Whole peripheral blood (2.5 ml) was collected using PAXgene Blood RNA tubes (BD Biosciences and Diagnostics). The tubes were inverted ten times immediately after drawing blood, stored upright at 15-25°C for 24 hours, followed by a -20°C freezer for 24 hours and then into a -80°C freezer for long-term storage. Transcriptomic analysis was conducted essentially as described in previous studies (35, 36). Total RNA was isolated using the PAXgene Blood RNA Kit (Qiagen) following the manufacturer’s recommendations. cRNA was synthesised and fluorescently labelled with Cy3-CTP from 200 ng of total RNA using the Low Input QuickAmp Labelling Kit, One Color (Agilent Technology). Labelled cRNA was hybridised on a Sure Print G3 Human Gene Expression 8 x 60K v2 microarray slide (Agilent Technologies). Standard manufacturer’s instructions for one colour gene-expression analysis were followed for labelling, hybridisation and washing steps. Extracted RNA was quantified using NanoDrop ND2000 spectrophotometer (Thermo Scientific). RNA quality and integrity was evaluated using either the Bioanalyzer 2100 or the TapeStation 4200 (Agilent Technologies). Only RNA samples with an RNA Integrity Number (RIN) of >7.0 were subjected to DNA microarray analysis. Microarrays were hybridised at 65°C for 17 hours in an Agilent Hybridization Oven on a rotisserie at 10 rpm. The washed microarrays were scanned using an Agilent Microarray Scanner with a resolution of 2 μm.

Raw scanned microarray images were processed using Agilent Feature Extraction software (v11.5.1.1) with the Agilent 039494_D_F_20140326_human_8x60K_v2 grid, and then imported into R for normalization and analysis using the LIMMA package (37). Normalised data for all participants (V3 – V1) was assessed by principal component analysis to screen for any batch effects ( Supplementary Figure 3 ). Microarray data were background-corrected using the ‘normexp’ method (with an offset of 50) and quantile normalised, producing expression values in the log base 2 scale. The processed data were then filtered to remove probes exhibiting low signals across the arrays, retaining non-control probes that are at least 10% brighter than negative control probe signals on at least 41 arrays (~20% of the arrays in the analysis). Data from identical replicate probes were then averaged to produce expression values at the unique probe level. Initial data exploration identified one sample (participant 0017, V3 time point) with array data that was a notable outlier from the group and therefore both the V1 and V3 microarrays for this subject were excluded from all subsequent analysis, reprocessing the data as above in their absence before proceeding.

Tests for differential expression were performed using LIMMA, applying appropriate linear model designs to identify: (i) significant differences in the transcriptional responses occurring across the 12-week V1 to V3 period of the study between the treatment and placebo groups for each ethnic cohort (example contrasts tested, in the format ethnicity_treatment_time: (WE_D2_V3 - WE_D2_V1) - (WE_P_V3-WE_P_V1) = 0, and (WE_D3_V3 - WE_D3_V1) - (WE_P_V3 - WE_P_V1) = 0); and ii) to determine significant changes for each ethnic cohort within each treatment group between the V3 and V1 sampling points, blocking on subject identity (example contrasts tested: WE_D2_V3 - WE_D2_V1 = 0, WE_D3_V3 - WE_D3_V1 = 0). Blocking on subject identity was used to control for inter-subject variability. Significance p-values were corrected for multiplicity using the Benjamini and Hochberg method, obtaining adjusted p-values (adj.P.Val).

Functional enrichment analysis of lists of genes of interest possessing valid ENTREZ gene identifiers was performed using the R package clusterProfiler (38). The software produces adjusted p-values (p.adjust) using the Benjamini and Hochberg correction method. Construction and analysis of protein-protein interaction networks from sets of genes was undertaken in Cytoscape (39) using the STRING plugin (40). Cytoscape was also used to construct and visualise commonality in the functional categories identified as being significantly enriched in the genes responding to the experimental treatments. To visualise categories identified from the D₂ or D₃ treatment groups but not from the placebo (e.g. Figure 4 ) significantly enriched categories (p.adjust < 0.01) from all groups were imported such that treatment group nodes (D₂, D₃ and placebo (P)) are linked to functional category nodes by edges assigned the corresponding p.adjust values. All nodes with an edge connection to the placebo treatment node were then removed and the resulting networks further filtered to retain only those nodes with at least one edge connection with p.adjust<=0.001 (p.adjust<=0.01 for Reactome pathways). The results were finally summarised as heatmap plots ( Figure 4 ).

