Enzymes Catalyzing Crotonyl-CoA Conversion to Acetoacetyl-CoA During the Autotrophic CO2 Fixation in Metallosphaera sedula

Autotrophic Crenarchaeota use two different cycles for carbon dioxide fixation. Members of the Sulfolobales use the 3-hydroxypropionate/4-hydroxybutyrate (HP/HB) cycle, whereas Desulfurococcales and Thermoproteales use the dicarboxylate/4-hydroxybutyrate cycle. While these two cycles differ in the carboxylation reactions resulting in the conversion of acetyl-CoA + 2 CO2 to succinyl-CoA, they have a common regeneration part in which succinyl-CoA is reconverted to two acetyl-CoA molecules. This common part includes crotonyl-CoA conversion to acetoacetyl-CoA, which has unequivocally been shown in Ignicoccus hospitalis (Desulfurococcales) and Pyrobaculum neutrophilus (Thermoproteales) to be catalyzed by a bifunctional crotonase/3-hydroxybutyryl-CoA dehydrogenase. It is a fusion protein consisting of an enoyl-CoA hydratase and a dehydrogenase domain. As the homologous bifunctional protein is present in Sulfolobales as well, its common functioning in the conversion of crotonyl-CoA to acetoacetyl-CoA was proposed. Here we show that a model autotrophic member of Sulfolobales, Metallosphaera sedula, possesses in addition to the bifunctional protein (Msed_0399) several separate genes coding for crotonyl-CoA hydratase and (S)-3-hydroxybutyryl-CoA dehydrogenase. Their genes were previously shown to be transcribed under autotrophic and mixotrophic conditions. The dehydrogenase Msed_1423 (and not the bifunctional protein Msed_0399) appears to be the main enzyme catalyzing the (S)-3-hydroxybutyryl-CoA dehydrogenase reaction. Homologs of this dehydrogenase are the only (S)-3-hydroxybutyryl-CoA dehydrogenases present in all autotrophic Sulfolobales, strengthening this conclusion. Two uncharacterized crotonase homologs present in M. sedula genome (Msed_0336 and Msed_0384) were heterologously produced and characterized. Both proteins were highly efficient crotonyl-CoA hydratases and may contribute (or be responsible) for the corresponding reaction in the HP/HB cycle in vivo.


