Many cohort studies have reported inverse associations between circulating 25-hydroxyvitamin D (25(OH)D) and mortality( Reference Chowdhury, Kunutsor and Vitezova 1 ). Vitamin D exists in two forms, ergocalciferol (vitamin D2) and cholecalciferol (vitamin D3); 25-hydroxyergocalciferol (25(OH)D2) and 25-hydroxycholecalciferol (25(OH)D3) can be quantified separately; and a large study of their relative prevalence in New York has been published( Reference Zhang, Stolla and Kwong 2 ). However, no cohort studies of mortality have presented associations separately for 25(OH)D2 and 25(OH)D3; many used assays such as RIA and chemiluminescent immunoassays( Reference Chowdhury, Kunutsor and Vitezova 1 ) that cannot quantify them separately.
Cholecalciferol is the natural form of vitamin D in man, endogenously synthesised following UVB irradiation of 7-dehydrocholesterol in skin( Reference Holick 3 ). Dietary sources of cholecalciferol include fatty fish, fortified foods and supplements( Reference Holick 3 ). Ergocalciferol, an exogenous form of vitamin D, is produced in fungi following UVB irradiation of ergosterol. People can obtain ergocalciferol from some fortified foods (e.g. margarine)( Reference Tripkovic, Lambert and Hart 4 ) and some species of mushrooms, and it is used in supplements in some countries( Reference Holick 3 , Reference Bjelakovic, Gluud and Nikolova 5 ). In Australia, margarine contains low levels of ergocalciferol (1 μg/100 g)( 6 ).
A systematic review and meta-analysis of randomised controlled trials (RCT) comparing the effects of ergocalciferol and cholecalciferol supplementation on 25(OH)D concentrations found that cholecalciferol is more effective than ergocalciferol at increasing serum 25(OH)D( Reference Tripkovic, Lambert and Hart 4 ), especially when given as large weekly or monthly doses. Ergocalciferol might also increase catabolism of cholecalciferol; several RCT have reported that ergocalciferol supplementation lowered 25(OH)D3 concentrations( Reference Tjellesen, Hummer and Christiansen 7 – Reference Nimitphong, Saetung and Chanprasertyotin 13 ). Reviews of RCT of the effects of vitamin D supplementation on mortality show that there was a modest decrease in mortality for trials of cholecalciferol supplementation, but no decrease for trials of ergocalciferol supplementation, although the ergocalciferol trials were few and of relatively low quality( Reference Chowdhury, Kunutsor and Vitezova 1 , Reference Bjelakovic, Gluud and Nikolova 5 ).
Given limited evidence from trials that ergocalciferol and cholecalciferol supplementation might have different effects on mortality, observational studies of 25(OH)D and mortality should separate the two forms. The advent of liquid chromatography–tandem mass spectrometry (LC–MS/MS) now makes this possible because it enables simultaneous separate measurement of 25(OH)D2 and 25(OH)D3. We report here analyses of 25(OH)D, 25(OH)D3 and 25(OH)D2 in relation to all-cause mortality in a prospective cohort study of middle-aged men and women. The large size of the study and the availability of data on self-reported health about four years after recruitment also enabled us to assess the potential impact of reverse causality, which has been a major limitation of existing cohort studies.
A case–cohort study to investigate vitamin D status and risk of cancer, diabetes and mortality was conducted within the Melbourne Collaborative Cohort Study (MCCS), a prospective cohort study of 41 514 participants (25 % Southern European migrants) aged predominantly between 40 and 69 years at study recruitment (1990–1994). Details of the study design, recruitment and procedures are published elsewhere( Reference Giles 14 ).
At baseline participants were interviewed about lifestyle and medical history, completed a self-administered 121-item FFQ developed for the MCCS( Reference Ireland, Jolley and Giles 15 ), and anthropometric measurements (height, weight, waist and hip circumferences) were performed according to a standard protocol. Blood samples were collected and from the second year of recruitment (approximately 75 % of participants), whole blood was spotted on to Guthrie cards that were air dried and stored in the dark at room temperature. About four years after baseline, participants were mailed a questionnaire that included a question about their general health status, rated as poor, fair, good, very good or excellent.
Eligibility for the case–cohort study was restricted to 29 206 participants who had dried blood spot samples and, to reduce the possibility of reverse causality, no cancer diagnosed before baseline. The present analysis was restricted to the 2410 people who died by 31 December 2007 and the sub-cohort, a sex-stratified random sample of 2996 participants, with sampling fractions of 7·8 % for women and 13·6 % for men, approximately double the number of incident breast cancer and prostate cancer cases, respectively (Fig. 1).
