The food additive EDTA aggravates colitis and colon carcinogenesis in mouse models

Inflammatory bowel disease is a group of conditions with rising incidence caused by genetic and environmental factors including diet. The chelator ethylenediaminetetraacetate (EDTA) is widely used by the food and pharmaceutical industry among numerous other applications, leading to a considerable environmental exposure. Numerous safety studies in healthy animals have revealed no relevant toxicity by EDTA. Here we show that, in the presence of intestinal inflammation, EDTA is surprisingly capable of massively exacerbating inflammation and even inducing colorectal carcinogenesis at doses that are presumed to be safe. This toxicity is evident in two biologically different mouse models of inflammatory bowel disease, the AOM/DSS and the IL10−/− model. The mechanism of this effect may be attributed to disruption of intercellular contacts as demonstrated by in vivo confocal endomicroscopy, electron microscopy and cell culture studies. Our findings add EDTA to the list of food additives that might be detrimental in the presence of intestinal inflammation, but the toxicity of which may have been missed by regulatory safety testing procedures that utilize only healthy models. We conclude that the current use of EDTA especially in food and pharmaceuticals should be reconsidered. Moreover, we suggest that intestinal inflammatory models should be implemented in the testing of food additives to account for the exposure of this primary organ to environmental and dietary stress.


