Pre-diagnostic leukocyte mitochondrial DNA copy number and colorectal cancer risk

Abstract

Mitochondrial DNA (mtDNA) is susceptible to oxidative stress and mutation. Few epidemiological studies have assessed the relationship between mtDNA copy number (mtDNAcn) and risk of colorectal cancer (CRC), with inconsistent findings. In this study, we examined the association between pre-diagnostic leukocyte mtDNAcn and CRC risk in a case–control study of 324 female cases and 658 matched controls nested within the Nurses’ Health Study (NHS). Relative mtDNAcn in peripheral blood leukocytes was measured by quantitative polymerase chain reaction-based assay. Conditional logistic regression models were applied to estimate odds ratios (ORs) and 95% confidence intervals (95% CIs) for the association of interest. Results showed lower log-mtDNAcn was significantly associated with increased risk of CRC, in a dose-dependent relationship (P for trend < 0.0001). Compared to the fourth quartile, multivariable-adjusted OR [95% confidence interval (CI)] was 1.10 (0.69, 1.76) for the third quartile, 1.40 (0.89, 2.19) for the second quartile and 2.19 (1.43, 3.35) for the first quartile. In analysis by anatomic subsite of CRC, we found a significant inverse association for proximal colon cancer [lowest versus highest quartile, multivariable-adjusted OR (95% CI) = 3.31 (1.70, 6.45), P for trend = 0.0003]. Additionally, stratified analysis according to the follow-up time since blood collection showed that the inverse association between mtDNAcn and CRC remained significant among individuals with ≥ 5 years’ follow-up, and marginally significant among those with ≥ 10 years’ follow-up since mtDNAcn testing, suggesting that mtDNAcn may serve as a long-term predictor for risk of CRC. In conclusion, pre-diagnostic leukocyte mtDNAcn was inversely associated with CRC risk. Further basic experimental studies are needed to explore the underlying biological mechanisms linking mtDNAcn to CRC carcinogenesis.

Pre-diagnostic leukocyte mitochondrial DNA copy number, a reflection of oxidative stress damage, was inversely associated with risk of colorectal cancer, and may be a long-term predictor of colorectal cancer risk.

Introduction

Colorectal cancer (CRC) is the third most common cancer among men and women in the USA (1). Across worldwide, CRC is the third most common cancer in men and the second most common cancer in women (2). The incidence and mortality of CRC have been decreasing over the past decade, likely due in part to the successful implementation of screening programs (3). However, the disease is still the third leading cause of cancer death in the USA and the fourth across the world among men and women (1,2), posing an enormous health and socioeconomic burden (4). Thus, identifying biomarkers for CRC risk that might inform prevention and early diagnosis is of great public health importance.

Mitochondria are essential eukaryotic organelles containing their own genome, i.e. mitochondrial DNA (mtDNA), which is usually maternally inherited (5). MtDNA consists of approximately 16,569 bp double-stranded circular DNA and encodes only 37 genes. Most mammalian cells contain between hundreds and over a thousand mitochondria per cell, and each mitochondrion has 2–10 copies of mtDNA (5). Compared to nuclear DNA, mtDNA has a higher mutation rate and is particularly susceptible to oxidative stress, probably due to its proximity to the source of reactive oxygen species and its lack of protective histones (6). Though the study of mtDNA repair pathways has lagged behind inquiries into nuclear DNA repair mechanisms, research has not only shown the existence of robust damage tolerance mechanisms in mitochondria, but also proposed various mtDNA repair pathways that may properly maintain the mitochondrial genome (7).

CRC is a heterogeneous disease associated with environmental and genetic factors through complicated interactions (8,9). Oxidative stress triggered by reactive oxygen species may initiate and promote carcinogenesis (including colorectal carcinogenesis) by inducing inflammation, DNA damage, gene mutations and genomic instability (10,11). Because mtDNA copy number (mtDNAcn) is a major biomarker for oxidative DNA damage and mitochondrial dysfunction, it has been hypothesized that altered pre-diagnostic leukocyte mtDNAcn may be associated with risk of developing cancers, including CRC.

