Genetic and geographic influence on phenotypic variation in European sarcoidosis patients

Introduction Sarcoidosis is a highly variable disease in terms of organ involvement, type of onset and course. Associations of genetic polymorphisms with sarcoidosis phenotypes have been observed and suggest genetic signatures. Methods After obtaining a positive vote of the competent ethics committee we genotyped 1909 patients of the deeply phenotyped Genetic-Phenotype Relationship in Sarcoidosis (GenPhenReSa) cohort of 31 European centers in 12 countries with 116 potentially disease-relevant single-nucleotide polymorphisms (SNPs). Using a meta-analysis, we investigated the association of relevant phenotypes (acute vs. sub-acute onset, phenotypes of organ involvement, specific organ involvements, and specific symptoms) with genetic markers. Subgroups were built on the basis of geographical, clinical and hospital provision considerations. Results In the meta-analysis of the full cohort, there was no significant genetic association with any considered phenotype after correcting for multiple testing. In the largest sub-cohort (Serbia), we confirmed the known association of acute onset with TNF and reported a new association of acute onset an HLA polymorphism. Multi-locus models with sets of three SNPs in different genes showed strong associations with the acute onset phenotype in Serbia and Lublin (Poland) demonstrating potential region-specific genetic links with clinical features, including recently described phenotypes of organ involvement. Discussion The observed associations between genetic variants and sarcoidosis phenotypes in subgroups suggest that gene–environment-interactions may influence the clinical phenotype. In addition, we show that two different sets of genetic variants are permissive for the same phenotype of acute disease only in two geographic subcohorts pointing to interactions of genetic signatures with different local environmental factors. Our results represent an important step towards understanding the genetic architecture of sarcoidosis.


Introduction
Sarcoidosis is a clinically heterogeneous disease, particularly regarding the affected organs, the type of onset and the course of the disease (1) and this clinical heterogeneity needs to be addressed therapeutically (2). Sarcoidosis is a complex disorder with a strong genetic background (3)(4)(5), for which a number of genetic risk factors have been identified by genome-wide association and candidate gene studies, e.g., various HLA-haplotypes (6), BTNL2 (7), ANXA11 (8), and IL23R (9), some of which confer susceptibility to certain sarcoidosis phenotypes (10). Sarcoidosis is, therefore, pathogenetically regarded as a disease resulting from a complex interplay of an unknown environmental causative agent or agents with a genetically permissive subject. Thus, genetic factors might have a potential to distinguish sarcoidosis phenotypes and should, in general, be most suitable as prospective biomarkers.
Many of the reported genetic associations with sarcoidosis or with a specific clinical phenotype of sarcoidosis could not be replicated in independent studies or showed contradictory results in a replication cohort (4). As an example, the combination of HLA-DRB1*01 and TNFA2 predisposes Europeans to favorable prognosis, but is associated with poor prognosis and cardiac manifestation of sarcoidosis in a Japanese cohort (11,12). Similarly, two ANXA11 variants are associated with sarcoidosis in African Americans, but not in European Americans (13). In some studies, insufficient statistical power due to small sample size could be a reason for this lack of replication. However, in other cases, this might reflect the fact that different causative agents encounter a different genetic architecture to cause sarcoidosis.
The aim of this study was to investigate whether genetic variants known or suggested to be associated with sarcoidosis or implicated in sarcoidosis pathogenesis, are associated with clinical phenotypes of sarcoidosis, including type of onset, phenotypes of organ involvement, specific organ involvements itself, or symptoms and to evaluate whether these associations are consistent throughout multiple European subcohorts or subcohort-specific.