Weighted gene co-expression network analysis was performed in R using the WGCNA (41) and CoExpNets (42) packages. A normalised expression data matrix generated by filtering to retain probes with signals more markedly above background (30% brighter than negative control probe signals on at least 41 arrays) was used as input, consisting of 12,169 unique probes. Signed scale-free networks were constructed separately for the data for the SA and WE ethnic groups using 50 iterations of the k-means clustering option in the CoExpNets ‘getDownstreamNetwork’ function to refine the clustering process ( Supplementary Data File 6 ). Pearson correlations between cluster module eigengenes and metadata for serum 25(OH)D₂, 25(OH)D₃, total 25(OH)D and PTH concentrations were calculated using pairwise complete observations.

GSEA was performed using the R package clusterProfiler (38), applying the default parameters which implement the fgsea algorithm and correct for multiple testing using the Benjamini and Hochberg method. Gene sets were obtained from MSigDB via the msigdbr package (43). Ranked gene lists for interrogation were derived from LIMMA analysis of the unfiltered microarray data, pre-ranking genes according to their t-statistic.

Following microarray data quality control, transcriptome data were available for both the V1 and V3 samples of 97 study participants ( Figure 1B ); the RNA from one participant failed quality control and was excluded. Similar to data reported previously for the entire study cohort (28), the sub-set of participants in the present study showed increased concentrations of 25(OH)D₂ and 25(OH)D₃ only after supplementation with vitamin D₂ and vitamin D₃, respectively, and higher total 25(OH)D in both treatment groups ( Figure 1C ). In the vitamin D₃ treatment groups the mean 25(OH)D₃ serum levels rose by 59% (WE) and 116% (SA) over the 12-week intervention. Conversely, mean serum 25(OH)D₃ concentrations fell across the 12 weeks of the study in the placebo groups who did not receive additional vitamin D; 25(OH)D₃ baseline vitamin D concentration in the SA ethnic group tended to be lower than for the WE group and dropped, respectively, by 23% and 29% relative to baseline V1 ( Figure 1C and Supplementary Data File 1 ). It is notable that, in the vitamin D₂ treatment group, serum 25(OH)D₃ concentration decreased to a greater extent over the 12-week period, by 52% (SA) and 53% (WE). Serum 25(OH)D₂ was low in the absence of specific supplementation, typically less than 5 nmol/L (152/162 samples analysed). Serum calcium concentration, appropriately adjusted for serum albumin concentration, was maintained within a normal clinical range across all sample groups, while the concentration of parathyroid hormone (PTH) was stable between V1 and V3 only within the SA placebo group, the WE vitamin D₂ and WE vitamin D₃ intervention groups. PTH concentrations increased at V3 compared with V1 in the WE placebo group, but decreased in both the vitamin D₂ and D₃ intervention groups in the SA ethnic cohort.

Filtering of the normalised microarray data to select for probes showing signals at least 10% above background in at least 41 (~20%) of the arrays yielded transcript abundance data from 20,662 probes corresponding to 17,588 different genomic features (12,436 of which were annotated with an ENTREZ gene identifier). Using the filtered data, we identified significant differences in the transcriptional responses occurring across the 12-week V1 to V3 period of the study between the vitamin D₂, vitamin D₃ and placebo treatment groups for each ethnic cohort. In a separate analysis, we determined significant changes within each group between the V3 and V1 sampling points ( Figure 2 and Supplementary Data File 2 ). The former was tested in a ‘difference-in-difference’ analysis according to the generalised null hypothesis [treatment1.V3 – treatment1.V1] - [treatment2.V3 – treatment2.V1] = 0 ( Figure 2A , rows 7-12), and the latter using treatment.V3 – treatment.V1 = 0 ( Figure 2A , rows 1-6).

