Diet composition and serum levels of selenium species: A cross-sectional study

legumes and dry fruits, while a negative correlation emerged with intakes of milk and dairy products and mushrooms. Concerning organic Se species, the most abundant one - Se-SelenoP - showed inverse correlation with food intake, and specifically with fish, fresh fruits, vegetables, and legumes, while Se-Cys positively correlated with fresh fruits, potatoes, legumes and mushroom intake. If confirmed in future studies, these findings may have major implications when assessing the nutritional and toxicological potential of foods with reference to Se.

The main source of Se exposure is dietary intake (Fairweather-Tait et al., 2011;Fan and Vinceti, 2015), though occupational environment, smoking, and air pollution may also contribute to Se exposure (Jablonska and Vinceti, 2015;Vinceti et al., 2013a). Detailed studies of dietary Se intake showed that cereals, meat, fish and dairy products are the main contributors to Se dietary intake in both European and Italian populations Gudmundsdottir et al., 2012;Lombardi-Boccia et al., 2003;Mariottini et al., 1995;Navarro-Alarcon and Cabrera-Vique, 2008).
The assessment of Se intake based on individual food consumption must consider the large variation of Se content in foods, influenced by environmental and geographic variations of Se content in soil and water (Mariottini et al., 1995;Shacklette and Boerngen, 2013), use of Se supplemented fertilizers (Fairweather-Tait et al., 2010), the differences in Se bioavailability in foods of plant origin (Baskett et al., 2001;Terry et al., 2000), and the effect of selfsupplementation with Se (Satia et al., 2006).
The few observational studies investigating Se content in food and its influence on Se blood levels have considered only the total amount of the trace element (Fairweather-Tait et al., 2011; M A N U S C R I P T A C C E P T E D ACCEPTED MANUSCRIPT 2 Mariottini et al., 1995;Pennington et al., 1984;Rayman, 2008), and not the single Se species.
Relatively few studies have assessed the levels of Se species in biological samples, including serum, cerebrospinal fluid and urine (Combs et al., 2011;Slotnick and Nriagu, 2006;Vinceti et al., 2015a;Vinceti et al., 2013b). These studies have shown large variation in Se species levels among individuals, probably related to the variations in dietary habits (Fairweather-Tait et al., 2011).
However, little is known about the Se species contained in different food items that characterize human diet, and even less about the relation between intake of specific foods and blood levels of Se species.
Recently, interest in Se speciation in foods has increased (Castro Grijalba et al., 2017;Ruizde-Cenzano et al., 2017), mirroring the increased attention to Se speciation in human health and disease. Distinguishing between Se species is of major relevance given the different properties of inorganic selenium species, such as selenite and selenate, and of the organic forms including for instance selenomethionine Lee and Jeong, 2012;Michalke et al., 2017b;Oliveira et al., 2017;Vinceti et al., 2017b;Weekley and Harris, 2013), the markedly different and not entirely elucidated metabolism of various selenium species, and the limited bioavailability of inorganic Se forms such as selenate (B'Hymer and Caruso, 2006;Gammelgaard et al., 2012;Jager et al., 2016;Weekley and Harris, 2013). The main form of Se in foods is selenomethionine (Weekley and Harris, 2013), which can be incorporated into proteins as a mimic for the correspondent sulfur-amino acid methionine (Schubert et al., 1987), a phenomenon which may cause protein misfolding and Se toxicity Vinceti et al., 2013b).
In this study, we sought to assess the influence of dietary intake on serum levels of Se species in a sample of a Northern Italian population.

