The plasma proteome of plant-based diets: analyses of 1463 proteins in 50,000 people

Study Population

The UK Biobank is a prospective cohort study of around 500,000 middle-aged people (recruitment target 40-69 years), recruited from across the United Kingdom between 2006 and 2010. The scientific rationale and design of the UK Biobank have been described in detail previously [5]. In brief, participants were identified from National Health Service registers, and were invited to join the study if they lived within travelling distance (∼25 km) of one of the 22 assessment centres in England, Wales and Scotland. People who consented to participate in the study attended a baseline visit at the assessment centre where they completed a touchscreen questionnaire which asked about their lifestyle (including diet, alcohol consumption, smoking status, physical activity), socio-demographic characteristics and general health and medical history. All participants were also given a verbal interview, and had their physical measurements and blood samples taken by trained staff. Permission for access to patient records for recruitment was approved by the Patient Information Advisory Group (now the National Information Governance Board for Health and Social Care) in England and Wales and the Community Health Index Advisory Group in Scotland, and all participants gave informed consent to participate using a signature capture device at the baseline visit.

Ethnicity classification

On the touchscreen questionnaire, participants were asked to self-identify their ethnicity as ‘White’, ‘Mixed’, ‘Asian or Asian British’, ‘Black or Black British’, ‘Chinese’, ‘Other ethnic group’, ‘Do not know’, or ‘Prefer not to answer’, with further sub-categories under each option. Participants were included for our analyses if they self-identified as ‘White’, or as ‘Asian or Asian British’ and subsequently as ‘Indian’, hereafter referred to as ‘white British’ and ‘British Indian’. The white British population was included as it made up the majority of the participants in UK Biobank, while the British Indian population was included because of the large proportion of vegetarians in this population group (25% compared to less than 2% in the overall population). The number of vegetarians in the other ethnic groups was too small to allow valid comparisons by diet groups, and therefore people of other ethnic groups were excluded from these analyses.

Diet group classification

Participants were classified into diet groups based on self-reported dietary data from the touchscreen questionnaire completed at recruitment. Participants were asked to report their frequency of consumption of processed meat (including processed poultry), unprocessed red meat (beef, lamb or mutton, pork), unprocessed poultry (such as chicken or turkey), oily fish, and other types of fish, with the possible responses ranging from “never” to “once or more daily”. Participants were also asked whether they never consumed dairy and eggs or foods containing eggs. Based on their responses to these questions, the white British participants were classified into six diet groups: regular meat eaters (red and processed meat consumption >3 times per week), low meat eaters (red and processed meat consumption ≤3 times per week), poultry eaters (participants who ate unprocessed poultry but no red and processed meat), fish eaters (participants who ate fish, but not red and processed meat, or poultry), vegetarians (participants who did not eat any meat or fish), and vegans (participants who did not eat any meat, fish, dairy products or eggs). The British Indian participants were classified into two diet groups: meat eaters (ate any combination of red or processed meat or poultry) and vegetarians; the numbers of fish eaters and vegans were low in the British Indians, and thus were not included in these analyses.

Assay for plasma proteins

Non-fasting blood samples were collected from all UK Biobank participants by trained personnel (either a phlebotomist or a nurse) except in a small proportion (0.3%) of participants who declined, were deemed unable to, or where the attempt was abandoned for technical or health reasons. Proteomic profiling was conducted using the Olink Proximity Extension Assay on blood plasma samples in 54,306 participants selected as part of the UK Biobank Pharma Proteomics Project (UKB-PPP). The samples were shipped on dry ice to Olink Analyses Service, Uppsala, Sweden for analysis; details of the selection procedures for inclusion in UKB-PPP and the technical details of the proteomics assays have been described elsewhere [6]. In brief, the Olink Explore 1536 platform used in this study is an antibody-based assay which measures the relative abundance of 1,472 protein analytes, including 1,463 unique proteins, distributed across four 384-plex panels: inflammation, oncology, cardiometabolic, and neurology. The Olink Explore platform is based on proximity extension assays that are highly sensitive and reproducible with low cross-reactivity. Relative concentrations of the 1,463 unique proteins were readout by next-generation sequencing. Measurements are expressed as normalized protein expression (NPX) values which are log-base-2 transformed. Protein values below the limit of detection (LOD) were replaced with the LOD divided by the square root of 2 [7]. Protein concentrations were subsequently inverse rank normal transformed. All results for differences in protein concentrations by diet groups may be interpreted as SD differences.

