Patient demographics
Normal breast tissue (> 3cm away from breast tumor) was analyzed from 34 patients with breast cancer (cases) and 50 healthy controls (controls), and peripheral blood samples were analyzed from 79 patients with breast cancer and 91 healthy controls. Matched breast and blood samples were available from 80 patients (Suppl Fig S1). The characteristics of the study population are summarized in Table 1. Overall, the breast cancer and control cohorts were very similar. The mean age of the breast cancer cohort was 51.4 years old and the mean age of the control cohort was 47.6 years old. Both cohorts included mostly non-Hispanic white patients (75% in the breast cancer cohort, and 88% in the control cohort), with similar BMIs (27.4 in the breast cancer cohort, and 27.7 in the control cohort), mostly non-smokers (57.5% in the breast cancer cohort, and 61.7% in the control cohort), with some alcohol consumption (65.5% in the breast cancer cohort, and 95.7% in the control cohort). The majority of cancers were hormone positive (63.7%), HER2 negative (72.6%), and BRCA negative (57.5%).
Table 1
Characteristics of the study population
|
Tissue sample
|
Blood sample
|
Characteristics
|
Breast cancer
N = 34
|
Control
N = 50
|
Breast cancer
N = 79
|
Control
N = 91
|
Age, years
|
48.3 ± 12.2
|
46.7 ± 20.8
|
54.6 ± 13.6
|
48.5 ± 17.7
|
Cohort (Komen/Yale/prevention clinic)
|
0/34/0
|
45/5/0
|
0/48/31
|
44/5/42
|
Age groups
|
|
|
|
|
< 50 years
|
21
|
27
|
32
|
44
|
≥ 50 years
|
13
|
23
|
47
|
47
|
Race/ethnicity
|
|
|
|
|
Non-Hispanic White
|
23
|
45
|
62
|
79
|
Hispanic White
|
3
|
2
|
3
|
4
|
Africa American
|
4
|
3
|
7
|
6
|
Others
|
4
|
0
|
7
|
2
|
Body mass index
|
26.6 ± 4.9
|
26.9 ± 5.3
|
28.2 ± 6.1
|
28.5 ± 6.7
|
Smoking
|
|
|
|
|
Non-smoker
|
21
|
31
|
44
|
56
|
Former smoker
|
12
|
14
|
30
|
25
|
Current smoker
|
5
|
5
|
10
|
Alcohol consumption
|
|
|
|
|
No
|
18
|
3
|
21
|
3
|
Yes
|
16
|
47
|
58
|
88
|
Age at first birth
|
25.5 ± 5.6
|
25.8 ± 3.4
|
26.0 ± 5.0
|
26.2 ± 4.5
|
Menopause, yes
|
|
|
|
|
No
|
24
|
25
|
35
|
37
|
Yes
|
10
|
25
|
44
|
54
|
Age at menses onset
|
12.3 ± 1.7
|
12.5 ± 1.1
|
12.7 ± 1.7
|
12.4 ± 1.4
|
Number of birth
|
1.8 ± 1.4
|
2.2 ± 0.6
|
1.9 ± 1.4
|
2.1 ± 1.0
|
Total pregnancy
|
2.3 ± 1.9
|
2.5 ± 1.0
|
2.3 ± 1.9
|
2.6 ± 1.5
|
ER/PR
|
|
|
|
|
+/+
|
23
|
−
|
49
|
−
|
+/-
|
1
|
−
|
2
|
−
|
-/-
|
1
|
−
|
10
|
−
|
NA
|
9
|
−
|
18
|
−
|
HER2
|
|
|
|
|
Positive
|
3
|
−
|
4
|
−
|
Negative
|
24
|
−
|
58
|
−
|
Not typed/insufficient tissue
|
7
|
−
|
10
|
−
|
NA
|
0
|
−
|
7
|
−
|
BRCA testing
|
|
|
|
|
Positive
|
2
|
−
|
2
|
0
|
Negative
|
16
|
−
|
49
|
33
|
Results unknown
|
4
|
−
|
4
|
0
|
Not tested
|
10
|
−
|
16
|
9
|
NA
|
2
|
−
|
8
|
49
|
Oral contraceptive pill use
|
|
|
|
|
Current and former user
|
5
|
3
|
26
|
36
|
Never
|
13
|
0
|
23
|
8
|
Unknown
|
15
|
1
|
26
|
1
|
NA
|
1
|
46
|
4
|
46
|
Ever hormone replacement therapy
|
|
|
|
|
Yes
|
3
|
45
|
12
|
20
|
NO
|
23
|
33
|
48
|
67
|
NA
|
8
|
2
|
19
|
4
|
Epigenetic age in normal breast tissue of breast cancer patients and controls
The correlation between epigenetic clocks (measured in breast tissue) and chronological age was examined for both cases and controls. When examining epigenetic aging in control samples, significant age correlations were found for five of the six clocks, ranging from r = 0.35 (Levine clock) to r = 0.68 (Lin clock), while the Yang clock did not exhibit a significant increase in epigenetic age as a function of chronological age in controls. On the other hand, no significant age correlations were found for any of the clocks when restricting samples to cases only (Suppl Figure S2), suggesting that chronological age did not differentiate epigenetic ages in normal breast tissue scores among women who had already developed breast cancer. When comparing epigenetic ages of cases to controls, we found that increased epigenetic age acceleration (age adjusted) was observed for the Levine (p = 5.7e-4) and Yang (p = 1.6e-2) clocks in the normal breast tissue from breast cancer patients (cases) compared to controls (Fig. 1). These results were maintained even after adjusting for the full list of confounders in our data (Suppl Figure S3).