Measurement of 25-hydroxyvitamin D
The mean time between blood sampling and 25(OH)D measurement was 18·1 (sd 0·8) years. Concentration of 25(OH)D3 has previously been shown to be stable in dried blood spots stored for 8–22 years( Reference Eyles, Anderson and Ko 16 ). Measurements of 25(OH)D2 and 25(OH)D3 were made in the laboratory of D.W.E. using LC–MS/MS( Reference Eyles, Anderson and Ko 16 ) over 15 months in thirty-one batches of approximately 230 samples each. This laboratory uses National Institute of Standards and Technology calibrants and participates in the Vitamin D External Quality Assessment Scheme. The absolute limits of detection were 1 nmol/l for 25(OH)D3 and 2 nmol/l for 25(OH)D2 ( Reference Eyles, Anderson and Ko 16 ). We have previously reported high reliability of dried blood spot 25(OH)D measurements using 493 replicates from the case–cohort study; within- and between-batch intraclass correlations were 0·82 (95 % CI 0·80, 0·85) and 0·73 (95 % CI 0·68, 0·78), respectively( Reference Heath, Williamson and Ebeling 17 ). To estimate the reliability of the 25(OH)D2 measurements, we calculated the intraclass correlation coefficient for the presence v. absence of 25(OH)D2 for the 493 replicate measurements. For a binary variable with two replicates, the intraclass correlation is the same as Cohen’s κ. We calculated bias-corrected CI using a bootstrap method with 10 000 replications. The intraclass correlation for detectability of 25(OH)D2 was 0·54 (95 % CI 0·25, 0·72).
Direct comparison between dried blood spot and serum/plasma 25(OH)D concentrations is problematic because 25(OH)D is excluded from erythrocytes. Therefore, for initial descriptions of the concentrations, we corrected the observed concentrations for mean sex-specific haematocrit( Reference Heath, Williamson and Ebeling 17 ). For all other analyses of 25(OH)D and 25(OH)D3, we used a calibration equation to predict plasma concentration( Reference Heath, Williamson and Ebeling 17 ). In a subset of sixty-two MCCS participants who had dried blood spot and matching plasma samples (two replicates for each biospecimen type), we found good agreement between dried blood spot 25(OH)D and plasma 25(OH)D concentrations (R 2=0·73) and developed an equation to estimate plasma 25(OH)D concentration from the dried blood spot concentration( Reference Heath, Williamson and Ebeling 17 ). Because few participants in that study had detectable 25(OH)D2, the calibration equation applies to total 25(OH)D and 25(OH)D3 only. The sensitivity and specificity of 25(OH)D2 detection in the dried blood spot samples compared with the plasma samples was respectively 100·0 (95 % CI 29·0 100·0) % and 94·9 (95 % CI 85·9, 98·9) % for the first replicate, and 50·0 (95 % CI 7·0, 93·0) % and 100·0 (95 % CI 93·8, 100·0) % for the second replicate.
Ascertainment of deaths
Deaths to 31 December 2007 were identified by linkage to the Victorian Registry of Births, Deaths and Marriages and the National Death Index at the Australian Institute of Health and Welfare.
Association between 25-hydroxyergocalciferol and 25-hydroxycholecalciferol
One participant whose 25(OH)D2 concentration was several times higher than the next highest value was excluded from these analyses. For participants with detectable 25(OH)D2, we estimated Spearman’s correlation between the raw 25(OH)D3 and 25(OH)D2 concentrations. We used linear regression to estimate the difference in mean plasma 25(OH)D and 25(OH)D3 concentrations for participants with and without detectable 25(OH)D2, with adjustment for sex. For the regression analyses, we used estimated plasma 25(OH)D and 25(OH)D3 concentrations derived from our calibration equation( Reference Heath, Williamson and Ebeling 17 ), but did not remove season and batch effects. The adjustment for sex was necessary because men had higher 25(OH)D and 25(OH)D3 concentrations and were more likely to have detectable 25(OH)D2.
Association between vitamin D and mortality
Participants with missing data for covariates (n 44) and those with extreme total energy intakes (<1st and >99th sex-specific percentiles; n 100) were excluded (Fig. 1).
