Exploiting the Mutanome for Tumor Vaccination

The advent of next generation sequencing (NGS) technology has added a new dimension to genome research by generating ultrafast and high-throughput sequencing data in an unprecedented manner (1). The first mouse tumor genome has been published (2), and the genomes and exomes (the RNA-encoding genomic sequence) of several human primary tumors and cell lines have been dissected (3–5). As cancerogenesis is driven by mutations, the capability of NGS to provide a comprehensive map of somatic mutations in individual tumors (the “mutanome”) provides a powerful tool to better understand and intervene against cancer.

The first views into the human mutanome revealed by NGS show that human cancers carry 10s to 100s of nonsynonymous mutations. Mutations shared among patients suffering from one tumor entity allow targeted approaches, as exemplified by small-molecule inhibitors targeting the bcr-abl translocation present in 90% of chronic myelogenous leukemia patients. However, shared mutations are rare and the great majority of mutations are patient specific, which has hindered exploitation of the mutanome for the development of broadly applicable drugs.

Once mutations have been identified, major efforts are typically invested to determine the mutation functional impact, such as cancer driver versus passenger status, to form a basis for selecting suitable therapeutic targets. However, little attention has been devoted to either define the immunogenicity of these mutations or characterize the immune responses they elicit. From the immunologic perspective, mutations may be particularly potent vaccination targets as they can create neoantigens that are not subject to central immune tolerance. Whereas for other tumor antigen categories there is the likelihood of downregulation in metastases, mutations that occur in early tumor development are sustained in advanced disease (5). Indeed, several immunogenic cancer mutations have been described in mice and patients. Some mutations are capable of inducing rejection of tumors in mice (6, 7) and others are targets of spontaneously occurring dominant immune responses in patients with malignant melanoma (8). However, it is not known whether these examples represent rare cases of incidental immunogenicity or the tip of the iceberg of potentially useful vaccine targets.

We addressed this question in a systematic approach by embarking on a study to determine the proportion of nonsynonymous mutations that are able to induce a target specific immune response in the syngeneic host. The aim is to define and validate molecular alterations in the B16F10 melanoma model, the most widely used model for experimental cancer therapies, and determine their immunogenicity in C57BL/6 mice. In doing so, we completed the first mouse tumor exome and identified 962 nonsynonymous somatic point mutations of high confidence, of which 563 occur in expressed genes. We show for the first time a correlation between tumor mutations and the epitope landscape by in vivo data, demonstrating that many nonsynonymous somatic mutations in tumors are immunogenic and confer antitumoral vaccine activity.

C57BL/6 mice (Jackson Laboratories) were kept in accordance with federal and state policies on animal research at the University of Mainz. B16F10 melanoma cell line was purchased in 2010 from the American Type Culture Collection (product: ATCC CRL-6475, lot number: 58078645). Early (third and fourth) passages of cells were used for tumor experiments. Reauthentification of cells has not been done since receipt.

DNA and RNA from B16F10 cells and DNA from C57BL/6 tail tissue were extracted in triplicate. Exome capture for DNA resequencing was carried out in triplicate with the Agilent mouse whole-exome SureSelect assay (9). Using 5 μg of total RNA, barcoded mRNA-seq cDNA libraries were prepared in triplicate using a modified Illumina mRNA-seq protocol. All libraries were sequenced on an Illumina HiSeq2000 to generate 50 nucleotide reads. Sequence reads were preprocessed according to the Illumina standard protocol. RNA reads were aligned to the mm9 reference genome (10) and transcriptome using bowtie (11), and gene expression was determined by comparison with RefSeq transcript coordinates, followed by normalization to RPKM units (RPKM, Reads which map per kilobase of transcript per million mapped reads; ref. 12). DNA reads were aligned to the reference genome with bwa (13). Mutations were identified by 3 algorithms and assigned a false discovery rate (FDR) confidence value (Löwer and colleagues, submitted). Detailed methods are described in Supplementary Information.

