Associations between vitamin D deficiency and disease risk have been widely reported by conventional epidemiological studies, including diseases which typically have a late-onset over the lifecourse, such as coronary artery disease, but also those which may be diagnosed in early life such as type 1 diabetes (T1D). As a result, vitamin D supplements are widely prescribed with an estimated 18% of adults in the USA reportedly taking supplements daily (Rooney et al., 2017). However, there is increasing evidence from the literature suggesting that vitamin D supplements may be ineffective at reducing disease risk in the population (Bouillon et al., 2022). Furthermore, these conventional association studies may have been susceptible to bias, given that vitamin D levels are known to differ amongst individuals based on various lifestyle factors, including their body mass index (BMI), as well as being prone to reverse causation, for example due to inflammatory processes which are known to lower vitamin D levels (Preiss and Sattar, 2019).

An approach to mitigate the influence of these sources of bias is Mendelian randomization (MR), a causal inference technique which exploits the random allocation of genetic variants at birth to evaluate the genetically predicted effects of modifiable exposures on disease outcomes and circulating biomarkers (Davey Smith and Ebrahim, 2003, Richmond and Davey Smith, 2022). For example, MR has been applied in recent years to suggest that vitamin D supplements are unlikely to have a beneficial effect on risk of T1D (Manousaki et al., 2021), whereas a recent study suggests that genetically lowered vitamin D levels may increase the risk of multiple sclerosis (Mokry et al., 2015). We previously extended the application of MR to investigate epidemiological hypotheses in a lifecourse setting (known to as ‘lifecourse MR’), by deriving sets of genetic variants to separate the independent effects of body size during childhood and adulthood within a multivariable framework (Richardson et al., 2020). Applying this approach has highlighted the putative causal role that early life adiposity may have on outcomes such as T1D risk (Richardson et al., 2022) and cardiac structure (O’Nunain et al., 2022). In contrast, we have demonstrated that its influence on other disease outcomes (e.g. cardiovascular disease; Power et al., 2021) is likely attributed to the long-term consequence of remaining overweight into later life.

Whilst these applications serve as powerful examples of lifecourse MR as an approach to separate the effects of the same exposure based on data derived from early and later life, it has yet to be applied to the same outcome when measured at different timepoints in the lifecourse. In this study, we applied lifecourse MR to investigate the independent effects of childhood and adult body size on 25-hydroxyvitamin D (25OHD) levels measured during childhood (mean age: 9.9 years, range = 8.9–11.5 years) using individual-level data from the Avon Longitudinal Study of Parents and Children (ALSPAC) (Boyd et al., 2013; Fraser et al., 2013) and during adulthood (mean age: 56.5 years, range = 40–69 years) using summary-level data based on individuals from the UK Biobank (UKB) study (Manousaki et al., 2020).

Firstly, using data measured at mean age 9.9 years in the lifecourse from the ALSPAC study, our analyses indicated that childhood body size directly influences vitamin D levels during childhood after accounting for the adult body size genetic instrument in our model (beta = −0.32 standard deviation change in natural log-transformed 25OHD per change in body size category, 95% CI = −0.54 to –0.10, p=0.004) (Figure 1). Using data from the UKB study, evidence of an effect of higher childhood body size on adulthood measured 25OHD levels (beta = −0.14, 95% CI = −0.10 to –0.03, p=2.4 × 10^–4) attenuated in a multivariable setting upon accounting for adulthood body size (beta = 0.04, 95% CI = −0.01 to 0.08, p=0.10). In contrast, strong evidence of an effect for higher adult body size on lower 25OHD levels measured in adulthood was found in the multivariable model accounting for the childhood body size score (beta = −0.17, 95% CI = −0.21 to –0.13, p=4.6 × 10^–17) (Figure 1C). This suggests that childhood body size has an indirect influence of 25OHD levels in adulthood, likely due to its persistent effect throughout the lifecourse (Figure 1D). The full sets of univariable and multivariable estimates derived in this analysis can be found in Supplementary file 1—Table 1 and Supplementary file 1—Table 2, respectively. Similar conclusions were drawn using estimates derived on adult 25OHD levels using findings by Revez et al (Revez et al., 2020; Supplementary file 1—Table 3).

