Epigenetic changes differentiate between endogenous ageing and photoageing in skin and can detect sun-protection factor cream effects in a 5-month interventional study

Abstract

Epigenetics is a rapidly expanding field in ageing research. This is partly because changes in DNA methylation are excellent indicators of overall tissue ageing. Here, in a cross-sectional observational study, methylation levels were compared in upper-inner (sun-protected) and lower-outer (sun-exposed) arm skin biopsies of young (21–30 years, n = 24) and older (45–65 years, n = 24) participants of either European or Chinese ancestry. In young sun-protected skin, there were 69 significantly differentially methylated sites (DMSs) between those of Chinese and European ancestry. In a comparison between young and older sun-protected skin (endogenous ageing) there were 3545 pan-ethnic DMSs with a preference for hypermethylation and enrichment at the RNF32, FOXG1, NEFM, and ZIC1 gene loci. For photoageing, 2208 pan-ethnic photoageing DMS were identified with a preference for hypomethylation and enrichment at the HOXB1-3, TWIST1, and TBX20 gene loci. From the photoageing DMSs, 156 CpG sites were selected that consistently differentiated photo-exposed from protected skin even in the young participants. The mean degree of change for these 156 CpG sites was subsequently measured over 5-months in an interventional study on participants’ lower outer-arm skin (n = 42). Participants had regular sun-protection factor (SPF) use on one arm and normal free-living conditions on the contralateral arm. The free-living arm demonstrated greater skin darkening and a significantly larger mean change in methylation compared with the SPF-treated arm. Overall, the results highlight endogenous and photoageing methylation differences in skin and suggest that specific DNA methylation sites can track short-term sun-exposure and its attenuation by SPF use.

Graphical Abstract

Graphical Abstract.

Introduction

With advancing age, a plethora of changes to DNA methylation levels occur in all tissues. Although useful for predicting the age of DNA samples for forensic purposes [1], the change in DNA methylation levels at particular CpG sites have also been demonstrated to be good markers of biological age (how well an individual’s tissues are ageing considering their chronological age). For example, epigenetic ages generated from blood sample DNA methylation data were demonstrated to predict survival in 13 different cohorts [2]. Although some of this survival prediction could be linked to changes with age in the proportions of different types of white blood cells, it also remained predictive whilst accounting for such cell-type admixture differences. This data indicates that there are methylation changes occurring at particular CpG sites within certain cell-types that reflect the degree of cellular ageing and will help give insights into the biological consequences of ageing to immune cells.

A few studies have investigated whole genome changes to DNA methylation levels in skin with age. Vandiver et al. [3] studied changes in epidermal and dermal samples and found that methylation ageing changes were more prevalent in epidermal rather than dermal samples, and that photoageing was mainly associated with large regions of hypomethylation (loss of DNA methylation across neighbouring CpG sites). In contrast to photoageing, they also found that hypermethylation was a more common change occurring in sun-protected sites (herein termed endogenous skin ageing). In support of this, Gronniger et al. [4] also found that photoageing was associated with a more prominent loss of methylation at CpG sites with age whereas endogenous ageing had a slight tendency for CpG sites to gain methylation with age. Vandiver et al. [3] also noted blocks of hypomethylation in colon cancer that overlapped with photoageing hypomethylated blocks, indicating that skin hypomethylation changes are not unique to skin. Indeed, in a study of methylation changes with age in blood [5], there was significant enrichment with ageing methylation changes in skin.

There are very few studies looking at epigenetic differences in skin across ethnicities. Winnefeld et al. [6] looked at skin ethnicity differences across participants with European, Asian, and African ancestry for 27 578 methylation sites, and found a number of CpG sites (n = 205) that were differentially methylated. Some of these sites that associated with ethnicity also changed with age in sun-exposed sites. This highlights difficulties in differentiating between early developmental driven and adult-lifestyle driven ethnicity differences in adults. It is also unclear whether there are many differences or similarities in the way epigenetic markers change with age across different ethnic groups.

