|Home | About | Journals | Submit | Contact Us | Français|
Like Ricin, Shiga, and Cholera toxins, yeast K28 is an A/B toxin that depends on endocytosis and retrograde trafficking for toxicity. Knowledge of the specific proteins, lipids, and mechanisms required for trafficking and killing by these toxins remains incomplete. Since K28 is a model for clinically relevant toxins, we screened over 5000 yeast mutants, identifying 365 that affect K28 sensitivity. Hypersensitive mutants revealed cytoprotective pathways, including stress-activated signaling and protein degradation. Resistant mutants clustered to endocytic, lipid organization and cell wall biogenesis pathways. Furthermore, GPI anchors and transcriptional regulation are important for K28-cell binding. Strikingly, the AP2 complex, which in metazoans links endocytic cargo to the clathrin coat, but had no assigned function in yeast, was critical for K28 toxicity. Yeast AP2 localizes to endocytic sites and has a cargo-specific function in K28 uptake. This comprehensive genetic analysis identified conserved processes important for A/B toxin trafficking and killing.
As a member of the A/B protein toxin family, the yeast K28 ‘killer’ toxin shows striking similarities to various clinically relevant toxins produced by plants and pathogenic bacteria. These A/B toxins usually consist of one or more β subunits that mediate cellular entry and intracellular targeting, and an α subunit that kills the invaded cell (Falnes and Sandvig, 2000). The K28 toxin is a virally encoded α/β heterodimer secreted by Saccharomyces cerevisiae infected with the M28 virus. After uptake by uninfected yeast, K28 follows a trafficking pathway that resembles those taken by numerous other A/B toxins, ultimately resulting in cell killing (Breinig et al., 2006). Studying A/B toxin pathways in a genetically tractable organism, such as yeast, promises to provide new insights into the interactions between A/B toxins and their molecular targets in mammalian cells.
Similar to Cholera toxin, Ricin, and Shiga toxin, K28 enters cells by endocytosis and traffics through endosomal and Golgi compartments to the ER, effectively bypassing degradation in the vacuole/lysosome (Eisfeld et al., 2000; Sandvig and van Deurs, 2002). The conserved ‘KDEL’ receptor, Erd2p in yeast, recognizes C-terminal K/HDEL sequences on K28 and Cholera toxins and directs their retrograde trafficking from the Golgi to the ER of their target cells. Some of the cellular ER retrotranslocation machinery is then co-opted to allow export into the cytosol (Heiligenstein et al., 2006; Sandvig and van Deurs, 2002; Yu and Haslam, 2005). Once released from the ER a disulfide bond connecting the α and β subunits of K28 breaks and the β subunit is degraded by the proteasome (Heiligenstein et al., 2006). The K28 α subunit elicits a pre-S phase cell cycle arrest and kills the target cell by an unknown mechanism, possibly by activating a yeast apoptotic-like cascade (Reiter et al., 2005; Schmitt et al., 1996).
Although there have been systematic efforts to uncover yeast mutants with altered sensitivity to the ionophoric K1 killer toxin and the tRNase toxin zymocin (Huang et al., 2008; Page et al., 2003), no systematic screens for the A/B toxin K28 have been reported. Previously, a genetic approach has been used to identify a limited number of K28 resistant mutants, such as the endocytic mutant sla2Δ and the Golgi-to-ER retention mutant erd2Δ, that have defects in toxin uptake and trafficking (Eisfeld et al., 2000). However, many questions remain spanning the entire proposed pathway of K28 uptake and killing. For example, the cell wall and plasma membrane factors important for K28 binding, the cellular components acted upon to induce cell death, and the cellular mechanisms used to mitigate the toxic effects of K28 are all unknown. Moreover, the extent to which common mechanisms are used by K28 and medically important toxins has not been fully investigated.
The 365 K28 resistant or hypersensitive mutants identified by our screen generate a detailed view of conserved A/B toxin trafficking and killing mechanisms. The mutants identified are statistically enriched in biological processes and components that are very likely to be important for K28 action. Our screen results also provide insights into the biology of cells targeted by A/B toxins. For example the yeast AP2 complex, which has not previously been associated with any endocytic functions in yeast (Yeung et al., 1999), was identified by our screen as having a critical role in K28 action. Our functional analysis of yeast AP2 suggests it has a previously unrecognized and conserved role in cargo-selective endocytosis. These results highlight the utility of K28 as a sensitive tool for analyzing cellular pathways and processes in intoxicated cells.
