The murine cytomegalovirus immunoevasin gp40 binds MHC class I molecules to retain them in the early secretory pathway

To escape from destruction by the cellular adaptive immune system, viruses inhibit almost every step of major histocompatibility complex (MHC)-class-I-mediated antigen presentation (Ambagala et al., 2005). Herpesviruses encode multiple interfering proteins (immunoevasins) (Basta and Bennink, 2003). The glycoprotein 40 kDa (gp40) of the murine cytomegalovirus (mCMV), encoded by the m152 gene, was the first inhibitor of MHC-class-I-mediated antigen presentation and natural killer cell function to be described in CMV (Krmpotić et al., 2002; Lu et al., 2006). Gp40 inhibits the transport of peptide-loaded class I molecules (proteins) to the cell surface and retains them in the early secretory pathway, but its molecular mechanism of action is still unknown (Pinto and Hill, 2005). Earlier investigations have shown that gp40-retained class I molecules are bound to high-affinity peptide but fail to proceed beyond the ER-Golgi intermediate compartment (ERGIC). In contrast, gp40 itself has been shown to progress to the lysosomes for degradation. The authors found an interaction of gp40 with calnexin but not with class I molecules (Ziegler et al., 2000,, 1997).

Gp40 also retains the class-I-related stress marker RAE-1. Recently, a crystal structure of gp40 in complex with RAE-1 has been published (Zhi et al., 2010). This has prompted us to re-examine the question of gp40–class-I interaction. We show here that, indeed, gp40 binds to class I molecules to retain them in the early secretory pathway, most likely by retrieval from the ERGIC or cis-Golgi, and we demonstrate that a sequence in the linker between the folded lumenal domain of gp40 and the transmembrane sequence is required for this retention.

To assess the effect of gp40 on murine class I molecules, we expressed m152 in K41 cells (murine fibroblasts) by lentiviral transduction. The surface levels of the endogenous class I allotypes H-2D^b (D^b) and H-2K^b (K^b) were reduced to background levels in most cells, as observed by flow cytometry with the allotype-specific beta-2 microglobulin (β₂m)-dependent antibodies B22.249 (for D^b) and Y3 (for K^b) (Fig. 1A). K^b, but not D^b, was fully resistant to gp40 in some cells (arrow), especially in confluent cultures, and there was no evidence of intermediate retention phenotypes. We conclude that whereas gp40-mediated retention of class I molecules can be highly effective, it varies between class I allotypes and growth conditions.

To find the intracellular steady-state location of the gp40-retained class I, we performed fluorescence microscopy with the same antibodies, and we also stained for gp40 with an antibody against a C-terminal hemagglutinin (HA) tag (Fig. 1B,C; Fig. S1). Gp40 was observed in a compact juxtanuclear location that colocalized well with the ERGIC-53 (also known as LMAN1) marker (Fig. 1C, top row) and with the Golgi marker GM130 (also known as GOLGA2) (Fig. S2). K^b was found in the same location (Fig. 1B, top row, Fig. 1C, center row; Fig. S1). In some cells (∼30%), D^b localized exclusively to the same juxtanuclear region (Fig. 1B, bottom row; Fig. S1), but in the majority of cells, D^b exhibited, in addition, a spread-out reticular pattern reminiscent of the ER (Fig. 1B, center row, Fig. 1C, bottom row, Fig. 1D; Fig. S1). In agreement with earlier data for H-2K^d (Ziegler et al., 1997), we conclude that D^b and K^b are indeed retained in an intracellular compartment where they colocalize with gp40, and we also find that the exact steady-state subcellular location of a gp40-retained class I molecule depends on the allotype (Kavanagh et al., 2001).

To understand the influence of gp40 on the kinetics of class I transport from the ER to the Golgi, we next performed radiolabeling and pulse-chase experiments with endoglycosidase F1 digests (EndoF1; see Materials and Methods). In wild-type cells, D^b and K^b glycans become resistant to digestion with EndoF1 in the medial Golgi due to the action of mannosidase II. When they reach the trans-Golgi, further carbohydrates including sialic acids are added; this sialylation results in an upward shift of the EndoF1-resistant band on SDS-PAGE gels (Fritzsche and Springer, 2013). In gp40-expressing cells, D^b showed no sialylation after 2 h of chase (Fig. 1E, bottom row); thus, it had not passed the trans-Golgi. In contrast, K^b showed a small amount of EndoF1 resistance and sialylation (Fig. 1E, second row, sia). This agrees with the above observation that in the presence of gp40, K^b can still reach the surface in some cells (Fig. 1A, arrow).