Here, genome-wide DNA methylation levels were measured in sun-protected and sun-exposed skin samples from 48 participants of either Chinese or European ancestry in the USA. Comparisons were made between ethnicities in young sun-protected skin, and between how methylation levels change with age in sun-protected and sun-exposed skin samples across the two ethnicities. Using these findings, 156 CpG sites that associated with photoageing were selected a priori to study how, in 42 participants, their DNA methylation levels changed over 5-months and if sun-protection factor (SPF) use could impact any such changes over the same period.

Results

Participant characteristics

The participants in the observational study were from two different age groups (21–30 and 45–65 years) and two different ethnicities (Supplementary Table 1). Participants with Chinese ancestry were less likely to be smokers and avoided the sun more, especially in the older age group. For the interventional study, 69% of participants identified as having European ancestry and 31% of participants identifying as Hispanic or Latino—for simplicity, this group is herein referred to as of European ancestry; participants were aged between 24 and 58 years (Supplementary Table 1).

Photoageing versus endogenous ageing DMSs in skin

There were 3545 significantly differentially methylated sites (DMSs) with a >5% delta-beta between young and older participants in upper inner arm skin (i.e. endogenous ageing) (Table 1). For the lower outer arm, there were 10 649 DMSs that had a >5% delta-beta between young and older samples. Of these, 2208 were also significant DMSs with a >5% delta-beta between protected and exposed sites in the older samples (i.e. photoageing). These sites had a concordance in whether they had a hypermethylation or hypomethylation change in the two comparisons—i.e. if a DMS was hypermethylated in the exposed–protected comparison then they were also hypermethylated in the older–younger lower outer arm sample comparison; likewise, for hypomethylated DMSs (Table 1). There were 391 sites in common between the endogenous and photoageing DMS lists, all changing with age in the same direction/manner.

Table 1.

Number of methylation sites that associate with endogenous and photo-ageing differences as well as ethnicity differences in young subjects.

Comparison		Direction of change	Significant at P-value 0.05 (FDR) and 5% delta beta	Significant at P-value 0.05 (FDR) and 10% delta beta
Endogenous	Young V Older protected sites	Hyper with age	1998	35
		Hypo with age	1547	56
Photo-ageing	Young V Older exposed sites & Protected V Exposed	Hyper with age	765	76a
		Hypo with age	1443	149a
Ancestral	Chinese V Caucasian young protected sites	Hyper in Chinese	39	19
		Hypo with Chinese	30	19

^aTen percent filter applies to young to older delta betas only for comparability to endogenous changes.

To determine whether cell-type changes with age could be having an influence on the results, cell-type deconvolution analysis was performed on the DNA methylation data which identified an increase in the proportion of fibroblasts and a decrease in the proportion of keratinocytes with age and similarly between exposed and protected sites [false discovery rate (FDR) < 0.05 and > 2% cell-type proportion change, (Supplementary Fig. 5)]. In a Spearman’s rho (p) correlation analysis with more homogenous cell-type skin methylation data, the photoageing delta-beta changes were highly significantly correlated (P = 0.87, P < 0.0001) with their delta-beta changes with age in an epidermal photo-exposed arm dataset [7] (n = 59). Furthermore, the endogenous changes were also highly significantly correlated (P = 0.63, P < 0.0001) with their delta-beta changes in an endogenous in-vitro keratinocyte basal cell dataset [8] (n = 9).

There were no significant differences between ethnicities in their endogenous or photoageing changes (P < 0.05 FDR), although the photoageing DMSs tended to be more pronounced in the European ancestral participants, likely reflecting the greater sun-seeking behaviour (Supplemental Fig. 2a and b). In young sun-protected skin, there were 69 significant DMSs (P < 0.05 FDR) between those of Chinese and European ancestry, including multiple DMSs located in or near the WD90 (n = 4) and IL6 (n = 3) gene loci (Supplemental Table 2). Of the 69, 3 were present and significantly different between ethnicities in a previous methylation study of inner forearms (epidermal suction blister samples) [6].

Next, the endogenous DMSs were used to identify differentially methylated regions (DMRs), and these were found around the gene loci RNF32, FOXG1, NEFM, and ZIC1 (Table 2a). In the case of photoageing DMSs, significant DMRs were found around the gene loci HOXB1-3, TWIST1, and TBX20 (Table 2b).

Table 2.

DMRs by either endogenous (a) or photo-ageing (b).