Patient recruitment and phenotyping
The Genetic-Phenotype Relationship in Sarcoidosis (GenPhenReSa) cohort includes a total of 2,163 patients from 31 European centers from 12 countries that were deeply phenotyped at three different time points over a course of 4 years (14). A positive vote by the Ethics Committee of the University of Freiburg was obtained prior to the initiation of the study, which was also registered in the German Clinical Trials Register (identifier DRKS00000045). In short, sarcoidosis patients were recruited to the study who had been diagnosed with sarcoidosis according to the ATS/ERS/ WASOG consensus statement (15) with a documented course of disease over at least 2 years prior to and 2 years following recruitment. As a Frontiers in Medicine 03 frontiersin.org prevention of migration bias only patients from families living for at least 3 generations in the same country were recruited. The 31 centers were grouped into 13 geographic regions based on geographic (16), clinical, hospital provision and sample size considerations. In general, centers were merged to form presumably homogenous regions with sample sizes of at least 24 patients per region. Thus, some countries had two or three geographically distinct regions (Poland with Lublin (n = 111) and Gdansk (n = 59), and Germany with North (n = 161), Central (n = 152), and South (n = 70)), while small centers from different countries were considered one region based on geographic proximity, e.g., Louven (n = 17) and Bobigny (n = 8) were merged with Amsterdam (n = 67) and Lille (n = 31) to form a Netherlands-Belgium-France region. A full list of recruitment centers and regional groups is given in the Supplementary material, and a detailed description of the GenPhenReSa cohort is given elsewhere (14). For this study, data from 1909 patients with complete information on basic variables (age, sex and region of origin) and documentation of the clinical course over 4 years were analyzed. Moreover, clinical and specific organ involvements such as central nervous system (CNS), skin or ocular involvement, and specific symptoms like night sweat, subfebrile temperature, and dyspnoea were collected using the definition of extrapulmonary organ involvement of the ACCESS study (17). Based on the course of disease in the 2 years prior to recruitment, patients were classified as acute or sub-acute onset phenotype (18). Except for this, all phenotypes used for analysis were defined based on the presentation at baseline, i.e., time of recruitment Patients were grouped into five previously described phenotypes of organ involvement (an abdominal, an ocular-cardio-cutaneous-CNS, musculoskeletal-cutaneous, a pulmonary-lymphonodal and an extrapulmonary phenotype) (14, 19).

Sample and data preparation
DNA from blood samples was isolated using the Qiagen Kit following the manufacturer's instructions. The amount and quality of all DNA samples was checked by agarose gel electrophoresis, and the DNA was amplified using the REPLI-g whole genome amplification kit prior to genotyping, which was performed using Sequenom Mass-ARRAY iPlex (Sequenom, Inc.) or Taqman (Thermo Fisher Scientific Inc.) technology. Initially, 137 SNPs were selected for investigation based on (i) a known or suggested association with sarcoidosis, (ii) an implication of the gene product in sarcoidosis pathogenesis or (iii) an association of the variant with a clinical phenotype. Allele calling was checked by visual inspection of cluster plots. SNPs or variants with a call rate (CR) < 90%, a minor allele frequency (MAF) < 1% (over all samples from all centers) or an extreme deviation from Hardy-Weinberg equilibrium (p < 10 −5 ) were excluded from analysis, leaving 116 SNPs for further statistical analysis. A full list of SNPs that were included in the analysis before filtering for these quality control parameters together with references is given in the Supplementary material.

Statistical methods
All evaluated phenotype variables were dichotomous (present or not), and their frequencies between study regions were compared by chi-square tests. As the main analysis, the overall association between geno-and phenotypes was investigated in the full cohort. Thus, sarcoidosis patients with a specific phenotype were compared to those without the specific phenotype. Taking the nested data structure into account, a meta-analysis using individual participant data (IPD) approach (20,21) was conducted, where the regional SNP-wise impact upon each phenotype was estimated by a logistic regression model and combined afterwards across all regions to an overall odds ratio (OR). Such a meta-analysis is necessary due to both clinical and genetic heterogeneity between regions (see results section) which could lead to spurious associations in a "megaanalysis" of simple pooled data (Simpson's paradox). In detail, for each region i (i = 1, …, 13), all SNPs were modelled separately as an independent variable using an additive genetic model (which has good power even under different true genetic models)In all these region-specific logistic regression models, sex, age and smoking status were included as possible confounders such that the resulting OR i are adjusted for these variables. For the meta-analysis, these individually calculated OR i , by phenotype and by region, were combined across the geographical regions in a weighted random effects model to get an overall OR, respectively for all phenotypes (22). Low phenotype prevalence and low MAF lead to very imprecise OR estimates and extremely large confidence intervals which cannot sensibly be included in a meta-analysis. However, this depends on sample size and in order to include as much information as possible in the meta-analysis, the region-specific ORs were only excluded from the meta-analysis (and further analyses) if the region-specific combination of phenotype prevalence, SNP MAF and sample size was insufficient to allow a reliable estimation of the OR i , i.e., lead to a width of the confidence interval more than 100. Reported ORs always refer to the minor allele: in case of an OR > 1, the minor allele increases the risk of the respective phenotype, while an OR < 1 indicates a reduced risk of the phenotype for the minor allele. Using Bonferroni correction for multiple testing, the significance level of 0.05 was divided by the number of tested SNPs which results in α* = 0.00043 for each phenotype and each SNP in the meta-analysis. Since the considered phenotypes are correlated, we do not correct for the number of phenotypes here.
In an explorative manner, the larger regional sub-cohorts (n > 100) were investigated individually in multi-locus models. In particular, for the acute onset phenotype, SNPs with univariate OR which had a p value smaller than 0.01 were included in a multiple model. The subsequent model selection was done by likelihood ratio criteria to identify a multi-locus model, which allowed interpretation as a genetic signature. In the explorative analyses of the sub-cohorts, all results with a nominal p < 0.01 are presented to balance power and false positive results. Point estimates of the OR are provided with 95% confidence intervals (given in square brackets). All statistical analyses were conducted in R version 3.62 (particularly library metafor (22)).