Mutations were selected that were: (i) present in all B16F10 and absent in all C57BL/6 triplicates, (ii) FDR ≤ 0.05, (iii) homogeneous in C57BL/6, (iv) occur in a RefSeq transcript, and (v) cause nonsynonymous changes. Further mutation selection criteria were occurrence in B16F10-expressed genes (median RPKM across replicates; ref. >10) and in an MHC-binding peptide based on the Immune Epitope Database (IEDB; ref. 14). For validation, variants were amplified from DNA from B16F10 cells and C57BL/6 tail tissue, subjected to Sanger sequencing, and results visually examined (Supplementary Information). DNA-derived mutations were classified as validated if confirmed by either Sanger sequencing or the B16F10 RNA-Seq reads. Algorithms SIFT (15) and POLYPHEN-2 (16) were used to predict mutation functional impact. They predict the functional significance of an amino acid on protein function using the location of protein domains and cross-species sequence conservation. Ingenuity IPA was used to infer gene function.

All peptides including vesiculo-stomatitis virus nucleoprotein (VSV-NP_52–59) and tyrosinase-related protein 2 (Trp2_180–188) were purchased from Jerini Peptide Technologies. Synthetic peptides were 27 amino acids long with the mutated (MUT) or wild-type (WT) amino acid on position 14. Polyinosinic:polycytidylic acid [poly(I:C); InvivoGen] was used as subcutaneously injected adjuvant.

Age-matched female C57BL/6 mice were injected subcutaneously with 100 μg peptide and 50 μg poly(I:C) formulated in PBS (200 μL total volume) into the lateral flank (5 mice per group). Every group was immunized on day 0 and day 7 with 2 different mutation-coding peptides, one peptide per flank. Mice were sacrificed 12 days after the initial injection and splenocytes isolated for immunologic testing.

Enzyme-linked immunospot (ELISPOT) assay (17) and generation of syngeneic bone marrow–derived dendritic cells (BMDC) as stimulators were previously described (18). BMDCs were either peptide pulsed (2 μg/mL) or transfected with in vitro transcribed (IVT) RNA coding for the indicated mutation or for control RNA (eGFP-RNA). Sequences representing 2 mutations, each comprising 50 amino acids with the mutation on position 25 and separated by a glycin/serine linker of 9aa were cloned into the pST1-2BgUTR-A120 backbone (19). In vitro transcription from this template and purification were previously described (20). For the assay, 5 × 10⁴ peptide or RNA-engineered BMDCs were coincubated with 5 × 10⁵ freshly isolated splenocytes in a microtiter plate coated with anti–IFN-γ antibody (10 μg/mL, clone AN18; Mabtech). After 18 hours at 37°C, cytokine secretion was detected with an anti–IFN-γ antibody (clone R4-6A2; Mabtech).

For tumor vaccination, 7.5 × 10⁴ B16F10 melanoma cells were inoculated s.c. into the flanks of C57BL/6 mice. Prophylactic immunization with mutation-specific peptide was carried out 4 days before and on days 2 and 9 after tumor inoculation. Therapeutic immunization with the peptide vaccine was administered on days 3 and 10 after tumor injection. The tumor sizes were measured every 3 days, and mice were sacrificed when tumor diameter reached 15 mm.

Our objective was to identify potentially immunogenic somatic point mutations in B16F10 mouse melanoma by NGS and to test these for in vivo immunogenicity by peptide vaccination of mice measuring elicited T-cell responses by ELISPOT assay (Fig. 1A). We sequenced the exomes of the C57BL/6 wild-type background genome and of B16F10 cells, each with triplicate extractions and captures. For each sample, more than 100 million single-end 50 nucleotide reads were generated (Supplementary Table S1). Of these, 80% align uniquely to the mouse mm9 genome and 49% align on target, showing successful target enrichment and resulting in over 20-fold coverage for 70% of the target nucleotides in each individual sample of the triplicates. RNA-Seq of B16F10 cells, also profiled in triplicate, generated a median of 30 million single-end 50 nucleotide reads for each sample, of which 80% align to the mouse transcriptome.

DNA reads (exome capture) from B16F10 and C57BL/6 were analyzed to identify somatic mutations. Copy number variation analysis (21) showed DNA amplifications and deletions in B16F10, including the homozygous deletion of tumor suppressor Cdkn2a (cyclin-dependent kinase inhibitor 2A, p16Ink4A). To identify possibly immunogenic mutations, we identified 3,570 somatic point mutations at FDR of 0.05 or less (Fig. 1B). The most frequent class of mutations was C>T/G>A transitions, as often results from UV light (22). Of these somatic mutations, 1,392 occur in transcripts with 126 mutations in untranslated regions. Of the 1,266 mutations in coding regions, 962 cause nonsynonymous protein changes and of these 563 occur in expressed genes (Fig. 1B, Supplementary Table S2).

