Sarcoidosis is a heterogeneous inflammatory disorder of unknown origin that may affect virtually any organ, although intrathoracic engagement is almost universal. Sarcoidosis may present rather dramatically as an acute disease, which usually resolves either spontaneously or with treatment, while other patients have an insidious onset and a chronic/progressive disease course. The different clinical phenotypes have led to the suggestion that sarcoidosis may consist of several separate entities. Yet, the characteristic immune response eventually leading to granuloma formation indicates that a number of features are common to all subgroups of the disease. Through a classical candidate gene approach, several genes of importance for sarcoidosis have been identified, and in some cases such gene variants associate with distinct clinical phenotypes. More recently, another approach to the search for sarcoidosis-associated genes has been applied, that is, through genome-wide association studies (GWAS). GWAS have led to the identification of a number of new genetic associations, although several of them need to be validated. Conversely, some of the previously identified human leucocyte antigen (HLA) associations with sarcoidosis have already been replicated in different cohorts and found to be quite strong, particularly in specific patient subgroups. In highly specialised centres such HLA associations already represent a useful aid in clinic practice for improving patient management. For the future, there is an urgent need for a better understanding, in particular, of gene–gene as well as gene–environmental interactions, both likely to be of importance for developing sarcoidosis.
- Diffuse parenchymal lung disease
- Respiratory Medicine
Statistics from Altmetric.com
Sarcoidosis is a chronic inflammatory disorder of unknown origin characterised histologically by the presence of non-caseating epithelioid granulomas in affected organs.1 Originally described as a disorder of the skin, sarcoidosis can involve any organ, although pulmonary manifestations are almost universal.2 Whatever the aetiology of the disease, its pathophysiology likely involves an aberrant immune response to environmental/infectious agents in genetically susceptible hosts.3 Clinical manifestations and prognosis range from benign, self-limited and often asymptomatic disease to severe loss of function and permanent damage of involved organs. An acute presentation of fever, bilateral hilar lymphadenopathy and erythema nodosum with or without periarticular inflammation of the ankles (referred to as Löfgren's syndrome) heralds an excellent prognosis.4
There is substantial evidence for a genetic predisposition to sarcoidosis: (1) monozygotic twins are more often concordant for the disease than dizygotic twins;5 ,6 (2) familial clustering occurs in approximately 5%–16% of patients;7 and (3) striking differences exist in the prevalence and clinical manifestations of the disease in different geographic areas and racial groups.8 In addition, a number of susceptibility loci have been identified from both linkage and association studies, with the human leucocyte antigen (HLA) class II alleles representing the main contributor to disease susceptibility across different ethnic populations.9 However, sarcoidosis is not due to defects in a single major gene or immunopathological pathway; instead, there are a multitude of genes, each contributing a relatively small effect and few, if any, being absolutely required for the disease to occur. Genetics is also likely to contribute to the wide variety of clinical manifestations and prognosis observed in sarcoidosis.
This review summarises current knowledge on genetics of sarcoidosis with emphasis on (1) recently performed genome-wide association studies (GWAS) and (2) influence of genetic factors on specific disease phenotypes.