Chromosome	Start	End	Width (bp)	Number of CpGs	Example CpG	Average delta-beta	Stouffer Z-score	Lowest CpG P-value	Nearest protein coding gene
a: Endogenous ageing
7	156 432 526	156 433 669	1144	13	cg00292135	0.011	<1*10−61	6.38E−56	RNF32
14	29 234 671	29237480	2810	27	cg02991338	0.035	2.06E−61	6.49E−131	FOXG1
8	24 770 342	24 773 148	2807	20	cg03169018	0.037	3.56E−56	4.49E−133	NEFM
3	147 125 287	147 128 157	2871	34	cg01227537	0.036	1.74E−46	1.95E−80	ZIC1
7	27 280 585	27 282 444	1860	23	cg04008703	0.027	6.38E−36	3.02E−66	EVX1
12	2 800 055	2 801 584	1530	14	cg00848945	0.041	1.24E−35	9.15E−89	CACNA1C
19	58 220 080	58 220 837	758	11	cg01268824	0.033	2.76E−35	1.21E−85	ZNF154
7	35 292 896	35 294 658	1763	15	cg08959216	0.023	4.11E−34	5.88E−94	TBX20
5	134 363 053	134 364 387	1335	9	cg00396667	0.058	9.52E−32	2.99E−48	PITX1
5	178 367 423	178 368 620	1198	12	cg03234732	0.024	1.13E−31	3.13E−81	ZNF454
b: Photoageing
17	46 607 828	46 608 874	1047	13	cg02497558	0.078	1.019E−49	1.3051E−68	HOXB1
17	46 656 543	46 658 170	1628	9	cg02132714	0.033	5.171E−36	3.9728E−47	HOXB3
17	46 628 224	46 629 804	1581	10	cg11085454	0.043	1.684E−35	2.2365E−55	HOXB3
7	19 152 017	19 152 335	319	4	cg24496841	0.018	6.883E−31	5.5368E−40	TWIST1
7	35 293 080	35 294 658	1579	14	cg08959216	0.024	5.564E−27	4.746E−47	TBX20
18	44 773 985	44 774 923	939	5	cg20192747	0.064	3.135E−26	3.1524E−29	SKOR2
22	19 754 125	19 754 815	691	4	cg01697719	0.024	8.975E−26	1.7174E−35	TBX1
5	92 929 319	92 930 052	734	5	cg17774851	0.063	5.089E−25	5.8345E−36	NR2F1
11	20 043 971	20 044 639	669	5	cg02404377	0.047	6.067E−22	1.0094E−31	NAV2
22	19 746 984	19 747 140	157	4	cg09574757	0.018	3.468E−21	1.3175E−31	TBX1

Interventional study

In a 5-month free-living and SPF interventional arm study, the SPF treated arm was found to be significantly lighter than the free-living arm at every monthly timepoint post baseline, indicating greater protection from UV exposure (Supplementary Fig. 3).

To determine whether the increased sun-exposure in the free-living arm impacted DNA methylation, 156 of the photoageing DMSs from the observational study were identified a priori that differed between sun-exposed and sun-protected skin in the younger participants as well as behaved in exactly the same manner in an independent observational study [3] (Supplementary Methods and CpG sites are listed in Supplementary Table 3). These criteria were intended to enrich for sites that show evidence of sun-exposure changes in younger individuals and therefore sensitive to shorter term-sun-exposure as opposed to solely decades of photoageing. Both arms had a similar proportion of the 156 sites that changed over the 5-months in the same direction (hypo- or hypermethylated) as the methylation difference between exposed and protected skin (free-living arm n = 134, 87% concordance; SPF arm n = 132, 85% concordance). For example, of the sites that were hypermethylated in exposed relative to protected skin, the majority were hypermethylated over the 5-months. The absolute degree of change for the 156 probes was significantly greater for the free-living arms than the SPF arms (difference in means of delta M-values of 0.0073, P = 0.0055) (Fig. 1), in keeping with a photoageing protective effect of the SPF treatment.

Figure 1.