To identify determinants of K28 sensitivity, we developed a ‘killer’ assay to systematically screen mutant yeast strains for K28 resistance or hypersensitivity. A dense spot of K28-secreting cells was plated onto a lawn of the strain to be tested, and the zone of clearance in the lawn around the spot was measured to give quantitative results. Typical K28 killer assay results for wild-type and hypersensitive and resistant control strains are shown in Figure 1A. We screened 4806 strains in the MATα deletion collection, which represents ~80% of all yeast genes, and an array of temperature sensitive (ts) alleles representing almost one-quarter of all essential yeast genes.
Our screen yielded 176 mutations causing K28 resistance (140 deletions, 36 ts-alleles) and 189 mutations causing K28 hypersensitivity (116 deletions, 73 ts-alleles) (Table S1). A manually annotated distribution of all K28 resistant or hypersensitive mutants identified is presented in Figures 1B and 1C. In general, K28 resistant hits support and expand upon the previously proposed pathway for K28 entry, trafficking, and killing (Schmitt and Breinig, 2006). The hypersensitive hits reveal the largely unexplored cellular pathways that resist toxin action. For example, over 25% of K28 hypersensitive mutations were found in genes related to protein translation. Although the mechanism of K28 killing is unknown, the ribosome is a common target for other toxins which follow a nearly identical trafficking pathway to K28 (Montanaro et al., 1973; Reisbig et al., 1981). Additionally, mutations in 21 genes of unknown function were found to alter K28 sensitivity, providing potential insights into the functions of these genes. The use of the ts strain collection allowed us to screen previously inaccessible aspects of yeast cell biology and identify novel determinants of K28 toxicity, such as the proteasome (hypersensitive) and glycophosphatidylinositol (GPI) -anchor biosynthetic machinery (resistant).
We next sought to gain an overview of K28 activity by identifying the cellular processes implicated repeatedly by different hypersensitive or resistant mutants. One unbiased approach to identify such processes is to calculate the enrichment of gene ontology (GO) terms associated with genes identified in the screen. GO annotations draw on many different types of data from the literature to organize genes into cellular processes, components and functions (The Gene Ontology Consortium, 2000). We identified 250 GO terms that were statistically enriched (Bonferroni corrected p <0.05) in our hit list (Table S2). The fold enrichments for K28 resistant and hypersensitive GO terms representing the major groups of enriched cellular processes or components are shown in Figures 1D and 1E, respectively.
The GO terms enriched in our dataset (Table S2) reveal the aspects of cell physiology that determine how K28 interacts with and affects cells. The resistant hits are enriched for genes involved in cell wall organization, endocytosis, and trafficking. In addition, specific cellular components never before identified as determinants of K28 killing, like the chromatin-silencing SIR complex and the clathrin adaptor AP2 complex, are enriched. The hypersensitive hits are enriched for genes involved in gene expression and translation, as well as genes that encode the protein sorting HOPS complex, and the chromatin remodeling SWR1 and ASTRA complexes (Shevchenko et al., 2008).
To assess the specificity of these enriched categories, we compared the GO terms enriched in our screen to those enriched in a screen for mutants resistant to the tRNase killer toxin zymocin (Huang et al., 2008). Other than GO terms relating to RNA processing and ribosome biogenesis, there was little overlap between GO terms enriched in the two screens (Table S3). This result suggests that the genes identified in our screen reflect the specific manner in which K28 interacts with cells.
In Figure 2 Osprey-generated networks (Breitkreutz et al., 2003) display examples of genes involved in processes and components important for K28 action based on enriched GO terms. These results are consistent with what is known about K28 action, but more comprehensively identify the cellular processes and components involved. For example, K28 is thought to initially interact with the yeast cell wall and possibly components of the plasma membrane (Heiligenstein et al., 2006; Schmitt and Radler, 1988; Schmitt and Radler, 1990), and we found a large number of mutations relating to cell wall and plasma membrane organization that were not previously linked to K28 resistance (Table S1 and Figure 2A). The importance of membrane lipids in toxin binding is exemplified by the toxin aerolysin, which binds to the GPI moiety of target cell surface proteins, and Cholera and Shiga toxins, which recognize modified sphingolipids (Table S5) (Heyningen, 1974; Jacewicz et al., 1986; Nelson et al., 1997). It is important to note that in addition to binding the toxin, the correct lipid environment could be needed for K28 endocytosis or trafficking (Beh and Rine, 2004; deHart et al., 2002).