Remarkably, at the 2-h time point, both D^b and K^b acquired partial (incomplete) EndoF1 resistance, visible as two (K^b) or three (D^b) intermediate bands between the EndoF1-resistant and -sensitive forms (Fig. 1E, arrowheads), corresponding to the number of glycosylation sites for each allotype. This was never seen in the gp40-free control samples, where D^b and K^b progressed directly from the EndoF1-sensitive form to the fully EndoF1-resistant and sialylated form. At longer chase times of up to 16 h, the partially EndoF1-resistant forms of D^b and K^b persisted (Fig. 1F, arrowheads), and sialylation still did not occur (D^b), or only for a small fraction of the molecules (K^b). To determine whether the partially EndoF1-resistant forms were still intracellular or had progressed to the cell surface, we pulse-labeled gp40-expressing cells and chased for 120 min. We then incubated the intact cells with biotinylated K^b-specific peptide for 120 min, washed, and lysed the cells, and we found that only the sialylated, but not the EndoF1-sensitive or the partially EndoF1-resistant forms of K^b, precipitated with streptavidin-conjugated agarose (Fig. S3A). Thus, the partially EndoF1-resistant forms of class I in gp40-containing cells (arrowheads in Fig. 1E,F) are not cell surface forms but are trapped in the cell interior, just like the partially EndoF1-resistant class I proteins in transporter associated with antigen processing (TAP)-deficient cells, which circulate between the ER and the cis-Golgi (Fritzsche et al., 2015). Given that at least D^b does not become EndoF1-resistant at all in gp40-containing cells, we conclude that the compact juxtanuclear steady-state location observed for D^b and K^b in Fig. 1B is a pre-medial-Golgi compartment, most likely the ERGIC and/or the cis-Golgi, as suggested previously (del Val et al., 1992), and in agreement with our microscopic analysis.

As shown in earlier reports, D^b and K^b were not rapidly degraded during the chase but instead were slightly more metabolically stable than in wild-type cells (Fig. S3B and Ziegler et al., 2000), suggesting that they are not rapidly routed to lysosomes in the presence of gp40.

Taken together with the microscopy results, the results of the pulse-chase experiments suggest that in the presence of gp40, D^b and K^b do not become transported to the medial Golgi but are present in the ERGIC and cis-Golgi (D^b and K^b) and the ER (D^b) at steady state, most likely circulating between these compartments.

We next wondered whether gp40 travels from the ER to the Golgi in synchrony with the class I molecules it retains, and we followed it in a pulse-chase experiment under the same conditions as above. Intriguingly, a form of gp40 that was fully EndoF1 resistant and sialylated (Fig. 1G, sia) and several partially EndoF1-resistant forms, resembling those of class I molecules (arrowheads), appeared after 30 min, and both fully and partially resistant forms persisted throughout the 2-h chase. Partial EndoF1 resistance during the chase was also visible for a variant of gp40 with a C-terminal cytosolic KKSQ sequence for ER retention (Fig. S3C). Thus, in our system, at least some gp40 is retained in a pre-medial-Golgi compartment, most likely in an ER–Golgi cycle, just like the class I molecules that it retains. Interestingly, the gp40-KKSQ variant retained both D^b and K^b very efficiently, which suggests that gp40 does not need to progress beyond the cis-Golgi to be effective (Fig. S3D).

Many viral proteins interfere with class I synthesis and folding or with peptide loading onto class I molecules, for example by inhibiting or removing TAP (Loch and Tampe, 2005; Momburg and Hengel, 2002). Given that our results suggested that gp40 acts in the early secretory pathway, we first asked whether it inhibits class I folding and peptide binding. We applied a 5-min radioactive pulse and followed the folding of the class I heavy chains by immunoprecipitating with β₂m-dependent antibodies at different chase times and quantifying total D^b or K^b. gp40 did not cause a significant change in in vivo folding kinetics for either allotype (Fig. 2A). To assess the binding affinity of the overall peptide load of D^b and K^b, we then performed thermal denaturation experiments by heating cell lysates to different temperatures and immunoprecipitating with Y3 and B22 (Gao et al., 2002; Garstka et al., 2011). There was no evidence of any impairment of peptide loading in the presence of gp40 for either allotype such as it is seen for tapasin or TAP deficiency (Fig. 2B; compare with Fig. 3B and with figures S2D and S3H in Fritzsche et al., 2015). Thus, in agreement with earlier reports (del Val et al., 1992; Ziegler et al., 2000), we find that gp40 does not impair class I folding, maturation, or peptide binding.

