Emergence and genomic diversification of a virulent serogroup W:ST-2881(CC175) Neisseria meningitidis clone in the African meningitis belt

Countries of the African ‘meningitis belt’ are susceptible to meningococcal meningitis outbreaks. While in the past major epidemics have been primarily caused by serogroup A meningococci, W strains are currently responsible for most of the cases. After an epidemic in Mecca in 2000, W:ST-11 strains have caused many outbreaks worldwide. An unrelated W:ST-2881 clone was described for the first time in 2002, with the first meningitis cases caused by these bacteria reported in 2003. Here we describe results of a comparative whole-genome analysis of 74 W:ST-2881 strains isolated within the framework of two longitudinal colonization and disease studies conducted in Ghana and Burkina Faso. Genomic data indicate that the W:ST-2881 clone has emerged from Y:ST-175(CC175) bacteria by capsule switching. The circulating W:ST-2881 populations were composed of a variety of closely related but distinct genomic variants with no systematic differences between colonization and disease isolates. Two distinct and geographically clustered phylogenetic clonal variants were identified in Burkina Faso and a third in Ghana. On the basis of the presence or absence of 17 recombination fragments, the Ghanaian variant could be differentiated into five clusters. All 25 Ghanaian disease isolates clustered together with 23 out of 40 Ghanaian isolates associated with carriage within one cluster, indicating that W:ST-2881 clusters differ in virulence. More than half of the genes affected by horizontal gene transfer encoded proteins of the ‘cell envelope’ and the ‘transport/binding protein’ categories, which indicates that exchange of non-capsular antigens plays an important role in immune evasion.