After endocytic uptake, K28 is proposed to traffic the secretory pathway in reverse and exit from the ER into the cytosol (Heiligenstein et al., 2006). Accordingly, we found both K28 resistant and hypersensitive mutants throughout the endomembrane system (Table S1, Figures 1B and 1C). Resistant mutants associated with GO terms related to intracellular trafficking (Figure 2B) show that both secretion and retrograde traffic are important for K28 toxicity. We speculate that mutants causing defects in secretory traffic might affect the amount of K28 receptor available for toxin binding, whereas retrograde trafficking defects might affect transport of K28 to the ER (Heiligenstein et al., 2006). Figure 2B also shows a group of hypersensitive mutants in the vacuolar HOPS complex. It is possible that vacuolar defects redistribute a factor that controls K28 killing. Alternatively a portion of internalized toxin might normally reach the vacuole, a lytic structure in yeast, where it is destroyed. When the vacuole is defective an excess of K28 might traffic to the ER, resulting in hypersensitivity. In support of the second hypothesis, the analogous organelle in metazoan cells, the lysosome, has been proposed to degrade a fraction of Ricin and Shiga toxins after endocytic uptake (Table S5) (Sandvig and van Deurs, 1996). It is also noteworthy that mutations in the ubiquitin-proteasome pathway cause K28 hypersensitivity (Figures 1C and 1E). Thus, our screen suggests that vacuolar targeting, and proteasomal degradation in the cytosol, may act in tandem to control the cytosolic K28 load.
We identified a number of hypersensitive and resistant mutants associated with enriched GO terms related to transcriptional regulation (Figure 2C). The identification of multiple subunits of several complexes illustrates the specificity, high coverage and sensitivity of our screen. For example in the mediator complex, which regulates transcription, loss of core subunits causes K28 hypersensitivity, while loss of the antagonistic CDK module subunits causes K28 resistance (Figure 2C) (Samuelsen et al., 2003). Similarly, mutation of multiple subunits of the chromatin silencing SIR complex causes K28 resistance while deletion of the antagonistic histone variant HTZ1 or SWR1 complex components causes K28 hypersensitivity (Figure 2C) (Meneghini et al., 2003). Further cell biological analyses should help elucidate the mechanistic roles for these complexes and others in K28 killing.
Finally, we generated double mutants to explore epistatic interactions between presumably early-acting resistant mutations and diverse hypersensitive mutations. To assess different parts of the toxin pathway we chose hypersensitive deletions of an osmolarity stress-activated kinase (HOG1), a vacuolar SNARE protein (VAM7) and a ribosomal subunit (RPS8A). Deletion of the cell wall biosynthetic enzyme MNN2 was epistatic to all hypersensitive mutants presumably because it abrogates K28 binding (Table S4 and Figure 3) (Schmitt and Radler, 1990). Deletion of the endocytic gene END3 or the gene APM4, which encodes a subunit of the AP2 complex, was epistatic to vam7Δ and rps8aΔ. Strikingly, deletion of HOG1 was epistatic to both end3Δ and apm4Δ. These results show the importance of the HOG pathway to resist K28 killing, and suggest either that the reduced amounts of toxin entering the endocytic mutant cells are sufficient for killing when HOG1 is absent, or that K28 might have a secondary killing mechanism, which is only revealed when the HOG pathway is absent. Strikingly, the metazoan HOG1 homolog p38 is activated in mammalian cells treated with Cholera, Shiga, and Ricin toxins, highlighting the relevance of our screen to toxin biology in more complex eukaryotes (Table S5) (Schnitzler et al., 2007; Tamura et al., 2003; Walchli et al., 2008).