INTRODUCTION
Autotrophic CO 2 fixation is the most important biosynthetic process in nature, being responsible for the primary production of organic matter. Autotrophy is arguably the oldest type of metabolism (Wächtershäuser, 1988;Russell and Martin, 2004;Fuchs, 2011), and the first metabolic pathway was probably the (autotrophic) reductive acetyl-CoA pathway (Russell and Martin, 2004;Fuchs, 2011;Weiss et al., 2016). During evolution, several fundamentally different CO 2 fixation pathways evolved. They vary not only in their reaction sequences, but also in their types of carboxylases, cofactors, and electron donors used to fix inorganic carbon into biomass (Berg et al., 2010a;Berg, 2011;Fuchs, 2011;Hügler and Sievert, 2011). Conversion of CO 2 , a greenhouse gas, into value-added products attracts great attention to autotrophic metabolism, rendering it one of the popular topics for fundamental and applied research.
The 3-hydroxypropionate/4-hydroxybutyrate (HP/HB) cycle was discovered in thermophilic aerobic Crenarchaeota of the order Sulfolobales (Figure 1; Ishii et al., 1996;Menendez et al., 1999;Berg et al., 2007). Later, a formally similar, but phylogenetically unrelated pathway was shown in aerobic ammonia-oxidizing Thaumarchaeota (Könneke et al., 2014;Otte et al., 2015). While the autotrophic cycle in Thaumarchaeota is less studied, with some characteristic enzymes of the cycle still being unidentified, all enzymes of the crenarchaeal cycle were identified and biochemically characterized ( Table 1).
In former studies, the model crenarchaeon Metallosphaera sedula was used; it is capable of growing autotrophically with hydrogen, sulfidic ores or S 0 under aerobic conditions. Furthermore, heterotrophic growth with, e.g., peptone or yeast extract and mixotrophic growth in the presence of organic compounds and inorganic electron donors is possible. The transcriptomic studies of M. sedula revealed that most of the FIGURE 1 | 3-Hydroxypropionate/4-hydroxybutyrate cycle in Sulfolobales (Berg et al., 2007). For the enzymes, see Table 1. enzymes of the cycle were up-regulated under mixotrophic and autotrophic growth conditions (Auernik and Kelly, 2010;Hawkins et al., 2013a;Ai et al., 2019). Nevertheless, there were several notable exceptions, as 3-hydroxypropionyl-CoA dehydratase, 4-hydroxybutyryl-CoA synthetase and bifunctional crotonyl-CoA hydratase/3-hydroxybutyryl-CoA dehydrogenase were not regulated, or even down-regulated in the presence of inorganic electron donors ( Table 1; Auernik and Kelly, 2010;Hawkins et al., 2013a;Ai et al., 2019). Two of these enzymes, 4hydroxybutyryl-CoA synthetase and bifunctional crotonyl-CoA hydratase/3-hydroxybutyryl-CoA dehydrogenase, stand out in comparative analysis of specific activities of the purified enzymes with the corresponding activities in cell extracts of M. sedula. This analysis allows to estimate the amount of the individual enzyme proteins required to account for autotrophic growth, based on the specific activities of the purified enzymes and the measured specific enzyme activities in cell extract. This estimate led to the surprising high number of 58% of total cellular protein required to operate the HP/HB cycle ( Table 1). 4-Hydroxybutyryl-CoA synthetase and crotonyl-CoA hydratase/3-hydroxybutyryl-CoA dehydrogenase alone were responsible for more than half of this amount (12.5 and 19.7%, respectively). Whereas 4hydroxybutyryl-CoA dehydratase appears to constitute a large fraction of cellular protein as well, it slowly catalyzes a mechanistically difficult radical reaction (Martins et al., 2004;Buckel, 2019) and is indeed present in high amounts in other organisms (Scherf and Buckel, 1993). Furthermore, the posttranscriptional inactivation of AMP-producing organic acid CoA-ligases by acetylation has been frequently reported and may result in down-estimation of the activity of recombinant 4-hydroxybutyryl-CoA synthetase (Crosby et al., 2010;Ramos-Vera et al., 2011). However, crotonyl-CoA hydratase catalyzes a relatively simple reaction, and its postulated high abundance in the cell is surprising.
Here we studied enzymes catalyzing the conversion of crotonyl-CoA to acetoacetyl-CoA in M. sedula. We found that this conversion might be catalyzed by a number of isoenzymes with different kinetic properties, and not by a single bifunctional enzyme, as was previously thought.

Preparation of Cell Extracts
Frozen cells were suspended in a double volume of 20 mM Tris-HCl (pH 7.8) containing 0.1 mg ml −1 DNase I. The cell suspensions were passed twice through a chilled French pressure cell at 137 MPa, and the cell lysates were centrifuged for 1 h (100,000 × g; 4 • C). The supernatant (cell extract) was used for protein purification or enzyme assays immediately.

Cloning of M. sedula Genes in E. coli
Standard protocols were used for purification, cloning, transformation, and amplification of DNA (Ausubel et al., 1987). Plasmid DNA was isolated with the Monarch Plasmid Miniprep Kit (NEB). Primers and restriction enzymes used for the cloning of M. sedula genes are listed in Supplementary  Table S1. The genes were amplified using Q5 polymerase (NEB). PCR conditions were as follows: 35 cycles of 10-s denaturation at 98 • C, 30-s primer annealing at 60 • C and 30-s elongation at 72 • C. The PCR products were treated with the corresponding restrictases (Supplementary Table S1), and the genes were separately ligated into the expression vector pET16b containing a sequence encoding an N-terminal His 10 -tag. The plasmids were transformed into E. coli DH5α for amplification, followed by their purification and sequencing.

Heterologous Expression of M. sedula
Genes in E. coli The amplified expression vectors were used to transform E. coli Rosetta 2 (DE3). The cells were grown at 37 • C in LB medium with ampicillin and chloramphenicol. Expression was induced at an optical density (OD 578 nm ) of 0.6 with 1 mM isopropyl-ß-Dthiogalactopyranoside (IPTG), and the temperature was lowered to 20 • C. The cells were harvested after additional growth for 4 h and stored at −20 • C until use.