INTRODUCTION
Neisseria meningitidis is an obligate commensal, which has adapted to colonize the upper respiratory tract of humans as its only ecological niche. Transmission requires close contact with a colonized person [1]. In colonized individuals meningococci have the potential to cross the epithelial barrier and to enter the bloodstream, where strains belonging to a limited number of hypervirulent lineages may multiply and cause invasive disease, such as sepsis and meningitis [2]. Factors that predispose individuals to invasive meningococcal disease include the lack of bactericidal antibodies and certain complement deficiencies [3]. The polysaccharide capsule covering the cell surface is the most important meningococcal virulence factor and the target for polysaccharide vaccines [3]. Of the 12 serogroups that can be distinguished based on the chemical composition of the capsule, serogroups A, B, C, W, X and Y are responsible for almost all cases of invasive meningococcal disease worldwide. Apart from the capsule, a large number of other genomic traits seem to influence fitness of meningococcal lineages to colonize a human population and to cause invasive disease. This includes proteins involved in adhesion to epithelial cells of the nasopharynx, such as the PilE and PilV components of Type IV pili, the opacity proteins, Opa, and Opc, NadA (Neisseria adhesin A), NhhA (Neisseria hia/hsf homologue) and App (adhesion and penetration protein) [4].
The genomic diversity of N. meningitidis is extensive and can be primarily attributed to the natural competence of these bacteria for transformation and homologous recombination. Co-colonization of meningococci with other meningococci and other commensal species of the genus Neisseria are the primary source of DNA for horizontal gene transfer (HGT) which leads to shared alleles among genetically distant strains. While HGT results in a highly dynamic population structure, isolates can still be readily grouped into clonal complexes (CC) by the standard seven-gene multilocus sequence typing (MLST) [5]. Sequence types (STs) are grouped into CC by their similarity to a central allelic profile. In the Neisseria database a CC includes all STs that match the central genotype at four or more loci. Additional sequence analysis of highly variable genes, such as porA, porB, fetA and fhbp, which encode proteins under positive selective pressure [6], and, in particular, comparative whole-genome sequence analysis, can reveal genomic diversity among outbreak strain isolates. An epidemic clonality model has been proposed, which postulates that meningococcal populations undergo occasional rounds of clonal propagation in an otherwise recombining population structure [7]. More recently it has been argued that a predominantly clonal evolution model fits best to the population genetics of N. meningitidis [8].
Most meningococcal invasive disease is caused by a limited number of hyperinvasive CC. However, not all strains belonging to a hyperinvasive CC have a similar phenotype and virulence, since the virulence gene content and antigenic makeup may be modified by HGT [9,10]. Due to the high frequency of recombination, clonal meningococcal populations may differ in recombination blocks acquired through HGT (mostly from other meningococci) and, to a lesser extent, by single-nucleotide polymorphisms (SNPs) caused by point mutations even during rapid spread in the course of an epidemic [10]. Out of the immensely diverse repertoire of genomic variants present in a spreading population, new highly successful 'fit genotypes' [11] or immune escape variants [10] may occasionally emerge. The emerging 'original clone', may rapidly diversify into distinct and recognisable clonal variants during its geographical spread [11].
While incidence rates of invasive meningococcal disease of 0.5-10 cases per 100 000 and per year have been reported for many regions worldwide, rates may be as high as 100-1000 per 100 000 during periodic epidemics occurring in the 'meningitis belt' of sub-Saharan Africa [12]. Historically, most of the large epidemics have been caused by serogroup A meningococci. Therefore, vaccination of the population of the meningitis belt with the meningococcal serogroup A capsule glycoconjugate vaccine MenAfriVac was initiated in 2010 [13]. Currently, meningococcal strains expressing the W, but also the C or X polysaccharide, represent the main aetiological agents of bacterial meningitis outbreaks in the meningitis belt [3].
After the first description of N. meningitidis serogroup W in 1968 [14]  W meningococci have been identified in Northern Europe and the USA [15]. In 2000, a first large outbreak caused by W meningococci was reported among Hajj pilgrims in Mecca [16]. Since then epidemics caused by W:ST-11 (CC11) meningococci have occurred in the African meningitis belt and outbreaks have been reported worldwide. Results of whole-genome sequencing analyses have indicated that all W:ST-11 isolates were likely to be descendants of a single ancestral strain [17]. Hyperinvasive meningococcal isolates belonging to CC11 express C, W, or less frequently B or Y capsule polysaccharide. The genomes of CC11 strains form a single main lineage comprising two sublineages, 11.1 and 11.2 [18]. Serogroup W:ST-11 isolates are confined within sublineage 11.1, forming several clusters in addition to the 'original' Hajj outbreak clone. It is assumed that the W:ST-11 clone has emerged in recent times through capsule switching. Results of analyses of the capsule synthesis gene cluster indicate that the capsular switch involved two separate recombination events and that the vast majority of W CC11 strains have retained both recombination blocks [19]. The Hajj clone has undergone additional recombination events involving the acquisition of virulence genes. Currently, appreciable amounts of invasive meningococcal disease cases worldwide are caused by W CC11 strains which are distinct from the Hajj clone. The distinct genotypes can only be identified by whole-genome sequencing analysis, and appear to translate into distinct phenotypes which can exhibit different clinical manifestations, such as a higher case fatality rate of infections caused by the 'South American/UK' W CC11 strains, in comparison to the original Hajj clone [18,19].
A highly virulent W meningococcal lineage unrelated to the CC11 meningococci and associated with ST-2881 belonging to CC175 [20] was first described in Niger, where it emerged in 2002 and caused in 2003 sporadic meningitis cases in Niamey [21,22] and a cluster of cases in the Illela district [23]. In a carriage study in the Illela district, ST-2881 strains with different capsule polysaccharides (W and Y), but the same pulsed-field gel electrophoresis pattern and (NT):P1.5,2 phenotype were isolated, indicating that the W: ST-2881 clone emerged from Y:ST-2881(CC175) bacteria by capsule switching [23]. In 2003 W:ST-2881 strains were also isolated in Benin, Nigeria [22] and Burkina Faso [24]. While W:ST-2881 meningococci represented in 2007 and 2008 the vast majority of meningococcal meningitis case isolates in Cameroon, since 2009 W:ST-11(CC11) strains have predominated [25]. Within the framework of two longitudinal colonization and disease studies [26,27]