No significant changes in probe signals (adj.P.Val<=0.05) were identified in the ‘difference-in-difference’ analysis for the data for the WE ethnic group, but five were found in the corresponding analysis of the SA cohort ( Figure 2A ). Furthermore, difference-in-difference analysis of the entire study data, ignoring ethnicity, did not yield statistically significant changes between the D₂ or D₃ treatment groups and the placebo control (data not shown). We note that another ‘difference-in-difference’ study on the influence of vitamin D on gene expression (the BEST-D trial) failed to identify any significant changes in gene expression following long-term vitamin D treatment in a Caucasian cohort (32). The observed log₂ fold-change (log₂FC) in signal abundances between V3 and V1 in the study typically fell well within +/- 1, and, in such a context, it is likely that the sample sizes used within the study restrict our ability to reliably identify individual gene expression responses using this highly conservative approach.

We consider that the strong interpersonal differences in gene expression across individuals in the study population are likely to hinder detection of small statistically significant gene expression changes across treatment groups from microarray-derived data. It is known that expression of any particular gene and its magnitude of change over time can vary widely across individuals in a human population, and the differences in response across time within an individual is considered more physiologically meaningful. This issue is considered further in the Discussion. In this study we observe that many statistically significant differences are detected when evaluating gene expression changes within an individual across time (as described below where we examine paired observations with two time points per subject, V1 and V3).

The five significant changes that were identified as occurring in response to treatment of the SA ethnic group with vitamin D₃, relative to the placebo, are summarised in Supplementary Figure 1 . These are driven primarily by the marked changes in the placebo group between the V1 and V3 sampling points, and may be associated with the sustained low serum 25(OH)D₃, or high PTH, concentrations observed in this group of subjects. One of these differentially expressed genes encodes the cAMP response element binding protein CREB1 ( Figure 2B ) which is part of the cAMP-PKA-CREB signaling pathway in bone cells which contributes to the regulation of skeletal metabolism in response to PTH concentrations (44).

In contrast to the difference-in-difference analysis, the direct within-group comparisons of transcript abundance measurements at V3 versus V1 per subject identified large numbers of significant changes in gene expression, most notably in the WE D₂, WE D₃ and WE placebo groups ( Figures 2A, C and Supplementary Data File 2 ). These statistically significant gene expression changes in the WE ethnic cohort arising in the vitamin D treatment groups are considered in more detail below. While there is some overlap in the groups of differentially expressed genes, only 13% of down-regulated differentially expressed genes (102 of 774) were common between the two vitamin D treatment groups. Conversely, 28% (216 of 774) and 59% (456 of 774) were uniquely down-regulated by D₂ and D₃, respectively (excluding those additionally down-regulated in the placebo group over the equivalent 12-week intervention)( Figure 2C ). As noted above, serum 25(OH)D₃ concentration fell over the 12-week intervention period in the vitamin D₂-treated group; thus, changes in gene expression observed in this group could a priori either be due to the influence of vitamin D₂ itself, or could be attributable to the depletion of the endogenous 25(OH)D₃ reserves. It was therefore important to have included comparative transcriptomic analysis of the placebo group in this study to distinguish gene expression changes resulting from vitamin D₂ per se from those changes arising due to the depletion of 25(OH)D₃ in the vitamin D₂ treatment group.

It is notable that expression of a large number of genes was altered between the first and last visits (V1 to V3) in the non-treated placebo group. While some of these changes will be attributable to the reduced synthesis of vitamin D that is seen in northern latitudes over the winter months, there is also significant over-representation of genes known to exhibit seasonal differences in gene expression (45) ( Supplementary Figure 2 ). There is however no unique association of seasonally expressed genes with the data for the WE placebo group; a similar significant enrichment in ‘seasonal genes’ is also observed for the significant changes in gene expression identified in the WE D₂ and WE D₃ treatment groups ( Supplementary Figure 2 ) and therefore it is unlikely that this apparent seasonal effect exclusively reflects vitamin D-specific effects. There was no overlap between genes significantly down-regulated exclusively in the placebo group and those significantly up-regulated exclusively in the vitamin D treatment groups (or vice versa). Consistent with the 12-week period of the study, seasonal changes in gene expression are therefore a background feature of all the data collected, and is therefore a factor that needs to be considered when undertaking such vitamin D supplementation studies.