Inclusion and exclusion criteria

Of the 54,306 participants selected for proteomic profiling, 52,705 participants remained after the quality control procedures. Participants were further excluded if they were not of white British or British Indian ethnicity (n=2964), could not be classified into one of the pre-specified diet groups (n=414), or had missing information for fasting time (n=1). After the exclusions, 49,326 participants (48,773 white British and 553 British Indian) were included in the analyses. A participant flow chart of the inclusion criteria is shown in Supplementary figure 1.

Statistical analyses

Baseline characteristics of UK Biobank participants included in these analyses were tabulated by six diet groups in the white British population and by two diet groups in the British Indian population, as mean (SD) for continuous variables and number (%) for categorical variables. We used multivariable-adjusted linear regressions to estimate differences in protein concentrations by diet group, separately by ethnicity, using regular meat eaters as a reference group in the white British participants, and meat eaters as a reference group in the British Indians. The model was adjusted for age at recruitment (5 year categories), sex, region (London, North-West England, North-East England, Yorkshire, West Midlands, East Midlands, South-East England, South-West England, Wales, Scotland), fasting status (0–1, 2, 3, 4, 5, 6–7, ≥8 hours), body mass index (BMI; <20, 20.0–22.4, 22.5–24.9, 25.0–27.4, 27.5–29.9, 30.0–32.4, 32.5–34.9, ≥35 kg/m², unknown), alcohol consumption (<1, 1–7, 8–15, ≥16 g/d, unknown), smoking status (never, previous, current <15 cigarettes/day, current ≥15 cigarettes/d, unknown) and physical activity (<10, 10-49.9, ≥50 excess metabolic equivalent of task, hr/wk, unknown). Wald tests were used to assess overall heterogeneity between diet group in each ethnicity, and for pairwise comparisons of each of the other diet groups compared to the reference group (regular meat eaters in white British participants/meat eaters in British Indians). Heterogeneity between vegetarians and vegans in the white British population was assessed based on post-estimation linear combinations of parameters for all proteins.

To account for multiple testing while considering the high correlations between the circulating proteins, we conducted a principal component analysis of the circulating proteins in the complete proteins dataset before exclusions for these analyses, and determined that the first 639 principal components explained 95% of the total variation in the exposure data. Consequently, the effective number of independent tests [8] was determined to be 639, and the statistical significance level was set to be p-value 0.05/639 = 0.000078. For the top proteins identified (the top 10 significant proteins in each pairwise comparison against the reference group, based on ranking of p-values), we conducted additional sensitivity analyses to evaluate the extent to which the associations may be influenced by key covariates including BMI, smoking and alcohol consumption, by presenting models with and without adjustment of these variables. Additionally, we also restricted the analyses to people who self-reported to be in good and excellent health. All analyses were performed using Stata version 17.0 (StataCorp, TX, USA). All figures were generated using R version 4.2.1, the forest plots using “Jasper makes plots” package version 2-266 [9].

Integrating publicly available information on gene expression

To further understand the biological context for the proteins of interest, we extracted single cell mRNA expression from the Human Protein Atlas [10] to describe expression for any proteins that were significantly different in one or more comparisons. Normalized expression levels were extracted for genes in 30 different human tissues and 82 cell-types. Gene expression specificity at the cell or tissue type level was calculated as the ratio of each gene cell-type or tissue expression to the total expression of each gene across all cell or tissue types. We subsequently identified genes for proteins that were majority expressed (more than 50% of total expression) in at least one cell or tissue type, and mapped these to their likely candidate cell and tissue of origin where possible.

Baseline characteristics

The baseline characteristics of the study population are shown by ethnicity and diet group in Table 1. Compared with white British regular meat eaters, the non-meat eaters were on average slightly younger, more likely to be women, and to reside in London. They were less likely to be overweight, reported lower alcohol consumption, lower current smoking, and more physical activity. Compared with British Indian meat eaters, the vegetarians were more likely to be women and reported lower alcohol consumption, lower current smoking and slightly lower physical activity, but BMI was not noticeably different between the two groups. Fasting time was not meaningfully different by diet group in both ethnicities.