Boxplot of mean change in absolute M-value per DMS by arm. The change in the M-values (mean absolute change in M-value from baseline) for each of the 156 probes in the free-living untreated and SPF treated arms are plotted; values for individual CpG sites are shown as points, with those for the same CpG site in the different arms connected by lines. If the line goes upwards from left to right, it represents a greater methylation change in the untreated arms whereas a greater change in the SPF arms is represented by a line that goes downwards from left to right. Overall, the free-living changes were significantly greater than the SPF changes (P < 0.0055).

To validate that the 156 CpG sites selected a priori were more sensitive to photoageing than other CpG sites, permutation testing was carried out. This tested whether the observed methylation difference between arms was more marked for the preselected 156 CpG sites compared to other CpG sites across the genome. Specifically, a null distribution of 10 000 random sets of 156 CpG methylation sites were generated and the mean difference in M-values between the free-living and SPF-treated arms was generated. Negative mean values represented a greater change over the 5-months in the free-living arm whereas positive mean values represented a greater change in the SPF arm. The delta-M difference between the SPF and free-living arm was significantly lower for the preselected 156 photoageing set (delta M-value = −0.0073) than the null distribution average (delta M-value = −0.0019) from the 10 000 permutation tests, P = 0.03, and see Supplementary Fig. 4. In other words, out of 10 000 randomized sets of 156 methylation sites, only 300 had an equal or greater methylation change in the free-living arm compared to the SPF arm.

Discussion

Through an observational DNA methylation study, endogenous skin ageing was found to be mainly associated with hypermethylation changes in skin, whereas photoageing was mainly associated with hypomethylation changes. Within the photoageing DMSs, 156 were identified that not only changed with age but were already different between sun-exposed and sun-protected skin in the young participants. The mean change in methylation levels of these 156 sites significantly differentiated between 5-months of free-living skin and SPF protected skin in an interventional study. Furthermore, in upper-inner arm skin, 69 significantly DMSs between young participants of Chinese descent and young participants of European descent were found.

One of the most potent drivers of skin ageing for those of European or East Asian ancestry is sun-exposure. However, it is difficult to differentiate between photoageing and endogenous ageing effects as both occur in sun-exposed skin. Here, photoageing changes were identified from endogenous changes by comparing not only DNA methylation changes in young and older sun-exposed skin, but also by comparing exposed to protected skin. Although there were many more DMSs changing in sun-exposed skin compared to protected, once an additional filter for exposed/protected differences was applied, there were similar numbers of DMSs that could be attributable more to photoageing than endogenous ageing. Around 18% of the photoageing sites overlapped with the endogenous sites indicating some commonality between these two skin ageing phenomena. However, photoageing had more hypomethylated DMSs, and endogenous ageing more hypermethylated DMSs, which has also been noted by others [4]. The biological mechanisms that drive this difference in hypermethylation versus hypomethylation in skin with age remains unknown, and further studies, such as direct UV radiation effects on these DMSs, are required.

Methylation changes with age can be due to changes to the proportion of cell-types in tissues as well as methylation changes within cell-types [2]. Although cell-type proportions were not measured here, a skin cell-type deconvolution method was used to estimate cell-type proportions in the observational data. More fibroblasts and less keratinocytes were estimated in the older skin samples suggestive that these cell-type changes could be influencing the results. The endogenous and photoageing DMSs, though, were also found to be changing in a similar manner in an epidermal dataset [7], and the epidermis is estimated to be composed of around 90%–95% keratinocytes [9]. This finding indicates that many of the endogenous and photoageing findings are occurring within keratinocytes although it also raises the possibility that a change to the proportion of basal keratinocytes to differentiated keratinocytes could be influencing the results. However, prior in-vitro comparisons of keratinocyte subtypes have reported relatively modest methylation differences between these populations [10], and the endogenous changes were also found to be highly correlated to ageing changes in an in-vitro keratinocyte basal-cell dataset [8]. Hence, although cell-type changes with age (including stem-cell to differentiated-cell proportions) could be influencing the results described here, a lot of the changes could also be detected within epidermal and keratinocyte basal-cell datasets highlighting that intra cell-type changes, notably in keratinocytes, were also present in the results. Future observational studies using single cell methylation techniques will enable cell-type proportion changes versus intra cell-type changes to be fully distinguished.