Probe A_33_P3275751, detecting SEC14L1 gene transcription, showed the largest decrease in response to both forms of vitamin D, but not placebo participants, in the group of probe signals with significantly reduced abundance in the V3 samples compared with V1 ( Figure 2D ). This probe was originally designed to detect ‘transcript variant 7’ and hybridizes to a location at the 3’ end of the SEC14L1 locus, detecting several splice variants. High expression of SEC14L1 is significantly associated with lymphovascular invasion in breast cancer patients, where transcript abundance correlates positively with higher grade lymph node metastasis, and poor prognostic outcome (46). Its overexpression is also frequent in prostate cancer where SEC14L1 has been identified as a potential biomarker of aggressive progression of the disease (47, 48).

As a complementary approach to the functional enrichment analysis of specific subsets of genes whose expression is altered by supplementation with vitamin D₂ and D₃ but not by placebo treatment, a comparative functional enrichment analysis of all significant changes in each treatment group was performed (see Methods). This aimed to identify functional categories that are more extensively affected by vitamin D₂, or by vitamin D₃, or by both vitamins D₂ and D₃, than by the placebo, and took into consideration the changes occurring in the placebo treatment group across the 12 weeks of the study ( Figure 4 and Supplementary Data File 4 ). Consistent with the earlier analysis, functional categories associated with immunity and immune response pathways are prominent among those genes repressed by vitamin D supplementation ( Figure 4A ), while mitochondrial, ribosomal and spliceosomal functions are prominent in the induced genes ( Figure 4B ). Although the two vitamin treatments share many common categories identified from this analysis, the results also highlight some differences between the respective effects of D₂ or D₃ supplementation. For example, ‘histone exchange’ is significant only in the vitamin D₂ up-regulated genes, and ‘chromatin modifying enzymes’ are significant only in the D₃ down-regulated genes.

Overall, the observed differences in gene expression from the blood transcriptome presented in this study suggest that the physiological effects of vitamin D₃ and D₂ may be dissimilar.

WGCNA quantifies both the correlations between individual pairs of genes or probes across a data set, and also the extent to which these probes share the same neighbours (41). The WGCNA process creates a dendrogram that clusters similarly abundant probes into discrete branches, and subsequent cutting of the dendrogram yields separate co-expression modules, representing putatively co-regulated sets of genes. The first principal component of the expression matrix of each module defines the expression profile of the eigengene for the module, and this can then be correlated with experimental metadata. By allowing phenotypic traits to be associated with relatively small numbers (tens) of module eigengenes, instead of thousands of individual variables (i.e. gene probes), WGCNA both alleviates the multiple testing problem associated with standard differential expression analysis and also directly relates experimental traits to gene expression data in an unsupervised way that is agnostic of the experimental design.

WGCNA was used to construct separate signed co-expression networks for the WE and SA ethnic groups as described in the Methods, and Pearson correlations between expression of the module eigengenes and serum 25(OH)D₂, 25(OH)D₃, total 25(OH)D and parathyroid hormone (PTH) concentrations were calculated ( Figure 5 and Supplementary data files 6 – 8 ). These results therefore ignore whether the data originates from the vitamins D₂, D₃ or placebo treatment groups and focus solely on the relationship between the serum metabolite concentrations and gene expression. A significant negative correlation was observed between 25(OH)D₂ concentration and expression modules in the WE co-expression network that are enriched for immune-associated functions (midnight blue and pink modules in Figure 5A ). This is consistent with the results from the different analytical approaches (presented in Figures 3 , 4 ). Similarly, the only module exhibiting a significant positive correlation with 25(OH)D₂ (and total 25(OH)D concentration), the red expression module, is enriched for GO categories associated with the ribosome and mRNA processing ( Supplementary Data File 6 ). No significant correlations with serum 25(OH)D₃ concentrations were detected in the WE cohort.