Traditionally, genetic studies of sarcoidosis have used a candidate–gene case–control approach. Candidate–gene studies rely on predicting the correct gene/s, usually on the basis of biological hypotheses (or the location of the candidate within a previously determined region of linkage). Therefore, when the fundamental defects underlying a given disease are unknown, this approach will clearly be inadequate to fully explain the genetic basis of that disease.10 Despite these limitations, candidate–gene studies have significantly improved our understanding of the genetic basis of sarcoidosis by identifying a number of robust associations, mostly with alleles located in the HLA region.9 ,11
HLA class I genes
Originally, HLA association studies in sarcoidosis concentrated on HLA class I alleles (HLA-A, -B and -C) using serological typing. Varying and often controversial results were obtained depending on the cohort and the ethnicity studied; however, HLA-B7 and -B8 were consistently found to be associated with disease risk.12–14 As more studies of sarcoidosis and HLA-B8 were published, it was noted that HLA-B8/DR3 genes are inherited as a unique haplotype in Caucasians, suggesting that disease associations with class I genes may simply be due to linkage disequilibrium (LD, defined as the tendency for genetic variants located close to each other on the same chromosome to be associated within a population more often than if they were unlinked) with class II genes.15 ,16 Nevertheless, a more recent study of Scandinavian patients clearly shows that HLA-B7 and -B8 increase the risk of sarcoidosis independently of class II genes. Therefore, class I alleles may have more influence on disease susceptibility and behaviour than previously thought.17
HLA class II genes
Our current understanding of sarcoidosis immunopathogenesis suggests that HLA class II molecules, expressed on the surface of antigen-presenting cells, present specific exogenous (or endogenous) antigenic peptides in such a way that recognition by CD4 T lymphocytes initiates an aberrant inflammatory response resulting in granuloma formation (figure 1).18 Thus, HLA class II genes are likely to be involved in sarcoidosis susceptibility. In addition, a restricted usage of T cell receptor α and β chain variable gene segments (Vα and Vβ) on lung accumulated highly activated T cells suggests recognition of discrete antigenic peptides presented by specific HLA class II variants (figure 1).23–25 Consistent with this hypothesis, the strongest genetic associations reported to date have been found with HLA class II genes—mostly DRB1 alleles—although the tight and highly variable degree of LD existing in this area makes it extremely difficult to determine which gene/s represents the primary association, and whether non-HLA genes located in this region (some of which are critical in inflammation and immune response) have any role. There appears to be a pattern of HLA class II associations across different ethnic groups with HLA-DRB1*01 and DRB1*04 being protective, while DRB1*03, DRB1*11, DRB1*12, DRB1*14 and DRB1*15 represent risk factors for sarcoidosis.26 Interestingly, in African Americans, HLA-DQB1 and not DRB1 alleles were suggested to be primarily associated with sarcoidosis.27 Extensive reviews of HLA associations with sarcoidosis can be found elsewhere.9 ,11 ,28 ,29
HLA class II alleles influence the risk for sarcoidosis and may associate with distinct clinical manifestations and disease course. Robust and consistent associations between class I (HLA-B8) and class II (HLA-DRB1*03) alleles and acute onset and good prognosis were described some 30 years ago.13 ,14
Subsequent studies, using more specific tools for HLA analysis and more rigorous disease definition, confirmed the association with Löfgren's syndrome (LS), a clinically related but genetically distinct sarcoidosis phenotype (figure 1).30 In addition to displaying distinct clinical manifestations, patients with LS can be further characterised according to the carriage of DRB1*03. In fact, disease resolution (defined as disease duration <2 years) occurs in 95% of DRB1*03positive patients, but only in 49% of DRB1*03negative.31 Interestingly, a clustering of disease onset in January, April and May has been observed only in DRB1*03positive patients, suggesting a key role for season-specific antigens in the development of LS. The mechanisms for DRB1*03 to influence disease prognosis in LS is unknown, although these patients display a less pronounced T helper 1 type immune response with reduced level of IFN-γ and TNF-α (figure 1).32
Genetic variants other than HLA alleles have also been specifically associated with LS. Carriage of a particular combination of polymorphisms within CC chemokine receptor 2 (CCR2) gene increases the risk for LS independently of HLA-DRB1*03.33 This association has been validated in two independent patient populations,34 but not replicated in a German case–control and family-based study, which however noted a positive linkage at 3p21 (where CCR2 lies), indicating a susceptibility gene in the surrounding chromosomal area.35 Whether the discrepancies are due to inconsistent case definition, subset ascertainment or true ethnic variation is unclear.