For the endogenous and photoageing methylation changes, there was enrichment in genes associated with development (e.g. RNF32, FOXG1, NEFM, and ZIC1 for endogenous ageing; HOXB genes, TWIST1, and TBX20 for photoageing) and not necessarily associated with skin ageing (e.g. a PubMed literature search only returned a total of 3 hits for skin ageing with any of these genes). The association of development loci with methylation changes is in keeping with enriched ageing loci in other tissues [11]; e.g. ZIC1, HOX genes A, C, and D, and TBX3/5 loci have been identified as DMRs changing with age in blood [5]. This suggests there might be a ‘disconnect’ for changes at the methylation level and nearby genes (i.e. in-cis). This was the case for methylation and RNA changes due to keratinocyte differentiation [10], and links between ageing methylation changes and gene expression in-trans have been found in blood [5]. In order to better understand the biological networks associated with the identified methylation changes, future studies should focus on regulatory networks disrupted by the ageing process that are driving the DMSs, such as changes in transcription factor binding to the DMSs loci.

Perhaps surprisingly, there were no significant young to older differences between ethnicities, although this could be down to a lack of statistical power when using interaction effects in the models used here. However, in principal component analysis (PCA) plots it was notable that the participants with European ancestry had greater photoageing changes than those of Chinese ancestry (Supplemental Fig. 2a and b). It is difficult to distinguish whether this observation is due to a greater susceptibility to sun-damage or greater sun-seeking behaviour as evident in the self-reported lifestyle questionnaires. A small number of DMSs, though, were significantly differentially methylated between those of Chinese and European ancestry in young sun-protected skin. This suggests that these DMSs are linked to intrinsic differences in skin physiology between ethnicities, since they were identified from young sun-protected sites. Although methylation sites that overlapped with common single nucleotide variations were removed from the analysis, it is possible that none-overlapping genotype variations between the two ancestral groups were influencing nearby methylation levels rather than these differences being driven by in-utero and/or childhood environmental effects. Three of these ethnicity DMSs were present in a previous inner forearm ethnicity study [6], and all were also significantly different by ethnicity in that study indicating these are robust ethnicity differences. Of the gene loci that were differentially methylated, WDR90 has an unknown function although it is expressed in skin (e.g. as per the GTEx database). IL6, in contrast, is a well-known inflammatory cytokine, which is upregulated in response to UV radiation [12] and, in keratinocytes, in response to irritants [13]. It is possible that such differences in methylation levels could be linked to differences in skin sensitivity to UV and irritants between ethnicities [14]. However, linking these methylation differences to, e.g. IL6 protein response to UV irradiation is required to determine whether these epigenetic differences relate to known ethnicity differences in skin biological responses.

There were 156 significantly DMSs between young and older sun-exposed samples that were also significantly different between sun-exposed and sun-protected samples from the young subjects in the observational study, as well as in exactly the same comparisons in a second previously published dataset [3]. In the interventional study, this a priori selected set of 156 methylation sites changed in the expected direction over 5-months in the free-living arm and, to a lesser extent, in the SPF-treated arm. This supports the hypothesis that these methylation sites are sensitive to photoageing over 5-months of sun-exposure as well years of sun-exposure.

To confirm these specific preselected CpG sites were particularly good at reflecting sun-exposure over the 5-months, they were compared to other randomly chosen sets of 156 CpG sites from across the genome. The preselected 156 set had a significantly greater change in the free-living arm than the SPF arm compared to the average from over 10 000 randomly selected sets. This result confirms that this preselected set of CpG sites are better than most other CpG sites at reflecting sun-exposure over months as well as years, and for testing the effectiveness of anti-ageing regimes/products. Along with a replication of these results, however, investigation of whether these sites associate with known photodamage effects (e.g. DNA damage, collagen cross-linking etc) is recommended before they are considered accurate biomarkers of skin photoageing. It is also not clear if these methylation changes revert back to pre-summer levels over the following winter months or maintain the changes over future years. Further studies of such dynamic changes over longer periods would help understand how changes, mainly found across decades of ageing, build over the years within tissues.