Supplementary Table S2 lists the 962 genes containing nonsynonymous somatic point mutations, subcellular localization, and gene type. Noteworthy, many of the mutated genes have been previously associated with cancer phenotypes. Mutations were found in established tumor suppressor genes, including Pten, Trp53 (also called p53), and Tp63. In Trp53, the best established tumor suppressor (23), the asparagine to aspartic acid mutation at protein position 127 (p.N127D) is localized in the DNA-binding domain and predicted to alter function. Pten contained 2 mutations (p.A39V and p.T131P), both of which are predicted to have deleterious impact on protein function. The p.T131P mutation is adjacent to a mutation (p.R130M) shown to diminish phosphatase activity (24).

Several mutations occur in genes associated with DNA repair pathways, such as Brca2 (breast cancer 2, early onset), Atm (ataxia telangiectasia mutated), Ddb1 (damage-specific DNA-binding protein 1), and Rad9b (RAD9 homolog B). Furthermore, we found mutations in other tumor-associated genes, including Aim1 (tumor suppressor “Absent In Melanoma 1”), Flt1 (oncogene Vegr1, fms-related tyrosine kinase 1), Pml (tumor suppressor “promyelocytic leukemia”), Fat1 (“FAT tumor suppressor homolog 1”), Mdm1 (TP53 binding nuclear protein), Mta3 (metastasis associated 1 family, member 3), and Alk (anaplastic lymphoma receptor tyrosine kinase). A mutation occurs at p.S144F in Pdgfra (platelet-derived growth factor receptor, α polypeptide), a cell membrane–bound receptor tyrosine kinase of the mitogen—activated protein kinase/extracellular signal—regulated kinase (MAPK/ERK) pathway, previously identified in tumors (25). Casp9 (caspase 9, apoptosis-related cysteine peptidase) proteolytically cleaves PARP, regulates apoptosis, and has been linked to several cancers (26); here, we found a mutation at p.L222V that may impact PARP and apoptosis signaling.

Interestingly, no mutations were found in Braf, c-Kit, Kras, or Nras. However, mutations were identified in Rassf7 (RAS-associated protein; p.S90R), Ksr1 (kinase suppressor of ras 1; p.L301V), and Atm (PI3K pathway; p.K91T), all of which are predicted to have significant impact on protein function. Trrap (transformation/transcription domain-associated protein) was identified earlier this year in human melanoma specimens as a novel potential melanoma target (27). In B16F10, a Trrap mutation occurs at p.K2783R which is predicted to disturb the overlapping phosphatidylinositol kinase–related kinase FAT domain.

From the 962 identified nonsynonymous mutations, we selected 50 mutations for PCR-based validation and immunogenicity testing. Selection criteria included low FDR, location in an expressed gene, and predicted immunogenicity (Methods). Noteworthy, all 50 mutations validated (Table 1, Fig. 1B) and can be visualized in Fig. 1C.

We employed long peptides as antigens for immunogenicity testing of these mutations. Long peptides have many advantages as induction of antigen-specific CD8⁺ as well as CD4⁺ T cells and processing to be presented on MHC molecules (28). Uptake is most efficiently done by dendritic cells, which are optimal for priming a potent T-cell response. Fitting peptides, in contrast, do not require trimming and are loaded exogenously on all cells expressing MHC molecules, including nonactivated B and T cells, leading to induction of tolerance and fratricide.

For each of the 50 validated mutations, we designed peptides of 27 amino acids length with the mutated or wild-type amino acid positioned centrally. Thus, all potential MHC class I and class II epitopes of 8 to 14 amino acid length carrying the mutation could be processed from this precursor peptide. As adjuvant for peptide vaccination, we used poly(I:C) which promotes cross-presentation and increases vaccine efficacy (29).