Another example is the recent finding of a genetic association between major histocompatibility complex (MHC)2TA (MHC2TA) gene polymorphisms, relating to the expression of MHC class II molecules, and Löfgren's syndrome (but not with non-Löfgren's sarcoidosis patients). Importantly, this association appears to be independent of HLA-DRB1*03, which is in line with the hypothesis that Löfgren's syndrome is a separate disease entity.36
Genetic studies in sarcoidosis have often produced conflicting results. Sato and coworkers applied consistent clinical definitions and HLA genotyping across three independent patient populations (from UK, The Netherlands and Japan) to test the hypothesis that inconsistencies across genetic studies may (also) be accounted for by interethnic differences in sarcoidosis susceptibility and disease manifestations.37 Despite clear (and expected) differences in the frequency of various disease phenotypes between the cohorts, they demonstrated that a number of HLA associations were similar across their ethnically different study populations. DRB1*01 was consistently associated with disease protection, DRB1*12 with disease risk and DRB1*1401/2 with lung-predominant sarcoidosis, while DRB1*0803 conferred disease risk only in Japanese patients—particularly in those with sarcoidosis-related uveitis. Conversely, DRB1*0301 and DQB1*0201, which were confirmed to be strongly associated with Löfgren's syndrome in Dutch patients, were absent in the Japanese population, in which, as expected, none of the patients had clinical features of LS. In addition, DRB1*0401-DQB1*0301 was found to be protective for disease overall in British but a clear risk factor in Japanese patients. Sarcoidosis-related uveitis was also confirmed to have distinct HLA genetics (figure 1).38 Of note, this study did not confirm the previously reported HLA-DRB1*11 association with disease risk.39 Another recent study showed a strong association between HLA-DRB1*04 and Caucasian sarcoidosis patients with Heerfordt´s syndrome, that is, the combination of uveitis (the most common symptom), cranial nerve palsy (most often the facial nerve), salivary gland enlargement and usually fever.40
Scanning the genome
During the past few years, GWAS have revolutionised human genetics and led to the identification of thousands of loci that affect susceptibility to complex diseases.41 ,42 Important insights from GWAS include identification of putative risk loci in or near genes not previously suspected of being involved in the pathogenesis of a given disease as well as associations with non-coding genomic regions. However, GWAS identify loci and not sequence variants per se and are unable to detect rare risk alleles. As such, a genetic association—regardless of its strength and biological plausibility—should always be considered the end of the beginning as it needs to be replicated in independent patient populations, ideally from different ethnicities. Yet, validation does not prove causation, which instead requires integration of genetic, gene expression and epigenetic data as well as functional (cellular and animal) studies.
In an early sarcoidosis study, 122 affected siblings from 55 German families were genotyped for seven polymorphisms flanking and covering the MHC region.43 Multipoint non-parametric linkage analysis showed linkage for the entire MHC region with the highest score (3.2, p=0.0008) observed at marker locus D6S1666 in the class III gene cluster. In a follow-up study from the same authors using 225 microsatellite markers in 63 families with affected siblings (n=138), the MHC region was again the most prominently identified with the highest peak observed in close proximity to the class II loci.44 A subsequent genome-wide, sib pair multipoint linkage analysis in 229 African American families (Sarcoidosis Genetic Analysis (SAGA) study) revealed a number of statistically significant peaks (but no positive signal within the HLA region), the most prominent one being located at D5S2500 on chromosome 5q11 (p=0.0005).45 In addition, there was only modest agreement for linkage between the scans performed in the German and African American populations. In a follow-up study on the same patient population using additional microsatellite markers to refine regions with significant linkage, sarcoidosis susceptibility alleles were identified on chromosome 5p15.2 and protective alleles on chromosome 5q11.2.46 A further linkage analysis of the SAGA sample stratified by genetically determined ancestry (in order to reduce ethnic confounding) confirmed the peaks on chromosome 5 and revealed additional-associated loci.47
A more rigorous scanning of the MHC region identified a novel sarcoidosis risk variant, which predicted a premature splice site at exon 5 of butyrophilin-like 2 (BTNL2) gene.48 The BTNL2-encoded protein is thought to act as a negative costimulatory molecule, owing to its structural similarities with CD80 and CD86 molecules; as such, non-functional BTNL2 could theoretically result in an exaggerated T lymphocyte activation, compatible with the proposed pathophysiology of sarcoidosis.49 In addition, and more importantly, the BTNL2 association with sarcoidosis appeared to be independent of HLA class II alleles, despite almost complete LD with HLA-DRB1. These findings have been replicated in white American (but not African American) patients as well as in an independent German population, although in this latter study the frequency of rs2076530 A allele was significantly increased only in patients with a chronic form of the disease.50 ,51 While the BTNL2 association with sarcoidosis has been convincingly reproduced, its independency of nearby HLA alleles in strong LD remains controversial. In fact, robust BTNL2 rs2076530 A associations with a number of diseases, such as ulcerative colitis, multiple sclerosis, type 1 diabetes, rheumatoid arthritis, systemic lupus erythematosus (SLE), Graves’ disease, tuberculosis, leprosy and Crohn's disease52–56 as well as sarcoidosis itself,57–59 appear to be driven by LD with various HLA-DRB1 alleles. Similarly, a recent GWAS of African and European American sarcoidosis patients replicated the BTNL2 association but did not confirm its independency of HLA class II alleles. In addition, in the combined African American dataset the strongest BTNL2 signal was at coding synonymous (ie, coding for an identical amino acid sequence) single nucleotide polymorphisms (SNP) rs9268480, emphasising the difficulties in pinpointing the causal variant/s when HLA genes and nearby loci in tight LD are considered simultaneously.60 More extensive genotyping across HLA class I and II regions is needed in order to define haplotypes accurately. However, at present, this genomic area is too complex for the current generation of microarray or DNA chips to genotype.