Differences in plasma proteins by diet group

The plasma proteins that were significantly different in white British vegetarians and vegans compared with regular meat eaters are shown as volcano plots in Figure 1. Significant differences in proteins in white British low meat eaters, poultry eaters and fish eaters compared with regular meat eaters, and in British Indian vegetarians compared with meat eaters are shown as volcano plots in Supplementary figures 2-6. Overall, 683 plasma proteins were significantly different by diet group (based on p-heterogeneity < 0.000078) in the white British participants, including 535 plasma proteins that were significantly different in one or more pairwise comparisons between regular meat eaters and at least one of the other diet groups. This includes 296 proteins (23 higher, 273 lower) in low meat eaters compared with regular meat eaters, 59 (9 higher, 50 lower) in poultry eaters, 157 (104 higher, 53 lower) in fish eaters, 219 (194 higher, 25 lower) in vegetarians and 15 (12 higher, 3 lower) in vegans; and 2 proteins were significantly different between vegetarians and vegans (AHSP and FOLR2 were both higher in vegans). The proteins identified are involved in a range of different biological pathways. Of the 535 plasma proteins that were significantly different in one or more pairwise comparisons, 97 are majority expressed (protein mRNA >50% of total body mRNA expression) in one tissue type, while the other proteins have limited tissue specificity. In particular, these included 39 proteins that are majority expressed in the digestive system (including 16 in the liver, 8 in the pancreas, 8 in the small intestine), 30 in the nervous system (27 in the brain), and 12 in the respiratory system (7 in the lung) (Supplementary figure 7).

• Download figure
• Open in new tab

Figure 1: Volcano plots of proteins in white British vegetarians (top) and vegans (bottom) compared with regular meat eaters.

The red dotted line signifies p-value threshold for statistical significance. Each dot represents one protein, which were colour-coded by whether the protein is majority expressed (>50%) in one tissue type. Results were based on the multivariable model adjusted for age at recruitment, sex, region, fasting status, body mass index, alcohol consumption, smoking status and physical activity.

In the British Indian population, only 1 protein (SUMF2, which is higher in vegetarians) was significantly different when comparing vegetarians and meat eaters, likely owing to the much smaller number of British Indians in the study. However, the differences in many proteins were directly consistent between the two ethnicities, as illustrated by the forest plots of top proteins by diet group and ethnicity, with the six diet groups in the white British participants and the two diet groups in British Indians (Figures 2-3, Supplementary figures 8-10). The complete results of all proteins are shown in Supplementary tables 1.

• Download figure
• Open in new tab

Figure 2: Top 10 proteins in vegetarians by diet group and ethnicity.

The top 10 proteins were selected by ranking the p-values of proteins comparing white British vegetarians with regular meat eaters, and sorted by betas in white British vegetarians, where the betas represent SD differences. Results were based on the multivariable model adjusted for age at recruitment, sex, region, fasting status, body mass index, alcohol consumption, smoking status and physical activity.

• Download figure
• Open in new tab

Figure 3: Top 10 proteins in vegans by diet group and ethnicity.

The top 10 proteins were selected by ranking the p-values of proteins comparing white British vegans with regular meat eaters, and sorted by betas in white British vegans, where the betas represent SD differences. Results were based on the multivariable model adjusted for age at recruitment, sex, region, fasting status, body mass index, alcohol consumption, smoking status and physical activity.

Generally, in the multivariable-adjusted model, many proteins showed a gradient effect in magnitude of differences across diet groups by degree of animal food exclusion, from regular meat eaters to vegans. Figure 2 shows the top 10 proteins in white British vegetarians (based on the ranking of pairwise p-values in vegetarians compared with regular meat eaters), and Figure 3 shows the top 10 proteins in white British vegans. Of the top proteins in vegetarians (Figure 2), compared with regular meat eaters, vegetarians and vegans both had significantly higher FGF21 (+0.40 SD in vegetarians; +0.74 SD in vegans), GUCA2A (+0.33; +0.79), FOLR1 (+0.32; +0.61) and IGFBP2 (+0.31; +0.55) after correction for multiple testing, whereas vegetarians but not vegans had significantly higher ART3 (+0.34), LAYN (+0.32), PCSK9 (+0.32), and lower CDHR2 (−0.26), HAVCR1 (−0.38) and CNDP1 (−0.39). In addition to the aforementioned proteins, vegans also had substantially higher FOLR2 (+0.69), GAS6 (+0.68), DSG2 (+0.68), THY1 (+0.62), MANSC1 (+0.58), but lower EGFR (−0.55) than regular meat eaters (Figure 3); the differences in these proteins were directionally consistent but less extreme in the vegetarians.