Patient characteristics
In total, 1909 patients from 13 European regions with a minimum number of 24 patients (in Romania) were included in this analysis. Table 1 lists the phenotypes used for the association analysis together with their prevalences in the largest cohorts. The other 323 patients from in total seven regions with a sample size of less than 100 patients Frontiers in Medicine 04 frontiersin.org each (Romania, n = 24, Czech Republic, n = 60, Dublin, n = 29, England, n = 29, Gdansk (Poland), n = 59, Iceland, n = 52, South Germany, n = 70) are here summarized in one separate category and are given in (Supplementary Table 2, see https://www.uni-kiel.de/medinfo/ mitarbeiter/freitag-wolf/download/). Frequencies of clinical manifestations (i.e., phenotypes) were significantly heterogeneous between regions (p < 0.006 for all characteristics), e.g., in total 755 of 1866 patients suffered from an acute disease onset (40.5%), which had the highest prevalence in the Serbian cohort with 53.6% (480/913) and the lowest prevalence in North Germany with only 18.4% (25/161). In addition to this, the occurrence of missing values differed depending on phenotype and region (23). Genetic heterogeneity, i.e., different minor allele frequencies (see (Supplementary Table 1, for details), between regions was also large, e.g., for rs4143332 at HLA-B the minor allele frequency was 7% in the Netherland-Belgian-France region (n = 125) and 16% in Serbia (n = 913), with even more extreme values in regions with smaller sample sizes, which required the statistical meta-analysis strategy as described above.

Individual participant data meta-analyses
In a meta-analysis approach all region-wise effects of each SNP upon the phenotypes were combined for the main analysis. After correction for multiple testing, this analysis did not yield any significant association of any of the investigated genetic variants to any phenotype (Supplementary Table 3, see https://www.uni-kiel.de/ medinfo/mitarbeiter/freitag-wolf/download/). The largest effect was found for rs4143332 at HLA-B associated with acute onset adjusted for sex, age and smoking status (OR = 1.79, p = 0.013), followed by rs1800629 at TNF (OR = 1.65, p = 0.017), Figures 1, 2.
Focusing on and summarizing the Central Europe regions (Germany and Netherlands-Belgium-France), the estimated effects were slightly smaller for the above mentioned two loci (OR = 1.58, p = 0.19 and OR = 1.42, p = 0.27, respectively). As an aside, the association with the smallest p-value in the pooled meta-analysis of the Central Europe regions was night sweat and rs2075800, a missense variant near HSPA1L (OR = 3.95, p = 0.0075).

Region-specific analyses
Because of the strong phenotypic and genetic heterogeneity, we also report results of the largest individual sub-cohorts as exploratory analyses. We investigated potential genetic associations with type of onset, phenotypes of organ involvement, specific organ involvements and symptoms within the defined regional sub-cohorts. Apart from acute onset, none of the tested phenotypes in the regions reached the required level of significance after correcting for multiple testing and all reported p values in the subsequent paragraphs are nominal p values. Figure 3 gives a map of the 12 included European countries with their most relevant phenotype-genotype relation in terms of their smallest p-values for each of the analyzed 13 regions. All potentially meaningful (p < 0.01) ORs, CI and p-values of all regions are given in Table 2 and all results with p < 0.05 are given in Supplementary Table 3. In the following, we report details on associations with p < 0.01.