In the first complete GWAS of sarcoidosis, Hofmann and coworkers genotyped 499 patients and 490 controls of German origin using 440 000 SNPs. The strongest signal mapped to the annexin A11 (ANXA11) gene on chromosome 10q22.3. This association was validated in an independent cohort of 1649 cases and 1832 controls. Finer mapping identified a non-synonymous SNP (rs1049550, R230C) as the variant most strongly associated with sarcoidosis.61 Of note, the ANXA11 association has been replicated in independent European populations as well as in African and European Americans.60 ,62–64 ANXA11 exerts a number of regulatory functions in calcium signalling, cell division, vesicle trafficking and apoptosis,65 and has been implicated in the pathogenesis of several autoimmune disorders—including rheumatoid arthritis, SLE and Sjögren syndrome—by giving rise to autoantibodies.66 While the functional relevance of rs1049550, if any, in sarcoidosis pathogenesis remains unknown, dysfunctional ANXA11 could theoretically affect the apoptosis pathway, hence modifying the balance between apoptosis and survival of activated inflammatory cells.
Studies in African Americans, who have an increased risk of sarcoidosis and more severe disease outcomes, suggest that focusing on genes of African ancestry may uncover novel ethnic-specific risk variants. Rybicki and colleagues conducted a genome-wide scan using a panel of highly ancestry markers in a large sample of African American sarcoidosis cases (n=1357) and healthy controls (n=703), and observed the most significant association at 6p22.3 (rs11966463).67 Other markers that demonstrated suggestive ancestry associations with sarcoidosis mapped on chromosome 8p12, which had the most significant association with European ancestry, 5p13 and 5q31, which correspond to regions previously identified through sib pair linkage analyses.
In a linkage study of German families (n=181), Fischer and coworkers genotyped 528 affected members for 3882 SNPs,68 and observed the two most prominent peaks at 12p13.31 and 9q33.1. Suggestive linkage to this latter region had already been identified in a previous study of African American sarcoidosis families.45 Because 9q33.1 showed linkage with chronic sarcoidosis (defined as disease lasting for ≥2 years), it is possible that this locus or an area nearby contains yet unidentified (probably rare) variants that predispose to more severe disease regardless of ethnicity. Of note, 10q22—the region where ANXA11 lies—was not associated with disease risk, while 6p21 that harbours, among others, HLA-DRB1 genes and BTNL2 showed only weak linkage.
Genome scan data from 381 German patients and 392 controls using 97 000 SNP revealed a novel sarcoidosis risk locus on chromosome 6p12.1.69 Extensive fine mapping of the region around rs10484410—the only marker associated with sarcoidosis after correction for multiple testing—pointed to RAB23 as the most likely risk factor. However, this association was not confirmed in European Americans.60 In addition, this study did not confirm previously reported associations with BTNL2 and ANXA11.48 ,61
A recent GWAS including 1.3 million markers identified a novel sarcoidosis risk locus on 11q13.1. This association was consistent across three independent European case–control populations (from Germany, Sweden and Czech Republic).70 Fine mapping of the region surrounding the lead SNP (rs479777) and expression analysis suggest CCDC88B as the most promising candidate, although its role in disease pathogenesis, if any, is unknown. Similarly, it is unclear whether this locus predisposes to sarcoidosis risk in populations other than Europeans. Interestingly, this locus has been associated with Crohn's disease, primary biliary cirrhosis and psoriasis, among others, thus suggesting a shared genetic background.71–73 An independent susceptibility locus was also identified at 12q13.3-q14.1 with the Osteosarcoma amplified 9 (OS9) representing the most likely candidate risk factor. This association has been validated in an independent German population and replicated by a meta-analysis of three independent cohorts of sarcoidosis patients from Germany, Czech Republic and Sweden.74 Data analysis stratified by disease course revealed a stronger association of the lead SNP rs1050045 with acute sarcoidosis. More recently, Adrianto and coworkers performed a GWAS in African and European Americans patients evaluating >6 million SNPs.60 In addition to replicating a number of previously reported associations, they identified a novel sarcoidosis-risk locus—NOTCH4—independent of neighbouring HLA genes as well as multiple independent signals within the HLA class II region. NOTCH4, which encodes a member of the NOTCH family that is involved in controlling cell fate decision during developmental processes and regulating T cell immune responses, has been associated with immune-related disorders, such as SLE and systemic sclerosis and, as such, represents an attractive biological candidate also in sarcoidosis.75 ,76 In this same study, some associations were shared between ethnicities, while others were unique to either African Americans or European Americans, confirming the existence of ethnic-specific genetic effects.60