Study population
The methodology for selection of study subjects has been described in detail elsewhere (Filippini et al., 2016;Vinceti et al., 2015a). Briefly, following approval of the study protocol by the Ethical Committee of the Policlinico University Hospital of Modena (n. 71/11), we randomly recruited fifty healthy subjects residing in Modena municipality without any relevant inflammatory disorder at the time of the recruitment, selected from sex-and age-specific subgroups through the databases of the Modena Municipality General Registry Office and using the Stata sample routine (Stata Corp., College Station, TX). Subjects were contacted from 2011 by land-line phone and invited to participate in the study. We recruited the 34% of contacted subjects and referred them to the Modena Health Unit to provide a fasting venous blood sample without any compensation apart from a complimentary breakfast.
Each participant completed a self-administered questionnaire collecting general demographic information, educational level, occupational history, smoking habits, and use of dietary supplements. In addition, we assessed their dietary habits through a validated semiquantitative food frequency questionnaire (FFQ) specifically designed for the Central-Northern Italian population within the EPIC study (Pala et al., 2003;Pasanisi et al., 2002). A general explanation of modality of completion of the FFQ was provided to each subject soon after recruitment, and instructions were also included on the questionnaire. In addition, within each section of the questionnaire (e.g. main courses, side dishes, beverages, etc.) detailed instructions about how to answer were included. This questionnaire assessed the average frequency and amount of consumption of 188 food items over the previous year, permitting estimation of consumption of macro-and micro-nutrients (including Se). We did this using an ad hoc software program (Malagoli et al., 2015;Malavolti et al., 2013). In particular the estimation of Se intake was implemented through ascertainment and analysis of foods characterizing the typical Italian diet (Bottecchi and Vinceti, 2012). Food intake was reported in g/day for each food.

Laboratory analysis
Blood samples were collected in a plastic tube (BD Vacutainer®, Becton Dickinson, Milan, Italy), immediately centrifuged for 10 min at 1000 x g and serum aliquots of 1 ml were stored at -15°C and kept continuously frozen until use. Selenium speciation analysis was performed at the Research Center for Environmental Health (Research Unit Analytical BioGeoChemistry, Neuherberg, Germany), using previously-described methodology (Solovyev et al., 2013;Vinceti et al., 2015a). To summarize, we slowly thawed samples in a refrigerator at 4°C, vortexed and subsequently analyzed them. Suprapure grade chemicals were used throughout. Selenite (Se(IV)), selenate (Se(VI)), selenomethionine-bound selenium (Se-Met), selenocysteine-bound selenium (Se-Cys), thioredoxin reductase-bound selenium (Se-TXNRD), glutathione peroxidase-bound selenium (Se-GPX), human serum albumin (HSA) and Tris buffer were from Sigma-Aldrich (Deisenhofen, Germany). We purchased certified Se and Rh stock standards (1000 mg/L) from CPI International, Santa Rosa, CA, USA, and we obtained ammonium acetate (NH 4 Ac) and acetic acid (HAc) from Merck, Darmstadt, Germany. Argon liq and methane (99.999% purity) were purchased from Air Liquide (Kleeve, Germany). We prepared stock solutions of Se(IV) and Se(VI) at a concentration of 1000 mg Se/L by dissolving in MILLI-Q water (18.2 MΩ cm, Milli-Q system, Millipore, Bedford, MA, USA). HSA was prepared at a concentration of 1000 mg/L. Preparation of Se-HSA was performed by mixing 10 mg Se/L selenite with this stock solution and incubating for at least 14 days. Working standards of Se species were prepared daily from their stock standard solutions by appropriate dilution with Milli-Q H 2 O. Since selenoprotein P-bound selenium (Se-SelenoP) is not commercially available as a standard compound, it can be prepared from serum using affinity chromatography (AFC) with the methodology previously described in detail (Solovyev et al., 2013).
We determined Se species Se(IV), Se(VI), Se-Met, Se-Cys, Se-TXNRD, Se-GPX, Se-SelenoP and Se-HSA in serum samples using anion exchange chromatography (IEC) coupled with M A N U S C R I P T A C C E P T E D ACCEPTED MANUSCRIPT 5 inductively-coupled plasma dynamic reaction cell mass spectrometry (ICP-DRC-MS) according to methodologies implemented for biological matrices (Michalke and Berthele, 2011;Solovyev et al., 2013). See Vinceti et al. (Vinceti et al., 2015a) for the detailed experimental settings, Se species determination, and quality controls. When the concentration of a Se compound was lower than the detection limit of 0.02 µg/L, we inputted in the database half that value (0.01 µg/L) (Whitcomb and Schisterman, 2008).