Interestingly, stronger and more numerous correlations were observed in the co-expression network for the SA cohort ( Figure 5B ). In agreement with the WE network, an expression module significantly associated with the ribosome (black module in Figure 5B ) showed a general positive correlation with total 25(OH)D concentration whereas a module enriched for histone binding (turquoise) had a significant negative correlation with 25(OH)D concentration. Strikingly however, modules enriched for immune response functions (midnight blue, orange, purple, yellow in Figure 5B ) were consistently positively correlated with total 25(OH)D and usually 25(OH)D₃, but not 25(OH)D₂ concentrations, in the SA network, and negatively correlated with PTH concentrations. This raises the possibility that vitamin D supplementation may exert different effects on the immune system depending on ethnicity of the individual, may indicate that PTH status has an influence on the outcome and may reflect physiological differences resulting from the low baseline vitamin D status in SA women. In this context we observed that PTH concentration at the V1 sampling point in the SA cohort was elevated compared with that in the WE cohort ( Figure 1C ).

The molecular signatures database (MSigDB) hallmark gene sets represent coherent gene expression signatures related to well-defined biological states or processes (https://www.gsea-msigdb.org/gsea/msigdb/). Gene set enrichment analysis (GSEA) was performed to identify statistically significant (padjust<=0.05), concordant changes in expression of these hallmark sets between the V3 and V1 sampling times for the placebo, vitamin D₂, and vitamin D₃ treatment groups ( Figure 6 and Supplementary Data File 9 ). Interestingly, in the WE cohort, the gene sets defining the signature responses to interferon alpha and gamma exhibited divergent behaviour in the vitamin D₂ and D₃ treatment groups. This is evident as significant up-regulation over the course of the intervention study in the D₃ group but significant down-regulation in the D₂ group (see also Supplementary Figures 4 , 5 illustrating the core genes accounting for the enrichment signals). This differed from the SA cohort where significant down-regulation was observed in the D₃ group (see also Supplementary Figure 6 ). It is acknowledged that some of the observed differences between the WE and SA groups could reflect the differences in the respective sample sizes. However, in cases where statistically significant differences are found, such as the reciprocal trend in the respective interferon responses these are not likely to be attributable to sample size differences.

Excluding genes that were also differentially expressed over the 12-week intervention in the placebo group, only 13% of down-regulated differentially expressed genes were common between the two treatment groups while 28% and 59% were uniquely down-regulated by vitamins D₂ and D₃, respectively ( Figure 2C ). For example, some biological processes such as histone modification and covalent chromatin modification are downregulated following vitamin D₃ supplementation only, while spliceosomal function are upregulated by vitamin D₂ only. Functional categories of genes enriched among the upregulated genes, following supplementation with either vitamin D₂ or D_3, include translation, mitochondrial and spliceosome function ( Figure 3B and Supplementary Data File 4 ); statistically enriched biological cellular component terms in these functional categories are ribosomal proteins, components of the mitochondrial respiratory chain, two subunits of the histone H4 and snRNP Sm protein components of the spliceosome assembly. It is known that, in addition to influencing transcription, vitamin D can also influence post-transcriptional events by recruiting co-regulators (54). In this context it is relevant that components of the spliceosome, such as snRNP Sm proteins that mediate both transcriptional control and splicing decisions, leading to alternatively spliced transcripts (55), were upregulated by vitamin D supplementation in this study.

In light of the finding that vitamins D₂ and D₃ influence expression of different genes in the human blood transcriptome, recently we undertook a parallel in vitro study with a model rat cell line (56). We examined the respective influences of the two physiologically active forms of vitamin D, 1,25(OH)D₃ and 1,25(OH)D₂ on differentiation and global gene expression in differentiating rat CG4 oligodendrocyte precursor cells and revealed considerable differences in the influence of the two types of vitamin D on gene expression at 24 h and after 72 h following onset of differentiation. We demonstrated that 1,25(OH)D₃ and 1,25(OH)D₂ respectively influenced expression of 1,272 and 574 genes at 24 h following addition of the vitamin to the culture, where many of the changes in expression were specific to one or the other form of the vitamin (56). This study provides evidence of the different direct effects of the two active vitamin D metabolites on gene expression in vitro in cultured cells and provides some evidence that the changes we observed in the in vivo study may reflect the influence of the physiologically active forms of the vitamin.