The few epidemiological studies that have assessed the relationship between mtDNAcn and risk of CRC have yielded inconsistent findings (12–14). A retrospective case–control study conducted by Qu et al. in a hospital setting in China first reported a positive association between mtDNAcn and CRC risk (12). Later, prospective case–control studies nested within the Shanghai Women’s Health Study (SWHS) (13) reported an inverse association, and the Singapore Chinese Health Study (SCHS) (14) reported a U-shaped relationship. Evidence supporting the relationship between mtDNAcn and CRC risk in western populations has been lacking. Therefore, in this study, we examined the association between pre-diagnostic leukocyte mtDNAcn and the risk of CRC in a case–control study of 324 CRC cases and 658 matched healthy controls nested within the Nurses’ Health Study (NHS), a long-term prospective cohort study of women in the USA.

Methods

Study population

The Nurses’ Health Study was initiated in 1976, when 121,700 female US registered nurses aged 30–55 years completed and returned questionnaires regarding their medical histories and baseline lifestyles. Biennially, participants completed self-administered follow-up questionnaires with updated information on their dietary habits and other lifestyle factors, medical history and disease diagnosis. In 1989–90, a total of 32,826 participants in the NHS provided blood samples. Details of the NHS have been previously published (15).

CRC case ascertainment and control selection

CRC diagnoses were based on the self-report by nurses on biennial questionnaires and then confirmed by a pathologist. All CRC cases were incident cases diagnosed after blood collection. In this nested case–control study, we randomly selected one to three controls from the same cohort (NHS) of participants who were free of cancer (excluding non-melanoma skin cancer) up to and including the questionnaire cycle in which the case was diagnosed. Control subjects were matched to each case based on year of birth (±1 year), race and fasting status at blood collection. A total of 324 CRC cases and 658 healthy controls were included. The study protocol was approved by the institutional review boards of the Brigham and Women’s Hospital and Harvard T.H. Chan School of Public Health, and those of participating registries as required.

Assessment of mtDNAcn

Details of the ascertainment and validation of leukocyte mtDNAcn for blood samples from the NHS participants and quality control procedures were described previously (16–18). More detail can be found in Supplementary Methods, available at Carcinogenesis Online. Briefly, genomic DNA was extracted from the buffy-coat leukocytes in peripheral blood, according to the QIAmp (Qiagen, CA) 96-spin blood protocol. Concentrations of DNA were measured by pico-green quantitation utilizing a Molecular Devices 96-well spectrophotometer. The quantitative polymerase chain reaction-based assay was used to determine the ratio of the copy numbers of mitochondrial ND2 gene to genomic single-copy gene (AluYb8) (N/S), which is proportional to the average number of mtDNAcn. The relative N/S ratio was then calculated by subtracting the N/S ratio of the calibrator DNA from the N/S ratio of each sample. The value of mtDNAcn was calculated as the exponentiated N/S ratio. Each sample was assayed in triplicate, and 10% replicate quality-control (QC) samples were included. The coefficients of variation (CVs) for ND2 and AluYb8 were less than 1% among QC samples.

Assessment of covariates

Covariate data were collected through self-administered questionnaires at baseline (1976) and during follow-up biennially. In this study, we used covariates data from the questionnaire cycle closest to blood collection (1989–90), including body mass index (BMI) calculated as height (m)/weight (kg)², smoking status, alcohol consumption, Alternate Healthy Eating Index (AHEI), physical activity, family history of CRC, regular use of aspirin and other non-steroidal anti-inflammatory drugs (NSAIDs), parity as well as menopausal status and postmenopausal hormone use. Specifically, height and weight were collected at baseline and then weight was updated at each biennial questionnaire; we used height at baseline and weight at blood collection to calculate BMI. We also included weight change from blood collection until 2 years before diagnosis of the cases and same cycle of their matched controls into the multivariable model; to minimize reverse causality of CRC on weight, we excluded the 2 years before diagnosis for the calculation of weight change. Participants who used aspirin (either standard or low dose) at least two times per week on average were classified as regular aspirin users. Regular users of non-aspirin NSAIDs were participants who responded ‘yes’ to regular use on questionnaires, defined as at least two times per week (19). Physical activity was represented by metabolic equivalent (MET) hours per week. Specifically, each activity was assigned a MET value, which refers to the metabolic rates for each specific activity divided by metabolic rates at rest. MET hours per week for each activity were calculated by multiplying average time per week in each activity by the MET of each activity. Then, total MET hours per week were derived by summing up MET hours per week for each activity (20). AHEI is a dietary score (0–100 points) measuring the adherence to a dietary pattern characterized by foods and nutrients most predictive of risk of diseases; higher AHEI indicates healthier dietary quality. Foods/nutrients involved in AHEI development include whole grains, sugar-sweetened beverages, fruit juice, vegetables, fruits, nuts and legumes, red and processed meat, poly-unsaturated fatty acids, trans fats, long-chain (n-3) fats (EPA + DHA) and sodium (21). AHEI used in the current study was derived from the food frequency questionnaire at blood collection. The validity and reproducibility of physical activity and dietary information from the food frequency questionnaire have been reported elsewhere (22,23).