Data analysis
In selecting food categories in the present study, we took into account both the typical dietary pattern of the Italian population (Turrini et al., 2001), the types of foods previously reported as being rich in Se and relevant sources of this metalloid (European Food Safety Authority, 2014; Filippini et al., 2018;Filippini et al., 2017a), and the ability of such plants, the so called 'Se accumulator' vegetables including garlic, onion and cabbage, to accumulate large amount of the element (Finley, 2005;Terry et al., 2000). The final list of foods we selected included cereals, meat (both processed and not-processed), milk and dairy products, eggs, fish and seafood, vegetables (further divided in 'Se-accumulator' and 'Se non-accumulator'), legumes, tubers (i.e. potatoes), mushrooms, fresh fruits, dry fruits, and sweets (including sugar, chocolate and cakes). We calculated the median and interquartile range (IQR) of the intake of these foods and food categories.
In order to reduce the influence of measurement error frequently associated with food frequency questionnaires, either under-reporting and over-reporting food intake, we adjusted the estimates for total energy intake using the Willett's residual method (Willett, 2013). We computed Pearson correlation coefficients along with 95% confidence intervals (CI) between serum Se species and food category intake, after testing for normality of the data. We also used both crude and multivariable linear regression models, taking into account variable such as sex, age and M A N U S C R I P T A C C E P T E D ACCEPTED MANUSCRIPT 6 storage time, smoking habits and Se supplement use. We also performed a sensitivity analysis restricting the sample to subjects not taking Se-containing supplements.
When considering Se species in the analysis, we left apart Se-HSA and Se unknown species from the standard classification in organic and inorganic Se, for the uncertainties related to these forms (Vinceti et al., 2015a;Vinceti et al., 2013b) . After detecting a subject with extremely high and implausible levels of inorganic hexavalent Se(VI), 88.7 µg/L, we considered this sample as an outlier and removed the subject from all data analyses.

Results
Demographic and lifestyle characteristics of study participants have been reported in detail elsewhere (Vinceti et al., 2015a). Briefly, we recruited 50 subjects aged 35-70 years, males and females were equally represented. Nine (18%) participants were current-smokers, while twentysix (52%) and fifteen (30%) were never-and former-smokers, respectively. Finally, nine (18%) subjects reported use of Se supplement at time of the blood sampling, with a mean additional intake due to these supplements of 28.6 μg/Se/day. The distribution of Se species in serum total population in selected subgroups is shown in Supplementary Table S1.  Table S1).
Usual intake of the main food categories is shown in Figure 2. Median consumption was 240.8 g/day (IQR: 181.6-340.2) for fresh fruits, 167.4 g/day (IQR: 77.4-214.7) for cereals, 161.6 g/day (IQR: 90.6-272.0) for milk and dairy products, followed by 124.4 g/day (IQR: 77.6-166.1) for meat and 119.5 g/day (IQR: 72.3-176.7) for vegetables. Figure 2 shows the foods yielding the highest Se contribution to overall dietary intake, primarily fish and seafood (∼30%), meat (∼20%) and cereals (∼13%), followed by milk and dairy products (∼11%) and vegetables (∼7%) (see Supplementary Table S2). Table 1 shows Pearson correlation coefficients between Se intake and total serum Se and its species. Se intake was inversely correlated with organic species, especially Se-GPX but directly with Se-TXNRD, while positively correlated with inorganic forms. Exclusion of subjects taking Se-containing supplements gave comparable results. Table 2 shows correlation coefficients between food categories and serum Se species.
Considering total Se concentrations, we found a positive correlation with cereals and fresh fruits intake, but an inverse correlation with dairy products, mushrooms, vegetables, legumes and potatoes intake. Overall, organic Se concentrations were directly correlated with intake of dairy products and inversely correlated with intake of fish and seafood, legumes and dry fruits. Focusing on single organic Se species, Se-SelenoP, the most abundant form, showed an inverse correlation with intake of fresh fruits, and to a lesser extent with vegetables, legumes, and fish and seafood.
Conversely, concentrations of Se-Cys, the only form being the one characterized in selenoproteins, were directly correlated with intake of cereals, eggs, vegetables, fresh fruits, legumes and mushrooms, while inversely correlated with intake of sweets. Se-Met showed a positive correlation with intake of dairy products, while Se-Met and Se-GPX negatively correlated with legume intake. Se-TXNRD showed positive correlation with cereal intake and opposite with sweet intake. Serum concentrations of inorganic Se, the most toxic Se species, were inversely correlated with consumption of dairy products and mushrooms, while directly correlated with intake of fish M A N U S C R I P T A C C E P T E D ACCEPTED MANUSCRIPT 8 and seafood, legumes and dry fruits. Finally Se-HSA showed positive correlations with intake of cereals and fresh fruits, and inverse correlations with eggs, milk and dairy products.
Sensitivity analyses carried out using linear regression in crude and multivariable models (Supplementary Table S3 and Figure S1) generated similar results, as did analyses excluding subjects taking Se supplements (Supplementary Table S4).