Since many cell wall mutants are known to have defects in binding to the toxin (Schmitt and Radler, 1990), we wanted to determine which mutants were resistant because of weak K28 binding. Applying a toxin-cell binding assay to our set of 176 K28 resistant mutants biochemically defined requirements for K28-cell binding. We found 20 of 140 resistant gene deletions and 8 of 36 resistant ts-alleles deplete K28 from culture supernatant significantly less well than wild-type cells (p <0.01; Figure 3). The gene deletions with the most severe phenotypes relate to glycosylation, especially protein mannosylation. Less severe binding defects occur in deletion strains with aberrant lipid biogenesis or transcriptional mutations (Figure 3). Similarly, the ts-alleles with K28 binding defects are mutated in genes affecting protein mannosylation and lipid biogenesis (Figure 3). In particular, half of all binding-defective ts-alleles affect attachment of GPI anchors to proteins. This assay sheds light on the mechanism of toxin resistance in >15% of the mutant strains and identified transcriptional networks likely to impact the structure of the cell wall. For example, the importance of the Cyc8p/Tup1p transcriptional repressor and the CDK module of the mediator complex in K28-cell binding is consistent with established roles for CYC8/TUP1 in manno-specific flocculation and the CDK module in regulation of other flocculation genes (Samuelsen et al., 2003; Stratford, 1992).
One intriguing result of our screen was the identification of all four AP2 clathrin adaptor subunit deletions (apl1Δ, aps2Δ, apl3Δ and apm4Δ) as strongly K28 resistant (Table S1 and Figure 4A). The AP2 complex functions in metazoans to link endocytic cargo proteins to the clathrin coat. However, in S. cerevisiae AP2 homolog deletions previously had no detectable endocytic phenotype and were thought to be unimportant for endocytosis (Sorkin, 2004; Yeung et al., 1999). To our knowledge K28 resistance is the first specific phenotype in S. cerevisiae for AP2 loss-of-function; large chemical genetic screens have found altered sensitivities, the basis for which is obscure (Hillenmeyer et al., 2008; Parsons et al., 2006).
To better understand the cause of toxin resistance in the four AP2 subunit deletions, we analyzed dynamics of the yeast AP2 β (Apl1p) and μ (Apm4p) subunits. The Apl1-GFP and Apm4-GFP strains exhibited wild-type toxin sensitivity, suggesting that the GFP fusions are functional (data not shown). Both Apl1-GFP and Apm4-GFP localize to cortical patches that resemble known sites of endocytosis (Figure 4B) (Kaksonen et al., 2003). The Apl1-GFP and Apm4-GFP patches are dynamic and appear and disappear with similar average lifetimes of 67 ± 32 seconds and 64 ± 29 seconds, respectively. The long and variable lifetimes of these proteins are similar to those of early arriving endocytic proteins clathrin (Kaksonen et al., 2005) and Ede1p (Toshima et al., 2006). The Apl1-GFP and Apm4-GFP patches internalize from the cortex of the cell before disappearing (95%, n=105 for both Apl1p and Apm4p) in a manner characteristic of endocytic coat proteins (Kaksonen et al., 2005). To confirm that this internalization corresponds to endocytic events, we recorded two-color movies of Apl1-GFP with the endocytic coat protein Sla1-mCherry or the late endocytic patch marker Abp1-mRFP. Apl1-GFP forms patches, which are joined subsequently by, and then internalized with, Abp1-mRFP (above) and Sla1-mCherry (below) (Figure 4C). Most endocytic patches contain AP2, as 91% (n=102) of Sla1-mCherry patches had detectable Apl1-GFP. These results indicate that AP2 is a part of the endocytic machinery in yeast and that it internalizes with the endocytic coat upon assembly of actin as marked by Abp1p. Because of this dynamic behavior, yeast AP2 can be included in the previously defined endocytic coat module (Kaksonen et al., 2005).
Like mammalian AP2, yeast AP2 subunits physically associate, and a deletion of one subunit prevents the other subunits from interacting (Yeung et al., 1999). Consistent with this finding, Apm4-GFP failed to localize to cortical endocytic patches in apl1Δ cells (Figure 4D). This observation suggests that AP2 subunits likely form a complex at endocytic sites, and that they are interdependent on each other for proper localization.
Together, our data place the AP2 complex within the modular framework set out by Kaksonen et al. (2005). Figure 4E shows all K28 resistant endocytic hits arrayed, where possible, along the temporal endocytic pathway from coat to WASP/myosin to actin to scission modules, and includes the newly assigned placement of AP2 in the coat module. Some of the other endocytic genes we identified cause only very mild endocytic phenotypes, such as abp1Δ and myo5Δ. In total, these results show that K28 can be used as an extremely sensitive tool to investigate cargo uptake by cells.