In common to both the D₂ and D₃ treatment groups, but not the placebo group, we found that many different pathways of the immune system are differentially expressed (largely down-regulated) by vitamin D, consistent with the notion that one physiological role of vitamin D is to restrain, or balance, the activity of the immune system (1, 2) ( Figures 3A , 4 and Supplementary Data Files 3 – 5 ). The immunomodulatory effects of vitamin D on both innate and adaptive immunity are well documented (57–59). In this regard, our observation that vitamin D₂ and D₃ supplementation in the WE cohort was associated with divergent patterns of expression for interferon alpha (type I) and interferon gamma (type II) signature gene sets stands out as particularly interesting ( Figure 6 ). Vitamin D appears to modulate type I interferon activity. For example, it enhances the effects of type I interferon treatment on mononuclear cells from patients with multiple sclerosis (60), which parallels evidence for modest benefits of vitamin D as an adjunct treatment with type II interferon in multiple sclerosis patients (61, 62). Vitamin D may also help suppress symptoms in autoimmune diseases such as systemic lupus erythematosus (63), which are associated with chronic over-activity of interferon signalling and tentatively designated interferonopathies. Moreover, type I interferons play a critical role in defence against viral infections. Basal expression of type I interferon-stimulated genes, in the absence of infection, is key to priming a rapid and effective response to viral infection (64). There is currently intense interest in vitamin D as both a potential prophylactic and a therapeutic agent for treatment of SARS-CoV-2 infection. One of the proposed modes of action of vitamin D is modulation of interferon activity (65); in this context our observation that vitamin D₃ (but not vitamin D₂) enhances the expression of genes involved in the interferon alpha response, is highly relevant to susceptibility to viral infection. Indeed, a recent genome-wide association study (GWAS) in 2,244 critically ill Covid-19 patients identified genetic variants leading to reduced interferon type I signalling that are associated with severe Covid-19 disease (66).

We have also found differences in gene expression response to vitamin D according to ethnicity (with the caveat that the sample size of the SA group was smaller than the WE group and the average baseline vitamin D status was significantly lower relative to the WE group). Unlike the white European group, differences in gene expression across treatment groups were found in the South Asian group ( Figures 2A, B ) who received supplementation with vitamin D₃, where three genes were significantly upregulated and two down-regulated after the 12-week intervention relative to the placebo group ( Figure 2 and Supplementary Figure 1 ). cAMP-responsive element binding protein 1 (CREB1), one of the three upregulated genes, encodes a transcription factor that binds as a homodimer to the cAMP-responsive element. CREB1 protein is phosphorylated by several protein kinases and induces transcription of genes in response to hormonal stimulation of the cAMP pathway. The protein kinase A (PKA) pathways are activated by the parathyroid hormone PTH in response to low serum calcium levels to maintain serum calcium homeostasis, primarily by promoting vitamin D 1α-hydroxylation in the kidney. Both PTH and 1,25 (OH)₂ vitamin D have similar effects in promoting the maturation of osteocytes and opposing the differentiation of osteoblasts into osteocytes (67, 68). It is relevant to note that those in the SA cohort had elevated serum PTH at baseline relative to the WE cohort.

Ethnic differences in response are also suggested from the Weighted Gene Correlation Network Analysis (WGCNA) and the Gene Set Enrichment Analysis (GSEA), where the response to vitamin D₃ intake appeared to have the opposite effect on the type I and II interferon pathways in the SA group compared with the WE group ( Figure 6 and Supplementary Figures 4 , 6 ). However, direct correlation between the stimulation of immune responses, other than the interferon pathways, with an increase in serum 25(OH)D₃ concentration was evident in the SA group only. This is the opposite of that observed in the WE group where the effect of vitamin D supplementation was to suppress several immune pathways.