Discussion
In our study population, we investigated correlations between food intake and serum is reduced to selenide, which in turn reacts with serine to form Se-Cys (Finley, 2005;Weekley and Harris, 2013). This was confirmed by Se speciation studies, which demonstrated that, in plants, organic Se species, especially selenomethionine and selenocysteine, are the predominant Se compounds, followed by inorganic species (Chan et al., 2010;Fox et al., 2002;Lyubenova et al., 2015;Montes-Bayon et al., 2002;Thavarajah et al., 2007;Whanger, 2002;Wolf and Goldschmidt, 2007). Accordingly, in our study we observed a strong positive correlation between intake of cereals, vegetables, legumes and fresh fruit and serum Se-Cys, while concerning inorganic forms only legumes demonstrated a positive correlation.
Our results showed little influence of meat intake on serum concentrations of Se species, despite the high Se content generally characterizing this food category, particularly liver and kidneys. Selenomethionine and selenocysteine appear to be the predominant species of Se in edible portions of meat (Bierla et al., 2008;Fairweather-Tait et al., 2011). However, the M A N U S C R I P T A C C E P T E D ACCEPTED MANUSCRIPT 9 bioavailability of the Se in meat is difficult to assess, and the literature has yielded inconsistent evidence on this topic (Bugel et al., 2004;Fairweather-Tait et al., 2010;van der Torre et al., 1991).
Milk and dairy products generally contain higher amount of organic Se compounds, mainly selenomethionine and selenocysteine (Muniz-Naveiro et al., 2007;Phipps et al., 2008).
Supplementation of animal feeding with organic Se species also demonstrated to be much more effective in increasing levels of Se in milk, while inorganic species showed limited effects (Ortman and Pehrson, 1999). Moreover, a study assessing the Se bioavailability of different types of milk showed that removal of the milk fat fraction increases its availability (Shen et al., 1996).
Interestingly, our results showed a positive correlation between milk intake and overall organic species, particularly Se-Met, but an inverse correlation with Se(IV).
In our study, egg consumption was strongly correlated with circulating Se-Cys levels, but did not influence serum concentrations of the other Se species, except for a slight increase in Se(IV) levels and inorganic Se overall. Hen's eggs contain from 3 to 25 µg Se per whole egg, and Se-Cys and Se-Met are the predominant species (Lipiec et al., 2010).
Similarly, a speciation study in fish samples from Spain and Portugal showed that Se-Met was the most abundant species detected, and its amount varied depending on fish type, with the higher level in swordfish (93%), followed by tuna (46%) and sardines (28%) (Cabanero et al., 2005). In line with these studies, in our study population intake of fish and seafood showed a positive correlation with this (only) organic form, Se-Met, and with inorganic Se and more specifically both its forms, Se(IV) and (VI). A study yielded in Sweden men yielded consistent results, showing that plasma Se increased slightly with the increasing consumption of fish, with no such increases in glutathione peroxidase and selenoprotein P (Huang et al., 1995). In general, Se bioavailability from seafood appears to be low (Moreda-Pineiro et al., 2013), and shrimps intake does not appear to M A N U S C R I P T