The 50 mutations were tested in vivo in mice for induction of T cells. Impressively, 16 out of 50 mutation-coding peptides were found to elicit immune responses in immunized mice. The induced T cells displayed different reactivity patterns (Table 2). Eleven peptides induced an immune response preferentially recognizing the mutated epitope, including mutations 30 (MUT30, Kif18b) and 36 (MUT36, Tm9sf3; Fig. 2A). ELISPOT testing revealed strong mutation-specific immune responses without cross-reactivity against the wild-type peptide or an unrelated control peptide (VSV-NP). With 5 peptides, including mutations 05 (MUT05, Eef2) and 25 (MUT25, Plod2; Fig. 2A), immune responses with comparable recognition of both the mutated as well as the wild-type peptide were obtained. The majority of mutated peptides were not capable of inducing significant T-cell responses, as shown by mutations 01 (MUT01, Fzd7), 02 (MUT02, Xpot), and 07 (MUT07, Trp53). Immune responses induced by several of the discovered mutations were in the range of the immunogenicity (500 spots per 5 × 10⁵ cells) generated by immunizing mice with the positive control MHC class I epitope from the murine melanoma tumor antigen tyrosinase-related protein 2 (Trp2_180–188; Fig. 2A; refs. 30, 31).

For peptides that induce a strong mutation-specific T-cell response, we confirmed immune recognition by an independent approach. Instead of long peptides, IVT RNAs coding for the mutated peptide fragments MUT17, MUT30, and MUT44 were used for the immunologic readout. BMDCs transfected with mutation-coding RNA or irrelevant RNA served as antigen-presenting cells (APC) in an ELISPOT assay, whereas spleen cells of immunized mice served as effector cell population. BMDCs transfected with MUT17-, MUT30-, and MUT44-encoding mRNA were specifically and strongly recognized by splenocytes of mice immunized with the respective long peptides (Fig. 2B). Significantly lower reactivity against control RNA–transfected BMDCs was recorded, which is likely due to the unspecific activation of the BMDCs by the single-stranded RNA (Student t test; MUT17: P = 0.0024, MUT30: P = 0.0122, MUT44: P = 0.0075). These data confirm that the induced mutation-specific T cells in effect recognize endogenously processed epitopes.

Two mutations that induce a preferred recognition of mutated epitopes are in genes Actn4 and Kif18b. The somatic mutation in ACTN4 (actinin, α 4) is at p.F835V in the calcium binding “EF-hand” protein domain. Although both SIFT and POLYPHEN predict a significant impact of this mutation on protein function, the gene is not an established oncogene. However, mutation-specific T cells against ACTN4 have been recently associated with a positive patient outcome (32). KIF18B (kinesin family member 18B) is a kinesin with microtubule motor activity and ATP and nucleotide binding that is involved in regulation of cell division (33; Fig. 3). The DNA sequence at the position encoding p.K739 is homogeneous in the reference C57BL/6, whereas B16F10 DNA reads reveal a heterozygous somatic mutation. Both nucleotides were detected in the B16F10 RNA-Seq reads and validated by Sanger sequencing. KIF18B has not been previously associated with a cancer phenotype. The mutation p.K739N is not localized in a known functional or conserved protein domain (Fig. 3, bottom) and thus most likely is a passenger rather than a driver mutation. These examples suggest a lack of correlation between the capability of inducing mutation-recognizing immune response and a functional relevance.

We assessed whether immune responses elicited in vivo translate in antitumoral effects in tumor-bearing mice with mutations MUT30 (mutation in Kif18b) and MUT44 as examples. Both mutations induce a strong immune reaction preferentially against the mutated peptide and are endogenously processed (Fig. 2A and B). The therapeutic potential of vaccinating with mutated peptides was explored by immunizing mice with MUT30, MUT44, or Trp2 and adjuvant 3 and 10 days after grafting with 7.5 × 10⁴ B16F10 cells. Growth of tumors was inhibited by both mutation encoding peptides as compared with the control groups (Fig. 4A). Remarkably, vaccinating with MUT30 and MUT44 peptides induced tumor growth inhibition equal to the established Trp2 peptide. Furthermore, as B16F10 is a very aggressive tumor, we also tested protective immune responses. Mice were immunized with MUT30 peptide, inoculated s.c. with 7.5 × 10⁴ B16F10 cells 4 days later and boosted with MUT30 immunization 2 and 9 days after tumor challenge. Complete tumor protection and survival of 40% of the mice treated with MUT30 were observed, whereas all mice in the control-treated group died within 44 days (Fig. 4B, left). In the mice that developed tumors despite immunization with MUT30, tumor growth was slower and resulted in a 6-day increase in median survival (Fig. 4B right). These data show that vaccination against a single mutation is able to confer antitumoral effects.