Acute vs. sub-acute onset
In the Serbian sub-cohort (Table 2), the strongest association of acute onset in the single locus model was found for rs1800629 (=TNFA-308G/A), a well-known sarcoidosis risk SNP that is located    Table 3).

Phenotypes, organ involvement and symptoms
We recently defined novel phenotypes of organ involvement: an abdominal, an ocular-cardio-cutaneous-CNS, a musculoskeletalcutaneous, a pulmonary-lymphonodal and an extrapulmonary phenotype (14). Here, we analyzed whether these phenotypes are associated with genetic variants and even though none of these reached statistical significance after adjustment for multiple testing, several are still noteworthy with large effect sizes in individual sub-cohorts (Supplementary Table 3). The extrapulmonary phenotype was associated with rs2015086 near CCL18 (p = 0.002; OR = 4.51), rs3800018 near RAB23 (p = 0.002; OR = 6.51) and rs7756421 near ZNF451 (p = 0.002; OR = 6.53) in the North German sub-cohort. The abdominal phenotype was associated with rs5007259 at BTNL2  Forest plot with a random effects meta-analysis model for the association between rs4143332 (HLA) and acute onset. Not included are Bucharest, Dublin, Czech Republic and Iceland due to a combination of low sample size, low MAF and low prevalence of acute onset (see statistical methods). Odds ratios (OR) are given with corresponding 95% confidence intervals in square brackets. Forest plot with a random effects meta-analysis model for the association between rs1800629 (=TNFA -308G/A) and acute onset. Not included are Bucharest and Iceland due to a combination of low sample size, low MAF and low prevalence of acute onset (see statistical methods). Odds ratios (OR) are given with corresponding 95% confidence intervals in square brackets.

Specific organ involvements
Sarcoidosis is a highly variable disease, especially variable in the distribution of affected organs. We hypothesized that specific genetic variants are associated with distinct organ involvements in sarcoidosis. In the Serbian sub-cohort we found noteworthy risk loci for neurosarcoidosis and ocular sarcoidosis. The minor alleles at three genetic variants are associated with increased risk of neurosarcoidosis: rs3775291 near TLR3 (p = 0.003; OR = 2.01), rs4921492 near IL12B (p = 0.008; OR = 1.88), while the minor allele at rs12793173 near LOC102723568 (p = 0.006; OR = 0.53) is protective. The following risk loci were observed for ocular sarcoidosis: rs2206593 near PTGS2 (p = 0.0096; OR = 2.20), and two variants near IL18, however with opposite directionality: rs1946518 (p = 0.006; OR = 1.61) and rs189667 (p = 0.004; OR = 0.59). In the Serbian cohort, skin involvement was associated with rs11730582 near SPP1 (p = 0.004; OR = 1.59), the aforementioned TNF variant rs1800629 (p = 0.008; OR = 0.55) and rs4143332 at 6p21 (p = 0.005; OR = 0.45). Again, borderline significance of these associations prevents any definite conclusions without further confirmation in other cohorts.

Multi-locus models
In the Serbian and Lublin (Poland) cohorts, two separate sets of three SNPs in six different genes showed association to the acute onset phenotype in multi-locus models (Table 3). In the Serbian cohort, after backward selection of all loci with p < 0.01, the individual impact of the analyzed SNPs on acute onset decreased for rs1800629 from OR = 2.19 in the single-locus model to 1.95 in the multi-locus model and for rs9271366 from OR = 1.69 to OR = 1.43, whereas the OR of rs4143332 increased slightly from 2.22 to OR = 2.34. A combination of the risk alleles at all three SNPs conferred an OR of 6.6 when considered as a genetic signature. In Lublin, all three SNPs had substantially stronger effects in a multi-locus model and the combined effect of a genetic signature comprising all three risk alleles corresponded to an OR of 37.9. Two of the SNPs in this genetic signature, namely rs1891467 near TGFB2 and rs7667298 near KDR, had been reported to be associated with an acute course of sarcoidosis before in a German cohort (4).
In amalgamate, two different sets of genetic variants are permissive for the same phenotype of acute disease onset resulting in combined ORs of 6.6 for the Serbian and of 37.9 for the Polish subcohort which points at their likely interactions with different local environmental factors resulting in the same phenotype.