Statistical analysis

Log-transformed mtDNA copy numbers [log (mtDNAcn)] of cases and controls were stratified into four categories based on the quartiles of log (mtDNAcn) among all controls. Conditional logistic regression was applied to estimate the odds ratio (OR) and 95% confidence interval (CI) for the association of log (mtDNAcn) with CRC risk. Two models were analyzed: Model 1, the crude model without covariate adjustment; and Model 2, the multivariable model that adjusted for potential confounders, including BMI (in tertiles: 0–23.2 kg/m², 23.2–26.6 kg/m², ≥ 26.6 kg/m²), physical activity (in tertiles, 0–8.2, 8.2–20.2, ≥ 20.2 MET hours per week), smoking status (never, former or current smokers), alcohol consumption (in tertiles, 0–0.8 g, 0.8–5.8 g, ≥ 5.8 g per day), menopausal status and postmenopausal hormone use (premenopausal, never and former users, current users), parity (0/1/2/3+children), AHEI (in tertiles, 0–42.8, 42.8–51.3, ≥ 51.3), regular aspirin use (yes/no), regular non-aspirin NSAIDs use (yes/no), and family history of CRC (yes/no) at blood collection and weight change from blood collection until 2 years before diagnosis (in tertiles, < 0 kg, 0–2.72 kg, ≥ 2.72 kg). Conditional logistic regression models were also used to estimate the association between mtDNAcn and CRC risk by anatomic subsites, including colon cancer (proximal, distal) and rectal cancer. Unconditional logistic regression with adjustment for matching factors and covariates was employed to further examine the effect of two-way interactions between mtDNAcn and potential confounders on the risk of CRC. The statistical significance of interaction was assessed using likelihood ratio test for cross-product terms of covariates and log (mtDNAcn).

To minimize the reverse influence of potential undetectable tumors if any at blood collection on mtDNAcn, we conducted a sensitivity analysis by removing cases diagnosed within 1 and 2 years after blood collection and their matched controls. In addition, to exclude any potential influence of colorectal polyps and inflammatory bowel disease (IBD) on mtDNAcn (especially among controls), we performed another sensitivity analysis by removing cases and controls who had colorectal polyps and/or IBDs before the time of CRC diagnosis. We also examined the association between mtDNAcn and CRC risk stratified by the follow-up time since blood collection to explore whether mtDNAcn has potential as a long-term predictive biomarker for risk of CRC. All statistical analyses were performed using SAS software, version 9.4 for UNIX (SAS Institute, North Carolina). All tests were two-sided and P < 0.05 was considered statistically significant.

Results

Basic characteristics of CRC cases (n = 324) and matched controls (n = 658) in this nested case–control study are presented in Table 1. Briefly, the mean age (standard deviation, SD) at blood collection for cases was 58.9 (6.7) years and for controls was 59.3 (6.6) years. Mean age (SD) at CRC diagnosis was 67.4 (7.5) years among cases. MtDNAcn was lower in cases than controls. Compared to controls, relatively fewer cases were regular users of aspirin or NSAIDs, or current users of postmenopausal hormones, while relatively more cases were current smokers, had family history of CRC, and consumed higher amounts of alcohol. We also present those basic characteristics according to mtDNAcn quartiles after age standardization among 658 control subjects (Table 2). Briefly, compared to the women in the highest quartile, percentages of current smokers and participants with family history of CRC were higher, while levels of total physical activity (MET hours per week) were lower among the women in the lowest quartile of mtDNAcn.

Table 1.