Potential genetic commonality with other chronic inflammatory disorders
As indicated above, several of the recent studies suggest that sarcoidosis may share genetic risk factors with other chronic inflammatory disorders. The potential role of inflammatory bowel disease risk loci in predisposition to sarcoidosis was investigated in a large cohort of German patients and controls.77 A non-synonymous SNP within HERC (15q13.1) was identified as a novel disease susceptibility marker, while 1q24.3 and rs11209026 (Arg381Gln) within interleukin (IL)-23 receptor (IL23R; 1p31.3) represented phenotype-specific risk loci for acute and chronic sarcoidosis, respectively. While the role of IL-23/Th17 pathways in the development of chronic inflammation in the context of inflammatory bowel disease is well established,78 ,79 its impact in sarcoidosis pathogenesis is less clear. Even less is known about HERC rs916977 variant and the risk loci on chromosome 1q24.3 and within IL23R and no potential functional relevance can be hypothesised at present.
AGWAS analysed 83 360 markers jointly in German patients with Crohn's disease and sarcoidosis—two diseases that share a number of clinical and immunological features.80 Fine mapping of the 225 kb surrounding the main association peak (10p12.2) by using 29 tagging SNPs, in addition to the lead SNP rs1398024 pointed to variants within C10ORF67 gene as the most likely underlying risk factors.81 This study suggests that the same gene—if not the same variant/s—predisposes to both Crohn's disease and sarcoidosis. However, the C10ORF67 association has not been replicated.60
Sarcoidosis genetic studies have been motivated by the hope that identifying risk alleles will help in understanding disease aetiology. They have often reported conflicting results, probably because of sampling methodology (population stratification, small sample size, patient misclassification) and trans-ethnic differences in disease mechanisms, though a number of HLA associations have consistently been reproduced irrespective of ethnicity (table 1). One clear message has emerged from the genetic studies performed thus far: genetics extends from determining overall disease susceptibility to also influencing its phenotypic expressions, including the course of the disease (eg, HLA-DRB1*03 is both a risk factor for sarcoidosis and a determinant of good prognosis) (figure 1). Consequently, it is essential that sarcoidosis genetic data are analysed according to clinical phenotype and not limited to a ‘generic’ disease susceptibility as stratification of data by specific disease subsets may reveal genetic associations which analysis of sarcoidosis susceptibility alone would fail to detect. Disease heterogeneity may have also hampered achieving highly statistically significant linkage signals in the genome scans performed thus far.
Scanning of the genome in large datasets and more robust study methodology have recently led to the identification of novel attractive candidates. Yet, they account for a small proportion of familial clustering and confer only relatively small increments in risk, although small ORs do not discard the possibility that a given allele is involved in a crucial disease pathway. A number of explanations for this missing heritability can be hypothesised, including much larger numbers of common variants with smaller effect yet to be found: rarer variants (possibly with larger effects) or structural variants (insertion, deletion, duplication, translocation or inversion of segments of DNA), both poorly captured by current genotyping assays; low power to detect gene–gene interactions; and inadequate accounting for environmental factors.90 While newer sequencing technologies (whole genome, whole exome, targeted sequencing) are likely to rapidly increase the number of genetic variants associated with many complex diseases,91 they may not be as successful in sarcoidosis for a number of reasons: the existence of a heterogeneous phenotype with potentially distinct genetic associations; the observation of clear ethnic-specific patterns of organ involvement; and the possibility that sarcoidosis has more than one cause. Indeed, many believe that sarcoidosis represents a family of separate entities (sarcoidoses) caused by different aetiologies. In this latter case, a myriad of rare risk alleles could potentially be involved in disease pathogenesis by interacting with environmental factors.92 This would also contribute to explain why so many studies have come to conflicting results.