Class I molecules loaded with suboptimal peptide ligands are retained in the early secretory pathway of wild-type cells by two different mechanisms (Springer, 2015): first, by the class-I-specific chaperone tapasin (Paulsson et al., 2002), and second, by the lectin calreticulin (Howe et al., 2009), probably in concert with the UDP-glucose:glycoprotein glucosyltransferase (UGT1) (Zhang et al., 2011). Both tapasin and calreticulin are members of the class I peptide-loading complex (PLC). To investigate whether gp40 somehow appropriates these or similar class I retention mechanisms that already exist in the cell, we studied the effect of gp40 in cell lines that lack functional calreticulin, tapasin, TAP, calnexin, or UGT1 (Fig. 2C). In every case, gp40 caused retention of both D^b and K^b to the same extent as in wild-type cells. We conclude that no single known class-I-associated protein of the early secretory pathway is required for gp40-mediated retention.

Class I molecules are released from the PLC after binding high-affinity peptide (Ortmann et al., 1994; Springer, 2015). We next asked whether this interaction is prolonged by gp40 to achieve class I retention, and so we performed a pulse-chase experiment, precipitated tapasin from the cell lysate, and analyzed the class I molecules bound to it in a re-precipitation experiment (Fig. 2D; schematic in Fig. S3H). Some D^b was indeed bound to tapasin throughout the chase, but this was the case both in the presence and absence of gp40. Binding of K^b to tapasin appeared stronger in the presence of gp40, but the K^b band intensity clearly decreased over time, which suggests eventual dissociation of K^b from tapasin and not an irreversible association. We conclude that gp40 does not use the PLC, or any of its proteins individually, to retain class I in the early secretory pathway.

Direct binding to class I is a hallmark of many viral immunoevasins (Bennett et al., 1999; Furman et al., 2002; Jones et al., 1996). To test whether gp40 interacts with class I molecules, we immunoprecipitated gp40–HA from lysates of radiolabeled cells and re-precipitated with an antiserum against a common sequence in the cytosolic tails of D^b and K^b (D^b/K^b serum). Both D^b and K^b co-precipitated with gp40; the interaction was also seen when we first immunoprecipitated with D^b/K^b serum and then re-precipitated with anti-HA antibodies (Fig. 3A). Given that D^b and gp40 have an almost identical molecular mass on SDS gels, we demonstrated in HeLa cells that the D^b/K^b serum does not cross-react with gp40 (Fig. S3E). Gp40 also co-precipitated with tapasin, calnexin, and calreticulin (Fig. S3F). Interaction with tapasin was genuine and not a post-lysis artifact (shown in a mixed lysate experiment, Fig. S3G), but it was not required for gp40-mediated class I retention because in tapasin-deficient cells, class I molecules still bound to gp40 (Fig. 3B) and were still retained (Fig. 2C).

Intriguingly, first-round immunoprecipitations with the monoclonal antibodies B22.249 and Y3 co-precipitated much less gp40 than the D^b/K^b serum. This suggests that gp40 might mask the epitopes of these antibodies, which lie in the α₁/α₂ superdomain of D^b and K^b. In contrast, the monoclonal antibody 28-14-8S, which binds to the α₃ domain of D^b, efficiently co-precipitated gp40 (Fig. 3C). Our data thus show that gp40 interacts with class I; the simplest interpretation is that it binds directly to the α₁/α₂ superdomain of D^b and K^b, as proposed previously for L^d (Wang et al., 2012).