Total 25(OH)D and 25(OH)D3 concentrations were adjusted for batch and season, then converted to plasma-equivalent values using our calibration equation( Reference Heath, Williamson and Ebeling 17 ). To remove batch effects, we first fitted a mixed-effects linear regression model with a random effect for batch to the sub-cohort observations. Next, for all observations, the predicted batch-specific deviations from the overall mean were subtracted from the observed values. Seasonal variation was removed by fitting linear regression models for 25(OH)D and 25(OH)D3 as a sinusoidal function of the time of year when blood was sampled( Reference Cox 18 ), then standardising measurements to represent 25(OH)D and 25(OH)D3 measurements on the dates when the annual mean was attained (15 May and 14 November; late autumn and late spring in Australia). Participants were grouped into sex-specific quintiles based on the distribution of batch- and season-adjusted 25(OH)D or 25(OH)D3 in the sub-cohort. For 25(OH)D2, batch and seasonal adjustments were not performed because there were too few participants with detectable 25(OH)D2.
Cox regression was used to estimate hazard ratios (HR) and 95 % CI. Barlow’s method, robust se ( Reference Barlow, Ichikawa and Rosner 19 ) and Wald tests were used to account for the case–cohort design. Age was the timescale and models were stratified by sex and country of birth. Follow-up began at baseline and ended at the date of death, date left Australia or 31 December 2007, whichever came first. The proportional hazards assumption was assessed by fitting interactions between each covariate separately and attained age. There was no evidence that any covariate violated the assumptions.
We estimated HR for all-cause mortality for each quintile of 25(OH)D and 25(OH)D3 relative to the lowest quintile and for the presence v. absence of detectable 25(OH)D2, adjusting for confounders described below. For 25(OH)D and 25(OH)D3 we also estimated the HR per 25 nmol/l by using the continuous forms of the variables. We fitted models with each vitamin D variable alone, a model including continuous 25(OH)D and its interaction with detectable 25(OH)D2, and a model including main effects for continuous 25(OH)D3 and detectable 25(OH)D2. The interaction model assesses whether the relationship between 25(OH)D and mortality depends on the source of the 25(OH)D. Effect modification by sex was assessed by fitting interactions between each vitamin D variable (continuous for 25(OH)D and 25(OH)D3) and sex.
The following confounders were selected using a causal diagram and a priori knowledge from existing literature: age, sex, country of birth (Australia/New Zealand/Northern Europe or Southern Europe), an area-based measure of socio-economic disadvantage (quintiles from most disadvantaged to least disadvantaged), highest education level attained (primary school, some secondary school, secondary school, tertiary qualification), total energy intake (sex-specific quartiles, kJ/d), Mediterranean diet score (three categories indicating dietary pattern, with the highest representing high adherence to a Mediterranean diet), alcohol (never, former and sex-specific tertiles of current intake), smoking (never, former, current), physical activity (four categories reflecting frequency and intensity of exercise in the past 6 months), waist circumference (sex-specific quartiles, centimetres), and self-reported history of diabetes mellitus, hypertension and CVD (angina, myocardial infarction or stroke).
We repeated the analyses for the continuous versions of 25(OH)D and 25(OH)D3 after excluding participants whose values were greater than the 99th percentiles to assess whether outliers had undue influence on the HR. We also assessed non-linearity of the dose–response relationships by fitting restricted cubic spline models with four knots at fixed, equally spaced percentiles (5 %, 35 %, 65 %, 95 %).
People in poor health may have had reduced sun exposure, leading to lower 25(OH)D3. They may also have taken supplements, leading to higher 25(OH)D2, since vitamin D supplements in Australia in the early 1990s contained ergocalciferol( Reference Nowson, Diamond and Pasco 20 ). Thus, observed associations between vitamin D and mortality could potentially be explained by reverse causality. To assess this possibility, a sensitivity analysis was conducted by restricting the analysis to people who reported having good, very good or excellent health on the 4-year follow-up questionnaire. For that analysis, follow-up began on the date of completion of that questionnaire.
Analyses were performed using the statistical software package Stata version 13·1.
Dried blood spots were not located for sixteen participants and seven deaths were identified after the study began, leaving 5108 for analysis (Fig. 1). All samples had detectable concentrations of 25(OH)D3 (range for haematocrit-corrected values, 2·4–146·9 nmol/l; to convert 25(OH)D in nmol/l to ng/ml, divide by 2·496), whereas 25(OH)D2 was detected in sixty-three samples (1·2 %; range for haematocrit-corrected values, 4·1–195·0 nmol/l). The median plasma-equivalent 25(OH)D concentration, adjusted for batch and season, was 48·4 nmol/l; 99 % of values were greater than 19·7 nmol/l and 1 % greater than 107·3 nmol/l.