Figure 1

Download asset Open asset

Our findings suggest that increased adiposity exerts a strong effect on lower vitamin D levels during both childhood and adulthood. Separating causal from confounding factors, particularly in a lifecourse setting, would have been extremely challenging to disentangle without the use of genetic variants as achieved in this study. In contrast, appropriately accounting for confounding factors in an conventional epidemiological setting is notoriously challenging, with previous studies reporting evidence of association between vitamin D and T1D in early life even after adjusting for factors such as birthweight (Hyppönen et al., 2001). Taken together with evidence from previous MR studies, which have found that childhood body size (Richardson et al., 2022), but not vitamin D levels (Manousaki et al., 2021), increases risk of T1D, our results suggest that adiposity may have acted as a confounding factor on the observed association between vitamin D and T1D. Evidence from this study therefore highlights the importance of developing a deeper understanding into the role that confounding factors, such as adiposity, may potentially have in distorting observational associations between vitamin D levels and disease risk. Future studies are therefore warranted to disentangle the causal factors which influence disease risk from non-causal confounding factors through the triangulation of multiple lines of evidence including those identified by robustly conducted MR studies.

Our conclusions are also supported by the outcomes of recent large-scale randomized controlled clinical trials (RCTs) of vitamin D supplementation. For instance, as highlighted by recent reviews (Ebeling, 2022; Bouillon et al., 2022), whilst observational studies have found that low vitamin D levels associate with increased risk of cardiovascular disease and cancer outcomes, there has been little convincing evidence from RCTs that a causal relationship may underly these findings. Additionally, recent RCTs have reported evidence of a potential interaction between vitamin D supplementation and BMI, such as the Finnish Vitamin D Trial (FIND) (Virtanen et al., 2022) and D-Health (Waterhouse et al., 2021) studies.

Effect estimates derived from MR studies are conventionally interpreted as ‘lifelong effects’ given that the germline genetic variants harnessed by this approach as instrumental variables are typically fixed at conception. The results of this study serve as a compelling demonstration that the timepoint at which both exposures and outcomes are measured across the lifecourse can meaningfully impact conclusions drawn by MR investigations. Whilst the childhood and adult body size genetic scores were used in this study as a proof of concept, our findings also help to further validate their utility in separating the temporal effects of adiposity within a lifecourse MR framework. Overall, these findings emphasise the importance of future MR studies taking further consideration into the age of participants that their genetic estimates are based on, as well as the age at which cases are diagnosed for disease outcome estimates, before interpreting and drawing conclusions from their findings.

Furthermore, our results suggest that conducting GWAS on populations of different age groups will add value in helping uncover time-varying genetic effects scattered throughout the human genome. Findings from these endeavours should facilitate studies applying techniques such as lifecourse MR, which can provide insight into the direct and indirect effects of modifiable early life exposures on disease outcomes by harnessing genetic estimates obtained from unprecedented sample sizes when conducted in a two-sample setting. That said, lifecourse MR requires careful examination of genetic instruments to ensure that they are capable of robustly separating the effects of an exposure at different timepoints over the lifecourse (Sanderson et al., 2022). Doing so may help to elucidate the critical timepoints whereby conferred risk by these exposures on disease outcomes starts to become immutable, which has important implications for improving patient care in a clinical setting.

All individual level data analysed in this study can be accessed via an approved application to ALSPAC (http://www.bristol.ac.uk/alspac/researchers/access/). Summary-level data on adulthood vitamin D levels can be accessed publicly via the OpenGWAS (https://gwas.mrcieu.ac.uk/). Figure 1 - raw estimates used to generate Figure 1A and Figure 1C can be found in Supplementary Data 1 & 2.