Since AP2 localizes to endocytic sites and is important for K28 toxicity, we tested its involvement in K28 internalization. Subcellular fractionation was used to follow K28 in cell lysates prepared from toxin-treated cells. The amount of K28 in each fraction was variable between experiments, and therefore these data cannot be quantitatively analyzed. As shown in Figure 4F, wild-type cells and deletions of APL2 or APL6, which are subunits of the AP2-related AP1 and AP3 complexes, accumulated K28 in the cytosol (S100 fraction). Conversely, deletion of AP2 subunits prevented K28 accumulation in the cytosol in each of three independent experiments (Figure 4F and data not shown). K28 was not detectable in the cytosolic fraction of end3Δ endocytic mutants in 3 out of 4 independent experiments, and appeared as a faint band in the fourth experiment, shown in Figure 4F. The presence of K28 in the P100 fraction (Golgi, endosomal and/or vesicle membranes) in end3 and apl1 mutants is likely caused by leakiness in the endocytic blocks of these mutants. To confirm that END3 and APL1 deletions had defects in K28 uptake, we measured K28 levels in cell-free culture supernatant of yeast spheroplasts incubated with K28. Figure 4G shows that wild-type but not end3 or apl1 mutant spheroplasts, deplete K28 from the supernatant in a time-dependent manner. Together these results suggest that AP2 mediates K28 uptake.
Mutating certain early-arriving endocytic proteins in yeast, such as clathrin or Ede1p, decreases both the lifetime of later patch components, such as Sla1p, and the number of endocytic sites in the cell (Kaksonen et al., 2005). Unlike ede1Δ yeast, there was no significant decrease in the Sla1-GFP lifetimes in apl1Δ yeast, nor were there decreases in Sla1-GFP patch number (Figure S1A). These data suggest that AP2 does not significantly impact the function of the core endocytic machinery. Finally, to test whether AP2 subunit deletions affect endocytosis of other cargos or the plasma membrane, we examined uptake of a fluorescent derivative of the peptide cargo α-factor and of the lipophilic dye FM4-64 (Toshima et al., 2006). Deletion of AP2 subunits did not affect the endocytosis of FM4-64 or α-factor (Figures S1B and S1C). Taken together, these results suggest yeast AP2 has a strong but cargo-specific effect on K28 endocytosis.
It is well established that mammalian AP2 mediates endocytic cargo recognition (Sorkin, 2004), and we have shown here that the yeast AP2 complex has an endocytic role in K28 uptake. An alternative hypothesis, that AP2 is involved in a non-endocytic trafficking event, is unlikely because Apl1-GFP and Apm4-GFP localize only to the cell cortex. Our data lead us to suggest that yeast AP2 acts as a specific endocytic adaptor for the killer toxin receptor. This is one of the first reports of a cargo-selective component of the yeast endocytic machinery (Burston et al., 2009). Which cargos other than K28 and its yet-to-be identified receptor are selected by AP2 remains to be discovered, but is of considerable interest, as a growing number of factors, such as a recently described family of arrestin-like proteins, seem to contribute selectivity in specification of the plasma membrane proteome composition (Lin et al., 2008).
Our results define interconnected protein complexes and pathways that impact K28 function and suggest testable hypotheses about virtually every aspect of K28-target cell interaction. This study and the literature (Broeck et al., 2007; Sandvig and van Deurs, 2002; Schmitt and Breinig, 2006) (Table S5) suggest that K28 has striking mechanistic similarities to clinically relevant toxins. By exploiting the genomic tools available in yeast we identified 365 genes that mediate K28 sensitivity, thereby providing a resource that promises to be useful in understanding A/B toxin action. This type of screen could not be done with toxins that target mammalian cells, but still identified conserved pathways. Therefore, the results are expected to be relevant to understanding how Cholera, Shiga and other toxins affect mammalian cells.
Yeast strains used in this study are listed in Table S6. The K28 toxin sensitivity screen was carried out using yeast deletion (Research Genetics) (Winzeler et al., 1999) and ts-allele collections (Ben-Aroya et al., 2008). GFP tagged AP2 subunits were constructed using homologous recombination as described (Longtine et al., 1998). Yeast were grown either in rich (YPD) or synthetic media essentially as described (Amberg et al., 2005).