Participants who were born in Southern Europe, had primary school education or less, were in the lowest energy intake quartile, did not consume alcohol (both lifetime abstainers and former drinkers), were current smokers, did not do any physical activity, were in the largest waist circumference quartile or had diabetes mellitus at baseline were more likely to have low 25(OH)D3 concentrations (Table 1). These patterns were almost identical for 25(OH)D (data not shown). Participants with high alcohol consumption and current smokers were less likely to have 25(OH)D2, while those who took multivitamins were more likely to have 25(OH)D2 (Table 2).
IQR, interquartile range; NZ, New Zealand.
Values are median and interquartile range for 25(OH)D3 concentrations and age, or number and percentage for categorical variables.
* To convert 25(OH)D in nmol/l to ng/ml, divide by 2·496.
† Quartiles of total energy intake (kJ/d): 1, females=3149–6301, males=3755–7399; 2, females=6302–8007, males=7400–9192; 3, females=8008–9827, males=9193–11 397; 4, females=9828–18 831, males=11 398–22 150.
‡ Categories of alcohol intake (g/d): low, females 0·1–4·2, males 0·1–9·2; medium, females 4·3–14·9, males 9·3–25·7; high, females 15·0–117·6, males 25·8–221·4.
§ Quartiles of waist circumference (cm): 1, females=52·7–70·6, males=62·0–85·9; 2, females=70·7–76·9, males=86·0–91·9; 3, females=77·0–85·9, males=92·0–98·4; 4, females=86·0–134·9, males=98·5–131·0.
NZ, New Zealand.
All values are number and percentage.
* Categories of alcohol intake (g/d): low, females=0·1–4·2, males=0·1–9·2; medium, females=4·3–14·9, males=9·3–25·7; high, females=15·0–157·9, males=25·8–232·2.
† Quartiles of waist circumference (cm): 1, females=47·0–70·5, males=62·0–85·9; 2, females=70·6–76·9, males=86·0–91·9; 3, females=77·0–85·4, males=92·0–98·3; 4, females=85·5–153·6, males=98·4–143·0.
The mean plasma 25(OH)D concentration was 11·9 (95 % CI 7·3, 16·6) nmol/l higher (P<0·001) when 25(OH)D2 was detected than when it was not detected, whereas the mean plasma 25(OH)D3 concentration was 4·9 (95 % CI −9·6, −0·3) nmol/l lower (P=0·04). For participants with detectable 25(OH)D2 there was a weak, non-significant inverse correlation between their raw 25(OH)D2 and 25(OH)D3 concentrations (Spearman’s correlation =−0·09, P=0·50).
Vitamin D and mortality
During a mean of 13·7 (sd 2·2) years of follow-up (maximum 16·9 years), 2410 (8·3 %) participants died. The present analysis included 4964 participants, 2307 of whom died (Fig. 1).
The HR for the highest compared with the lowest total 25(OH)D quintile was 0·67 (95 % CI 0·54, 0·84) and the HR per 25 nmol/l increment in 25(OH)D was 0·86 (95 % CI 0·78, 0·96; P=0·007; Table 3). There was no evidence that the splines fitted better than a linear trend (P=0·44) and the HR was unchanged after excluding outliers (data not shown).
IQR, interquartile range; ref., reference category.
* To convert 25(OH)D in nmol/l to ng/ml, divide by 2·496.
† Sub-cohort participants.
‡ Adjusted for age, area-based index of socio-economic disadvantage, education, total energy intake, Mediterranean diet pattern, alcohol intake, smoking status, physical activity, waist measurement, diabetes mellitus at baseline, history of hypertension, history of angina, history of myocardial infarction, history of stroke, and stratified by sex and Southern European migrant status.
§ Adjusted for batch and seasonal effects.
|| Interaction between 25(OH)D and 25(OH)D2.
¶ Mutual adjustment for 25(OH)D2 and 25(OH)D3.