A C C E P T E D ACCEPTED MANUSCRIPT
10 influence plasma and glutathione peroxidase activity (Bugel et al., 2001). In addition, though Se in seafood generally demonstrates good absorption and high bioavailability after consumption of Serich fish (Fox et al., 2004), plasma Se tends to be quickly cleared from other organs (Bugel et al., 2001). Conversely other studies supported a low bioavailability of selenium from fish (Meltzer et al., 1993), possibly due to its interaction with heavy metals, particularly mercury and arsenic (Burger and Gochfeld, 2013;Cabanero et al., 2007;Copat et al., 2014;Olmedo et al., 2013).
Another reason explaining its bioavailability is the low conversion rate of inorganic Se into organic Se in the human, also taking into account the peculiar metabolic and catabolic pathways of inorganic Se (Cummins and Martin, 1967;Jager et al., 2016;Marschall et al., 2016;Suzuki et al., 2006).
Commercially and wild-grown mushrooms generally contain high Se levels (Cocchi et al., 2006;Pounis et al., 2014) and a wide spectrum of Se forms, i.e. Se-Met, Se-methylselenocysteine, Se-Cys, Se(IV) as well as several unidentified seleno-compounds (Falandysz, 2013). In our sample, we unexpectedly found null or inverse correlations of mushroom intake with most Se species, except for the slight positive association with Se-Cys. Possible explanations of these discrepancies are the different concentrations of Se species in the fruiting bodies of mushrooms, the biotransformation process of Se into organic compounds (Piepponen et al., 1983), and the speciation extraction procedures implemented (Diaz Huerta et al., 2006). For example, studies carried out on different species of Se-enriched mushrooms identified a major content alternatively of Se-Met or Se(IV) and Se-Cys (Bhatia et al., 2013;Stefanka et al., 2001). A low bioavailability of Se in mushrooms may also explain the low correlation we observed between intake of this food item and serum Se levels, possibly due to the high heavy metal content of mushrooms, thereby favoring Se excretion with the same mechanisms hypothesized for smoking (Vinceti et al., 2013a;Vinceti et al., 2000). The limited consumption of mushroom in the total diet characterizing our M A N U S C R I P T A C C E P T E D ACCEPTED MANUSCRIPT 11 free-living study subjects likely contributes to explain the limited influence of this food item on serum Se levels.
Considering sweet products, intake in our study population was inversely correlated with Se-Cys, and more slightly with Se-TXNRD, but no relevant relation emerged with any other Se species. Previous studies have generally assessed only total Se serum concentrations in relation with sweet intake, though a study carried out in multiple sclerosis patients showed results consistent with our findings (Socha et al., 2014), although no such relation emerged in controls.
However, other studies yielded conflicting results (Borawska et al., 2004;Letsiou et al., 2010), also depending on type of sweet products considered and on Se exposure status, with an inverse correlation below serum Se <70 µg/L and a positive one above that cut-point (Borawska et al., 2004).
We found a positive correlation between overall Se intake, as assessed through the food frequency questionnaire, and inorganic serum Se species, while no such correlation emerged for the other Se species. This could be due to the large influence of one dietary category (i.e. fish) on overall Se intake ( Figure 2). However, other factors could have hampered the correlations. First, biosynthesis of selenoproteins and therefore requirement of Se are only partially regulated by Se intake and status (Sunde et al., 2009), and many other factors such as pro-oxidant stressors, intake of other factors such as methionine may be involved (Vinceti et al., 2013a). This means that foods containing pro-oxidant factors might induce higher body retention and levels of organic Se due to the induction of a reactive proteomic response, which upregulate the antioxidant Se-containing proteins and therefore increase circulating Se-Cys levels (Jablonska and Vinceti, 2015). Secondly, uptake and retention by peripheral tissues tend to be considerably higher for the Se organic forms than for inorganic ones (Jablonska and Vinceti, 2015;Labunskyy et al., 2014;Vinceti et al., 2009).
In addition, two endocytic receptors are implicated in the tissue-specific uptake of SelenoP (Chiu-Ugalde et al., 2010;Olson et al., 2008;Olson et al., 2007), possible explaining the generally null or M A N U S C R I P T A C C E P T E D ACCEPTED MANUSCRIPT 12 even inverse correlation of Se-SelenoP with most of food as opposed to Se-Cys. Finally, Se intake evaluated through the food frequency questionnaire reflects the long term dietary exposure over the last year, while levels in blood are influenced by the short-term dietary intake and also by other environmental exposures such as smoking and air pollution (Di Dato et al., 2017;Vinceti et al., 2015a). In addition, since the relative distribution of Se species can differ between serum and other body compartments (Behne et al., 1996;Michalke et al., 2017a;Misra et al., 2012;Vinceti et al., 2017a), the associations we observed are specific to serum and do not necessarily apply to other matrices. Moreover, it is also worth noting that one (organic) Se species, Se-Met, was below the detection limit for half the sample, and therefore the results concerning this compound were incomplete and imprecise.
A few limitations of our study warrant discussion. First, a general caveat must be outlined: in this study, we assessed correlations between diet composition and blood Se species levels, but we did not have available information about food content of specific Se chemical forms, neither about the exact origin of the circulating Se species, which could derive either from foods and from endogenous metabolism of Se species, including ex-novo synthesis of organic Se forms (Bulteau and Chavatte, 2015;Cummins and Martin, 1967;Esaki et al., 1981;Gladyshev et al., 2016;Schmidt and Simonovic, 2012). Therefore, the observed correlations between food intake and circulating Se species could have been confounded by Se metabolism (Gladyshev et al., 2016;Schmidt and Simonovic, 2012), and in turn, influenced by diet composition, life-style variables and genetic factors (Vinceti et al., 2015a). Other factors that could have influenced correlations between food intake and serum levels of Se species in our study were intake of heavy metals, which are known to alter Se absorption, retention and bioavailability (Jablonska and Vinceti, 2015;Vinceti et al., 2013a).
Moreover, dietary information was collected with a self-administered food frequency questionnaire that, although validated for the Northern Italian population (Pasanisi et al., 2002),