Basic characteristics of colorectal cancer cases and controls in the nested case–control study within the NHS

Characteristics	Cases (n = 324)	Controls (n = 658)
Age at blood draw, mean (SD)	58.9 (6.7)	59.3 (6.6)
Age at diagnosis, mean (SD)	67.4 (7.5)	-
Caucasians, %	98.2	99.7
log-mtDNAcn, mean (SD)	−0.1 (0.3)	0.01 (0.3)
Regular users of aspirin, %	38.0	47.0
Regular users of non-aspirin NSAIDs, %	12.7	19.8
BMI, kg/m2, mean (SD)	25.7 (5.0)	25.6 (4.7)
Weight change from blood collection until 2 years before diagnosis, kg, mean (SD)	1.0 (6.5)	1.8 (6.9)
Physical activity, MET hours/ week, mean (SD)	18.1 (18.0)	18.6 (19.1)
AHEI score, mean (SD)	46.4 (9.2)	47.3 (9.6)
Smoking status, %
Past smokers	39.2	42.3
Current smokers	17.3	12.2
Alcohol consumption, g/d, mean (SD)	7.3 (12.3)	6.9 (10.5)
CRC in a parent or sibling, %	17.3	15.5
Parity, %
0 child	6.5	4.4
1 child	5.6	8.4
2 children	25.3	23.9
3+ children	62.0	62.3
Postmenopasual status, %	86.1	88.6
Current postmenopausal hormone use, %	34.0	43.7

Values are means (SD) for continuous variables and percentages for categorical variables; percentage of current postmenopausal hormone use is calculated among postmenopausal women.

Table 2.

Age-standardized basic characteristics by mtDNAcn quartiles among controls in this nested case–control study within the NHS

Characteristics	Quartile 1	Quartile 2	Quartile 3	Quartile 4
Number of participants	164	165	165	164
log-mtDNAcn, mean (SD)	−0.4 (0.1)	−0.1 (0.0)	0.1 (0.0)	0.5 (0.2)
Age at blood draw, mean (SD)*	59.4 (6.9)	59.4 (6.5)	59.2 (6.4)	59.0 (6.7)
Regular users of aspirin, %	53.1	47.3	49.4	48.9
Regular users of non-aspirin NSAIDs, %	19.4	18.5	13.9	23.3
BMI, kg/m2, mean (SD)	25.6 (3.4)	25.7 (3.2)	26.1 (4.1)	25.7 (4.1)
Weight change from blood collection until 2 years before diagnosis, kg, mean (SD)	1.7 (5.1)	2.5 (6.6)	0.8 (4.3)	1.3 (4.2)
Physical activity, MET hours/week, mean (SD)	16.9 (12.4)	18.6 (12.9)	19.2 (15.3)	20.9 (17.8)
AHEI score, mean (SD)	48.2 (8.0)	48.3 (7.0)	48.6 (7.3)	47.3 (8.3)
Smoking status, %
Past smokers	46.1	40.0	44.8	43.2
Current smokers	13.9	13.2	7.0	9.0
Alcohol consumption, g/d, mean (SD)	6.7 (8.2)	6.8 (7.8)	7.6 (9.0)	6.9 (8.6)
CRC in a parent or sibling, %	19.3	17.2	16.0	16.8
Parity (≥2 children), %	88.5	88.6	81.0	88.8
Postmenopausal women, %	90.1	92.1	91.2	93.4
Current postmenopausal hormone use, %	47.5	42.1	39.3	42.6

Values are means (SD) for continuous variables and percentages for categorical variables, and are standardized to the age distribution of the study population.

Percentage of current postmenopausal hormone use is calculated among postmenopausal women.

*Value is not age adjusted.

For the association between mtDNAcn and CRC risk, we found that lower log-mtDNAcn level was significantly associated with an increased risk of CRC, with a dose-dependent relationship in both the crude and multivariable-adjusted models; compared to the crude model, results did not change materially after adjusting for a list of covariates (Table 3). Compared to the highest (fourth) quartile, multivariable-adjusted OR (AOR, 95% CI) was 1.10 (0.69, 1.76) for the third quartile, 1.40 (0.89, 2.19) for the second quartile, and 2.19 (1.43, 3.35) for the first quartile (P for trend < 0.0001). In the further analysis of CRC by anatomic subsite, we observed a significant inverse association for proximal colon cancer [lowest versus highest quartile, AOR (95% CI) = 3.31 (1.70, 6.45), P for trend = 0.0003] (Table 3). The inverse association was not statistically significant for distal colon cancer and rectal cancer, which may be due to the small number of cases with cancer at those subsites.