Purification of Recombinant Proteins
Recombinant crotonyl-CoA hydratase/(S)-3-hydroxybutyryl-CoA dehydrogenase Msed_0399 was produced untagged. The protein was purified as described previously (Ramos-Vera et al., 2011). The heterologously produced His 10 -tagged (S)-3-hydroxybutyryl-CoA dehydrogenases Msed_1423 and Msed_0389 as well as crotonyl-CoA hydratases Msed_0336 and Msed_0384 were purified using affinity chromatography. The corresponding cell extracts were applied at a flow rate of Frontiers in Microbiology | www.frontiersin.org 0.5 ml min −1 to a 1-ml Protino Ni-NTA column (Macherey-Nagel) that had been equilibrated with 20 mM Tris-HCl containing 150 mM KCl (pH 7.8 for Msed_1423 and Msed_0389, pH 7.6 for Msed_0384, and pH 8.0 for Msed_0336). The column was washed with the same buffer containing 50 mM imidazole at a flow rate of 0.5 ml min −1 to elute unwanted protein.
The recombinant enzymes were eluted with the same buffer containing 500 mM imidazole. The enzymes were concentrated using 10K Vivaspin Turbo 4 (Sartorius, Göttingen, Germany) and stored at −20 • C with 50% glycerol. The identity of the purified recombinant proteins was confirmed using gel digestion by trypsin followed by LC-MS/MS (Supplementary Table S2).

Enzyme Assays
The 3-hydroxybutyryl-CoA dehydratase and 3hydroxypropionyl-CoA dehydratase activities were measured at 42 • C. All other enzyme activities were measured at 65 • C.

Other Methods
Apparent K m and V max values were calculated using GraphPad Prism5 software. Protein concentration was determined using the Bradford method (Bradford, 1976) with bovine serum albumin as a standard. DNA sequence determination was performed by Eurofins (Ebersberg, Germany). Sodium dodecyl sulfatepolyacrylamide gel electrophoresis (SDS-PAGE) (12.5%) was performed as described by Laemmli (1970). An unstained protein MW marker (Thermo Scientific, 14.4-116 kDa) and a prestained protein marker (Cell Signaling, were used as molecular mass standard. Proteins were visualized by Coomassie blue staining (Zehr et al., 1989). Protein identification was performed at the IZKF Core Unit Proteomics Münster based on tryptic in-gel digestion and mass spectrometric analysis using Synapt G2 Si coupled to M-Class (Waters Corp.). Query sequences for the database searches were obtained from the National Center for Biotechnology Information (NCBI) data base. The BLAST searches were performed via NCBI BLAST server 1 (Altschul et al., 1990).

Catalytic Properties of Bifunctional Crotonyl-CoA Hydratase/(S)-3-Hydroxybutyryl-CoA Dehydrogenase Msed_0399
Msed_0399 is a bifunctional crotonyl-CoA hydratase/(S)-3hydroxybutyryl-CoA dehydrogenase consisting of two domains, a C-terminal enoyl-CoA hydratase domain (∼20 kDa) and an N-terminal dehydrogenase domain (∼40 kDa) (Ramos-Vera et al., 2011). This enzyme is widely distributed among Archaea and has been predicted to be present in the common ancestor of the TACK superphylum, i.e., it was already present in the common ancestor of Cren-and Thaumarchaeota (Williams et al., 2017). The enzyme was purified formerly from autotrophically grown M. sedula cells, following the reduction of NAD with crotonyl-CoA, i.e., the presence of both crotonase and dehydrogenase reactions simultaneously (Ramos-Vera et al., 2011). The identified msed_0399 was cloned and heterologously produced, and the recombinant enzyme was characterized (Ramos-Vera et al., 2011;Hawkins et al., 2014). However, the characterization was preliminary, as the activity was measured only as crotonyl-CoA or (S)-3-hydroxybutyryl-CoAdependent NAD reduction (Ramos-Vera et al., 2011;Hawkins et al., 2014). The measured hydratase activity in the coupled assay was necessarily limited by the level of the much lower dehydrogenase activity.
In order to fill this gap, we heterologously produced and purified the bifunctional protein Msed_0399, and characterized the enzyme by measuring dehydrogenase and crotonase reactions separately. In hydratase reaction, Msed_0399 was promiscuous and could hydrate not only crotonyl-CoA, but also (E)-2octenoyl-CoA and acrylyl-CoA (Table 2). Similarly, it acted 1 http://www.ncbi.nlm.nih.gov/BLAST/ as dehydrogenase with both C 4 and C 8 (S)-3-hydroxyacyl-CoAs (Table 2). Interestingly, the enzyme was most active with octenoyl-CoA, thus suggesting that Msed_0399 may participate in β-oxidation, in addition to autotrophic CO 2 fixation. Importantly, Msed_0399 is the only 3-hydroxyacyl-CoA dehydrogenase that was active with C 8 (3-hydroxy)acyl-CoAs (Tables 2, 3), further suggesting its participation in heterotrophic metabolism.