1. 1. Bouillon R
   2. Manousaki D
   3. Rosen C
   4. Trajanoska K
   5. Rivadeneira F
   6. Richards JB

   (2022) The health effects of vitamin D supplementation: evidence from human studies

   Nature Reviews. Endocrinology 18:96–110.

   https://doi.org/10.1038/s41574-021-00593-z

   • PubMed
   • Google Scholar
2. 1. Bowden J
   2. Davey Smith G
   3. Burgess S

   (2015) Mendelian randomization with invalid instruments: effect estimation and bias detection through Egger regression

   International Journal of Epidemiology 44:512–525.

   https://doi.org/10.1093/ije/dyv080

   • PubMed
   • Google Scholar
3. 1. Bowden J
   2. Davey Smith G
   3. Haycock PC
   4. Burgess S

   (2016) Consistent estimation in mendelian randomization with some invalid instruments using a weighted median estimator

   Genetic Epidemiology 40:304–314.

   https://doi.org/10.1002/gepi.21965

   • PubMed
   • Google Scholar
4. 1. Boyd A
   2. Golding J
   3. Macleod J
   4. Lawlor DA
   5. Fraser A
   6. Henderson J
   7. Molloy L
   8. Ness A
   9. Ring S
   10. Davey Smith G

   (2013) Cohort Profile: the ’children of the 90s’--the index offspring of the Avon Longitudinal Study of Parents and Children

   International Journal of Epidemiology 42:111–127.

   https://doi.org/10.1093/ije/dys064

   • PubMed
   • Google Scholar
5. 1. Brandkvist M
   2. Bjørngaard JH
   3. Ødegård RA
   4. Åsvold BO
   5. Davey Smith G
   6. Brumpton B
   7. Hveem K
   8. Richardson TG
   9. Vie GÅ

   (2020) Separating the genetics of childhood and adult obesity: a validation study of genetic scores for body mass index in adolescence and adulthood in the HUNT Study

   Human Molecular Genetics 29:3966–3973.

   https://doi.org/10.1093/hmg/ddaa256

   • PubMed
   • Google Scholar
6. 1. Burgess S
   2. Butterworth A
   3. Thompson SG

   (2013) Mendelian randomization analysis with multiple genetic variants using summarized data

   Genetic Epidemiology 37:658–665.

   https://doi.org/10.1002/gepi.21758

   • PubMed
   • Google Scholar
7. 1. Burgess S
   2. Davies NM
   3. Thompson SG

   (2016) Bias due to participant overlap in two-sample Mendelian randomization

   Genetic Epidemiology 40:597–608.

   https://doi.org/10.1002/gepi.21998

   • PubMed
   • Google Scholar
8. 1. Davey Smith G
   2. Ebrahim S

   (2003) “Mendelian randomization”: can genetic epidemiology contribute to understanding environmental determinants of disease?

   International Journal of Epidemiology 32:1–22.

   https://doi.org/10.1093/ije/dyg070

   • PubMed
   • Google Scholar
9. 1. Ebeling PR

   (2022) Does vitamin D supplementation reduce cardiovascular events and cancer?

   The American Journal of Clinical Nutrition 115:1255–1256.

   https://doi.org/10.1093/ajcn/nqac050

   • PubMed
   • Google Scholar
10. 1. Fraser A
    2. Macdonald-Wallis C
    3. Tilling K
    4. Boyd A
    5. Golding J
    6. Davey Smith G
    7. Henderson J
    8. Macleod J
    9. Molloy L
    10. Ness A
    11. Ring S
    12. Nelson SM
    13. Lawlor DA

    (2013) Cohort profile: The avon longitudinal study of parents and children: Alspac mothers cohort

    International Journal of Epidemiology 42:97–110.

    https://doi.org/10.1093/ije/dys066

    • PubMed
    • Google Scholar
11. 1. Hemani G
    2. Zheng J
    3. Elsworth B
    4. Wade KH
    5. Haberland V
    6. Baird D
    7. Laurin C
    8. Burgess S
    9. Bowden J
    10. Langdon R
    11. Tan VY
    12. Yarmolinsky J
    13. Shihab HA
    14. Timpson NJ
    15. Evans DM
    16. Relton C
    17. Martin RM
    18. Davey Smith G
    19. Gaunt TR
    20. Haycock PC