Table 3.

Associations of mtDNAcn with the risk of overall colorectal cancer, as well as cancers at anatomic subsites

	Fourth quartile	Third quartile	Second quartile	First quartile	P for trend
		OR (95% CI)	OR (95% CI)	OR (95%CI)
Colorectal cancer
Cases/controls (324/658)	57/164	68/165	78/165	121/164
Model 1	ref	1.22 (0.78, 1.90)	1.42 (0.93, 2.18)	2.19 (1.47, 3.27)	<0.0001
Model 2	ref	1.10 (0.69, 1.76)	1.40 (0.89, 2.19)	2.19 (1.43, 3.35)	<0.0001
Colon cancer
Cases/controls (253/509)	44/132	51/119	62/128	96/130
Model 1	ref	1.28 (0.77, 2.12)	1.50 (0.94, 2.42)	2.27 (1.45, 3.55)	0.0002
Model 2	ref	1.15 (0.66, 1.99)	1.53 (0.92, 2.55)	2.28 (1.40, 3.71)	0.0003
Proximal colon cancer
Cases/controls (151/300)	22/76	36/72	30/78	63/74
Model 1	ref	1.76 (0.89, 3.50)	1.50 (0.77, 2.91)	3.06 (1.67, 5.61)	0.0003
Model 2	ref	1.71 (0.80, 3.69)	1.70 (0.82, 3.51)	3.31 (1.70, 6.45)	0.0003
Distal colon cancer
Cases/controls (90/185)	18/48	14/43	31/45	27/49
Model 1	ref	0.87 (0.38, 2.00)	1.70 (0.81, 3.54)	1.26 (0.58, 2.74)	0.28
Model 2	ref	0.51 (0.16, 1.56)	1.20 (0.46, 3.15)	1.15 (0.42, 3.12)	0.47
Rectal cancer
Cases/controls (71/149)	13/32	17/46	16/37	25/34
Model 1	ref	1.01 (0.40, 2.58)	1.13 (0.43, 2.94)	1.90 (0.78, 4.59)	0.10
Model 2	ref	0.73 (0.23, 2.26)	0.94 (0.30, 2.92)	1.91 (0.67, 5.46)	0.12

ref, reference group.

Model 1: Conditional logistic regression model, no covariates adjustment;

Model 2: Conditional logistic regression model, adjusting for body mass index (in tertiles: 0–23.2 kg/m², 23.2–26.6 kg/m², ≥ 26.6 kg/m²), physical activity (in tertiles, 0–8.2, 8.2–20.2, ≥ 20.2 MET hours/week), weight change from blood collection until 2 years before diagnosis (in tertiles, < 0 kg, 0–2.72 kg, ≥ 2.72 kg), smoking status (never, former, or current smokers), alcohol consumption (in tertiles, 0–0.8 g, 0.8–5.8 g, ≥ 5.8 g per day), menopausal status and postmenopausal hormone use (premenopausal, non-current users, current users), parity (0/1/2/3+children), Alternate healthy eating index (AHEI) (in tertiles, 0–42.8, 42.8–51.3, ≥ 51.3), regular aspirin use (yes/no), regular non-aspirin NSAIDs use (yes/no), family history of colorectal cancer (yes/no).

In the further sensitivity analysis to test any potential influence of colorectal polyps and IBD on mtDNAcn/CRC, the inverse association remained significant after removing cases and controls who had colorectal polyps and/or IBD before CRC diagnosis [lowest versus highest quartile, AOR (95% CI) = 2.51 (1.52, 4.13), P for trend < 0.0001)]. In another sensitivity analysis examining the possible reverse influence of potential undetectable tumors (if any) at blood collection on mtDNAcn, the results did not change materially after removal of cases diagnosed within 1 and 2 years after blood collection and their matched controls, indicating minimal reverse causation [follow-up ≥ 1 year, lowest versus highest quartile, AOR (95% CI) = 2.08 (1.33, 3.24), P for trend = 0.0006; follow-up ≥ 2 year, lowest versus highest quartile, AOR (95% CI) = 1.92 (1.22, 3.02), P for trend = 0.004].