Study sites
The serogroup W N. meningitidis isolates investigated in this study have been collected in the KND of Ghana and in the NHD of Burkina Faso. Case strains were isolated from the cerebrospinal fluid (CSF) of meningitis patients, and carriage strains were isolated from throat swabs collected during longitudinal carriage surveys. Isolation and characterization of bacterial isolates has been described previously [26,28,29].
The NHD lies in the Kossi region of Burkina Faso. The Nouna Health and Demographic Surveillance System (Nou-naDSS) covers a part of the NHD with a current population size of about 99 000 people. The NounaDSS is distributed into 58 villages and Nouna city. The population of the district is mainly rural, excluding the inhabitants of Nouna city [27]. The NHD has dry savannah vegetation with a long dry season lasting from October to May and a short rainy season during the rest of the year. For a longitudinal meningococcal colonisation study with a sample size of about 300 participants, 37 compounds were selected by proportional cluster sampling from the NounaDSS population. A second longitudinal study was initiated in a village located close to Ira outside the NounaDSS area after a focal serogroup A meningococcal meningitis outbreak in 2006 [29]. The KND lies in the Upper-East Region of Ghana within the Guinea savannah woodland and has two major seasons: a short wet season from June to October and a long dry season for the rest of the year. The district population is about 140 000, mainly rural, except for the 20 000 inhabitants of Navrongo town. In the KND, people live in compounds with an average of ten inhabitants. For the colonization study with a sample size of about 300 participants, 37 residential compounds from the complete listing of the district population were selected using the Navrongo Demographic Surveillance System (NavrongoDSS) [26]. In all three colonization studies throat swabs were taken twice annually from all consenting members of the selected compounds present at the time of the visit. All isolates included in the whole genome sequencing study (Table S1) came from different persons. Preparation and sequencing of genomic DNA N. meningitidis strains (Table S1) were grown in liquid Brain-Heart Infusion (Bacto) medium and chromosomal DNA was prepared as described previously [30,31]. Briefly, bacterial pellets were resuspended in 0.5 ml TES buffer [50 mM Tris, 20 mM EDTA, 50 mM NaCl (pH 8)]. A 2 µl volume of RNase A (Qiagen 100 mg ml À1 ) and 20 % SDS were added to a final concentration of 1 % and cells were lysed for 5 min at 42 C. After two phenol/chloroform/ isoamyl alcohol (25 : 24 : 1, Sigma) extractions and one chloroform/isoamyl alcohol (24 : 1, Sigma) extraction, DNA was precipitated in two volumes of isopropanol followed by suspension in TE buffer [10 mM Tris, 1 mM EDTA (pH 8)]. Ammonium acetate was added to a final concentration of 2.5 M and the DNA was precipitated in two volumes of ethanol. After two 70 % ethanol wash steps and subsequent drying, the DNA was diluted in TE buffer [10 mM Tris, 0.1 mM EDTA (pH 8)].

Ethical clearance
Multiplexed genomic DNA libraries were prepared with an insert size of 200 using 24 unique index tags. Libraries were combined into pools of 24 and sequenced on an Illumina HiSeq for 75 cycles from each end to produce paired-end reads plus an eight-base index sequence read. Downstream analysis used the index tags to assign reads to individual samples.

Read alignment and SNP detection
To provide a high-quality reference sequence, the genome of the W:ST-2881 strain 2039 (Table S1) was sequenced using single-molecule, real-time (SMRT) technology on a PacBio RS II (Pacific Biosciences) with the P5-C3 chemistry. One SMRT cell was used with a movie length of 180 min. Strain 2039 was isolated in 2005 in Burkina Faso and was selected as a reference strain because it was the oldest isolate among our W ST-2881 strain collection. The genome was assembled de novo using the hierarchical genome assembler (HGAP in SMRTAnalysis v2.3.0) [32]. The assembly contained two contigs with a large overlap, which were manually joined and trimmed to remove end overlaps caused by the circularity of the chromosome. The sequence was then corrected using Illumina sequence data for the same isolate using RATT [33] and annotated using PROKKA v. 1.11 [34]. Sequence data has been deposited in the ENA under study accession number ERP002590 and sample accession number ERS323054. Comparison and visualization of the strain 2039 genome (Fig. S1) was carried out using BRIG (http://sourceforge.net/projects/brig/) [35].
Variation in the form of single-nucleotide polymorphisms (SNPs) among the sequenced W:ST-2881 isolates (Table S1) was detected using a mapping approach. The paired-end Illumina reads were mapped against the N. meningitidis serogroup W:ST-2881 strain 2039 as reference (accession numbers ERR369484) with an insert size of between 50 and 400 bp using SMALT (www.sanger.ac.uk/resources/software/smalt/). SNPs were identified using SAMtools [36] as previously described [37]. SNPs called in phage sequences and repetitive regions of the N. meningitidis reference genome were excluded. Repetitive regions were defined as exact repetitive sequences of !50 bp in length, identified using rep repeat-match. If 10 % of the genomes had an indetermination in a called SNP, these positions were removed from the analysis.