Our study was motivated by the hypothesis that mutations provide a rich source for novel vaccine targets. Although immunogenic mutations have been reported in mouse and man, there is a complete lack of knowledge about the fraction of nonsynonymous mutations processed and presented in an immunologic relevant manner. To address this question, we carried out the first mouse tumor exome capture study, identifying B16F10 somatic point mutations, followed by testing of mutations for their capability to elicit T-cell immunogenicity. Our result is the first map of the B16F10 “T-cell–druggable mutanome.”

The B16 melanoma cell line originated spontaneously in a C57BL/6 mouse in 1954 and was propagated by the Jackson Memorial Laboratory in Bar Harbor. In vivo passage of B16 cells resulted in formation of the highly metastatic daughter clone B16F10 (34). B16 cells express melanoma differentiation antigens and present MHC class I restricted epitopes from gp100, MART1 (35), tyrosinase (36), TRP1 (37) and TRP2 (30). Use of the B16F10 melanoma model has gained momentum in conjunction with vaccine research (38–41) and is today one of the most widely used cell lines for scientific validation of T cell–based immunotherapies (over 1,700 citations) and general cancer studies (over 13,000 citations). Mutation analyses of B16 cells have focused on prominent candidate genes, showing that Cdkn2a (cyclin-dependent kinase inhibitor 2A, p19^Arf and p16^Ink4a) is homozygously deleted and that neither activating mutations in Braf nor inactivating mutations in Trp53 are prevalent (42), but no unbiased genome-wide studies have been carried out. Thus, the frequent use of B16 as a preclinical model stands in sharp contrast to the lack of knowledge of the genetic alterations that underlie the malignant phenotype and contribute to the repertoire of potential targets for T cell–based therapies.

We carried out a meticulous mutation search with a novel NGS protocol and biostatistical algorithm we developed to identify mutations with high confidence. The B16F10 and parental (germ line) C57BL/6 exomes were sequenced in triplicate (biological replicates), resulting in over 100 million reads per sample (over 620 million in total). The lack of reads aligning to the Y chromosome revealed that the B16F10 cells were derived from a female mouse. We identified a considerably higher number of mutations in B16F10 melanoma compared with primary human solid tumors, for which 16 to 302 nonsynonymous mutations have been reported (26, 43, 44). Methods to identify and validate somatic mutations from NGS data are still in development, with recent studies reporting true positive rates of 54% (45) and largely unknown false negative rates. To confirm authenticity of mutation hits and exclude sequencing artifacts, we selected 50 of the identified somatic point mutations and were able to validate all 50 without exception. A likely explanation for the high number of mutations is that B16F10 has accumulated mutations over the 5 decades since it was established.

Whole-exome sequencing in combination with transcriptome profiling enables the discovery of the expressed protein coding mutanome and provides an insight into the molecular nature of potential driver mutations in B16F10 melanoma. The KEGG (Kyoto Encyclopedia of Genes and Genomes; ref. 46) melanoma pathway (hsa05218) identifies 4 melanoma oncogenes (Braf, Nras, Cdk4, and Mitf) and 3 melanoma tumor suppressors (Pten, Cdkn2a, and Trp53). We found all 3 tumor suppressors to be mutated in B16F10. In contrast, none of the 4 KEGG-identified melanoma oncogenes displayed a mutation. However, we discovered mutations in associated proteins of the respective pathways: Rassf7 (RAS pathway), Ksr1 (RAS pathway), Atm (PI3K/AKT pathway), and Pdgfra (RAS-MAPK/ERK pathway). Moreover, although not mutated, we detected overexpression of Mitf, suggestive of a transcriptional dysregulation. Furthermore, we identified a somatic mutation of Trrap, a gene identified earlier this year as a potential novel human target for malignant melanoma (27). Additional potential driver or enabling mutations in B16F10 include those in DNA-repair machinery (Brac, Rad9b, and Atm) and those associated with melanoma and cancer (e.g., Alk, Aim1, Flt1, Pml, Mta3, and Fat1).

For systematic immunogenicity testing, we selected 50 confirmed nonsynonymous mutations that were abundantly expressed in B16F10 melanoma. We used long synthetic peptides for vaccination to simulate natural antigen processing and to allow presentation of both MHC I and II restricted epitopes (47, 48). RNA-transfected DCs as APCs for ELISPOT assays were used as independent method to confirm that immune responses recognize endogenously processed epitopes, additionally validating that the respective antigen fragment encoded by the RNA is processed and naturally presented.