The transcriptome results differ considerably between the two different ethnic groups. While some of the differences may be attributable to differences in sample sizes studied between the two groups it is also possible that some of the differences represent genuine differences in the respective physiological responses of the two ethnic groups. Alternatively, the differences may reflect the starting physiological status of the SA participant group, who had considerably lower baseline vitamin D concentrations (and higher PTH concentrations) than the WE group. The importance of clarifying these different responses is given particular impetus by the emerging evidence for the interplay between ethnicity, skin tone, vitamin D status and susceptibility to viral infection, Covid-19 being of particular current relevance here (12, 13, 69, 70). The transcriptomic data presented in this study might provide a useful context for further studies aimed at understanding the role of vitamin D in influencing the immune response to SARS-CoV-2 infection, particularly in relation to severe Covid-19. It is notable that the recent GenOMICC study of severe Covid-19 disease highlighted a link between severe Covid-19 and reduced type I interferon signalling (66). Our finding that vitamin D₃ appears to stimulate type I interferon signalling could be relevant in the context of its use as a prophylactic treatment. It should be noted here that in order to have a beneficial effect, a vitamin D replete status would be required prior to exposure to the virus, contributing innate immunity; administration of vitamin D following admission to hospital, even at high bolus doses, would not provide any clinical benefit because the virus would have already established itself. Recent clinical trials have borne out this latter point.

Our ability to detect differences in the effects of vitamin D₂ and vitamin D₃ may have been negatively impacted by the relatively low statistical power and by the inclusion of two different ethnic groups among the 97 participants, since the transcriptome results from the two ethnic groups are clearly different; the number of participants represents a weakness of the present study. It will be important to replicate this study using a larger cohort in order to verify, or otherwise, the key findings from this study. From power calculations it is considered that for a whole human microarray-based gene expression study of this nature, and with gene expression changes of the magnitude we observe, at least 400 participants of each ethnic group should be recruited for each treatment, giving a total cohort size of 2,400. In this context, the biological interpretation of our findings should be considered as preliminary, requiring independent verification. A second limitation of this study was our failure to take account of the contribution of potential changes in blood cell composition across the seasons and across ethnicity; it is known that blood cell composition can vary significantly throughout the year (51, 71).

It is difficult to compare the findings of the present study with other reported in vivo studies because of the considerable differences in experimental design, including the use of supra-physiological vitamin D doses (up to 2,000 μg single bolus doses), different human population types and small sample sizes, which make statistical analysis not feasible [e.g. (33, 34)]. Furthermore, our study is unique in that it compared gene expression in participants given either of the two commonly used forms of vitamin D, vitamin D₂ and vitamin D₃. The present study evaluated overall expression changes in a complex blood cell population, which is also known to change in cellular composition across seasons (51). Furthermore, this study has focussed on identification of statistically significant pathway, or gene set, enrichment following sustained vitamin D supplementation, rather than a gene-focussed analysis; the latter approach is limited given the effect sizes and sample size of this study.

This study suggests that a more detailed consideration of the system-wide effects of these two forms vitamin D is warranted. This is perhaps of particular importance to at-risk ethnic groups, including black and South Asian populations who reside in northern latitudes. The studies would need to be time course-based (temporal) to track gene expression changes within individuals from baseline to defined sampling times and would need in-built control to account for seasonal gene expression changes. They would need to be designed specifically to answer whether different ethnicity or different vitamin D baseline concentrations give rise to different responses to vitamin D supplementation.

Since some pathways appear to be regulated specifically by vitamin D_3, or in some cases, in opposing directions by vitamin D₃ and D₂, future studies should investigate whether vitamin D₂ supplementation might counteract some of the benefits of vitamin D₃ on human health. This possibility is prompted by the findings from this cohort that the circulating concentration of 25(OH)D₃ within vitamin D₂-treated participants was significantly lower after the 12-week intervention than in the placebo group, who received no vitamin D supplements – suggesting that the former might be depleted by the latter. The results from this study suggest that guidelines on food fortification and supplementation with specific forms of vitamin D may need revisiting.