We next hypothesized that gp40 and the class I molecules bound to it might be retained as a complex in the early secretory pathway, and so we decided to investigate the EndoF1 resistance pattern of class-I-associated gp40 in pulse-chase experiments with re-precipitations (Fig. 4). Association of gp40 with D^b and K^b was observed at the start of the chase, suggesting that it occurs shortly after synthesis of the three proteins. Those gp40 molecules that were bound to class I (Fig. 4D) became partially EndoF1-resistant over time, but they showed little sialylation (as compared to the entire cohort of gp40 at the same time point, Fig. 4C), suggesting that most did not progress beyond the ER–Golgi cycle. Likewise, in the class I molecules bound to gp40, the sialylated band of K^b was very weak (Fig. 4B, arrow; compare Fig. 4F, arrow), suggesting that gp40 associates only with the intracellular forms of class I. The simplest explanation of these data is that gp40 and class I form a complex that is retained in the ER, ERGIC and cis-Golgi, probably by cycling through these compartments.

We next decided to investigate the mechanism of retention of gp40 in the early secretory pathway. Given that the sequence of gp40 contains no known retention signal, we tested a panel of mutants (data not shown). In one such mutant, we replaced the 43-amino-acid linker between the folded luminal domain and the transmembrane domain (Wang et al., 2012) by a (glycine₄-serine)₉ sequence to yield the gp40-(G₄S)₉ mutant (Fig. 5A). In a pulse-chase, gp40-(G₄S)₉ was much more strongly sialylated than wild-type gp40, suggesting fast progress to the trans-Golgi, and it was much shorter-lived, suggesting rapid degradation in lysosomes (Fig. 5B). The rapid export of gp40-(G₄S)₉ indicates that it passes the cellular quality control and that the introduced mutation does not lead to misfolding. Gp40-(G₄S)₉ no longer decreased the steady-state surface levels of D^b or K^b (Fig. 5C,D), and it bound to class I molecules much more weakly than wild-type gp40 (Fig. 5E).

To find out whether this weak binding of gp40-(G₄S)₉ to class I molecules was the cause or the consequence of its rapid export from the early secretory pathway, we forced it to remain in the ER together with class I molecules by treating the cells with brefeldin A (BFA). Immunoprecipitation showed that under these conditions, it bound to class I molecules with the same strength as the wild type (Fig. 5F); this suggests that its rapid export deprives gp40-(G₄S)₉ of the opportunity to bind tightly to class I molecules, resulting in lack of class I retention.

Taken together, these experiments suggest that retention of the gp40–class-I complex in the early secretory pathway is achieved through a sequence in the linker of gp40, and that the linker does not mediate binding of gp40 to class I molecules, but it is required for the retention of gp40 itself.

We demonstrate here for the first time a physical interaction of murine class I molecules with mCMV gp40 that causes their retention in the cell interior. From a crystal structure of gp40 with the class-I-like protein RAE-1, Margulies and collaborators have predicted that gp40 binds directly to the top of the α₁/α₂ superdomain of class I molecules (Wang et al., 2012). Our data in Fig. 3C support this hypothesis: both the monoclonal antibodies Y3 and B22.249, which bind to the α₁/α₂ superdomain of class I molecules, preclude gp40 co-precipitation, whereas the antibody 28-14-8S, which binds to the α₃ domain, co-precipitates gp40 (Allen et al., 1984; Nathenson et al., 1989). In the structural model of Wang et al. (2012), there is space for class-I-bound peptide, which agrees with the observation that gp40 does not inhibit peptide binding (Fig. 2A,B; del Val et al., 1992). In previous co-precipitation experiments, a gp40–class-I interaction was not found (Ziegler et al., 2000). Given that the authors of that study did not re-precipitate, they were unable to use the heavy chain signal to detect co-precipitated class I (it appears at the same molecular mass as gp40); thus, they depended on the β₂m signal, which might have been too weak to observe in their system. We have also found that gp40 does not bind to the human class I molecules in HeLa cells, which might explain why gp40 does not retain human class I molecules (Ziegler et al., 1997; data not shown).

We think that in the cell, the class-I–gp40 complex is restricted to a compartment prior to the medial Golgi because class-I-bound gp40 and gp40-bound class I both acquire little sialylation over time (which would signify arrival in the trans-Golgi) and do not even become completely EndoF1 resistant (Fig. 4). As assessed by microscopy, in the presence of gp40, D^b is mostly found in the ER, whereas K^b is mostly in the ERGIC and cis-Golgi (Fig. 1B,C). Given that K^b is exported from the ER with greater efficiency than D^b (Fritzsche et al., 2015), this suggests a dynamic scenario of class I retention by retrieval of gp40–class-I complexes from the cis-Golgi to the ER; in such an export-and-retrieval cycle, the higher ER exit rate of K^b might shift its steady-state location further towards the ERGIC and cis-Golgi. The faster anterograde transport of K^b might also cause the escape of K^b from gp40-mediated retention that we observed in some cells (Fig. 1A, arrow). When K^b is forced by gp40 to circulate between ER and Golgi, its increased concentration in the early secretory pathway might cause the tighter association with the PLC that we observe (Fig. 2D).