    (2018) The MR-Base platform supports systematic causal inference across the human phenome

    eLife 7:e34408.

    https://doi.org/10.7554/eLife.34408

    • PubMed
    • Google Scholar
12. 1. Hyppönen E
    2. Läärä E
    3. Reunanen A
    4. Järvelin MR
    5. Virtanen SM

    (2001) Intake of vitamin D and risk of type 1 diabetes: a birth-cohort study

    Lancet 358:1500–1503.

    https://doi.org/10.1016/S0140-6736(01)06580-1

    • PubMed
    • Google Scholar
13. 1. Manousaki D
    2. Mitchell R
    3. Dudding T
    4. Haworth S
    5. Harroud A
    6. Forgetta V
    7. Shah RL
    8. Luan J
    9. Langenberg C
    10. Timpson NJ
    11. Richards JB

    (2020) Genome-wide association study for vitamin d levels reveals 69 independent loci

    American Journal of Human Genetics 106:327–337.

    https://doi.org/10.1016/j.ajhg.2020.01.017

    • PubMed
    • Google Scholar
14. 1. Manousaki D
    2. Harroud A
    3. Mitchell RE
    4. Ross S
    5. Forgetta V
    6. Timpson NJ
    7. Davey Smith G
    8. Polychronakos C
    9. Richards JB

    (2021) Vitamin D levels and risk of type 1 diabetes: A Mendelian randomization study

    PLOS Medicine 18:e1003536.

    https://doi.org/10.1371/journal.pmed.1003536

    • PubMed
    • Google Scholar
15. 1. Mokry LE
    2. Ross S
    3. Ahmad OS
    4. Forgetta V
    5. Davey Smith G
    6. Goltzman D
    7. Leong A
    8. Greenwood CMT
    9. Thanassoulis G
    10. Richards JB

    (2015) Vitamin D and risk of multiple sclerosis: A mendelian randomization studyRisk of Multiple Sclerosis: A Mendelian Randomization Study

    PLOS Medicine 12:e1001866.

    https://doi.org/10.1371/journal.pmed.1001866

    • PubMed
    • Google Scholar
16. 1. O’Nunain K
    2. Park C
    3. Urquijo H
    4. Leyden GM
    5. Hughes AD
    6. Davey Smith G
    7. Richardson TG

    (2022) A lifecourse mendelian randomization study highlights the long-term influence of childhood body size on later life heart structure

    PLOS Biology 20:e3001656.

    https://doi.org/10.1371/journal.pbio.3001656

    • PubMed
    • Google Scholar
17. 1. Power GM
    2. Tyrrell J
    3. Frayling TM
    4. Davey Smith G
    5. Richardson TG

    (2021) Mendelian randomization analyses suggest childhood body size indirectly influences end points from across the cardiovascular disease spectrum through adult body size

    Journal of the American Heart Association 10:e021503.

    https://doi.org/10.1161/JAHA.121.021503

    • PubMed
    • Google Scholar
18. 1. Preiss D
    2. Sattar N

    (2019) Research digest: vitamin D supplementation

    The Lancet. Diabetes & Endocrinology 7:91.

    https://doi.org/10.1016/S2213-8587(19)30007-5

    • PubMed
    • Google Scholar
19. 1. Revez JA
    2. Lin T
    3. Qiao Z
    4. Xue A
    5. Holtz Y
    6. Zhu Z
    7. Zeng J
    8. Wang H
    9. Sidorenko J
    10. Kemper KE
    11. Vinkhuyzen AAE
    12. Frater J
    13. Eyles D
    14. Burne THJ
    15. Mitchell B
    16. Martin NG
    17. Zhu G
    18. Visscher PM
    19. Yang J
    20. Wray NR
    21. McGrath JJ

    (2020) Genome-wide association study identifies 143 loci associated with 25 hydroxyvitamin D concentration

    Nature Communications 11:1647.

