Accuracy of age prediction among donors
Among donors, there was a strong correlation between the calendar age and the predicted age (r = 0.96, r
2 = 0.93, p = 2 × 10−9, Fig. 1a). Mean absolute deviation (MAD) was 3 years with min. and max. of 0 and 7 and quartile range from 1.5 to 5 years. The calendar age of 14 out of 16 donors (87.5 %) was predicted with an absolute error of 5 years or less. The mean difference between calendar and predicted age was 1 year, with min. and max. of −7 and 7 years, and quartile range from −1.5 to 4.5 years.
Among HSCT recipients, the predicted age did not depend on the recipient calendar age but strongly correlated with calendar and predicted age of the donor.
Among recipients post HSCT, there was no correlation between the calendar and the age predicted from the methylation analysis of the blood DNA (r = 0.134, r
2 = 0.018, p = 0.62, Fig. 1b). Of note, analysis of the recipients’ blood revealed that there was a strong correlation between the predicted age assessed by the methylation signature of blood DNA with the calendar age of the donor (r = 0.94, r
2 = 0.89, p = 4 × 10−8, Fig. 1c) as well as the predicted age of the donor assessed by analysis of the donor’s blood (r = 0.97, r
2 = 0.94, p = 5 × 10−10, Fig. 1d).
The methylation level indicates lower predicted age in the recipients compared to the predicted age in donors after HSCT
Despite the very strong correlation between predicted age of the recipient and the predicted age of the donor, we noted that in all but one case the age predicted from the post-HSCT methylation signature of recipient’s blood was lower than the age predicted from the methylation signature in the donor of transplanted material (Fig. 1e). In the case which constituted the exception the age of the recipient predicted from post HSCT DNA methylation status was 2 years higher than the age predicted from DNA methylation status in the respective donor.
The mean predicted age among recipients was 29.2 years with standard deviation (SD) of 12.3; whereas among the donors, the respective value was 32.9 years with SD of 13.81. The mean difference between the predicted age of the donor and the predicted age of the recipient was 3.7 years with SD of 3.52 (p = 0.00078, t test for paired samples).
The apparent lower predicted age in the recipients compared to the predicted age in donors seen at the DNA methylation level (i.e., the difference between the predicted age of donor and the predicted age of recipient) did not depend on sex (p = 0.49), the age of recipients (p = 0.95), the length of period between HSCT and blood sampling in the recipients group (p = 0.82), or the difference in the calendar age between the donors and the recipients (p = 0.14). However, we noted a trend (p = 0.064) towards a correlation of the size of difference between predicted age in the recipients and donors, with donor’s age (r = 0.47) suggesting that this effect may be more pronounced when HSCT comes from an older donor (i.e., the older the donor the greater the difference between the his/her predicted age and the predicted age of the recipient).
The apparent lower predicted age in the recipients compared to the predicted age in donors after HSCT is driven by selective hypermethylation of CpGs within the C1orf132 locus
We further investigated whether the consistent difference of ~4 years between the predicted age of recipient and the predicted age of donor was caused by coherent effects of all CpG predictors or the direction of DNA methylation changes was different in particular loci. In Additional file 1: Table S2, we show the results of comparison of recipient vs. donor methylation differences for all the 5 CpG included in the age prediction model as well as for the remaining 27 CpGs whose methylation was assessed in our study. We found that among the CpG sites included in the model, only CpG_1 at C1orf132 (C1orf132 CpG_1) showed different methylation among recipients vs. donors (mean difference 8.2 %; SD = 6.9 % (p = 0.00026; P
corrected = 0.008).
In order to better estimate the size of this effect, we developed an age prediction model based solely on CpGs in C1orf132. The model included C1orf132 CpG_1 and C1orf132 CpG_3, and it had a reasonable performance (r = 0.89, r
2 = 0.8, and MAD = 6.1 years). Using the “C1orf132 only” model, the mean predicted age of recipients was 22.4 years and that of donors was 33.8 years, yielding a difference of 11.3 years (p = 0.000016, paired t test).
Although we did not detect individual recipient-donor DNA methylation differences in other loci than C1orf132, these loci could still have weak effects possibly amounting to detectable joint influence. In order to test this, we developed an age prediction model including all loci except the C1orf132 (r = 0.96, r
2 = 0.92, MAD = 4.1 years). Using this “non-C1orf132 model” the mean predicted age of recipients was 32.1 years and that of donors was 33.1 years, yielding a difference of 1.1 year which was not statistically significant (p = 0.28, paired t test).
Size of the increase in methylation of C1orf132 CpG_1 in the recipient vs. donor correlates with calendar age of the donor
The size of the increase in methylation of C1orf132 CpG_1 in the recipients vs. the donors correlated with the calendar age of the donors (r = 0.76, r
2 = 0.58, p = 0.0006) but not with sex (r = −0.40, p = 0.12), the recipient’s age (r = 0.13, p = 0.62), the length of time between HSCT and blood sampling in the recipient group (r = −0.004, p = 0.99), or diagnosis (AML vs. other, 9 vs. 6.4 years, respectively, p > 0.5). The size of the increase in methylation of C1orf132 CpG_1 in the recipients vs. the donors nominally correlated also with the recipient-donor calendar age differences (r = −0.51, p = 0.043), but this effect was no longer observed after adjustment for calendar age of the donors (p = 0.7). Correlation with the age of the donors was also found with the recipient vs. donor age differences calculated using the “C1orf132 only” model (r = −0.7, r
2 = 0.5, p = 0.003).
The increase in the methylation in the recipients vs. the donors was also apparent for two other C1orf132 CpGs studied; for C1orf132 CpG_2, the difference was 13.4 % (SD = 7.54. p = 0.000003, P
corrected = 0.0001) and for C1orf132 CpG_3 it was 13.8 % (SD = 7.8. p = 0.000004, P
corrected = 0.00012, Fig. 1f). However, no statistically significant correlation was observed between the size of these effects and the donor’s age or other analyzed variables (data not shown).
The apparent effect of the donor’s age on the methylation of C1orf132 CpG_1 in the recipient group could be secondary to the differences in the cellular composition of the reconstituted hematopoietic system in the recipients receiving transplant from young vs. older donors. Indeed, when we analyzed donors age vs. white blood cell (WBC) or counts/percentages of granulocytes/lymphocytes/monocytes in the recipients, we found that donor’s age correlated with lymphocyte numbers in the recipient group (r = 0.78, r
2 = 0.61, p = 0.0003). Since lymphocyte numbers in the recipient also correlated with the size of the increased C1orf132 CpG_1 methylation in the recipient vs. donor (r = 0.63, r
2 = 0.40, p = 0.009), we analyzed the effect of lymphocyte numbers and the increase in C1orf132 CpG_1 methylation on donor’s age by multiple linear regression. We found that lymphocyte numbers and the difference in C1orf132 CpG_1 methylation between the recipient and donor were independent predictors of donor’s age (beta = 0.51, p = 0.016 and beta = 0.44, p = 0.033, respectively).