Because few participants had 25(OH)D2, the HR for 25(OH)D3 were essentially identical to those for total 25(OH)D (Table 3). The splines fitted no better than a linear trend (P=0·42) and excluding outliers did not affect the HR (data not shown). The HR for the presence of detectable 25(OH)D2 was 1·80 (95 % CI 1·09, 2·97; P=0·02; Table 3). One batch of measurements had a disproportionate number of samples with 25(OH)D2; results were similar when this batch was excluded (HR=1·88; 95 % CI 1·07, 3·28; P=0·03).
When 25(OH)D3 and detectable 25(OH)D2 were included in the same model, their HR changed little (Table 3). The interaction between the continuous 25(OH)D and 25(OH)D2 was significant (P=0·02). From this model, the HR per 25 nmol/l increment in 25(OH)D concentration was 0·85 (95 % CI 0·77, 0·95) when there was no detectable 25(OH)D2 and 1·06 (95 % CI 0·87, 1·29) when there was detectable 25(OH)D2.
There was little evidence that the HR for any of the vitamin D variables varied by sex (interactions: 25(OH)D, P=0·30; 25(OH)D3, P=0·28; 25(OH)D2, P=0·25).
Vitamin D and mortality in people reporting good health
When restricted to the 3159 participants (1136 of whom died) who reported good, very good or excellent health four years after baseline blood sample collection, forty-six had detectable 25(OH)D2. The HR per 25 nmol/l increment in 25(OH)D and 25(OH)D3 were 0·85 (95 % CI 0·73, 0·98; P=0·021) and 0·83 (95 % CI 0·72, 0·95; P=0·009), respectively. The HR for detectable 25(OH)D2 was 1·75 (95 % CI 0·98, 3·10; P=0·06). The HR from the model including 25(OH)D and the interaction with detectable 25(OH)D2 were also similar to those for all participants (data not shown).
Higher 25(OH)D3 and 25(OH)D concentrations were associated with a moderately reduced risk of death, whereas the presence of 25(OH)D2 was associated with a moderately increased risk of death, despite participants with detectable 25(OH)D2 having higher mean 25(OH)D concentration. The inverse association between mortality and 25(OH)D was restricted to those with no detectable 25(OH)D2. The HR for 25(OH)D3 and detectable 25(OH)D2 changed little when they were included in the same model. These associations were similar when the analysis was restricted to people who reported being in good to excellent health about four years after blood draw.
Strengths of the current study include its prospective design, long follow-up, use of highly sensitive LC–MS/MS for separate measurement of both 25(OH)D forms, availability of extensive information on potential confounders, accurate and complete mortality data, and broad generalisability. Measurement error is unlikely to explain the associations: the samples were analysed in random order and blind to outcome status and so any measurement error would be non-differential and attenuate associations. We had no data on vitamin D supplementation, but only 16 % of participants reported using multivitamins at baseline (which at the time in Australia contained ergocalciferol only, but which now contain cholecalciferol only). We controlled for history of disease and several risk factors, but cannot rule out the possibility of residual confounding. Given the small number of people with detectable 25(OH)D2, chance is also a potential explanation for its positive association with mortality.
Inclusion of middle-aged adults with no pre-baseline cancer reduced the possibility of reverse causality explaining the results. However, reverse causality might explain the positive association with 25(OH)D2 and inverse association with 25(OH)D3 if people in poor health were more likely to take supplements and spend less time outdoors, thus having lower concentrations of 25(OH)D3. Our sensitivity analysis, restricted to participants in good to excellent health several years after baseline, was designed to assess this possibility. While it is not possible to rule out reverse causality, the sensitivity analysis showed no evidence of its occurrence. The HR for total 25(OH)D and 25(OH)D3 were very similar; this might not be the case in populations with more ergocalciferol use, such as in the USA( Reference Zhang, Stolla and Kwong 2 ). Finally, the absolute 25(OH)D values reported should be interpreted cautiously as these were plasma-equivalent 25(OH)D concentrations estimated from measurements of 25(OH)D in dried blood spots and with adjustment for batch and season.