Identification of recombination events
Ancestral sequences were reconstructed onto each node of the phylogeny using PAML [38]. From these ancestral sequences, SNPs were reconstructed onto branches of the tree. To identify recombination events, we applied the moving window approach as previously described [39]. The Gubbins (Genealogies Unbiased By recomBinations In Nucleotide Sequences) approach is used to identify recombination blocks using an algorithm that iteratively identifies loci containing elevated densities of base substitutions while concurrently reconstructing a phylogeny based on the putative point mutations outside of these regions. Sequences of all strains with possible recombination fragments were assembled using velvet v1.0.12 [40] de novo assemblies with contigs realigned by Abacas [41]. Draft genomes were used to query the 2039 genome using BLASTN, and comparison files were generated and viewed using the Artemis Comparison Tool (ACT) [42]. In order to confirm the recombination fragments and to identify the most closely related alleles, the predicted recombination fragments and their upstream and downstream flanking regions were extracted from the assembled genomes and aligned with their homologues from 15 publicly available genomes of strains of N. meningi- TCDC-NG08107 (PRJNA61377) and TCDC-NG08107 (CP002440.1)] and one N. lactamica [020-06, (FN995097.1)] genomes. The alignments were used to reconstruct maximum-likelihood phylogenetic trees with RAxML v7.0.4 [43] using a general time-reversible (GTR) substitution model with g correction for among-site rate variation. Support for nodes on the trees was assessed using 100 bootstrap replicates. We considered a recombination fragment confirmed when the cluster differed from the clusters belonging to the upstream and downstream flanking regions. Recombination hotspots were defined as regions in a genome that exhibit elevated rates of recombination, relative to a neutral expectation [44,45].

Phylogenetic analyses
A maximum-likelihood phylogenetic tree was generated from the whole-genome sequence alignment without SNPs associated with recombination events. Phylogenetic trees were reconstructed as described above. and the ST-8447 strain 9281, the strains were compared by using the Genome Comparator tool located in the pubMLST database (https://pubmlst.org/neisseria) with the 'All loci' option [46]. We calculated a distance matrix between the strains based on the number of loci that differed between each pair of isolates. The Genome Comparator tool conducts a gene-by-gene comparison of isolates. A distance matrix is generated as the result of allelic variation the isolates exhibit at the different loci included in the comparison. The NeighbourNet algorithm [47] resolves the relationship among the isolates into a network of relatedness, visualized in SplitsTree 4 (version 4) [48]. Alleles of the MLST genes and of the fetA, porB, Nhba, opcA and fHbp genes were determined with the tool 'Sequence query -Neisseria profile/sequence definitions' (https://pubmlst.org/ neisseria). Whole-genome comparison of W:ST-2881 strains with other meningococci Using the tool Genome Comparator located in the pubMLST database (pubmlst.org/neisseria), we identified N. meningitidis strains with available genome data that were most closely related to the W:ST-2881 isolates. Using the genome sequence of the W:ST-2881 strain 2039 from Burkina Faso, for which the genome was analysed using the PacBio sequencing platform, as a reference and published N. meningitidis genome sequencing data, we identified strains belonging to CC167 and CC175 (Table 1) as being most closely related. Based on the number of loci that differed between each pair of isolates (allelic differences) a distance matrix was calculated using the Genome Comparator tool, which represents the equivalent of a whole-genome MLST profile. A NeighborNet graph generated from the distance matrix (Fig. 2) showed that the W:ST-2881 (CC175) strain 2007461,which was isolated in 2007 in Togo (Table 1), was the most closely related isolate with published genome sequencing data. However, the W:ST-2881 strains from Ghana and from Burkina Faso were more closely related to each other than to this W:ST-2881 strain from Togo (Fig. S2). The most closely related CC was CC175 comprising a cluster of Y and C isolates from South Africa and another cluster of non-groupable ST-175 strains from Europe (Fig. 2). Y and W strains belonging to CC167 were less closely related to the W:ST-2881 strains than CC175 (Fig. 2) . SNPs associated with recombination blocks were removed as described previously [39] and a maximum-likelihood phylogenetic tree was reconstructed using a General Time Reversible (GTR) evolutionary model based on the 145 SNPs not associated with recombination blocks. Hence, 91.1 % (1470/1615) of all called SNPs were associated with recombination blocks and only 8.9 % of them thus represent putative point mutations.