The top 10 proteins in white British low meat eaters, poultry eaters and fish eaters compared with regular meat eaters are shown in Supplementary figures 8, 9 and 10 respectively. Of these top proteins, none was uniquely different in one diet group only when compared with regular meat eaters. Similar to the observations in vegetarians, all three groups had substantially higher DSG2 and lower HAVCR1 compared with regular meat eaters; in both low meat eaters and poultry eaters HAVCR1 is the protein that exhibited the largest magnitude in difference across all proteins, while in fish eaters FGF23 showed the biggest difference. In addition, low meat eaters and poultry eaters both had lower OSM and CD99L2, the latter of which was higher in vegetarians. Low meat eaters, fish eaters and vegetarians all had lower ACP5 than regular meat eaters; this protein was lower but not significantly different in poultry eaters. In sensitivity analyses of the top proteins with different levels of covariate adjustment and limited to people of good and excellent health, results were similar (Supplementary tables 2).

FGF21

There are only a limited number of studies examining vegetarian and vegan diets with individual proteins. In support of our findings, higher concentrations of FGF21 in vegetarians and vegans have been shown in a small study of 36 each of omnivores, vegetarians and vegans [21]. FGF21 has been hypothesised to be high in vegans as a downstream effect of low intake of methionine [15], an essential amino acid that is known to be limiting in vegan diets [3]. Studies of FGF21 administration in rodents have found favourable effects on adiposity, lipid profiles, and non-alcohol fatty liver disease, but possible adverse effects on bone homeostasis [22]; randomised trials of FGF21 analogues also support the reduction in triglycerides [23] and improvement of fibrosis in patients with non-alcoholic steatohepatitis [24]. Consistent with these observations, we have previously reported lower BMI and body fat [25], more favourable lipid profiles [26] and lower heart disease risk [27], but also lower bone mineral density [25] and higher fracture risk [28] in vegetarians and vegans when compared with meat eaters; another study has reported lower odds of non-alcohol fatty liver disease in vegetarians than non-vegetarians [29].

GUCA2A

GUCA2A was substantially higher in all diet groups compared to regular meat eaters, with the exception of low meat eaters. GUCA2A activates the Guanylate Cyclase C receptor, which has a role in the maintenance of gut physiology including the increase of water movement into the intestinal lumen [30]. Previous studies have found that fasting plasma levels of GUCA2A are significantly lower in patients who have been diagnosed with Crohn’s disease compared with healthy controls [31], and that loss of GUCA2A expression may be an important determinant of colorectal cancer [32]. Consistent with the observations in this study, a previous EPIC-Oxford study reported a higher frequency of bowel movements in vegetarians and vegans [33].

FOLR1 and FOLR2

We found both FOLR1 and FOLR2 to be higher in non-meat eaters, with vegans having particularly high levels. As folate receptors, FOLR1 and FOLR2 both bind to and import folic acid into cells, and are usually downregulated with repletion of folate, however, this process may also be mediated by homocysteine in a positive direction [16]. While vegetarians and vegans typically have high folate, they may also have high homocysteine due to low vitamin B12 intakes [34, 35], though no other studies have reported on folate receptor expression in vegetarians and vegans for comparison.

IGFBP2

Similar to our findings, a previous study in EPIC-Oxford has found that compared with meat-eaters, both vegetarians and vegans had higher concentrations of IGFBP2, a binding protein for IGF-1 [13]. The mechanism for higher IGFBP2 in vegetarians and vegans is not well established but has been suggested to be related to lower dietary intakes of essential amino acids [13], protein (especially from dairy) [14], or lower total energy intake [36], which are all characteristic of vegetarian and vegan diets [3, 25, 37].