Discussion
Sarcoidosis is a highly variable and heterogeneous disease with a need for understanding the genetic background of this heterogeneity. In total 116 selected candidate SNPs were analyzed for their association with clinical traits in sarcoidosis patients of the GenPhenReSa cohort (14). The GenPhenReSa patient collection represents a large sarcoidosis cohort of almost 2,000 patients of European origin that was deeply phenotyped over 4 years at three points in time, i.e., every 2 years (14). In line with the known heterogeneity, we did not observe any genotype-phenotype association that was uniformly present in the whole European cohort, when analyzed with an appropriate meta-analysis. However, different unique associations were present in different geographic sub-cohorts. With our approach, we were able to identify specific SNPs that are associated with course of the disease (acute onset) in the Serbian and Lublin (Poland) sub-cohorts which result in increased ORs if used as a set. In addition, we found some genetic associations with the phenotypes of organ involvement and specific symptoms, mostly restricted to individual geographic cohorts ( Figure 3). In these smaller samples, effect sizes were large but did not reach the predefined, stringent significance levels.
Most strikingly and of potential prognostic value, marker rs1800629 (=TNF-308G/A), a SNP in the TNF promoter region, was associated with acute onset. The A-allele of this SNP is associated with acute onset with an OR of 2.18 in the Serbian sub-cohort, which is in line with previous reports that describe the A-allele being associated with a favorable prognosis (24). Likewise, associations with acute onset were observed for the HLA variants rs4143332 (OR = 2.22) and rs9271366 (OR = 1.69). The latter variant is located between HLA-DRB1 and HLA-DQA1 and is associated with inflammatory bowel disease, multiple sclerosis, systemic lupus erythematosus and others (25-27), but has not yet been reported with regards to sarcoidosis. Whether or not this finding might be dependent on HLA haplotypes that are known to be associated with certain sarcoidosis subtypes has to be analyzed in future studies. Noteworthy, used in combination, these variants are associated with acute onset sarcoidosis with an OR of 6.6 ( Table 3) which could be used as a genetic signature.
Interestingly, the estimated effects of SNPs near TNFRSF1B, TGFB2, and KDR on acute onset in the Polish sub-cohort from Lublin increased in the multi-locus models and result in an OR of 37.9 (Table 3) when considered as a genetic signature. TGFB2 and KDR variants were reported to be associated in a German cohort with an acute course of sarcoidosis (4) and variants of TNFRSF1B are known to influence the response to anti-TNF therapy in sarcoidosis (28). Which leads to the concept of a possible regulation of acute sarcoid inflammation by these genetic variants.
The effects of SNPs near TNFSFR1B, TGFB2 and KDR on acute onset are increased in the multi-locus models. The latter two SNPs were reported to be associated with an acute course of sarcoidosis before (4). Different roles of the variant gene products in the regulation of inflammation can be assumed on the basis of ex-vivo data. The SNP rs1148459 is located at the position −9,298 of the 5′-flanking region of TNFRSF1B and it is speculated that this SNP may regulate the expression of the receptor (29). TNFSFR1B is associated with the generation of Treg (30). Lack of TNFSFR1B expression of Treg may lead to non-functional Treg which suggests a functional influence on the effect of anti-TNF therapy of sarcoidosis (28). The SNP rs1891467 is located in the beginning of the second intron of TGFB2, however, the consequences of the SNP are unclear. Members of the TGFβ family are anti-inflammatory mediators (31) and in bronchoalveolar lavage cell culture supernatants from patients suffering from active sarcoidosis with spontaneous remission we found increased levels of TGFβ indicating an anti-inflammatory activity of TGFβ also in sarcoidosis. However, it is currently not clear whether or not this is related with SNP rs1891467. The SNP rs7667298 is located in the 5'UTR of the gene KDR which codes for the VEGF-R2. It has been demonstrated that rs7667298 is related with diminished expression of VEGF-R2 (32). In contrast, in sarcoidosis increased levels of VEGF have been detected (33). VEGF-R2 is a strong inducer of vascularization whereas VEGF-R1 exerts pro-inflammatory stimulation. Thus, lack of VEGF-R2 may lead to abnormal vascularization (33) but fosters inflammation. Thus, relevance of these genetic variants on the up-and down-regulation of sarcoid alveolitis is most likely.