Moreover, we performed a stratified analysis according to the follow-up time since blood collection (i.e. mtDNAcn testing) (Table 4). The inverse association between mtDNAcn and CRC remained significant among individuals with ≥ 5 years’ follow-up since mtDNAcn testing [lowest versus highest quartile, AOR (95% CI) = 1.98 (1.18, 3.33), P for trend = 0.009]. The inverse association was also marginally significant among those with ≥ 10 years’ follow-up [lowest versus highest quartile, AOR (95% CI) = 1.92 (0.94, 3.95), P for trend = 0.06]. These data suggest that mtDNAcn could serve as a long-term predictive marker for the risk of CRC.

Table 4.

Associations between mtDNAcn and colorectal cancer risk by time of follow-up since blood collection

	Fourth quartile	Third quartile	Second quartile	First quartile	P for trend
		OR (95%CI)	OR (95%CI)	OR (95%CI)
<5 years (n = 330)
Cases/controls (94/236)	18/56	19/70	18/64	39/46
Model 1	ref	0.89 (0.40, 1.97)	0.94 (0.42, 2.08)	3.03 (1.39, 6.60)	0.002
Model 2	ref	0.80 (0.33, 1.90)	0.86 (0.36, 2.06)	3.44 (1.43, 8.27)	0.003
≥5 years (n = 652)
Cases/controls (230/422)	39/108	49/95	60/101	82/118
Model 1	ref	1.47 (0.86, 2.50)	1.76 (1.06, 2.93)	1.96 (1.23, 3.13)	0.004
Model 2	ref	1.48 (0.82, 2.66)	1.72 (0.98, 3.00)	1.98 (1.18, 3.33)	0.009
≥8 years (n = 492)
Cases/controls (176/316)	34/91	35/66	43/72	64/87
Model 1	ref	1.45 (0.81, 2.62)	1.67 (0.96, 2.93)	1.95 (1.17, 3.26)	0.01
Model 2	ref	1.58 (0.79, 3.17)	1.59 (0.82, 3.07)	2.09 (1.14, 3.83)	0.02
≥10 years (n = 339)
Cases/controls (123/216)	23/62	25/43	32/47	43/64
Model 1	ref	1.56 (0.77, 3.15)	1.97 (1.00, 3.90)	1.76 (0.95, 3.24)	0.07
Model 2	ref	1.38 (0.58, 3.30)	2.13 (0.92, 4.92)	1.92 (0.94, 3.95)	0.06

ref, reference group.

Model 1: Conditional logistic regression model, no covariates adjustment;

Model 2: Conditional logistic regression model, adjusting for body mass index (in tertiles: 0–23.2 kg/m², 23.2–26.6 kg/m², ≥ 26.6 kg/m²), physical activity (in tertiles, 0–8.2, 8.2–20.2, ≥ 20.2 MET hours/week), weight change from blood collection until 2 years before diagnosis (in tertiles, < 0 kg, 0–2.72 kg, > 2.72 kg), smoking status (never, former, or current smokers), alcohol consumption (in tertiles, 0–0.8 g, 0.8–5.8 g, ≥ 5.8 g per day), menopausal status and postmenopausal hormone use (premenopausal, non-current users, current users), parity (0/1/2/3+children), Alternate healthy eating index (AHEI) (in tertiles, 0–42.8, 42.8–51.3, ≥ 51.3), regular aspirin use (yes/no), regular non-aspirin NSAIDs use (yes/no), family history of colorectal cancer (yes/no).

We also examined the effect of interactions between mtDNAcn and potential confounders on the risk of CRC. We observed an effect modification of AHEI on the association between mtDNAcn and CRC risk (P for interaction = 0.03). In the stratified analysis by AHEI, a significant inverse association between mtDNAcn and CRC risk was shown among individuals in the lowest AHEI (i.e. less healthy diet) tertile group [lowest versus highest mtDNAcn quartile, AOR (95% CI) = 3.79 (1.77, 8.13), P for trend = 0.001]; the inverse associations were weaker and not statistically significant among those in the second AHEI tertile group [lowest versus highest mtDNAcn quartile, AOR (95% CI) = 1.62 (0.79, 3.35), P for trend = 0.11] and third AHEI tertile group [AOR (95% CI) = 1.75 (0.82, 3.74), P for trend = 0.10]. No significant effect modification appeared for other potential confounders, including BMI, physical activity, weight change from blood collection until 2 years before diagnosis, smoking status, alcohol consumption, postmenopausal hormone use, parity, regular aspirin and non-aspirin NSAID use and CRC family history (P for interactions >0.05) (data not shown).