MS300c yeast secreting K28 were spotted onto methylene blue agar (MBA) plates spread with a lawn of the mutant strain being tested (for details, see Supplemental Experimental Procedures). Deletion mutants were grown at room temperature, while replicates of the ts-alleles were grown at 25, 30 and 34°C. The distance from the spot of MS300c cells to the lawn of sensitive cells was measured at 48 and 72 hours. MATα (BY4742) deletion mutants that tested as resistant or hypersensitive were retested, and reproducible hits were confirmed with further testing in the MATa or homozygous diploid deletion collections, or by PCR (see Table S1).
Gene ontology processes, functions and components annotated to the 176 K28 resistant or 189 K28 hypersensitive genes (Table S1) were compared to a background set of all 5015 mutations for which we generated K28 sensitivity scores according to the method described in Boyle et al., (2004) (Generic GO Term Finder, http://go.princeton.edu/cgi-bin/GOTermFinder).
10 OD/ml of wild-type or mutant cells were incubated in K28-containing supernatant for 15 minutes at 4°C, as previously described (Breinig et al., 2002) (for details, see Supplemental Experimental Procedures). The cells were removed and the remaining supernatant was tested for activity. The zone of clearance in the lawn could be converted into a percentage of the undepleted toxin activity.
Imaging of GFP-labeled proteins and two-color imaging of GFP and mRFP or mCherry labeled proteins was done using an Olympus IX-71 or IX-81 microscope essentially as described (Kaksonen et al., 2005). For single channel imaging, neutral density filters were used to reduce the intensity of excitation light and time-lapse movies were typically acquired at rate of 1 frame every 2 seconds.
For cell fractionations yeast were grown to early exponential growth phase, spheroplasted, washed in 0.8 M sorbitol buffer, and incubated for 1.5 h at 20°C in the presence of 2×105 U/ml K28 toxin (Eisfeld et al., 2000; Schmitt and Tipper, 1990; Seaman et al., 1998) (for details, see Supplemental Experimental Procedures). Thereafter, cells were washed, lysed, and subjected to differential centrifugation: P13, plasma, endosomal, Golgi and ER membrane fraction (13,000 g); P100, endosomal and Golgi membrane and vesicle fraction (100,000 g pellet); S100, cytosolic fraction (100,000 g supernatant). Protein samples were analyzed by Western blot using antibodies directed against K28 (Heiligenstein et al., 2006), Pfk1/2p, actin-β (mAbcam 8224), Pep12p (2C3-G4, Molecular Probes), and/or Sec61p.
Toxin uptake by yeast cell spheroplasts was determined via SDS-PAGE and Western analysis as previously described (Heiligenstein et al., 2006). Briefly, anti-K28 antibodies detected the amount of toxin remaining in the cell-free culture supernatant after incubating 2.5×108 spheroplasts at 30°C with purified K28 over 4 hours.
Imaging data was collected using Metamorph software (Molecular Devices) and processed using Image J (http://rsbweb.nih.gov/ij/index.html) to subtract background fluorescence and to normalize for the effects of photo-bleaching. Network diagrams were generated using Osprey 1.2.0 (Breitkreutz et al., 2003). Graphs and tables were generated using Microsoft Excel and all figures were generated using Adobe Illustrator CS or CSII.
Table S1. Mutant strains with an altered K28 killer phenotype
Table S2. Gene Ontology terms for K28 resistant and hypersensitive genes
Table S3. Comparison of enriched GO terms from zymocin and K28 screens
Table S4. Epistatic relationships among mutants with altered K28 sensitivity
Table S5. Examples of K28-related cellular pathways implicated in the function of mammalian toxins
Table S6. Yeast strains used in this study
Document S1. Supplemental Results and Supplemental Experimental Procedures
We thank Phil Hieter for providing the collection of ts-alleles, Randy Schekman and Jürgen Heinisch for antibodies, Aysha Chemparathy for technical assistance, Yidi Sun and Christopher Buser for critical reading of the manuscript, and members of the Drubin and Barnes labs, Liz Conibear and Mike Eisen for helpful discussions. P.C.S. acknowledges a postdoctoral fellowship from the Natural Sciences and Engineering Research Council of Canada. D.G.D and M.J.S. acknowledge funding from the National Institutes of Health (GMR01 50399) and from the Deutsche Forschungsgemeinschaft (Schm 541/12-1 and GRK845), respectively.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.