The main weaknesses to these findings are that they are associative in nature and might not have any causal relationship with skin ageing. Furthermore, there was no SPF placebo control in the interventional study meaning that the 5-month SPF product changes cannot be solely attributable to sun-exposure, as it is not clear what effect the SPF product ingredients might have had beyond the SPF effect. In addition, links to physiological markers of ageing are now required to further validate the utility of these markers. The interventional study ran from summer into the autumn and as suggested by the melanin index changes, this limited the sun-exposure of the participant arms. As the reduced sun-exposure should have limited any free-living versus SPF differences, it suggests that greater differences may have been evident if the study was run in-totality over peak summer months.

In summary, we have carried out extensive analyses of DMSs in skin from different skin sites, age groups and ethnicities. We have identified DMSs that associate with either photoageing or endogenous skin ageing and identified a photoageing specific set of methylation sites that change as expected over 5-months of free-living and to a significantly lesser extent in SPF treated skin. Follow-up studies to investigate mRNA and protein associations with the DMSs identified here are now required to ascertain a greater biological understanding of these epigenetic changes and their link to physiological ageing characteristics.

Materials and methods

Study design and sampling

The young–older observational study was an IRB approved single centre, cross-sectional biopsy study of 48 female US nationals, half of Chinese and half of European descent (all grandparents had to have ancestry primarily from these geographies). All participants gave written informed consent and completed a health history, eligibility, and lifestyle questionnaire. Skin biopsies (3 mm) were taken from the lower-outer arm and the upper-inner arm.

The interventional study was an IRB approved, five-month split arm exploratory study with one product treatment conducted in the USA between June and December. All participants gave written informed consent and were in general good health, reported they go outside for greater than 10 h per week and normally had exposed forearms when going out during warmer months. A sunscreen product with a broad-spectrum SPF (SPF 30) was applied at least twice daily to one randomly pre-selected forearm and participants requested to carry on doing what they normally do (‘free-living’) with the other forearm. Three millimetres skin biopsies were taken from both forearms at baseline and after 5-months. Spectrophotometer CM2600D (Konica Minolta) assessments were obtained monthly on the arm; melanin index was created from spectrophotometer readings as per Stamatas et al. [15]. The study completed with 42 female participants with a Fitzpatrick Skin Type of II-III identifying as primarily Hispanic/Latino or European ancestry.

DNA methylation measurements

DNA was extracted from whole skin punches using first phenol chloroform then a Qiagen DNA extraction Kit as per the manufacturer’s instructions. Interrogation of the DNA samples were performed by Expression Analysis, Inc. (Durham, NC, USA), using Illumina’s Infinium Methylation assay and run on Infinium Human Methylation 450 Bead Kit (for the observational study) and the EPIC array (for the interventional study) as per the manufacturer’s instructions. Gene annotations for the methylation sites were provided by Hansen 2016 [16] and 2017 [17].

Data analyses

Instrumental measures

For the interventional study, analysis of the melanin data was done by a linear model analysing change from baseline by month, including baseline values as a covariate, treatment as a fixed effect, and subject as a random effect.

Methylation data analyses

Beta and M values from the Illumina chips were calculated (Supplemental Methods) and raw data normalized using SWAN and quantile normalization; the M-values were used for statistical modelling whereas beta-values were used to filter the statistical results for degree of differences between groups—the delta-beta values.

For the observational study, smoking, menopausal status and holiday sun-exposure in the past year had significant associations with PCA of the data and, hence, were used as covariates in the statistical models where appropriate (Supplemental Methods). To differentiate between endogenous and photoageing changes, we used comparisons contrasting young and older participants as well as protected and exposed sites (Supplementary Fig. 1). To detect differences in ageing between ethnicities, we included an ethnicity*young/older interaction effect (Supplemental Methods).

To examine influences of cell-type changes on the observational results, two types of analysis were performed. First, differences in cell-types between young and old and protected exposed samples were computed using the cell-type deconvolution method EpiDISH [18] and applying a skin-specific reference matrix [19] (also see Supplementary Methods). Second, Spearman’s correlation analysis rho (p) was performed between the significant endogenous delta-beta changes and their delta-beta changes in an epidermal (lower outer arm) ageing DNA methylation dataset [7] as well as an endogenous (abdomen sourced) keratinocyte ageing dataset [8], n = 3531 and n = 3240 sites overlapped between datasets, respectively. The photoageing DMS delta-beta changes were compared to their delta-beta changes in the epidermal dataset, n = 2205 sites overlapped between the two datasets.