    https://doi.org/10.1038/s41467-020-15421-7

    • PubMed
    • Google Scholar
20. 1. Richardson TG
    2. Sanderson E
    3. Elsworth B
    4. Tilling K
    5. Davey Smith G

    (2020) Use of genetic variation to separate the effects of early and later life adiposity on disease risk: mendelian randomisation study

    BMJ 369:m1203.

    https://doi.org/10.1136/bmj.m1203

    • PubMed
    • Google Scholar
21. 1. Richardson TG
    2. Mykkänen J
    3. Pahkala K
    4. Ala-Korpela M
    5. Bell JA
    6. Taylor K
    7. Viikari J
    8. Lehtimäki T
    9. Raitakari O
    10. Davey Smith G

    (2021) Evaluating the direct effects of childhood adiposity on adult systemic metabolism: a multivariable Mendelian randomization analysis

    International Journal of Epidemiology 50:1580–1592.

    https://doi.org/10.1093/ije/dyab051

    • PubMed
    • Google Scholar
22. 1. Richardson TG
    2. Crouch DJM
    3. Power GM
    4. Morales-Berstein F
    5. Hazelwood E
    6. Fang S
    7. Cho Y
    8. Inshaw JRJ
    9. Robertson CC
    10. Sidore C
    11. Cucca F
    12. Rich SS
    13. Todd JA
    14. Davey Smith G

    (2022) Childhood body size directly increases type 1 diabetes risk based on a lifecourse Mendelian randomization approach

    Nature Communications 13:2337.

    https://doi.org/10.1038/s41467-022-29932-y

    • PubMed
    • Google Scholar
23. 1. Richmond RC
    2. Davey Smith G

    (2022) Mendelian randomization: concepts and scope

    Cold Spring Harbor Perspectives in Medicine 12:a040501.

    https://doi.org/10.1101/cshperspect.a040501

    • PubMed
    • Google Scholar
24. 1. Rooney MR
    2. Harnack L
    3. Michos ED
    4. Ogilvie RP
    5. Sempos CT
    6. Lutsey PL

    (2017) Trends in use of high-dose vitamin d supplements exceeding 1000 or 4000 international units daily, 1999-2014

    JAMA 317:2448–2450.

    https://doi.org/10.1001/jama.2017.4392

    • PubMed
    • Google Scholar
25. 1. Sanderson E
    2. Davey Smith G
    3. Windmeijer F
    4. Bowden J

    (2019) An examination of multivariable Mendelian randomization in the single-sample and two-sample summary data settings

    International Journal of Epidemiology 48:713–727.

    https://doi.org/10.1093/ije/dyy262

    • PubMed
    • Google Scholar
26. 1. Sanderson E
    2. Richardson TG
    3. Morris TT
    4. Tilling K
    5. Davey Smith G

    (2022) Estimation of causal effects of a time-varying exposure at multiple time points through multivariable mendelian randomization

    PLOS Genetics 18:e1010290.

    https://doi.org/10.1371/journal.pgen.1010290

    • PubMed
    • Google Scholar
27. 1. Virtanen JK
    2. Nurmi T
    3. Aro A
    4. Bertone-Johnson ER
    5. Hyppönen E
    6. Kröger H
    7. Lamberg-Allardt C
    8. Manson JE
    9. Mursu J
    10. Mäntyselkä P
    11. Suominen S
    12. Uusitupa M
    13. Voutilainen A
    14. Tuomainen T-P
    15. Hantunen S

    (2022) Vitamin D supplementation and prevention of cardiovascular disease and cancer in the Finnish Vitamin D Trial: a randomized controlled trial

    The American Journal of Clinical Nutrition 115:1300–1310.

    https://doi.org/10.1093/ajcn/nqab419

    • PubMed
    • Google Scholar
28. 1. Waterhouse M
    2. Sanguineti E
    3. Baxter C
    4. Duarte Romero B
    5. McLeod DSA
    6. English DR
    7. Armstrong BK
    8. Ebeling PR
    9. Hartel G
    10. Kimlin MG
    11. O’Connell RL
    12. Pham H
    13. van der Pols JC
    14. Venn AJ
    15. Webb PM
    16. Whiteman DC
    17. Neale RE