A recent meta-analysis of cohort studies of vitamin D and all-cause mortality found a relative risk (RR) of 1·16 (95 % CI 1·08, 1·23) for a 10 ng/ml decrease in 25(OH)D, which is equivalent to 0·86 (95 % CI 0·81, 0·93) for a 25 nmol/l increase, and thus identical to the estimate from our study for 25(OH)D( Reference Chowdhury, Kunutsor and Vitezova 1 ). There was substantial heterogeneity between studies (I 2=90 %) that did not appear to be explained by any of the study characteristics used to form subgroups (no formal meta-regression was performed). The association was similar when stratified by latitude (≤40° v. >40°) and by age at baseline (<70 v. ≥70 years)( Reference Chowdhury, Kunutsor and Vitezova 1 ). It was slightly stronger for males, whereas we found similar associations for men and women. The association was stronger in the first five years after baseline( Reference Chowdhury, Kunutsor and Vitezova 1 ). One possible explanation for this finding is reverse causality – our study is one of few that have been able to assess this possibility by performing an analysis restricted to people in good health several years after baseline. Our results for 25(OH)D are also consistent with a recent meta-analysis of individual data from eight cohort studies from Europe and the USA, in which the RR for mortality in the lowest v. the highest quintile was 1·57 (95 % CI 1·36, 1·81), which is equivalent to RR of 0·64 (=1/1·57) for the highest v. the lowest quintile (our corresponding HR was 0·67)( Reference Schöttker, Jorde and Peasey 21 ). In that study, there was also little heterogeneity by age and sex( Reference Schöttker, Jorde and Peasey 21 ). Some cohort studies have found U-shaped or reverse J-shaped relationships between 25(OH)D and all-cause mortality( Reference Durup, Jørgensen and Christensen 22 – Reference Zittermann, Iodice and Pilz 25 ) but we found no such evidence. However, because few of our participants had concentrations above 100 nmol/l, our ability to assess curvilinear relationships was limited.
We are not aware of any cohort studies of 25(OH)D2 and 25(OH)D3 and mortality. According to a Cochrane review of RCT of vitamin D supplementation, cholecalciferol reduced all-cause mortality by 6 % (RR=0·94; 95 % CI 0·91, 0·98; P=0·002). Ergocalciferol had no apparent beneficial effect (RR=1·02; 95 % CI 0·96, 1·08; P=0·54), but in RCT involving participants with vitamin D insufficiency, the risk of mortality was elevated (RR=1·20; 95 % CI 1·05, 1·37; P=0·008)( Reference Bjelakovic, Gluud and Nikolova 5 ). The authors commented that ‘the effect of vitamin D2 may be neutral or even detrimental’( Reference Bjelakovic, Gluud and Nikolova 5 ). In another meta-analysis of RCT, cholecalciferol supplementation reduced all-cause mortality (RR=0·89; 95 % CI 0·80, 0·99). Overall, ergocalciferol had no beneficial effect (RR=1·04; 95 % CI 0·97, 1·11), but increased mortality risk in RCT using doses of 15–50 µg/d (600–2000 IU/d; RR=1·20; 95 % CI 1·05, 1·38), in RCT with shorter mean intervention periods (<1·5 years; RR=1·20; 95 % CI 1·05, 1·37) and in RCT with shorter follow-up (<2·5 years; RR=1·19; 95 % CI 1·04, 1·36)( Reference Chowdhury, Kunutsor and Vitezova 1 ). The median follow-up from the included RCT was only 1·4 years, the median age of participants was 77 years, there were relatively few deaths in each study, and thirteen out of the twenty-two RCT involved participants with pre-existing chronic disease( Reference Chowdhury, Kunutsor and Vitezova 1 ). Thus, existing RCT are not necessarily generalisable to younger/middle-aged community-dwelling populations.
Our findings that participants with detectable 25(OH)D2 had higher mean 25(OH)D but lower mean 25(OH)D3 concentration is consistent with the Osteoporotic Fractures in Men Study (MrOS)( Reference Swanson, Nielson and Shrestha 26 ). These findings complement evidence from RCT showing that ergocalciferol might interfere with production of 25(OH)D3 and/or induce processes that degrade 25(OH)D3 ( Reference Tjellesen, Hummer and Christiansen 7 – Reference Nimitphong, Saetung and Chanprasertyotin 13 ).
Several potential mechanisms could explain differences in effects of ergocalciferol and cholecalciferol. The presence of a methyl group on C24 of ergocalciferol metabolites may slightly alter their ability to bind to, or their affinity for, vitamin D binding protein( Reference Hollis 27 ), vitamin D hydroxylases and the vitamin D receptor( Reference Jones, Assar and Harnpanich 28 ). The lower affinity of 25(OH)D2 for vitamin D binding protein relative to 25(OH)D3 means its free fraction in blood is likely to be higher, consistent with the faster turnover of 25(OH)D2 ( Reference Armas, Hollis and Heaney 8 , Reference Hollis 27 ). The shorter circulating half-life of ergocalciferol metabolites may lead to up-regulation of enzymes that not only metabolise ergocalciferol, but also concomitantly induce accelerated catabolism and clearance of 25(OH)D3 ( Reference Armas, Hollis and Heaney 8 , Reference Logan, Gray and Peddie 11 ).