RESULTS
The generated phylogenetic tree reflects a highly structured clonal population composed of three closely related phylogenetic variants associated with the geographic origin of the strains (Fig. 3). The strains from the NHD of Burkina Faso formed two separate variants (designated A and B in Fig. 3) with variant A containing all three strains from the NDSS area in the south of the district and variant B containing all six strains from the Ira Health Centre in the north of the NHD. Variant C, formed by the 65 strains from the KND of Ghana, was divided into five clusters, designated C1-C5 (Fig. 3). There was no association of the clusters with the year of isolation, but remarkably, all 25 Ghanaian disease isolates clustered together with 23 of the Ghanaian isolates associated with carriage in cluster C1. The remaining 17 carried isolates were distributed over clusters C2-C5.
Within the analysed set of W:ST-2881 strains, the highly variable porA, porB, fetA, nba, opcA and fhbp genes, which are often used as additional molecular epidemiological markers, showed only very limited diversity ( Table 1). The isolates from Ghana and Burkina Faso harboured the same . However, diversity was found in fhbP, which is related to a recombination event involving the fhbP gene (Fig. S7). While the fhbP allele 1 present in the Togo isolate was also found both in all isolates from Burkina Faso and in all cluster C4 and C5 strains from Ghana, Ghanaian isolates belonging to the clusters C1, C2 and C3, harboured the allele 774. Among the CC167 and CC175 strains, the non-groupable ST-175 isolates from Europe (Fig. 3) had the most similar allelic profile, with four out of six identical alleles (Table 1). Only one point mutation was found in the six genes: strain 2959 has a non-synonymous SNP at position 578 (AfiC) producing an amino acid exchange (Qfi P) in the porA gene.

Analysis of the phylogenetic history of recombination blocks
To elucidate the role of recombination events in the genomic diversification of the W:ST-2881 meningococci, we investigated their recombination history. A total of 17 recombination blocks with high SNP density were predicted by the method of Croucher et al. [39] and confirmed by phylogenetic analysis (Figs S3-S18). The size of the identified recombination blocks, which are spread over the genome (Fig. S1), varied between 312 and 9323 bp, with a mean length of 3460 bp (Table 2), affecting altogether 2.7 % of the total genome length of 2 185 600 bp.
The locations of the recombination events, which were all associated with the genomic diversification within cluster C, across the phylogenetic tree, are depicted in Fig. 3. Strains belonging to clusters C1, C2 and C3 have six recombination blocks in common, and all strains of clusters C1 and C2 shared one additional recombination block. Also, all C4 and C5 strains have six recombination blocks in common; all C4 strains shared two and all C5 strains one additional recombination block. Furthermore, one block was present in only a single C5 strain (3261). In order to elucidate the putative origins of the recombination blocks, we performed an a posteriori phylogenetic analysis (Figs S3-S18). The most closely related genes were found in one case in N. gonorrhoeae, in two cases in N. lactamica and in the other 14 cases in strains of N. meningitidis ( Table 2).
The 17 recombination blocks comprised altogether 61 genes (Figs S3-S18). Functional classification of the genes affected by the recombination events revealed that genes encoding proteins of the 'cell envelope' (20/61) and the 'transport/ binding protein' (12/61) categories were over-represented, since 32.8 and 19.7 % of the affected genes belonged to these two categories, while these comprised only 16.3 and 6.3 % of all genes, respectively [49]. One genomic region, the yggS-pilT-pilU locus was associated with two independent recombination events (Fig. S13). While a 6645 bp fragment spanning the genomic region from dskA to yccS was common to all C4 and C5 strains, a smaller (2619 bp) fragment was common for all C1 and C2 strains (Fig. 3). This locus encodes the PilU and PilT proteins, which are involved in the regulation of type IV pilus expression and has been identified before as recombination hotspot in a comparative whole-genome analysis of 100 serogroup A strains [10].