Crotonyl-CoA Hydratase and (S)-3-Hydroxybutyryl-CoA Dehydrogenase Activities in M. sedula Cell Extract
The apparent K m value of Msed_0399 for crotonyl-CoA in the crotonase/(S)-3-hydroxybutyryl-CoA dehydrogenase reaction was four times higher than the corresponding value in M. sedula cell extract (Table 2) suggesting the presence of additional crotonase(s). Furthermore, the apparent K m value of Msed_0399 for (S)-3-hydroxybutyryl-CoA in the dehydrogenase reaction was three times higher than the corresponding value in M. sedula cell extract ( Table 2). These data suggest that Msed_0399 may not be the only enzyme responsible for crotonyl-CoA hydratase and (S)-3-hydroxybutyryl-CoA dehydrogenase activity in autotrophically grown M. sedula. Therefore, we cloned and expressed other candidate genes for these enzymes (Table 1) in E. coli and characterized the catalytic activities of the resulting enzymes.

3-Hydroxybutyryl-CoA Dehydrogenases Msed_0389 and Msed_1423
The purified recombinant Msed_0389 and Msed_1423 catalyzed the NAD-dependent oxidation of (S)-3-hydroxybutyryl-CoA to acetoacetyl-CoA and were highly specific for their substrate (K m 2.6 and 5 µM, respectively; Table 3). Both enzymes were not active with (S)-3-hydroxyoctanoyl-CoA. Their k cat /K m values were 3-and 20-fold higher than the corresponding value of Msed_0399 in (S)-3-hydroxybutyryl-CoA dehydrogenase reaction (Table 3). Interestingly, both these genes were shown to be up-regulated in auto-and mixotrophically grown cell (see Table 1; Auernik and Kelly, 2010;Ai et al., 2019). The Msed_0389 gene clustered together with various β-oxidation genes in the M. sedula genome. In contrast, msed_1423 clustered with several genes encoding specific enzymes of the HP/HB cycle, suggesting its involvement in autotrophic CO 2 fixation (Figure 2).

Crotonyl-CoA Hydratases Msed_0336 and Msed_0384
The purified recombinant Msed_0336 and Msed_0384 catalyzed the hydration of crotonyl-CoA to (S)-3-hydroxybutyryl-CoA (Table 4); the k cat /K m values of both enzymes for the reverse reaction were ∼2.5 times higher. Both enzymes were also active with (E)-2-octenoyl-CoA and, to a lesser extent, with acrylyl-CoA and methacrylyl-CoA. The V max values of these enzymes for crotonyl-CoA hydration were lower than the V max of Msed_0399; yet their apparent crotonyl-CoA K m values were also much lower, resulting in a catalytic constant k cat /K m similar to that  The V max values are normalized to 75 • C based on the assumption that a 10 • C rise in temperature doubles the reaction rate. No activity was detected with (S)-3hydroxyoctanoyl-CoA (with 0.5 mM NAD; detection limit ≤0.2 U mg −1 for Msed_1423 and ≤0.1 U mg −1 for Msed_0389). a k cat was calculated in s −1 mM −1 for the activities at 75 • C.
of Msed_0399. The clustering of msed_0384 with β-oxidation genes conforms with its higher k cat /K m value for octenoyl-CoA. Nevertheless, it is Msed_0399 which has the highest k cat /K m value for this substrate. In the published transcriptomic analyses, Msed_0384 was shown to be down-regulated in autoand mixotrophically grown cell, while Msed_0336 showed no significant regulation in these cells (see Table 1; Auernik and Kelly, 2010;Ai et al., 2019).