Our finding of localization of gp40 in the ERGIC and cis-Golgi at steady state agrees with previous microscopy results that show that it had an excellent colocalization with p58 (also known as LMAN-1) (Ziegler et al., 1997). In that work, the steady-state location of gp40 shifted to the lysosomes if protein degradation was inhibited with leupeptin (Ziegler et al., 2000); this suggests that, eventually, most gp40 leaves the ER–Golgi cycle to become degraded in the lysosomes. In our experiments, when expressed from a retroviral vector, gp40 is rather long-lived (Fig. 1G), whereas in Ziegler et al. (2000), where cells are transfected with an expression plasmid, the bulk of gp40 is degraded after 2 h (without cycloheximide) or is EndoH resistant, suggesting transit through the medial Golgi. We think that overexpression of gp40 might saturate the retention mechanism that holds gp40 in the early secretory pathway and thus lead to the transport of excess gp40 to the lysosomes for degradation. Even in our expression system, retention of gp40 is not complete because we observe some sialylated gp40 in the pulse-chase experiments (Fig. 1G). In class-I-deficient murine fibroblasts, this sialylated gp40 fraction was the same (data not shown), which suggests that class I molecules do not determine the rate of exit of gp40 from the early secretory pathway.

Taken together, our data and those from the literature suggest that gp40 is temporarily held in the early secretory pathway by a saturable retention mechanism. This retention clearly depends on an amino acid sequence in the gp40 linker (Fig. 5), but its molecular mechanism is not obvious because gp40 lacks any known retention or retrieval motif. The linker itself might contain such a motif, perhaps in a three-dimensional structural element; alternatively, the linker sequence might structurally support, or contribute to, a motif that is elsewhere in the gp40 protein.

As we and others have shown, gp40 interacts with calnexin and calreticulin, which have ER retention signals (Ziegler et al., 2000; data not shown and Fig. S3F), but neither protein is required for gp40-mediated class I retention (Fig. 2C). Perhaps gp40 can bind alternatively to several different proteins to remain in the early secretory pathway, in analogy to the ‘dynamic matrix’ model (Nehls et al., 2000).

So how does gp40 retain class I molecules? Our data suggest that the major fraction of gp40 is retained in the ER and associates with class I very soon after its synthesis (Fig. 4). The complex then circulates in the early secretory pathway for several hours (Figs 4 and 1F). Retention of gp40 itself in this cycle is necessary for class I retention (Figs 5 and 6). Eventually, the complex of class I and gp40 travels to the lysosomes for rapid degradation, perhaps by direct transfer from the trans-Golgi and so avoiding the cell surface. Some gp40 molecules, in contrast, are not retained in the early secretory pathway, and due to their rapid exit, they do not associate with class I molecules. In earlier published work, the large amount of overexpressed gp40 that was rapidly transported to the lysosomes might have obscured those gp40 molecules that remained in the early secretory pathway to retain class I molecules.

In this work, we have demonstrated a new mechanism used by cytomegaloviruses in their multi-pronged attack on class I antigen presentation. Future work might focus on further exploring the intriguing trafficking path of gp40 and the factors that govern it. Gp40-mediated retention or retardation of class I might be used to study the connection between the rate of cell surface transport of class I and its binding of peptide ligands (Praveen et al., 2010), and perhaps as a tool in a therapeutic context to manipulate class I peptide selection in autoimmunity, cancer and immunotherapy.