The number of significant associations between SNPs and clinical variables in the entire cohort of 1909 Caucasians fell short of expectations, which may suggest that permissive genetic signatures depend strongly on specific environmental or occupational events in a particular geographic region or cohort to cause a particular phenotype as suggested by the ACCESS-study (19,34). This conclusion seems likely since all patients came from families that lived for at least three generations in the same region. Nevertheless, the application of genetic markers as predictors for certain phenotypes has turned out to be challenging for the majority of complex diseases and various methods have been applied to identify risk loci sets [reviewed in (35)]. A critical point in our study may have been the choice of markers that were genotyped which had been selected because of previous reports in sarcoidosis or other granulomatous diseases. As this was not a genome-wide association study, we might have missed important genetic associations. Furthermore, an association with the investigated phenotypes had not been established before for the majority of the analyzed SNPs and therefore replication in independent cohorts of diverse ethnicities and with different study designs is warranted. Within the entire cohort, which comprised a large, genetically and phenotypically heterogeneous population there were no clearly significant genetic associations of the phenotypes of organ involvement; however, the observed effects were large (ORs >2 or <0.5) within specific and more homogeneous geographic areas.
In this multicentre study, the number of patients and prevalences of specific phenotypes for the clinical variables were not large enough to detect significant genetic effects on most phenotypes. Indeed, in terms of statistical considerations, with the overall sample size of more than 1900 patients, an OR larger 1.3 could have been found with a power above 90% (additive model, alpha set to 0.00043, allele frequency assumed to be 0.25), but the genetic and phenotypic heterogeneity of the included patients turned out to be most challenging. While the genotype frequencies of investigated markers show at least a slight gradient within Europe (16), which was addressed by the meta-analysis, the phenotypic heterogeneity was so large that it could not be fully compensated by statistical methods. This large phenotypic heterogeneity between regions might be due to differences in health care systems and recruitment settings (different age distributions of cases, different commonly used treatment) and different frequencies of individual non-genetic risk factors such as smoking which has been reported to increase, e.g., the susceptibility in Sweden by a gene-environment interaction (36). Moreover, about 30% of the sarcoidosis incidence is attributed to occupational exposures (37) and many associations of sarcoidosis with environmental, infectious or occupational exposures have been reported (19,38,39). Unfortunately, individual exposure information is not available for the GenPhenReSa cohort. Chronic Beryllium Disease (CBD) is a perfect phenocopy of sarcoidosis and SNPs associated with CBD exert janiform functions depending on disease stage which might also be the case in sarcoidosis which demands more detailed phenotyping in future studies (40). Thus, gene -environment / occupation interactions have to be considered which cause a complex setting of exposures to be analyzed. An example would be the specific T-cell response to a peptide of Aspergillus nidulans documented in Swedish HLA-DR3 positive Löfgren's syndrome patients (41). Further genetic studies are needed and our data presented in the (Supplementary Tables will be very helpful in designing future studies. Taken together, we confirmed the association of rs4143332 at HLA-B and rs1800629 at TNF locus with acute onset of sarcoidosis in the Serbian sub-cohort and describe potential genetic links with several clinical features, including the recently described phenotypes of organ involvement (14,19). Furthermore, we demonstrate that the concept of genetic signatures needs to be complemented with environmental and occupational data. The observed associations of genetic and clinical features suggest that the clinical phenotype is highly dependent on subcohort-specific environmental interactions with genetics and is more complex than anticipated. This view is supported by the fact that in the Polish and German geographic subcohorts different gene variants exhibited highest ORs (Figure 3). Most likely different genetic backgrounds in combination with local exposures are responsible for identical sarcoid phenotypes in different environmental and regional settings of Europe (14, 19). Our finding represents an important and necessary step towards understanding the genetic architecture of the heterogeneous and complex disease sarcoidosis. Local signatures comprising genetic, occupational, environmental and infectious parameters will be necessary and seem achievable to predict courses of the disease.

Data availability statement
The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found at: https://www.uni-kiel.de/medinfo/ mitarbeiter/freitag-wolf/download/.

Ethics statement
The studies involving human participants were reviewed and approved by Ethik-Kommission der Albert-Ludwigs-Universität Freiburg, email: ekfr@uniklinik-freiburg.de. The patients/ participants provided their written informed consent to participate in this study.