Distribution of Crotonase and (S)-3-Hydroxybutyryl-CoA Dehydrogenase Genes in Sulfolobales
The genes of the HP/HB cycle catalyzing reactions from acetyl-CoA to crotonyl-CoA are present in all fully sequenced Sulfolobales genomes (Supplementary Table S3), indicating that they have the potential to grow autotrophically. The only exception was Sulfodiicoccus acidiphilus that does not possess most of the key genes of the HP/HB cycle. This species is only distantly related to other Sulfolobales and is not capable to grow autotrophically Kurosawa, 2017, 2019). On the contrary, most of the other Sulfolobales are autotrophs, with the exception of Sulfolobus acidocaldarius and Saccharolobus solfataricus that nevertheless encode all key genes of the HP/HB cycle in their genomes (Supplementary Table S3). Homolog(s) of bifunctional crotonyl-CoA hydratase/(S)-3hydroxybutyryl-CoA dehydrogenase Msed_0399 genes were found in 75% of the genomes, whereas four autotrophic The V max values are normalized to 75 • C based on the assumption that a 10 • C rise in temperature doubles the reaction rate. NA, not applicable. a k cat was calculated in s −1 mM −1 for the activities at 75 • C. b No activity was detected with (R)-3-hydroxybutyryl-CoA (detection limit ≤0.3 U mg −1 Msed_0336, ≤1 U mg −1 Msed_0384). c Activity was measured at 42 • C. d Specific activity.
species (Acidianus ambivalens, Acidianus hospitalis, Acidianus sulfidivorans, and Stygiolobus azoricus) do not possess the corresponding gene (Table 5). Notably, the functioning of the HP/HB cycle was experimentally shown for S. azoricus (Berg et al., 2010b), indicating that the presence of the bifunctional protein is not essential for autotrophic CO 2 fixation. The only two genes that are (almost) universally distributed in Sulfolobales are the crotonase Msed_0566 gene and the dehydrogenase Msed_1423 gene (Table 5 and  Supplementary Table S3). They are absent only in the genome of S. acidocaldarius.