DSG2

Low and non-meat eaters, particularly vegans, had higher DSG2 in the current study. DSG2 belongs to the family of desmosomal cadherins, which includes both desmogleins (DSGs) and desmocollins (DSCs), with DSG2 and DSC2 being the major isoforms in humans, and the only isoforms expressed in the intestinal epithelium [12]. Consistently, the current study also showed that DSC2 was higher in non-meat eaters particularly vegans. Though the two proteins are structurally and functionally similar, laboratory evidence in mouse models suggest that DSG2 may have a more prominent role in the maintenance of the integrity of the intestinal epithelial barrier than DSC2 [38]. In humans, several small studies on IBD patients have found that lower intestinal protein levels of DSG2 in IBD patients, suggesting a role of the protein in IBD pathogenesis [12].

HAVCR1

HAVCR1 has been recognised as a biomarker of kidney injury and possibly an early diagnostic marker of kidney disease [11]. In mice, high-fat diets or the saturated fatty acid palmitate have been shown to upregulate expression of HAVCR1 in the proximal tubules of the kidney [39–41]. In humans, vegetarians and vegans tend to have lower intake of saturated fat, which is consistent with the observation of lower HAVCR1 in the non-meat eaters (though not significantly so in vegans) in the current study. Additionally, high protein diets, especially diets high in protein from animal sources, have been suggested to have adverse effects on kidney health [42]. A previous study suggested a lower risk of chronic kidney disease in vegetarians after accounting for baseline risk of diabetes and hypertension [43]. Furthermore, as HAVCR1 is expressed on the surface of immune cells, it is also believed to have a role regulating immune responses via activation and proliferation of immune cells [11].

PCSK9

The observation of higher PCSK9 in vegetarians is of interest. PCSK9 is a well-established drug target for treating hypercholesterolaemia, whereby PCSK9 inhibitors lowers low-density-lipoprotein cholesterol (LDL-C) concentrations by preventing the binding of PCSK9 to LDL receptors, which subsequently enables the LDL receptors to bind and remove LDL-C from the bloodstream instead [44, 45]. As a result, higher PCSK9 is often accompanied by high LDL-C, whereas the vegetarians in the current [26] and other studies [46] typically have significantly lower LDL-C, likely partly due to vegetarians having lower saturated fat intake [47]. It may therefore be plausible that the relatively higher PCSK9 expression in people with low LDL-C may be related to maintaining the minimum level of LDL-C necessary to support cellular function and structural integrity, but further studies are needed to confirm this.

Other proteins

The functions of many other proteins identified are not well established. CNDP1 is a dipeptidase, and one previous study reported higher CNDP1 in people on high protein diets [48], which would be consistent with our observation of lower CNDP1 in non-meat eaters (not statistically significant in vegans) who generally have lower protein intake. Several of the proteins identified such as ART3, GAS6, THY1 and EGFR are involved in a range of signalling pathways and progression of tumour tissues [49–52], while other proteins including LAYN and CDHR2 are believed to have roles in cell adhesion [53, 54]. AHSP, which was significantly higher in vegans than vegetarians, has a role in stabilising haemoglobin and its loss of function has been linked to inherited anaemia [55], but its relevance for nutritional anaemia is unclear. Overall, the reasons that these proteins appear to be influenced by diet and their implications require further research. As the first study on diet group and a large panel of circulating proteins, the current study focused on the biggest differences by diet groups, and many other differences were observed that could not be described in detail, but still deserve further investigation. The study demonstrates the utility of investigating protein differences by diet groups; future research should measure these and other proteins using multiple technologies in relation to dietary factors, to replicate and expand on our findings.

Strengths and limitations

The key strength of this study was that it showed for first time differences in circulating proteins between people of different habitual diet groups in a large population, and many of the findings were completely novel. We have defined six diet groups in the white British population and two diet groups in the British Indian population, which allowed a detailed comparison by both dietary habits and ethnicity. Of the limitations, as with all observational studies, some level of self-selection bias may be present, which limits the generalisability of the findings. While we have adjusted for several important confounders, residual confounding by other dietary and non-dietary factors may still be present, though sensitivity analyses showed that results were consistent across all adjustment models. We are also unable to infer causality due to the cross-sectional nature of the studies.

Conclusions

In this large population-based cohort in the United Kingdom, many differences in circulating protein concentrations were observed between different diet groups. These proteins are involved in a range of different biological pathways and processes, particularly related to gastrointestinal tract and kidney function. These differences likely reflect physiological differences between the different diet groups, and the implications of these differences for future disease risk require further investigation.