Discussion

In our nested case–control study, we report that pre-diagnostic leukocyte mtDNAcn was inversely associated with subsequent CRC risk in a dose-dependent manner. Our findings are in line with results from a case–control study of 444 CRC cases (mean baseline age = 58.6) and 1423 controls (mean baseline age = 55.2) nested within the Shanghai Women’s Health Study, in which Huang et al. found that lower mtDNAcn was associated with higher risk of CRC [lowest versus highest tertile, OR (95% CI) = 1.44 (1.06–1.94), P for trend = 0.02] (13). In another case–control study nested within the Singapore Chinese Health Study of women and men, Thyagarajan et al. reported a U-shaped relationship between mtDNAcn and CRC risk among 422 cases (mean baseline age = 66.1) and 874 controls (mean baseline age = 57.6) [lowest versus second quartile, OR = 1.81 (1.13–2.89), highest versus second quartile, OR = 3.40 (2.15–5.36), P for curvilinearity < 0.0001] (14).

Besides CRC, several other cancers have also been inversely associated with mtDNAcn in epidemiological studies. For example, Meng et al. studied the association between mtDNAcn and melanoma in a case–control study (272 cases and 293 controls) nested within the NHS, and found an inverse association among the high cumulative UV exposure group [low versus high mtDNAcn, OR (95% CI) = 3.40 (1.46–7.92), P for trend = 0.004] (16). In another study by Meng et al. using both NHS (women) and the Health Professionals Follow-Up Study (HPFS, men), among current smokers, those with median mtDNAcn levels were found to have higher risk of lung cancer than those with high mtDNAcn levels [median versus high mtDNAcn, OR (95% CI) = 2.09 (1.12–3.90)] (17). Also, Xie et al. found an inverse association between mtDNAcn and soft tissue sarcoma among 325 patients and 330 healthy controls (age, sex, ethnicity matched); among both men and women, lower mtDNAcn was associated with a significantly increased risk of soft tissue sarcoma [< median versus ≥ median, AOR (95% CI) = 2.71(1.94–3.82)] (24). However, mixed results (including both positive and null associations) were also reported for the relationship between mtDNAcn and other cancers, such as renal cell carcinoma and non-Hodgkin lymphoma (25,26). Considering the complexity of carcinogenesis, it is possible that the relationship between mtDNAcn and cancer risk may be site-specific, depending on the specific organ or tissue of origin.

Elevated oxidative stress may affect the abundance of mitochondria and mtDNAcn as well as mitochondrial function (27,28). Recent evidence has shown the existence of various DNA-repair pathways in mitochondria, such as mismatch repair, base excision repair, homologous recombination and non-homologous end joining, lesion bypass and mtDNA degradation (6,7). However, when the rate of oxidative damage overwhelms the ability of these mechanisms to repair mtDNA efficiently, mtDNA may proliferate, followed by the eventual loss of mtDNA (27). Specifically, when mtDNA is impaired by excessive oxidative stress, healthy mitochondria may first increase their DNA copy number to counteract the metabolic defects in injured mitochondria (27). However, when the damage exceeds the limitation of the feedback mechanism, increasing mtDNAcn can no longer cope with the stress. This results in a net decrease in mtDNAcn, because mtDNA undergoes degradation by the inner cellular enzyme system to prevent excessive accumulation of oxidative stress damage (28). These mechanisms may explain the inconsistent findings from previous studies, and the inverse association we observed might be because extensive oxidative stress may have surpassed mitochondrial capacity to compensate for oxidative damage. Moreover, previous studies have demonstrated a strong positive association between mtDNAcn and telomere length (18,29), a crucial marker of cellular aging and the cumulative burden of oxidative stress (30). With each cell division, telomeres undergo shortening, and oxidative stress could increase this erosion (31). The positive correlation between mtDNAcn and telomere length also implies the potential of pre-diagnostic mtDNAcn serving as a biomarker for predicting oxidative stress-related outcomes.