Chemicals were purchased from AppliChem (Darmstadt, Germany) or Carl Roth (Karlsruhe, Germany). Mouse monoclonal hybridoma supernatants Y3 (Hammerling et al., 1982), B22.249 (Lemke et al., 1979) and HA 12CA5 (Niman et al., 1983) were as described previously. Tapasin serum 2668 was kindly donated by Xiaoli Wang (Dept. of Pathology and Immunology, Washington University School of Medicine, USA). Rabbit anti-calnexin serum was kindly provided by David Williams (Dept. of Biochemistry, University of Toronto, Toronto, Canada). PA3-900 antiserum (Pierce Antibodies), monoclonal antibody against GM130 (BD Transduction Laboratories, No. 610822), and Cy3-conjugated ERGIC-53 antiserum (Sigma-Aldrich) were purchased. Rabbit antiserum against H-2D^b and H-2K^b was generated by Charles Rivers Laboratories (Kisslegg, Germany) against the peptide C-RRRNTGGKGGDYALAPGSQ corresponding to the membrane proximal C-terminal cytosolic tail of both H-2D^b and H-2K^b (residues 331–349, Swiss-Prot P01901); biotinylated SIINFEKL peptide was synthesized by Genecust (Luxembourg).

K41 and K42 cells (Gao et al., 2002) were kindly provided by Tim Elliott (Faculty of Medicine and Institute for Life Sciences, University of Southhampton, UK), MEF TPNd (Grandea et al., 2000) and MEF TAPd were kindly provided by Luc van Kaer (Dept. of Pathology, Microbiology and Immunology, Vanderbilt University School of Medicine, USA), MEF CNXd (Kraus et al., 2010) were a gift from Jody Groenendyk (Dept. of Biochemistry, University of Alberta, USA). MEF UGT1d (Solda et al., 2007) were a gift from Tatiana Soldà (Institute for Research in Biomedicine, Protein Folding and Quality Control, Switzerland Università della Svizzera Italiana, Switzerland). Cells were grown at 37°C and 5% CO₂ in high-glucose (4.5 g/l) Dulbecco's modified Eagle's medium (DMEM) medium (GE Healthcare) supplemented with 10% FCS (Biochrom, Berlin, Germany), 2 mM glutamine, 100 U/ml penicillin and 100 µg/ml streptomycin. For flow cytometry experiments with MEF TAP- and MEF TPN-deficient cells, cells were incubated before staining at 25°C in CO₂-independent medium (Life Technologies, Darmstadt, Germany) supplemented as above to accumulate MHC class I at the cell surface (Ljunggren et al., 1990).

Retroviral expression, microscopy and flow cytometry were performed as in Hein et al., 2014.

Pulse-chase experiments were performed as in Fritzsche and Springer, 2014. For Fig. S3A, 5 µM K^b-specific biotinylated peptide (ovalbumin 357-64 peptide, sequence SIINFEK_bioL, with biotin attached to the lysine side chain by a 6-aminohexanoic acid linker) was added to the cells, followed by lysis in the presence of 10 µM non-biotinylated peptide (to prevent post-lysis binding), precipitation with streptavidin–agarose and SDS-PAGE (lane 3). For the other lanes, biotinylated peptide was added only after lysis (without non-biotinylated peptide) in order to precipitate all forms of K^b. Precipitation was with streptavidin–agarose or monoclonal antibody Y3 and protein-A–agarose, as indicated. For Fig. 5F, cells were radiolabeled for 15 min in medium containing 10 µg/ml BFA. Afterwards, cells were lysed, and gp40–class-I complexes were precipitated as described below.

Labeling, pulse chase and co-immunoprecipitation were performed as in Halenius et al., 2011 except that protein A agarose (Merck Millipore) was used instead of protein-A–sepharose. Precipitated proteins were then eluted from the agarose beads by boiling in 50 µl denaturation buffer (1% SDS, 2 mM DTT) at 95°C for 10 min. Samples were cooled on ice, and SDS was neutralized with a 20-fold volume (1 ml) of 0.1% Triton X-100 in PBS. Samples were centrifuged at 1000 g for 10 min, and 900 µl were transferred to protein-A–agarose beads prebound to the respective antibody for the re-immunoprecipitation and incubated for 1 h at 4°C rotating. The beads were washed twice in PBS with 0.1% Triton X-100, and precipitated proteins were eluted by boiling in 20 µl denaturation buffer at 95°C for 5 min for SDS-PAGE.

We would like to thank Hartmut Hengel for advice on the project and the manuscript; Constanze Wiek and Helmut Hanenberg for the retroviral expression system; Peter Reinink for computational biology support; those mentioned in the Materials and Methods and Hesso Farhan for donating reagents; Uschi Wellbrock for excellent technical assistance; and Ina Huppertz, Maria Bottermann, Andrei Iosif S¸mid, and Florin Tudor Ilca for preparatory and additional laboratory work on the project.