Patent Application
20210207201
This invention relates to methods of detecting and analysing patterns of cytosine methylation in genomic DNA. More specifically, it relates to detecting and analysing patterns of cytosine methylation in specific sites in genomic DNA in order to determine the intrinsic age and health of skin.
It is well known that ageing is a multifactorial process predominantly driven by the age of the individual. Skin ageing in an especially multifactorial phenomenon driven by both intrinsic and extrinsic factors. In terms of intrinsic factors, the chronological age of an individual is the most well-known but other intrinsic factors such as an individual's metabolism, diet, stress and underlying health also contribute to the age if the skin. In addition to these intrinsic factors, the skin is exposed to external challenges such as UV radiation, pollution, drying conditions and extremes of temperature. These extrinsic factors therefore also contribute to the age on an individual's skin.
It is therefore clear that there are two distinct forms of skin age: Extrinsic age, which is dominated by the accumulation of ageing caused by extrinsic factors (i.e. originating from outside the exterior surface of the stratum corneum and that then penetrate into the skin through the stratum corneum), especially sun exposure (photo-ageing); and Intrinsic age, which is the degree of ageing in skin due to factors that originate endogenously; in other words ageing not due to extrinsic factors. For the sake of understanding, it is helpful to consider 2 different types of skin of an individual. One from a site normally protected by clothing (such as the buttock area or upper inner arm area). Another from a sun exposed site (such as the face or back of the hand). The protected site will have far less exposure to extrinsic aging factors and therefore any aging will be due to intrinsic factors. The exposed site will been fully exposed to extrinsic aging factors and therefore the age of this area aging will be due to a combination of both the inherent intrinsic age caused by the intrinsic factors but also the aging due to the extrinsic factors.
The present invention is directed towards the development of an epigenetic method to estimate the intrinsic age of an individual's skin.
DNA methylation is an epigenetic determinant of gene expression. Patterns of CpG methylation are heritable, tissue specific, and correlate with gene expression. The consequence of methylation, particularly if located in a gene promoter, is usually gene silencing. DNA methylation also correlates with other cellular processes including embryonic development, chromatin structure, genomic imprinting, somatic X-chromosome inactivation in females, inhibition of transcription and transposition of foreign DNA and timing of DNA replication. When a gene is highly methylated it is less likely to be expressed. Thus, the identification of sites in the genome containing 5-meC is important in understanding cell-type specific programs of gene expression and how gene expression profiles are altered during both normal development, ageing and diseases such as cancer. Mapping of DNA methylation patterns is important for understanding diverse biological processes such as the regulation of imprinted genes, X chromosome inactivation, and tumor suppressor gene silencing in human cancers.
Horvath S. et al “DNA methylation age of human tissues and cell types” (Genome Biology 14 (2103) R115) reports the use of a transformed version of chronological age that was regressed on CpGs using a penalized regression model (elastic net). The elastic net regression model selected 353 CpGs which were referred to as epigenetic clock CpGs since their weighted average (formed by the regression coefficients) was said to amount to an epigenetic clock. This study is referred to as the “Horvath Study” in this patent.
However, we have now found that for sun-exposed skin sites the predicted ages based on these 353 loci were approximately 9 years younger than their actual (“chronological”) age, indicating they do not detect sun-induced damage in skin. Additionally, sun-protected skin samples were found to have an age 4 years younger than the chronological age which is a underestimation of the age of the sun-protected skin which would be expected to be approximately the same as the chronological age of the subject that the sample was taken from. These 353 loci therefore fail to recognize the difference between photo-damaged and photo-protected skin types, underestimate the age of sun-protected skin, and predict photo-damaged skin as younger than photo-protected. It can therefore be appreciated that this model is not capable of assessing the different forms of aging—extrinsic and intrinsic ageing
The present invention therefore aims to address the poor performance of this prior art ageing model and to provide an improved method for evaluating the intrinsic age of skin.
We have surprisingly found that a different, specific set of methylation sites provide enhanced accuracy for the prediction of intrinsic skin age. In particular, the sites are capable of predicting the age of protected skin and are also capable of giving an intrinsic age for exposed skin that is surprisingly not influenced by extrinsic factors.
Accordingly, in a first aspect the invention provides a method for obtaining information useful to determine the intrinsic age of skin of an individual, the method comprising the steps of:
(a) obtaining genomic DNA from skin cells derived from the individual; and
(b) observing cytosine methylation of >30 CpG loci in the genomic DNA selected from the group consisting of:
so that information useful to determine the intrinsic age of the skin of the individual is obtained.