M A N U S C R I P T
A C C E P T E D ACCEPTED MANUSCRIPT 13 was originally implemented nearly 10 years before the administration to study population (Pala et al., 2003), and this might have introduced some slight misclassification in the assessment of dietary habits. In addition, the small sample size of the study was associated with a limited statistical precision of the point estimates, as shown by their wide confidence intervals. Finally, other sources of selenium exposure could have affected the correlation between food intake and serum Se, such as airborne Se from anthropogenic activities including motorized traffic pollution (Fan and Vinceti, 2015;Heck et al., 2014), volcanic emissions (Ferrante et al., 2013;Tolu et al., 2014), smoking (Bogden et al., 1981), and the use of shampoos containing selenium (LeBlanc et al., 1999), at least in some subjects. This may have induced some exposure misclassification for particularly for inorganic Se serum concentrations, which may be found at higher levels in highly polluted urban environments and have been associated to risk of chronic disease (Heck et al., 2014;Ito et al., 2011).
Our study is the first to investigate, to the best of our knowledge, the relation between dietary habits, extensively assessed through a validated food frequency questionnaire, and the circulating levels of Se species Michalke, 2016;Michalke et al., 2009;Vinceti et al., 2013b), a relevant health issue taking into account the highly specific and in some cases even opposite biological effects of Se species. Another strength is that study subjects were randomly recruited from Modena residents, and their demographic characteristics and lifestyles (such as body mass index, educational attainments, occupational status and smoking habits) were similar to those characterizing the population of the entire Modena municipality, and more generally of Italy (Ferrante et al., 2012;Gruppo Tecnico PASSI Emilia-Romagna, 2012).

Conclusions
In this study, we found several correlations between food intake and serum Se species. The most toxic Se species, the inorganic ones, showed a positive correlation with intakes of fish, M A N U S C R I P T A C C E P T E D ACCEPTED MANUSCRIPT 14 legumes and dry fruits, while a negative correlation emerged with intakes of milk and dairy products and mushrooms. Concerning organic Se species, the most abundant one -Se-SelenoPshowed inverse correlation with food intake, and specifically with fish, fresh fruits, vegetables, and legumes, while Se-Cys positively correlated with fresh fruits, potatoes, legumes and mushroom intake. If confirmed in future studies, these findings may have major implications when assessing the nutritional and toxicological potential of foods with reference to Se.

M A N U S C R I P T
A C C E P T E D ACCEPTED MANUSCRIPT 15    -Selenium species concentrations in serum shows different correlation pattern with food intake -Inorganic selenium positively correlates with intake of fish, legumes and dry fruits, and negatively with dairy products -Organic selenocysteine-bound selenium positively correlated with intake of cereals, eggs, vegetables and fresh fruits, and negatively with sweets