As expected, only a subset of the B16F10 mutations is immunogenic. No apparent correlation was observed between immunogenicity with potential oncological relevance of the protein, structural features of the respective gene, or subcellular localization of the encoded protein. Importantly, we revealed the immunogenicity of novel somatic mutations that are not known to promote a malignant phenotype, such as those in Actn4 (MUT46) and Kif18b (MUT30). Thus, regardless of their function, immunity against these mutations, as long as they are stably expressed in a relevant malignant cell population, may mediate tumor control. Indeed, a point mutation in the human Actn4 ortholog was recently described in a lung cancer patient. The mutation created a tumor-specific HLA-A2 epitope that was presented and recognized by a mutation-specific CTL population that persisted over years in the blood of the patient, potentially influencing the positive course of the disease (32).

Our study allows estimation of the entire space of immunogenic tumor mutations (the T-cell–druggable mutanome) in B16F10. We found one-third (16 of 50) of the mutated epitopes to be immunogenic. Nearly half of these epitopes induced a strong T-cell response that matches the intensity observed for the H-2Kb–restricted Trp2_180–188 epitope, which is one of the most immunogenic B16 melanoma target antigens. Thus, this study adds validated immunodominant epitopes to the target antigen space of B16 melanoma that qualify as candidates for antitumor vaccines. B16F10 contains 563 expressed, nonsynonymous somatic mutations at FDR of 0.05 or less, with 430 of those predicted to be presented on MHC. We estimate that the B16F10 tumor mutanome, relative to the C57BL/6 mouse, comprises more than 180 immunogenic mutations with approximately 80 of them being able to mount strong immune responses.

B16 cells are poorly immunogenic and vaccination with irradiated tumor cells does not protect mice from subsequent challenge with living B16 cells (49, 50), in part due to the low expression of MHC class I and absence of MHC class II molecules (51). Nevertheless, in our vaccination studies, we see antitumor effects for 2 of the tested mutated epitopes. The strong immune responses are able to compensate for the nearly complete downregulation of MHC molecules in B16F10 melanoma cells (52). This interpretation is supported by the fact that after immunizing a group of mice with the immunodominant Trp2 epitope, we observed comparable antitumor activity (Fig. 4A).

This study has implications for human cancer therapies. Mutations in human TAAs that elicit both CD8⁺ and CD4⁺ T-cell responses have been described (8, 52). Human cancers carry on average 100 to 120 nonsynonymous mutations. Segal and colleagues predicted 40 to 60 HLA class I restricted epitopes per patient derived from tumor-specific somatic mutations (53). Given our identification of 563 expressed protein mutations in B16F10, the in silico prediction of Segal and colleagues matches to our estimate of 180 T-cell–druggable mutations in B16F10 melanoma that is supported by experimental in vivo data. Even considering patient and tumor variability, this estimate suggests a rich T-cell–druggable mutanome, which can be tapped by combining deep sequencing with systematic immunogenicity analysis of mutations. As a consequence, the approach opens a new dimension for individualized immunotherapy and adds to tailored vaccine concepts that were previously suggested by us and others (54–57). Every patient's tumor bears a highly individual mutation “signature,” and more than 95% of mutations are unique and patient specific (1). Thus, a vaccine concept based on the multiplicity of mutated epitopes would require profiling of each patient's tumor to determine the unique mutation signature. Dramatically reduced costs and time required for genome-wide discovery of cancer-specific mutations opens the door for individualized immunotherapy of cancer patients. In particular, in advanced disease with tumor genomes becoming more unstable, individualized T-cell therapies may outdo other treatment options as accumulation of mutations allows to combine even more antigens, thereby counteracting the selection of antigen loss variants during immunotherapy and tumor evolution.

U. Sahin, C. Huber, C. Britten, and A. Kuhn are associated with BioNTech AG (Mainz, Germany), a company that develops RNA-based cancer vaccines. U. Sahin, S. Kreiter, Ö. Türeci, A. Selmi, J. Castle, J. Diekmann, M. Löwer, C. Britten, and M. Diken are inventors on patent applications, in which parts of this article are covered. The other authors disclosed no potential conflicts of interest.

The authors thank M. Holzmann, R. Roth, M. Wagner, C. Kneip, T. Bukur, B. Renard and V. Boisguérin for assistance.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.