The genomic DNA is obtained from skin cells derived from the individual. The skin sample preferably comprises the epidermis, either alone or in combination with the dermis.
Preferably >40 sites from this group are used, more preferably >45, >50, >55, >60, >65, >70, >75, >80, >85, most preferably all 89 sites of this group are used.
Preferably the loci that are observed are:
More preferably the loci that are observed are:
In an alternative embodiment, the cytosine methylation in the genomic DNA is assessed wherein the genomic DNA is within 20 kBp of the CpG locus designation listed above, preferably within 15 kBp, more preferably within 10 kBp, yet more preferably within 5 kBp, even more preferably within 1 kBp, most preferably within 0.5 kBp.
In a second aspect, the invention provides a kit for obtaining information useful to determine the intrinsic age of the skin of an individual, the kit comprising:
and
Preferably the primers or probes are specific for >40 of the genomic DNA sequences in a biological sample, more preferably >45, >50, >55, >60, >65, >70, >75, >80, >85, most preferably the primers or probes are specific for all 89 sites of this group.
Preferably primers or probes are specific for genomic DNA sequences in a skin sample, most preferably a skin sample comprising the epidermis, either alone or in combination with the dermis.
Preferably the primers or probes are specific for the following CpG locus designations:
More preferably the primers or probes are specific for the following CpG locus designations:
In an alternative embodiment, the cytosine methylation in the genomic DNA is assessed wherein the genomic DNA is within 20 kBp of the CpG locus designation listed above, preferably within 15 kBp, more preferably within 10 kBp, yet more preferably within 5 kBp, even more preferably within 1 kBp, most preferably within 0.5 kBp.
Preferably the kit comprises a methylation microarray.
Preferably the kit comprises a DNA sequencing method.
As discussed, the aging process in skin is a highly multifactorial phenomenon that also varies across the body. For example, protected skin is exposed to far fewer insults than exposed skin and it is therefore apparent that different areas of skin from the same individual will have different levels of damage and therefore different “ages”.
In the present invention we consider two forms of skin age: Intrinsic age; and Extrinsic age.
In terms of intrinsic age, the chronological age of an individual is predominant but other endogenous factors such as an individual's metabolism, diet, stress and underlying health also contribute to the age of the skin. Therefore, in the context of the present invention, intrinsic age means the age of the skin caused by endogenous factors.
In terms of extrinsic age, the inherent age will still be a fundamental component but in addition, exogenous factors such as UV radiation, pollution, drying conditions and extremes of temperature will also contribute. Therefore, in the context of the present invention, extrinsic age means the age of the skin caused predominantly by exogenous factors.
For the sake of clarity: Extrinsic age is dominated by the accumulation of ageing caused by extrinsic factors (i.e. originating from outside the exterior surface of the stratum corneum and that then penetrate into the skin through the stratum corneum), especially sun exposure (photo-ageing); whereas Intrinsic age is the degree of ageing in skin due to factors that originate endogenously; in other words ageing not due to extrinsic factors.
The present invention is directed towards the development of an epigenetic method to estimate the intrinsic age of an individual's skin.
This application utilised three epigenetic datasets.
The first dataset (Identification) was a single centre, cross-sectional biopsy study involving 24 Chinese and 24 Caucasian female participants in which 24 young and 24 old females had enrolled. Samples of skin were collected from two different areas of each subject: samples from exposed area of the skin; and samples from protected area of the skin. Sites designated as exposed were located on the lower outer arm. Protected sites were located on the upper inner arm, typically half way between the elbow and axilla area.
The second training dataset (Training) was a publicly available dataset (Bormann F. et al: Reduced DNA methylation patterning and transcriptional connectivity define human skin aging. Aging Cell (2016) 1-9. Array express id: EMTAB-4385). The dataset comprised a total of 108 epidermis samples, 48 samples had been isolated from punch biopsies that had been obtained from the outer forearm of 24 young (18-27 years) and 24 old (61-78 years). 60 samples had been obtained as suction blister roofs from the outer forearm of 60 volunteers aged 20-79 years. All volunteers were female, Caucasian, and disease-free.
The final test dataset (Testing) was a publicly available dataset (Vandiver A. R. et al.: Age and sun exposure-related widespread genomic blocks of hypomethylation in nonmalignant skin. Genome Biology (2015) 16:80) Gene Expression Omnibus accession number: GSE51954). The dataset contained epidermal samples (N=38) from 20 Caucasian subjects. Paired punch biopsy samples, 4 mm in diameter, had been collected under local anaesthesia from the outer forearm or lateral epicanthus (exposed area) and upper inner arm (protected area).