    (2021) Vitamin D supplementation and risk of falling: outcomes from the randomized, placebo-controlled D-Health Trial

    Journal of Cachexia, Sarcopenia and Muscle 12:1428–1439.

    https://doi.org/10.1002/jcsm.12759

    • PubMed
    • Google Scholar

Decision letter after peer review:

Thank you for submitting your article "Associations between vitamin D and disease risk may be attributed to the confounding influence of adiposity during childhood and adulthood: a lifecourse Mendelian randomization study" for consideration by eLife. Your article has been reviewed by 2 peer reviewers, and I personally oversaw the evaluation in my joint role of Reviewing Editor and Senior Editor. The following individual involved in review of your submission has agreed to reveal their identity: Peter Ebeling (Reviewer #1).

The reviewers have discussed their reviews with one another and with me. I drafted this decision letter to help you prepare a revised submission. Please submit a revised version that addresses these concerns directly. Although we expect that you will address these comments in your response letter, we also need to see the corresponding revision clearly marked in the text of the manuscript. Some of the reviewers' comments may seem to be simple queries or challenges that do not prompt revisions to the text. Please keep in mind, however, that readers may have the same perspective as the reviewers. Therefore, it is essential that you attempt to amend or expand the text to clarify the narrative accordingly.

Essential revisions:

1) As recommended by the two reviewers, please include a discussion on recent trials or studies that shed light on the same topic of the interplay of BMI and vitamin D on health outcomes.

2) Please acknowledge the limitations and additional caveats pointed out by reviewer #2

3) In addition, please revise the manuscript as per the additional recommendations by both reviewers.

Reviewer #1 (Recommendations for the authors):

This reviewer has the following recommendations for the authors:

1. It might be worth including data from recent large RCTs showing interactions of BMI with outcomes related to vitamin D supplementation, cardiovascular events and falls, highlighted in a recent editorial (PMID: 35348579). This would support their data.

2. Examples are the FIND (PMID: 34982819) and D-Health (PMID: 34337905) studies with the specific outcomes of increased cardiovascular events and falls, respectively, in those supplemented participants with a BMI <25 kg/m2.

Reviewer #2 (Recommendations for the authors):

Introduction:

– Line 50-52: I would suggest replacing the Murai et al. 2021 reference by a paper providing a more general overview, as the Murai et al. 2021 publication describes a RCT for vitamin D supplementation in covid-19 patients specifically. PMID 34815552 (Bouillon et al. Nat Rev Endocrinol 2022) might be more suited here

– Line 53: Mokry et al. 2015 is a Mendelian randomisation study, which is further described in the second paragraph of the introduction (starting from line 61). It might be more suited to incorporate this reference in line 64-65.

Results:

– Line 82: 'has a largely indirect influence' needs to be removed here?

– Line 87: Manousaki et al. 2020 GWAS was used to select instrumental variables for adulthood vitamin D. I was wondering whether similar results are obtained when selecting variants from the Revez et al. 2020 GWAS (PMID 32242144) ? Both are conducted in UK Biobank, though there are small differences in sample sizes and number of variants (especially due to differences in minor allele frequency filtering).

– Line 93: Multivariate while in line 96: multivariable. Is there a difference here?

– Line 96-97: perhaps add 'upon accounting for childhood body size'? (as in line 94), for further clarification on analyses

Figure 1: Number is lacking in the legend (mean age: X years)

Materials and methods:

– Line 168-171: What was the percentage of sample overlap? Although sample overlap does not seem to influence the findings, it may be worth mentioning the numbers of sample overlap, aiding readers in interpreting the degree of overlap.

– Line 173-182: It is stated in line 159-160 that genetic instruments for childhood and adulthood body size are derived from Richardson et al. 2020. Would it be possible to add some more details on selection of instrumental variables (in line with STROBE-MR guidelines), as initially 295 variants for childhood and 557 variants for adulthood body size have been identified in Richardson et al. 2020 while respectively 253 and 488 are included in the MR analyses. Are palindromic SNPs retained? Are proxy SNPs identified for SNPs missing in the outcome data?