Differential specificities of the various cytochrome P450 (CYP) enzymes that metabolise vitamin D may also help to explain the apparent effect of ergocalciferol on 25(OH)D3 concentration. While the microsomal vitamin D 25-hydroxylase, CYP2R1, 25-hydroxylates both ergocalciferol and cholecalciferol equally well( Reference Cheng, Motola and Mangelsdorf 29 , Reference Shinkyo, Sakaki and Kamakura 30 ), mitochondrial CYP27A1 25-hydroxylates cholecalciferol( Reference Sawada, Sakaki and Ohta 31 ), but not ergocalciferol( Reference Guo, Strugnell and Back 32 ).
Another enzyme, CYP3A4, 25-hydroxylates ergocalciferol and not cholecalciferol, but also 24-hydroxylates several vitamin D metabolites( Reference Gupta, Hollis and Patel 33 – Reference Xu, Hashizume and Shuhart 35 ). In particular, CYP3A4 plays a major role in 23- and 24-hydroxylation of the biologically active form of cholecalciferol, 1,25-dihydroxycholecalciferol (1,25(OH)2D3), in the liver and small intestine( Reference Xu, Hashizume and Shuhart 35 ). CYP3A4 is the most abundant liver cytochrome P450 enzyme and has broad specificity, catabolising more than 50 % of known drugs( Reference Gupta, He and Patrick 34 , Reference Prosser and Jones 36 ). Many drugs inhibit or induce CYP3A4, thereby potentially increasing vitamin D turnover( Reference Gupta, Hollis and Patel 33 , Reference Xu, Hashizume and Shuhart 35 ). It is possible that ergocalciferol similarly induces CYP3A4, leading to enhanced 23- and 24-hydroxylation of 1,25(OH)2D3; inactivating cholecalciferol and ultimately eliminating it from the body. Thus, a plausible explanation for the inverse association between circulating 25(OH)D2 and 25(OH)D3, and any potential detrimental effect of ergocalciferol on health, is enhanced CYP3A4-induced catabolism of 1,25(OH)2D3 in the presence of ergocalciferol.
In summary, our results confirm those of previous studies that higher levels of 25(OH)D are associated with reduced mortality. For our study at least, this was due to the inverse association between 25(OH)D3 and mortality. Our findings with respect to 25(OH)D2 require confirmation in other populations with higher exposure to this form of vitamin D.
Acknowledgements: This study was made possible by the contribution of many people, including the original investigators, the Programme Manager, and the diligent team who recruited the participants and who continue working on follow-up. The authors would like to express their gratitude to the many thousands of Melbourne residents who continue to participate in the study. Participants’ vital status was ascertained through the Victorian Cancer Registry (VCR) and the Australian Institute of Health and Welfare (AIHW), including the National Death Index. Financial support: Cohort recruitment was funded by VicHealth and The Cancer Council Victoria. This work was supported by the National Health and Medical Research Council (project grant 623208). A.K.H. is supported by an Australian Postgraduate Award. The funding source had no role in the design, analysis or writing of this article. Conflict of interest: None. Authorship: D.W.E. and D.R.E. made equal contributions. A.K.H. formulated the research question, conducted the literature search, analysed and interpreted the data, and wrote the first draft of the report. E.J.W. oversaw the design, statistical analysis, interpretation of the data and drafting of the report. D.K. carried out laboratory measurements of vitamin D status. A.M.H., P.R.E. and R.E.N. reviewed the manuscript. L.B. was involved in the design of the case–cohort study, obtained funding and reviewed the manuscript. G.G.G. conceived and designed the cohort study and reviewed the manuscript. D.W.E. obtained funding, oversaw measurements of vitamin D status and reviewed the manuscript. D.R.E. obtained funding, oversaw the data analysis and interpretation, and reviewed the manuscript. All authors read and approved the final manuscript. Ethics of human subject participation: This study was conducted according to the guidelines laid down in the Declaration of Helsinki and all procedures involving human subjects were approved by the Cancer Council Victoria’s Human Research Ethics Committee. Written informed consent was obtained from all participants.