Predicting susceptibility to sarcoidosis and the course that the disease will take by means of a simple genetic test is an attractive idea with potentially tremendous implications for future healthcare delivery. Such a test would allow offering closer surveillance and medical attention to those patients likely to develop aggressive disease as well as avoiding overtreatment in those more likely to remit. In addition, identifying functionally relevant gene variants that influence susceptibility to or manifestations of sarcoidosis would allow the development of targeted therapies. Should a respiratory physician consider genetic testing in sarcoidosis? Although the answer remains no, we believe that the time is approaching when the use of genetic analysis will be practical and clinically relevant.93 In this regard, in highly specialised centres (eg, Karolinska Institute and Karolinska University Hospital, Sweden), patients investigated for sarcoidosis are routinely genotyped for HLA-DRB1/DQB1 alleles as the results help in predicting disease behaviour with clear implications for treatment and monitoring of patients.
The development of sarcoidosis seems to depend on exposure to certain—yet to be identified—antigen/s in a genetically susceptible host. We thus need a better understanding of how risk alleles interact with the environment in causing the disease, as well as how this may influence its differential phenotypic expression. In other inflammatory diseases such as rheumatoid arthritis, a very strong synergistic effect on the disease risk has been shown for patients who are smokers and also have a certain HLA background. This observation has resulted in a new hypothesis on the development of the disease, where smoking is believed to result in post-translational modifications of self-proteins, leading to immune responses against self-proteins in a HLA-dependent manner and autoimmunity.94 It is thus imperative that meticulous databases of phenotypically well-defined sarcoidosis patients continued to be created and extended and that such databases include careful information on environmental exposure.
The time has come for us to start thinking beyond the significance of individual genes/loci and to include information on patient phenotype, environmental exposure and gene–gene interactions in order to eventually unravel the sarcoidosis pathobiology, which has proven to be far more complicated than previously thought.
The strong associations between HLA class II and sarcoidosis, together with detailed studies of the host immune response, suggest the presence of several specific antigenic triggers in sarcoidosis. Because antigens are presented in the form of peptides in the context of HLA molecules on antigen presenting cells to CD4 T cells, thus generating specific immune responses, the challenge is now to identify what peptides are presented by the HLA class II molecules. Novel methodologies are starting to shed light on the repertoire of HLA-bound peptides, which conceivably result from both genetic background (eg, HLA class II alleles) and exposure/s to yet unknown environmental factors.95 In turn, mapping of such peptide repertoire may help identify novel sarcoidosis-associated antigens, which may differ across distinct patient subsets. Advancing our understanding of the events transpiring at the interface between the host's immune system and environment is likely to also provide the key to understanding the pathogenesis of sarcoidosis, thus leading to the development of more effective therapies. Ideally, treatment could become personalised according to specific sarcoidosis subtypes. One reason for the limited number of studies on genetic associations with specific disease phenotypes is the difficulty in recruiting large cohorts of patients. Collaborative studies of many institutions, biobanks and careful patient phenotyping will hopefully overcome this issue.
In the last decade, the development of newer genetic technologies has allowed covering close to the entire genome. However, because sarcoidosis is a highly heterogeneous disease, careful clinical characterisation remains critical. In the future, with patients subgrouped according to clinical features, we are likely to obtain stronger and more precise genetic associations. We have incomplete knowledge of gene–gene and gene–environment interactions. Future studies should also address these deficiencies. It is hoped that such studies will eventually identify markers to predict disease development and behaviour and improve our understanding of the pathogenesis of this fascinating disease.
Contributors PS conceived of the study and drafted the manuscript. JG drafted the manuscript and revised it for important intellectual content. Both authors read and approved the final manuscript.
Funding This work was supported by the Swedish Heart-Lung Foundation grant number 20100254, the Swedish Research Council grant number K2013-57X-14182-12-3 and through the regional agreement on medical training and clinical research (ALF) between Stockholm County Council and Karolinska Institutet grant number 20120025.
Competing interests None.
Provenance and peer review Not commissioned; externally peer reviewed.
Data sharing statement This is not an original research article and does not include unpublished data.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.