The choice of datasets was guided by the following criteria. First, the training and test data needed to be from epidermal skin, either skin biopsy or epidermis only. The chosen Training data (Bormann et al.) was from skin biopsy and suction blister of the outer forearm and epidermis samples were available for the Testing (Vandiver et al.) dataset. Second, the Training data needed to be on continuous ages and the Testing data needed to have both exposed and protected samples across both young and old age groups. Third, the mean age in the Training dataset (47 years, standard deviation=21) needed to be, and was, comparable to that of the Testing dataset (51 years, standard deviation=25).
All three datasets used bisulphite converted DNA hybridized to Infinium 450k human methylation beadchip.
The methylation data from all DNA samples in the Identification dataset passed quality checks based on three array quality metrics (MAplot, Boxplot, Heatmap). Beta-values were calculated as B=R/R+G and M-values were calculated as M=log 2(R/G), where R represents methylated signals and G unmethylated signals. An offset of 60 was added to the denominator. M-values were used to create the expression matrix. Raw data were normalized using quantile normalization. Beta-values were used for subsequent modelling and filtering the statistical results.
Quality control and pre-processing of the Training dataset was done from raw .idat files in ‘minfi’ R package. Raw data was normalized using Subset-quantile Within Array Normalization (SWAN).
For the Testing dataset, the raw .idat files that are necessary for performing SWAN were unavailable. Therefore, the Illumina pre-processed beta values that were provided were used for subsequent analysis. The quality control and pre-processing applied on the data was also done using ‘minfi’ R package.
Exploratory analysis using principle component analysis (PCA) on the Identification dataset was carried out. It was found that the between-array replicates did not cluster together, likely due to batch effect linked to array number. Clustering analysis of the Testing dataset revealed a similar array batch effect. No technical batch effect was seen on the Training dataset.
The array batch effects observed in the Identification and Testing datasets was adjusted using the ComBat method (Johnson W. E. et al.: Adjusting batch effects in microarray expression data using empirical Bayes methods. Biostatistics 8(1) (2007) 118-127) following quality control, normalization and averaging of within-array replicates. The resulting datasets after batch correction showed no clustering on array. The remaining biological effects were still present and tended to be the main effects in the data.
As used herein, CpG loci refer to the unique identifiers found in the Illumina CpG loci database (as described in Technical Note: Epigenetics, CpG Loci Identification ILLUMINA Inc. 2010, https://www.illumina.com/documents/products/technotes/technote_cpg_loci_identification.pdf). These CpG site identifiers therefore provide consistent and deterministic CpG loci database to ensure uniformity in the reporting of methylation data.
The age predictor from the Horvath Study (which uses the 353 CpG sites discussed above) was run against the exposed (Se) and protected (Sp) samples of the Testing dataset. The performance of the Horvath model was assessed using Linear Regression from which an R2 (“pho” or “p”) was obtained. Median Error (Predicted vs. Actual Age) was also calculated. The results are provided in Table 1.
It can be seen that for 15 out of the 19 subjects the Horvath model calculated exposed samples as being younger than protected samples which is not correct because samples subjected to exposure such as UV radiation are expected to be older than those protected from UV damage.
Average age acceleration on the predicted age reveals the sun-exposed skin sample to have an age 9 years younger than the chronological age which goes against the known physiology that sun that exposure, especially sun-exposure, causes premature ageing of skin. In addition, for a model that is related to intrinsic ageing only, this would be expected to give approximately similar ages for both the protected and exposed samples.
Additionally, the protected skin samples were found to have an age 4 years younger than the chronological age which is a underestimation of the age of the protected skin which would be expected to be approximately the same as the chronological age of the person from which the sample was taken.
It can therefore be concluded that the 353 CpG sites from the Horvath Study are not able to recognize the difference between exposed and protected skin types, nor intrinsic ageing effects in exposed skin, incorrectly predict sun-damaged skin as younger than sun-protected, and underestimate the age of the protected samples.
It was also found that the 353 CpG sites identified by the Horvath Study performed poorly in terms of the accuracy score for protected samples.
The accuracy score for protected samples was:
ρ=0.93 (error=16.6 years).
It can therefore be appreciated that an improved epigenetic method for determining the intrinsic age of skin is required.