For the interventional study, one individual was missing methylation data for one of the four sampling points and as a result all observations for this participant were excluded from the datasets. Batch correction was performed using ComBat [20] to correct for the variation in the data attributable to the EPIC arrays, due to its association in PCAs.

Differential methylated regions

To determine whether particular regions of the genome were enriched in the methylation lists, we used DMRcate [21]. The t-statistics from the differential methylation analysis of the CpG sites were smoothed over a 1000 bp bandwidth using a Gaussian kernel smoothing method. The smoothed distribution of t-statistics was then used to compute new P-values for each CpG. These P-values were adjusted for FDR and the CpGs reaching statistical significance (FDR-adjusted P-values <0.05) within a 1000 bp interval were agglomerated into a DMR. The FDR-adjusted P-values of the CpG in a DMR were summarized using Stouffer’s Z transformation.

156 sun-specific methylation site changes

The observational study data was used to identify sun-specific methylation changes in the observational study and in an epidermal dataset (Vandiver et al. [3]). Methylation sites had to change in the same direction from young to older (sun-exposed sites) and from protected to exposed skin in the young groups for both studies (Supplementary Methods) to help identify sites likely changing over months rather than just decades. In addition, sites were excluded if not on both the 450K and EPIC arrays, and some sites were excluded if <50 Kb apart, leaving 156 sites that passed filtering criteria.

The absolute change in M-value over the 5-months (delta-M) for each of the 156 sites was calculated per participant, separately for the free-living and SPF treated arms. A statistical comparison between the two arms was performed using a Wilcoxon signed-rank test in JMP 14.1.0., as the delta-M values violated a Goodness-of-fit test for a normal distribution. These 156 sites were selected from whole genome data as the best biomarkers of photoageing. To test how good this set of sites was at detecting photoageing, the absolute methylation change (delta-M value) for the free-living arm was subtracted from the absolute delta-M value change in the SPF protected arm over the 5-months for this preselected set of 156 sites. This value was then compared to the same comparison for randomly chosen sets (10 000 sets) of 156 CpG sites. First, a null distribution for the mean delta-M differences between the free-living and SPF arms for the 10 000 sets was constructed. Then, the mean delta-M difference between the arms for the preselected 156 set was evaluated against this null distribution of means and its average. The empirical P-value was defined as the proportion of the 10 000 random sets with a mean difference less-than or equal-to the observed value. In this case, a negative value (less than) would mean a greater methylation change in the free-living arm relative to the SPF arm, as would be expected from greater sun-exposure in the free-living arm.

Author contributions

Ashley Furtak (Data curation [equal], Formal analysis [supporting], Investigation [equal], Methodology [equal], Writing – original draft [supporting]), Laura Ramos del-Cano (Formal Analysis [equal], Writing – review & editing [equal]), Jeremy Shen (Formal analysis [equal], Writing – review & editing [equal]), Arthika Bappal (Formal analysis [equal], Writing – review & editing [equal]), Tony Dadd (Conceptualization [equal], Formal analysis [equal], Writing – review & editing [equal]), Fei Ling Lim (Data curation [equal], Formal analysis [equal], Resources [supporting], Writing – review & editing [equal]), Sheila Rocha (Conceptualization [equal], Funding acquisition [equal], Investigation [equal], Project administration [equal], Resources [equal], Supervision [equal], Writing – review & editing [equal]), and David Gunn (Conceptualization [equal], Formal analysis [equal], Funding acquisition [equal], Investigation [equal], Methodology [equal], Project administration [equal], Resources [equal], Supervision [equal], Visualization [equal], Writing – original draft [lead])

Conflicts of interest

The SPF product is part of a Unilever brand portfolio and some of the methylation site changes are subject to Unilever patents.

Funding

Funding was provided by Unilever PLC.

Data availability

DNA methylation data for the cross-sectional observational study is freely available at ArrayExpress under E-MTAB-14619 and for the interventional study under E-MTAB-14620.