In this large group of individuals without diabetes of Western European descent, we estimated the heritability of SAF to be 33%, taking into account the known (in part environmental) factors which may influence SAF levels. We identified four novel loci with 5 SNPs, one on chromosomes 11 and 15, two on chromosome 16, and one locus on chromosome 8 near the previously identified locus NAT2, that were associated with SAF.
Most of the studies on SAF have reported on environmental factors associated with this measurement [31, 33]. SAF increases with ageing, and is elevated in people with type 2 diabetes compared with age-matched controls [11, 34]. It was demonstrated that SAF is already elevated in people without diabetes but with metabolic syndrome, and is associated with some of its individual components [35]. Furthermore, SAF is strongly related to current smoking status as well as smoking history, coffee consumption, and renal function [36]. In the general population SAF is a strong predictor of the development of cardiovascular disease and mortality [37]. In addition, higher SAF levels predicted a higher risk of development of type 2 diabetes, and algorithms have been developed to predict individuals with the highest risk of developing type 2 diabetes based on age, BMI, and SAF levels [37]. SAF reflects fluorescent properties present in skin and skin collagen half-life is about 15 years [38]. As a result, SAF is a marker for cumulative accumulation over a much longer period of time than for instance haemoglobin [39]. Compared with excreted molecules or markers, SAF may therefore be a more appropriate tool for screening for diseases that develop over time.
Because of the growing interest to use SAF to screen for type 2 diabetes, diabetes complication risks, cardiovascular disease and mortality [12, 37], more information on possible genetic determinants of SAF is warranted, especially because studies of serum AGEs have estimated heritability of 63–74% [14, 15]. Understanding the genetics of a quantitative trait can improve performance of the tool. Unfortunately, there are only a few genetic studies of SAF [16, 17].
Our previous study reported the genetic association between a SNP in NAT2 and SAF, in a first set of approximately 9000 participants of the Lifelines Cohort Study, that overlaps partly with those reported in the CytoSNP cohort here [16]. We also reported the association of a locus on chromosome 1 and skin intrinsic fluorescence, but this association was only observed in individuals with type 1 diabetes [14]. The present study has combined the earlier dataset of the Lifelines Cohort Study, and added a new set of over 17,000 individuals genotyped recently on the Illumina Infinium Global Screening Array-24 (GSA) version 1.
In meta-GWAS, we identified five additional loci which are associated with SAF in individuals without diabetes.
The association of the locus on chromosome 16 was the most significantly associated with SAF after NAT2. rs12931267 was significantly associated in all GWAS models, however the SNP-effect was slightly attenuated in model 4 where SR was added as a covariate. The SNP is near MC1R, in which multiple independent coding variants have been associated with skin colour and pigmentation traits [40, 41]. In GWAS of SR measures from the same individuals we identified multiple genome wide significant signals in the MC1R region as well. Through internal software in the AGE Reader, SAF is adjusted for variation in skin colour by SR [42]. However, in multivariable models SR and SAF are still correlated. This association is complex and requires further investigation.
On chromosome 11, rs2846707 which results in a Met30Val missense variant in MMP27 exon 1 was associated with SAF. The pathologic significance of the association could be by the altering of an amino acid, however the SNP is also a cis-eQTL and associates with the expression level of several nearby genes. It is not clear whether there is allelic imbalance as the functional impact of this specific variant in MMP27 has not been investigated. Previous research did demonstrate an interplay between matrix metalloproteinases (MMPs) and AGEs, specifically because of the known collagen degrading activity of MMPs [43–45]. There was no association of this SNP with SR. In PheWeb, comprising the genetic (TOPMed imputed) and clinical information from the White British participants of the UK Biobank, the trait with the smallest P-value for rs2846707 was ‘Diffuse diseases of connective tissue’ (p = 0.002) and the second most associated trait was ‘Diabetic retinopathy’ (p = 0.002) [46, 47].
The effect of rs2470893 on chromosome 15 was attenuated when coffee intake -an important environmental factor influencing SAF- was included as a covariate. It can be postulated that some SNPs which are associated with daily coffee consumption may also be associated with SAF levels, based on the fact that higher coffee consumption is associated with higher SAF [36, 48]. Indeed, in our post-hoc evaluation, we found that multiple SNPs associated with coffee consumption were also nominally associated with SAF with similar direction of the effect (additional File 12, Table S8). A variety of factors potentially resulting from the roasting process of coffee beans (Maillard Reaction) may be involved in the association of increased coffee consumption and higher SAF, examples are the loss of chlorogenic acid, the presence of melanoidins or caffeine [33, 36, 49]). The relative contribution of each of these fluorescent factors, and potentially others, on SAF is unknown.
We identified an additional independent locus in the NAT2 region that was associated with SAF, which underlines the robust association of NAT2 variants with SAF. Previous research identified that NAT2 is also associated with insulin resistance [50]. This shared association of insulin resistance as well as SAF with NAT2 may, in part, be explanatory for the predictive value of SAF for type 2 diabetes development.
We examined all SNPs in GTEx project and found that rs12931267, rs2846707, rs2470893 and rs3764257 are all cis-eQTLs for various nearby genes. However, if the SNP-effect on SAF is explained by the association of the SNP with gene-expression, requires further investigation. For rs1495741 and rs576201050 we found no cis-eQTLs effects in GTEx. This may be caused by low minor allele frequency of rs576201050, since cis-eQTLs are only described for variants with MAF > 1% [51]. Since rs1495741 is a tag-SNP for the acetylator phenotype which well explains the association with SAF, it is reasonable that the association is not mediated through cis-eQTL effects.
All newly identified loci explain a small percentage (below 1%) of the variance in SAF. In combination with the previously reported NAT2 SNP they explain less than 5% of the total SAF variation. However, insight into the genetic factors associated with SAF variation does provide more knowledge of the complex pathology of formation and accumulation of AGEs. As the estimated genetic heritability of SAF was 33%, the majority of genetic variants that explain SAF variability remain undiscovered. It seems reasonable that numerous SNPs each explain a small fraction of SAF variation, which could be identified by larger GWAS studies to increase power for detection of loci with even smaller effect sizes.
We also evaluated genome-wide signals for skin reflectance, an additional measure from the AGE Reader. SR is important for the proper measurement of SAF. Using the current AGE Reader only SAF values can be reliably measured in combination with SR measurements above 6%. As mentioned, SR is a proxy for skin colour or pigmentary phenotype. All genetic regions associated with SR have indeed earlier been shown to be associated with skin colour traits and perceived age, including SLC45A2, IRF4, BNC2, TYR, HERC2, MC1R and RALY/ASIP [41, 52]. Furthermore, variants in the MC1R region have been shown to be strongly associated with higher susceptibility of cutaneous melanoma [53, 54].
Limitations of the current study are the fact that we used two different genotyping arrays on different subjects and we used HRC imputed data while the TOPMed panel comprises a larger reference panel and results in more imputed SNPs as well as better quality imputation [46]. However, for imputation we are dependent on the Lifelines research group who have not yet released the TOPMed imputed Lifelines data. Another limitation is that we only analyzed autosomal SNPs while there is a considerable possibility that some of the SAF variance is explained by variation on the sex chromosomes. Finally, our study comprises only white individuals from European descent and the results are therefore not generalizable to the non-white population.