Additionally, as a reflection of oxidative stress levels, mtDNAcn may be especially closely associated with risks of obesity-related cancers, such as CRC (30). Overall, obesity and abdominal adiposity may lead to increases in oxidative stress and systemic inflammation (32), and have been associated with elevated CRC risk (33). Our previous work showed that in healthy women, mtDNAcn was inversely associated with BMI even after adjusting for telomere length (18). Recently, Hang et al. also found that mtDNAcn tends to decrease continuously and persistently with adiposity over the life course (34). In addition, other environmental exposures such as exercise and smoking may also be involved in the regulation of mtDNAcn (35,36). For example, our own group found that duration and pack-years of smoking were inversely associated with mtDNAcn in leukocytes, while consumption of whole fruits and intake of flavanones (a group of antioxidants abundant in fruits) were positively associated with mtDNAcn (37). Smoking and physical inactivity are well-established risk factors for CRC (9,38), while fruit and vegetables may reduce risk (39). These prior data suggest that mtDNAcn could be a marker or mediator of the accumulating environmental exposures and associated systemic inflammation, and may exert an indirect influence on CRC risk.

To the best of our knowledge, this is the first prospective study examining the association between mtDNAcn and CRC risk in a western population. Our study has several strengths, including its prospective design, long-term follow-up, pre-diagnostic assessment of mtDNAcn and a comprehensive list of covariates. In addition, we include only incident CRC cases diagnosed after blood collection, which avoids the potential reverse influence of cancer progress and treatment effects on leukocyte mtDNAcn levels. In the sensitivity analysis, results barely changed after removing CRC cases diagnosed within 2 years after blood draw, suggesting that the observed association is unlikely the result of undiagnosed CRC present at blood draw.

Notably, in our study, we observed a significant inverse association for proximal colon cancer but not for distal colon or rectal cancer. This may be due to the small number of cancer cases at the latter sites. However, research has shown that clinical, pathological/histological and molecular features differ between colon and rectal cancer, as well as between distal (left side) and proximal (right side) colon cancer (40–42). For example, proximal colon cancers are more likely to be microsatellite instability-high (MSI-high) tumors, while distal colon cancers are more likely to be chromosomal instability-high tumors (42). Also, previous research has demonstrated that associations between environmental factors and CRC risk may be modified by tumor molecular subtypes (43–45). Recently, van Osch et al. found that, compared to other CRC tissues, mtDNAcn was significantly lower in CRC tissues with BRAF mutation (a mutated gene typically in MSI-high tumors) and those with high-level MSI, while mtDNAcn was higher in CRC tissues with KRAS mutation (a mutated gene typically in chromosomal instability-high tumors) (46). Whether these molecular features interact with mtDNAcn in modifying risk of CRC at subsites requires further investigation.

We acknowledged some limitations of our study. One is the relatively modest sample size in stratified subgroups, which limited the statistical power of interaction and stratified analyses. Another limitation is the lack of detailed clinical–pathological characteristics and molecular classifications of these tumors. Future research investigating the mtDNAcn/CRC relationship by cancer molecular subtypes according to established markers and somatic profiles is needed.

In summary, in this nested case–control study, we found a significant inverse association between pre-diagnostic leukocyte mtDNAcn and CRC risk. Further investigations are warranted to explore whether mtDNAcn could become a valuable and long-term biomarker in evaluating the risk and prognosis of CRC. Importantly, additional basic experimental studies are needed to explore the biological mechanisms underlying the relationship between mtDNAcn and CRC carcinogenesis.

Supplementary material

Supplementary data are available at Carcinogenesis online.

Supporting information: Supplementary methods: mtDNAcn ascertainment and validation.

Glossary

Abbreviations

AHEI

Alternate Healthy Eating Index

BMI

body mass index

CI

confidence interval

CRC

colorectal cancer

HPFS

Health Professionals Follow-Up Study

IBD

inflammatory bowel disease

MET

metabolic equivalent

MSI

microsatellite instability

mtDNA

mitochondrial DNA

mtDNAcn

mitochondrial DNA copy number

NHS

Nurses’ Health Study

NSAID

non-steroidal anti-inflammatory drugs

OR

odds ratio

SCHS

Singapore Chinese Health Study

SWHS

Shanghai Women’s Health Study

Funding

This work is supported by NIH grants UM1 CA186107, R01 CA49449, and P01CA87969. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. Dr. Hongmei Nan is partially supported by the Walther Cancer Foundation.

Conflict of Interest Statement: None declared.