– Line 177-178: The two-sample MR design allowed the performance of additional analyses (weighted median and MR-Egger, beyond the inverse variance weighted method). Now it seems that the IVW method allowed the additional analyses rather than the two-sample MR design. Would it be possible to rephrase this sentence accordingly?

Essential revisions:

1) As recommended by the two reviewers, please include a discussion on recent trials or studies that shed light on the same topic of the interplay of BMI and vitamin D on health outcomes.

We have added some discussions to address this to page 6 of the manuscript as described in detail below.

2) Please acknowledge the limitations and additional caveats pointed out by reviewer #2

We have added some discussion of page 7 to address this point directly in response to recommendation to reviewer #2 below.

Reviewer #1 (Recommendations for the authors):

This reviewer has the following recommendations for the authors:

1. It might be worth including data from recent large RCTs showing interactions of BMI with outcomes related to vitamin D supplementation, cardiovascular events and falls, highlighted in a recent editorial (PMID: 35348579). This would support their data.

Thank you for sharing this review with us which corroborates the findings of our study. We have discussed this as recommended on page 6 of the Discussion:

‘Our conclusions are also supported by the outcomes of recent large-scale randomized controlled clinical trials (RCTs) of vitamin D supplementation. For instance, as highlighted by a recent review (Ebeling, 2022), whilst observational studies have found that low vitamin D levels associate with increased risk of cardiovascular disease and cancer outcomes, there has been little convincing evidence from RCTs that a causal relationship may underly these findings.’

2. Examples are the FIND (PMID: 34982819) and D-Health (PMID: 34337905) studies with the specific outcomes of increased cardiovascular events and falls, respectively, in those supplemented participants with a BMI <25 kg/m2.

Evidence of an interaction between vitamin D supplementation and BMI reported by these two RCTs has now been discussed on page 6:

‘Moreover, recent RCTs have reported evidence of a potential interaction between vitamin D supplementation and BMI, such as the Finnish Vitamin D Trial (FIND) (Virtanen et al., 2022) and D-Health (Waterhouse et al., 2021) studies.’

Reviewer #2 (Recommendations for the authors):

Introduction:

– Line 50-52: I would suggest replacing the Murai et al. 2021 reference by a paper providing a more general overview, as the Murai et al. 2021 publication describes a RCT for vitamin D supplementation in covid-19 patients specifically. PMID 34815552 (Bouillon et al. Nat Rev Endocrinol 2022) might be more suited here

Many thanks for this suggestion. We have updated this reference as suggested.

– Line 53: Mokry et al. 2015 is a Mendelian randomisation study, which is further described in the second paragraph of the introduction (starting from line 61). It might be more suited to incorporate this reference in line 64-65.

As recommended, we have moved this reference to the MR section of this work (Page 3):

‘For example, MR has been applied in recent years to suggest that vitamin D supplements are unlikely to have a beneficial effect on risk of type 1 diabetes (Manousaki et al., 2021), whereas a recent study suggests that genetically lowered vitamin D levels may increase risk of multiple sclerosis (Mokry et al., 2015).’

Results:

– Line 82: 'has a largely indirect influence' needs to be removed here?

Thank you. We have now removed this text.

– Line 87: Manousaki et al. 2020 GWAS was used to select instrumental variables for adulthood vitamin D. I was wondering whether similar results are obtained when selecting variants from the Revez et al. 2020 GWAS (PMID 32242144) ? Both are conducted in UK Biobank, though there are small differences in sample sizes and number of variants (especially due to differences in minor allele frequency filtering).

We have repeated our analysis using the Revez et al. GWAS results (unadjusted for BMI) and added results to Supplementary Table 3. These results did not change the overall conclusions of this paper as reported on page 4:

‘Similar conclusions were drawn using estimates derived on adult 25OHD levels using findings by Revez et al. (Revez et al., 2020) (Supplementary Table 3).’