Identification of Methylation Sites Associated with Protected Sites (from the Identification Dataset)
A total of 5 comparisons, using different linear models were performed on the normalized batch corrected data for the purpose of generating extrinsic and intrinsic age lists (Table 2). A statistical cut-off set at multiple testing corrected lists (adjust P-value—adjP, benjamini Hochberg)<0.05 together with a delta-beta >=0.05 was applied.
A high number of differentially methylated CpG sites were detected for the comparison of young versus old in exposed sites (Comparison 1: n=10,649). Relatively fewer differentially methylated CpG sites were identified for the comparison of age group versus site interaction (Comparison 5: n=233).
To identify CpG sites that capture intrinsic ageing only, Comparison 2 (Young vs. Old protected sites) results were filtered to remove probes changing by site in young or old (Comparisons 3 & 4), to remove any aging changes in protected skin that might be additionally influenced by extrinsic factors.
The resulting list was 1,575 CpG sites. PCA analysis on these 1,575 sites allowed identification of sites contributing to maximum variance in classifying protected sites into young and old groups across both ethnicities. PCA loadings were used to select these variable probes, a cut-off of 0.030 loading applied to the first component resulted in 322 probes capturing the maximum variability between the age groups.
Intrinsic Age Predictor from Protected Sites
The 322 CpG sites identified to capture intrinsic age changes from the Identification dataset were used to build an intrinsic age model in which the same elastic net as that used in the Horvath Study was utilised on the Training dataset with 10 sets of size n/10 (train on 9 datasets and test on 1). These were repeated 10 times and a mean “accuracy” for each iteration was obtained to give a model for calculating age, and a coefficient for each probe.
Lists of predictors were arrived at by running several iterations of the model. The first iteration identified the best set of predictors. For each subsequent iteration, the identified predictors from the previous iteration were excluded from the training set to identify the next-best set of predictors. The iterations were repeated until the predictive accuracy, measured in terms of rho and error margin was found to be less accurate than that of the Horvath model as described above.
For the intrinsic sites, 3 iterations were performed. The first identified 36 sites, the second identified 53 sites, the third identified 25 as shown in Table 3.
Resultant models where the sites from each of these 3 iterations were removed from the final intrinsic age list of 322 CpG sites were used to estimate the age of the protected samples from the Testing dataset. The results are shown in Table 4. In addition, the average ages for both sun-protected and sun exposed samples were calculated for the resultant models. The results are shown in Table 5. The accuracy of the model using 353 sites from Horvath study for predicting sun-protected age is also shown in Tables 4 and 5 (in italics) for reference.
According to the accuracy measures shown in Table 4 the models of intrinsic age that included the sites identified in iterations 1 and 2 performed with higher or equivalent accuracy (R2=0.96, error 5.7 years and R2=0.94, error=12.6 years) and better error than that the models using the 353 Horvath sites (R2=0.93, error=16.6 years). The remaining 208 sites (which included the 25 sites from iteration 3) performed with lower accuracy than the 353 Horvath sites. Therefore, the 89 sites of iterations 1 and 2 were better at predicting intrinsic age than the Horvath model.
It is expected that the intrinsic age of samples from sun-exposed sites will be similar to that of samples from sun-protected sites. As can be seen from Table 5, the models from this study have a smaller difference between average age for sun-exposed and sun-protected sites than the Horvath model. This demonstrates that the models described herein are better than the Horvath model in predicting intrinsic age.
It can therefore be seen that the use of CpG sites selected from those of iterations 1 and 2 as shown in Table 3 delivers better accuracy when determining the intrinsic age of skin. Therefore, the present invention provides >30 of these 89 sites for use in predicting the intrinsic age of skin. The invention also provides the 53 sites of iteration 2 as a preferred group. The invention further provides the 36 sites of iteration 1 as the most preferred group.
It is an alternative of the invention that the foregoing CpG sites may also be replaced and the closest gene used instead.
Table 6 provides annotations of the 105 sites identified in Iterations 1 & 2 (as described in Price et al. Epigenetics & Chromatin 2013, 6:4, “Additional annotation enhances potential for biologically-relevant analysis of the Illumina Infinium HumanMethylation450 BeadChip array” using Human Genome version HG19), including the closest gene names.
| Number | Date | Country | Kind |
|---|---|---|---|
| 18177976.0 | Jun 2018 | EP | regional |
| Filing Document | Filing Date | Country | Kind |
|---|---|---|---|
| PCT/EP2019/065709 | 6/15/2019 | WO | 00 |