DISCUSSION
Genome analysis shows that M. sedula and most other Sulfolobales have multiple homologs of genes responsible for the conversion of crotonyl-CoA into acetyl-CoA, and our data indeed suggest that crotonyl-CoA hydratase and (S)-3-hydroxybutyryl-CoA dehydrogenase reactions may be catalyzed by several proteins in vivo. It is probably also true for the β-ketothiolase reaction cleaving acetoacetyl-CoA into two molecules of acetyl-CoA, as at least one of the homologous genes (msed_0386) is up-regulated under mixo-and autotrophic growth conditions, in addition to the characterized β-ketothiolase Msed_0656 gene ( Table 1; Auernik and Kelly, 2010;Hawkins et al., 2013a;Ai et al., 2019). The studied crotonases are homologous to 3-hydroxypropionyl-CoA dehydratase Msed_2001 and possess low 3-hydroxypropionyl-CoA dehydratase activity (Table 2), thus probably also contributing to 3-hydroxypropionyl-CoA dehydratase activity in the cells. 3-Hydroxypropionyl-CoA dehydratase, in turn, has crotonase activity (Teufel et al., 2009). Although analysis of deletion mutants would be very helpful to clarify the role of individual proteins in autotrophic CO 2 fixation, genetic system is unfortunately not available for autotrophic Sulfolobales.
In contrast to the current understanding, our data show that the bifunctional crotonyl-CoA hydratase/(S)-3-hydroxybutyryl-CoA dehydrogenase is not the main enzyme that catalyzes these reactions in M. sedula. This is corroborated by the fact that its gene is absent in many other autotrophic Sulfolobales. In fact, msed_0399 is located in the cluster with the genes homologous to various β-oxidation genes (Figure 2A) and may thus be primarily involved in, e.g., fatty acid metabolism rather than CO 2 fixation ( Table 2). Although the growth of M. sedula on fatty acids has not been shown, some Sulfolobales are capable to use these compounds as a sole carbon source . Furthermore, reversibility of β-oxidation enzymes in Archaea and their involvement in fatty acid synthesis was proposed (Dibrova et al., 2014).
Interestingly and in contrast to Sulfolobales, the homologous bifunctional protein is the only enzyme capable to covert crotonyl-CoA into acetoacetyl-CoA in autotrophic Desulfurococcales (Ignicoccus hospitalis) and Thermoproteales (Pyrobaculum neutrophilus) that use the dicarboxylate/4hydroxybutyrate cycle for CO 2 fixation (Huber et al., 2008;Ramos-Vera et al., 2009. This cycle differs from the HP/HB cycle in the carboxylation reactions (ferredoxindependent pyruvate synthase and phosphoenolpyruvate carboxylase) but shares the regeneration part, i.e., the conversion of succinyl-CoA into acetyl-CoA that includes crotonase and (S)-3-hydroxybutyryl-CoA dehydrogenase reactions. The anaerobic dicarboxylate/4-hydroxybutyrate cycle is probably the ancestral CO 2 fixation pathway in Crenarchaeota, and the presence of the bifunctional protein widely distributed among Archaea seems to be an ancestral feature for the pathway.
Nevertheless, even if several enzymes catalyze one and the same reactions in M. sedula, our biochemical data (Table 3) and the results of our gene distribution analysis (Table 5) together with the published transcriptomic data (see Table 1) show that Msed_1423 is the main (S)-3-hydroxybutyryl-CoA dehydrogenase in the HP/HB cycle in Sulfolobales. This protein is clustered together with two specific enzymes of the HP/HB cycle, succinic semialdehyde reductase Msed_1424 and acryloyl-CoA reductase Msed_1426 (Figure 2B), further supporting its involvement in autotrophic CO 2 fixation. The only species that does not possess the corresponding gene, S. acidocaldarius, is not capable to grow autotrophically (Zeldes et al., 2019). The crotonase homolog, Msed_0566, was the only monofunctional crotonase (besides the promiscuous 3hydroxypropionyl-CoA dehydratase) that was present in all Sulfolobales except for the heterotrophic S. acidocaldarius (Table 5). Unfortunately, all our attempts to heterologously express msed_0566 (as well as msed_0385) were not successful and resulted in insoluble protein. It is to be shown whether this protein is the primary crotonyl-CoA hydratase in M. sedula and especially in S. azoricus, A. ambivalens, and A. hospitalis (Table 5), or this reaction is also catalyzed by a promiscuous 3-hydroxypropionyl-CoA dehydratase, as was shown for the HP/HB cycle in Thaumarchaeota (Könneke et al., 2014).
The inability of S. acidocaldarius to grow autotrophically is well-known (Berg et al., 2010b;Zeldes et al., 2019). It has recently been suggested that the obligate heterotrophy of this organism is rooted rather in gene regulation than in the biochemical capacity (Zeldes et al., 2019). Our data suggest that this inability is likely due to the absence of the Msed_1423 (and possibly Msed_0566) homologs in this species.
Comparison of the activities of the characterized (S)-3hydroxybutyryl-CoA dehydrogenase Msed_1423 and crotonase Msed_0384 with the activities of these reactions in cell extract of M. sedula leads to the values 1.7 and 7.1% of total cellular protein, respectively. The last value was calculated based on the activity of crotonase found in our work ( Table 2), which is twice higher than the previously measured activity (Berg et al., 2007; see Table 1). With these corrected values, the estimated abundance of the HP/HB cycle enzymes is ∼47% of total cellular protein, i.e., 11% less than was calculated in Table 1. This value may further be corrected (reduced?) after the characterization of the further crotonase homologs in M. sedula.
The studied enzymes were promiscuous, differing in their kinetic properties. Their simultaneous presence not only increases the activity of particular reactions, but also allows fast and efficient fine-tuning of metabolism and adaptation to different growth conditions, varying metabolite concentrations and growth rates. This strategy may be especially advantageous for metabolically versatile organisms that are not limited in energy source.
The HP/HB cycle is a perspective route for the conversion of CO 2 into liquid fuels and industrial chemicals and can be engineered for the production of value-added compounds like 3-hydroxypropionate or n-butanol (Hawkins et al., 2011(Hawkins et al., , 2013bKeller et al., 2013;Fuchs and Berg, 2014;Loder et al., 2016). The identification of highly specific thermophilic enzymes involved in this pathway expands opportunities for the application of the pathway. Simultaneous usage of several isoenzymes differing in their kinetic properties is an interesting concept for biotechnology that may allow stable production under varying conditions.

DATA AVAILABILITY STATEMENT
The datasets generated for this study are available on request to the corresponding author.

AUTHOR CONTRIBUTIONS
IB designed the experiments. LL and HH performed the experiments. LL, HH, and IB analyzed the data. LL and IB wrote the manuscript. All authors read and approved the final manuscript.

FUNDING
This work was funded by the Deutsche Forschungsgemeinschaft (BE 4822/5-1) and China Scholarship Council (201504910718 to LL).