– Line 93: Multivariate while in line 96: multivariable. Is there a difference here?

To be consistent we have now changed ‘multivariate’ to ‘multivariable’ on line 93.

– Line 96-97: perhaps add 'upon accounting for childhood body size'? (as in line 94), for further clarification on analyses

This sentence has now been updated to incorporate your suggestion:

‘In contrast, strong evidence of an effect for higher adult body size on lower 25OHD levels measured in adulthood was found in the multivariable model accounting for the childhood body size score (Β=-0.17, 95% CI=-0.21 to -0.13, P=4.6x10^-17) (Figure 1C).’

Figure 1: Number is lacking in the legend (mean age: X years)

Thank you for spotting this – we have now updated the text to say ‘mean age 56.5 years’

Materials and methods:

– Line 168-171: What was the percentage of sample overlap? Although sample overlap does not seem to influence the findings, it may be worth mentioning the numbers of sample overlap, aiding readers in interpreting the degree of overlap.

Although it is challenging to determine the exact percentage of sample overlap in the GWAS of our exposure variables compared with the vitamin D GWAS undertaken by Manousaki et al., we have estimated the proportion of overlap using individual level from UKB on page 7:

‘Despite 431,074 of the 476,169 participants with measures of childhood and adult body size in UKB also having measures of vitamin D levels (90.5%), there was little evidence of inflated type 1 error rates (based on the calculator at https://sb452.shinyapps.io/overlap) (Burgess et al., 2016)

– Line 173-182: It is stated in line 159-160 that genetic instruments for childhood and adulthood body size are derived from Richardson et al. 2020. Would it be possible to add some more details on selection of instrumental variables (in line with STROBE-MR guidelines), as initially 295 variants for childhood and 557 variants for adulthood body size have been identified in Richardson et al. 2020 while respectively 253 and 488 are included in the MR analyses. Are palindromic SNPs retained? Are proxy SNPs identified for SNPs missing in the outcome data?

We have added the following to the Methods section on page 8 to add further clarity to our instrument selection:

‘Genetic estimates were harmonised using the ‘TwoSampleMR’ R package (Hemani et al., 2018) which by default removes palindromic variants and attempts to identify proxies for any instruments whose estimates are not available in the outcome dataset.’

– Line 177-178: The two-sample MR design allowed the performance of additional analyses (weighted median and MR-Egger, beyond the inverse variance weighted method). Now it seems that the IVW method allowed the additional analyses rather than the two-sample MR design. Would it be possible to rephrase this sentence accordingly?

Thank you – this sentence on page 8 has now been rephrased accordingly:

‘MR analyses to estimate effects on adulthood 25OHD were undertaken in a two-sample setting using the inverse variance weighted (IVW) method (Burgess et al., 2013), as well as the weighted median and MR-Egger methods (Bowden et al., 2015, Bowden et al., 2016) (Supplementary Table 1).’

Author details

1. Tom G Richardson

   MRC Integrative Epidemiology Unit (IEU), Population Health Sciences, Bristol Medical School, University of Bristol, Bristol, United Kingdom

   Contribution

   Data curation, Software, Formal analysis, Visualization, Methodology, Writing – original draft, Writing – review and editing

   For correspondence

   Tom.G.Richardson@bristol.ac.uk

   Competing interests

   is employed part-time by Novo Nordisk on work outside of the work reported in this study

   "This ORCID iD identifies the author of this article:" 0000-0002-7918-2040

2. Grace M Power

   MRC Integrative Epidemiology Unit (IEU), Population Health Sciences, Bristol Medical School, University of Bristol, Bristol, United Kingdom

   Contribution

   Formal analysis, Methodology, Writing – review and editing

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0002-5702-7728

3. George Davey Smith

   MRC Integrative Epidemiology Unit (IEU), Population Health Sciences, Bristol Medical School, University of Bristol, Bristol, United Kingdom

   Contribution

   Conceptualization, Supervision, Funding acquisition, Methodology, Writing – review and editing

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0002-1407-8314

• 1,062

  Page views

• 216

  Downloads

• 2

  Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.