin (Fig. 4B). Perforin alone had no significant effect on caspase 3 activity.
DNA fragmentation is widely thought to be caspase dependent, but there are instances in which apoptotic changes and cell death may occur without caspases (37,38). Apoptosis-inducing factor and endonuclease G are two molecules reported to induce DNA fragmentation without the need for caspases (39,40). We tested whether caspase activity was required for granzyme- and perforin-induced DNA fragmentation in islet cells by analysis of DNA fragmentation after blocking caspase activity. Cells were preincubated with the pancaspase inhibitor zVAD.fmk or the cathepsin B inhibitor zFA.fmk as a control for the nonspecific protease inhibitory effects of fluoromethylketones. Cells pretreated with zVAD.fmk had a significant reduction in perforin/granzyme B–induced DNA fragmentation compared with cells pretreated with zFA.fmk (Fig. 4C). There was no effect of blocking caspase activity on membrane damage induced by perforin alone, as measured by release of 51Cr from islet cells (Fig. 4D).
Granzyme B–mediated cytochrome c release in pancreatic islet cells is independent of caspase activity.
The relationship between caspase activation and mitochondrial dysfunction due to granzyme B has not been studied in pancreatic β-cells. We tested whether granzyme B–induced mitochondrial dysfunction measured by release of cytochrome c was secondary to caspase activation or whether it could occur independently in pancreatic islet cells. To do this, release of cytochrome c was measured by flow cytometry. Islet cells were treated with the caspase inhibitor zVAD.fmk before exposure to perforin and granzyme B to distinguish between caspase-dependent and -independent cytochrome c release. Islet cells treated with perforin and granzyme B demonstrated a significant reduction in cytochrome c staining, consistent with its release after mitochondrial disruption (Fig. 5A and B). This was also seen in islets exposed to perforin and granzyme B in which caspase activity was inhibited by zVAD.fmk, suggesting that mitochondrial dysfunction in islet cells induced by granzyme B does not require caspase activation. This was true in both diabetes-susceptible NOD mouse islets (Fig. 5A) and nonsusceptible C57Bl/6 islets (not shown). In contrast, Fas ligand–mediated cytochrome c release was dependent on caspase activity (Fig. 5C).
Islet cells deficient in Bid are resistant to granzyme B–mediated apoptosis.
The proapoptotic BH3-only protein Bid is a substrate for cleavage by granzyme B in hemopoietic cells (18). Wild-type islet cells treated with perforin/granzyme B had increased cytochrome c release consistent with mitochondrial damage (Fig. 6A). Release of cytochrome c in wild-type islets was not dependent on caspase activation because it was not prevented by preincubation with zVAD.fmk. Bid deficiency resulted in a block of cytochrome c release (Fig. 6A). These data support a critical role for granzyme B–mediated cleavage of Bid in perforin- and granzyme B–induced mitochondrial membrane dysfunction.
Deficiency of Bid also blocked caspase 3 activity induced by perforin and granzyme B (Fig. 6B). These data demonstrate that perforin- and granzyme B–induced caspase 3 activation is downstream of Bid and mitochondrial membrane dysfunction. Consistent with the lack of cytochrome c release and caspase 3 activation in Bid-deficient islets, we also observed protection from perforin- and granzyme B–induced DNA fragmentation in these islets (Fig. 6C).
The protection from diabetes but not insulitis in NOD mice deficient in perforin (9,10) implies that CTLs specific for β-cell antigens destroy β-cells at least in part by releasing the contents of the cytolytic granule. The current study confirms that isolated primary islet cells are susceptible to perforin and granzyme B when these are added as recombinant molecules. Islet cells were susceptible to perforin on its own at concentrations that induced membrane damage and cell lysis. However, current models of the mechanism of perforin action suggest that it acts at a lower concentration than that required for cell lysis, by facilitating granzyme B–dependent apoptosis. Islet cells were killed by combinations of perforin at low concentration and granzyme B in assays that measure DNA fragmentation, caspase 3 activation, and cytochrome c release, all features of apoptosis.
Given that perforin and granzymes appear to be the main mechanism of β-cell death in autoimmune diabetes, it was interesting to find that islet cells are relatively resistant to these molecules compared with other cell types. We found islet cells required longer exposure time and higher concentrations of perforin/granzyme B than hemopoietic cells for apoptotic changes to appear. The membrane-lytic activity of recombinant perforin on islet cells was similar to hemopoietic cells, and entry of fluorescently labeled granzyme B was also similar. It therefore is likely that resistance to perforin and granzyme B action may be at the level of intracellular cell death pathways. Differentiated end cells with limited potential for regeneration are relatively resistant to cell death, and β-cells are robust in the absence of specific immune attack. Whether islet cells resist apoptotic death by inherently low levels of proapoptotic molecules, high levels of anti-apoptotic molecules, or both is unknown. Bcl-2 and Bcl-XL have been observed in islet cells, which may account for their increased resistance (41,42).
We showed that only the outer cells of whole islets treated with perforin had membrane damage. These data indicate that a physical barrier may be formed by the cells and connective tissue of the outer layers of the islet. However, this barrier must be permeable to other proteins because intact islets treated with cytokines with or without Fas ligand demonstrate changes in gene expression in all cells and >60% DNA fragmentation that would involve accessing even those cells deep within the islet (43). Perforin has a high affinity for lipid (44), thus cells on the periphery of the islet could act as a sink for active perforin. The pathophysiological significance of this is unclear because perforin would be delivered to cells by the invasion of CTLs into the islet, not released at a distance. This is consistent with the histopathology observed during progression to diabetes in which β-cell destruction is associated with lymphocytic infiltrate inside the islet rather than peri-insulitis.
In islet cells exposed to perforin and granzyme B, DNA fragmentation was dependent on caspase 3 activation, even though caspase-independent cell death can occur in some cell types (37,38). Despite preventing DNA fragmentation, blockade of caspase 3 activity in islet cells did not alter granzyme B–induced mitochondrial permeability, consistent with observations made in hemopoietic cells (18,45,46). Mitochondrial disruption has been attributed to granzyme B directly activating Bid (20). This differs from Fas ligand–mediated mitochondrial perturbation that is dependent on Bid cleavage by caspase 8 and abolished by caspase inhibition (47,48).
Our findings that Bid-deficient islets are protected from mitochondrial perturbation evidenced by failure to release cytochrome c are consistent with Bid playing a crucial role in granzyme B–induced apoptosis in β-cells. We have recently shown that granzyme B–deficient alloreactive CTLs have significantly reduced ability to kill β-cells, indicating a role for this granzyme in β-cell death (13). This is in contrast to viral infections or clearance of tumors, where deficiency of perforin leads to a more serious phenotype than absence of granzyme B, which does not generally affect morbidity (49,50). We have therefore identified key steps in the molecular pathway for specific intervention in CTL-mediated β-cell death.
The costs of publication of this article were defrayed in part by the payment of page charges. The article must therefore be hereby marked “advertisement” in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
This work was supported by grants from the National Health and Medical Research Council of Australia and the Juvenile Diabetes Research Foundation.
We thank Dr. Nigel Waterhouse and Dr. Ricky Johnstone for reagents and helpful discussions. We thank Melanie Rowe, Kylie Tolley, and Rochelle Ayala-Perez for animal husbandry and technical assistance.