In the peripheral blood, all six epigenetic clocks correlated with age in both cases and controls, and we observed much higher age correlations than were observed in breast tissue (Figure S4). For instance, age correlations in controls ranged from r = 0.43 (Yang) to r = 0.87 (Horvath2). Similarly, age correlations in cases ranged from r = 0.26 (Yang) to r = 0.84 (Hannum and Horvath2). However, when comparing epigenetic age acceleration between cases and controls, no clocks showed statistically significant differences (Fig. 2).
Finally, we compared the predicted epigenetic age acceleration values in breast tissue vs. peripheral blood using matched samples. Results showed no significant associations across any of the six clocks (Fig. 3). This suggests that aging in the breast tissue is likely discordant with aging in the blood and therefore, peripheral blood would not serve as a useful surrogate measure when it comes to assessing the epigenetic age in breast tissue.
Epigenetic age acceleration seen in normal breast tissue from breast cancer patients is driven by polycomb related genes
The Levine and Yang clocks that were associated with breast cancer and demonstrated significant age acceleration share a characteristic in that they are both enriched for CpGs associated with polycomb-group (PcG) protein targets. For instance, the Yang clock was intentionally developed to comprise only CpGs found in PcG regions of the genome, while the Levine clock had no CpG preselection, but was found to be enriched for PcG regions, particularly among its CpGs that exhibit hypermethylation age. Given that the Levine clock 1) showed the strongest association with case versus control status, and 2) is comprised of CpGs that both are and are not associated with PcG target genes, we set to test whether the associations with breast cancer differed between the signal coming from PcG related CpGs versus the remaining (non-PcG) CpGs. To do this, the Levine clock score was recalculated twice, but in each case, CpGs belonging to either PcG or non-PcG related regions were set to zero, such that they dropped out of the model. This left us with two clocks scores that together sum to the full Levine clock score, yet enabled us to test the associations between breast cancer and these two distinct types of epigenetic age changes.
When only the CpGs associated with polycomb related genes (n = 55) from the Levine clock are used, we found that the score was still highly correlated with the original Levine clock score in both cases and controls. Interestingly, this correlation is higher in breast tissue for both controls (r = 0.86, Fig. 4A) and cases (r = 0.84, Fig. 4B) than it is in peripheral blood (Fig. 4D & 4E). Furthermore, this aging score based on CpGs that are associated with only PcG related genes has an even higher association with cases than controls in breast tissue (p = 0.00012) compared to the entire Levine clock (Fig. 4C).
When calculating the epigenetic clock using the CpGs that are not associated with PcG genes (n = 458), we find that these scores are highly correlated with the Levine clock scores in both breast and peripheral blood of cases and controls (range r = 0.89 to r = 0.99). However, we do not observe a significant difference between cases and controls (Fig. 5C and 5F). Finally, when comparing the breast vs. blood concordance in matched samples for these two subclocks, again, no significant association is found, suggesting blood and breast may be aging asynchronously (Fig. 6).