DISCUSSION
The only habitat of N. meningitidis is the human nasopharynx and many meningococcal lineages have evolved that compete for colonization of this niche. In meningococcal carriage studies in Europe diverse and relatively temporally stable colonizing meningococcal populations have been found that primarily consist of strains belonging to CC that generally exhibit low virulence [50,51]. In contrast, in the African meningitis belt, clonal waves are a characteristic feature of meningococcal colonization [26] and outbreaks of invasive disease occur, when the dominating colonizing population belongs to a hypervirulent clone. The genomic basis of hyperinvasiveness of certain clones is not clear, a core pathogenome has not been identified and comparative whole-genome sequencing studies, including the present analysis, have not revealed systematic differences between clonally related disease and carriage isolates [10].
Phylogenetic studies have revealed that meningococcal populations are structured as consequence of clonal decent, HGT and point mutations. In particular HGT leads to the development of an immensely diverse repertoire of genomic variants during clonal spread and new 'fit genotypes' [11] may emerge occasionally.  [52]. While the enzymes encoded by cssA-cssC that synthesize CMP-N-acetylneuraminic acid are conserved among serogroups Y and W, the sialyltransferases encoded by the csw and csy genes, which catalyse the transfer of sialic acid to galactose or glucose, respectively [53], are serogroup-specific. Due to the similarity of the region A of the cps locus, we were not able to dissect the recombination event(s) associated with the Y to W capsule switch in the progenitor of the W:ST-2881 bacteria. CC175 strains with the Y capsular polysaccharide were described for the first time in 1999 in South Africa [54] and apart from the Y and W isolates, B and C strains belonging to this CC have also been isolated (pubmlst.org).
During the spread of a clonal epidemic, individual examples of the clone can exhibit micro variation due to HGT and point mutations. The 'original clone' can thus diversify into distinct and recognisable variants, as it spreads over time. Variants may inhabit different niches of the nasopharyngeal ecosystem, as has been indicated for pneumococcal lineages [55]. These variants may exhibit differences in genes encoding proteins, which may alter the strain's capacity to cause invasive meningococcal disease. It has been suggested that W:      [56]. In fact, in spite of their widespread occurrence in the meningitis belt, the total number of meningitis cases caused by the W:ST-2881 clone seems to be much more limited than those caused by W:ST-11 meningococci. On the other hand, substantial W:ST-2881 outbreaks have been recorded in Cameroon and in Ghana. The phylogenetic analysis presented here has revealed a highly structured clonal population with several distinct clusters. The fact that all Ghanaian disease isolates clustered with corresponding carriage isolates in one of the five variant C clusters is indicative of different virulence capacities within the W:ST-2881 population. This underscores the necessity of comprehensive and continuous strain characterization using whole-genome sequencing, to distinguish the minor genotypic differences among circulating strains which demonstrated differential capacities to cause invasive meningococcal disease. It remains to be seen whether certain W:ST-2881 variants will persist as colonizers in the population of the African meningitis belt and whether these have epidemic potential.
Based both on the analysis of point mutations and the presence of recombination blocks, the W:ST-2881 variant C, which colonized the population of the KND in 2009/ 10, could be differentiated into five clusters. Clusters C1, C2 and C3 had six recombination blocks in common and the two clusters C4 and C5 had another six blocks in common. In addition, five more recombination events were identified among the 65 variant C isolates. Similar to results describing clonal serogroup A meningococcal isolates [10], the pilT-pilU locus, regulating type IV pilus expression, was identified as a recombination hotspot. While the highly variable porA, porB, fetA, nhba and opcA loci showed no variation, the C1, C2 and C3 strains harboured a different fhbP allele from the C4 and C5 strains. The fhbP gene encodes factor H-binding protein, a key virulence factor and vaccine antigen. More than half of the genes affected by the 17 identified HGT events encoded proteins of the 'cell envelope' and the 'transport/ binding protein' categories. The accumulation of changes in cell-surface structures may be the result of selection imposed by natural immunity developing in the local host population. Genomic epidemiology studies may therefore represent an eminent approach for identifying targets of natural immune responses, with major implications for the design of next-generation protein-based subunit vaccines. Furthermore, improvement of our understanding of meningococcal population dynamics, and the basis of meningococcal virulence, will lead to the development of better vaccines and enhance our ability to recognize outbreaks and epidemics earlier, which in turn may lead to earlier and more appropriate interventions.

Funding information
The research leading to these results has received funding from the Swiss National Science Foundation (grant number 310030-152840) and from the Volkswagen Foundation.