Results
We were primarily interested in comparing iron compounds at a dose mimicking oral iron replacement therapy, for their potential to aggravate colitis and colitis-associated carcinogenesis. Several studies have implied a negative effect of oral iron replacement and associated high luminal iron concentrations in IBD [7][8][9][10][11][12] , although the effect in the human setting is controversial 13,14 . We examined the effect of oral iron compounds with different chemical properties on clinical and histological inflammation as well as on tumorigenesis in the azoxymethane-dextran sodium sulfate (AOM/DSS) mouse model 15,16 , as well as the interleukin 10-knockout (IL10 −/− ) mouse 17 . The selected compounds were: ferrous sulfate (the most commonly used ferrous salt for oral iron replacement), ferric maltol (a novel trivalent iron compound that has been licensed for iron replacement therapy in IBD 18 ), plant iron (an extract from the curry leaf plant (Bergera koenigii)), and Fe-EDTA (a product with high bioavailability which is recommended by the World Health Organization as an iron fortificant 19 ). Control animals received ferrous sulfate at 45 mg elemental Fe/kg chow, i.e. the iron content of the standardized rodent chow AIN76A. The test substances were mixed into an iron-depleted chow to achieve a tenfold iron content of AIN76A (Supplementary Table S1). We were surprised to find that only Fe-EDTA exacerbated colitis and massively increased tumour burden, while the other iron compounds did not differ from the control (Fig. 1, Supplementary Fig. S1). In the AOM/DSS-model, the mice fed Fe-EDTA were only able to tolerate the first and fourth DSS cycle due to massive weight loss and diarrhea. Similarly, in the IL10 −/− model, the Fe-EDTA group developed severe intestinal inflammation leading to premature sacrifice at week 8 compared to week 31 for the other groups.
Next we tested whether the intestinal toxicity is due to EDTA or is only specific to Fe-EDTA. We compared Fe-EDTA, Ca-EDTA and Na-EDTA added to a regular chow to a control group. Two EDTA doses, 173 mg EDTA/kg bw, corresponding to the NOAEL in rodents, and 21 mg EDTA/kg bw, mimicking the human ADI dose (Supplementary S1) were used. Also, a less aggressive treatment with longer recovery phases after DSS was applied. The results demonstrated that, in the AOM/DSS model, all EDTA compounds led to higher colitis activity compared to the control group (Fig. 2). No tumours were found in the control group compared to EDTA groups. Histological activity was low in all groups, partially because mice were sacrificed after 11 days of recovery from DSS. In the IL10 −/− model, the colitis activity was also higher in EDTA groups. A dose-related effect could be observed, with groups treated with 173 mg/kg having higher colitis activity than groups treated with 21 mg/kg. EDTA groups, especially those treated with 173 mg/kg also had a significantly higher tumour burden than the control group (Fig. 2, Supplementary Fig. S2). These results confirm that the aggravating effect on colitis and colitis-associated carcinogenesis was not specific to a certain EDTA compound but rather conferred by EDTA itself. The omitted DSS cycles in the Fe-EDTA group due to high colitis activity are marked with an asterisk. (c,d) Mean DAI over the full time course (i.e. weeks 2-9 for AOM/DSS (c), weeks 1-8 for IL10 −/− (d)); (e,f) Histological activity index (HAI) for AOM/DSS (e) or IL10 −/− (f); (g,h) Tumour burden (i.e., total tumour area per mouse) for AOM/DSS (g) or IL10 −/− (h); exemplary image of hematoxylin-eosin-stained intestines of control (i,k) and Fe-EDTA-fed (j,l) animals. (i) DSS-induced increased inflammatory infiltrate (arrow) with partial loss of crypts in a control animal from the AOM/DSS model. (j) massive inflammation (double arrow) with complete crypt destruction and a single regeneratory layer of epithelial cells covering the lamina propria (single arrow) in an Fe-EDTA-treated animal from the AOM/DSS model. On the lower magnification image (left side), an invasive tumour (*) is seen; the point of invasion through the lamina mucularis mucosae is marked with **. (k) Inflammatory infiltrate (double arrow) and cryptitis through invading neutrophils (single arrow) in a control animal from the IL10 −/− model. (l) Marked hyperplasia, crypt abscess (single arrow) and massive inflammatory infiltrate (lymphocyte aggregates; double arrow) in an Fe-EDTA-treated animal from the IL10 −/− model. Error bars represent standard deviations. Asterisks (*: p < 0.05; **: p < 0.01; ***: p < 0.001) denote statistically significant results compared to the control group.  www.nature.com/scientificreports/ EDTA is a chelator with high affinity for di-and trivalent cations including Ca 2+ and Mg 2+ . Many of the intracellular contacts are directly (cadherin-based contacts, i.e. adherens junctions (AJs) and desmosomes) or indirectly calcium-dependent (tight junctions (TJs) and integrin-based contacts, i.e. hemidesmosomes 20,21 ). Of note, EDTA is widely used in biomedical research to detach adherent cultured cells or to isolate epithelia from organs 22,23 . Hence, we hypothesized that the toxic effect of EDTA may be conferred by disruption of the mucosal barrier and intercellular contacts. Therefore, we examined different components of the intestinal barrier (mucus layer, TJs, AJs and desmosomes) in the intestines from different groups of EDTA-treated animals. No changes were observed in expression of ZO-1 (marker of TJs) or desmoglein-2 (desmosomes); however, markers of AJs (β-catenin and E-cadherin) showed modestly altered expression. A loss of membranous β-catenin in distal colon (where disease activity was highest in AOM/DSS mice) was observed ( Supplementary Fig. S3). E-cadherin also tended to reduced membranous expression with increased cytoplasmic localization. A similar trend was observed were not conclusive about EDTA-specific effects in the setting of mild histological disease activity at the time of sacrifice as well as regenerative inflamed mucosa with hyperplastic and dysplastic tissue architecture.
Intestinal inflammation itself is known to weaken the intestinal barrier 23,24 . In order to examine the EDTA effect independently of colitis, we exposed cultured monolayers of T84 cells to EDTA compounds. Also, to mimic inflammation, a pretreatment with tumour necrosis alpha (TNFα) and interferon gamma (IFNγ) was administered. The integrity of the intercellular contacts was analyzed by immunofluorescence of junctional proteins. EDTA treatment alone caused breaks and mislocalization of intercellular contact proteins (Supplementary Fig. S4). The effect of Fe-EDTA and Ca-EDTA was comparable, whereas Na-EDTA detached the whole cell monolayer at the same concentrations. TNFα + IFNγ alone had a similar effect compared to EDTA, and enhanced the disruption of the epithelial barrier components by EDTA. We next assessed the integrity of the barrier by measuring paracellular FITC-dextran permeability in T84 cell monolayers. The results showed an additive effect of inflammation and EDTA on increased permeability (Fig. 3a). To further confirm the direct disruption of epithelial barrier by EDTA, we studied the effect of topical EDTA application in vivo using endomicroscopy. Healthy mice received two applications of Na-EDTA rectally for 10 min. Upon fluorescein injection, endomicroscopy showed increased permeability of the intestine for fluorescein. (Fig. 3b,c). Electron microscopy of samples obtained from these animals demonstrated the appearance of gaps on the lateral cell-to-cell contact surface that rarely reached the luminal or the basolateral surface, corresponding to disrupted AJs (Fig. 3d,e). There were few desmosomes present, which were intact. The adhesion to the basal membrane showed no abnormalities. Altogether, these in vitro and in vivo studies demonstrated a direct effect of EDTA on epithelial barrier disruption.
Next we examined the composition of the bacterial flora in EDTA-treated mice by bacterial 16S-rRNA gene amplicon sequencing. The microbiome analysis revealed clear separation of EDTA groups in both AOM/DSS and IL10 −/− models, combined with a decrease in diversity (Fig. 4). The most notable change was a > 10-fold increase in the abundance of Akkermansia muciniphilia, a mucin-degrading bacterium of the phylum Verrucomicrobia. A. muciniphilia is typically reduced in patients with active IBD 25 , although some studies associate it with proinflammatory 26 and procarcinogenic properties 27 . This is one of the first species able to colonize the intestine after fecal microbiota transplantation 28,29 , therefore, its increase might simply reflect a regeneratory state. A slight increase in Peptostreptococcae was also present. Peptostreptococcus anaerobius has previously been attributed procarcinogenic properties 30 , the exact functional consequence of our finding remains unknown. www.nature.com/scientificreports/

Discussion
Our results demonstrate that EDTA is toxic to the intestine when inflammation is present in doses that were not expected to cause any adverse effects. The addition of EDTA compounds to the food strongly enhances intestinal inflammation and colorectal carcinogenesis in two biologically different models of IBD. We show that EDTA disrupts various components of the intestinal barrier and increases intestinal permeability. This effect is also present in healthy animals and is likely massively aggravated in the presence of inflammation, leading to impaired wound healing and perpetuation of the inflammatory cascade. Dysbiosis is also induced by EDTA and it may contribute to the toxic effects.
No noticeable toxicity of EDTA in very high doses has been shown in multiple safety testings in healthy animals, leading to the recommended safety doses for human use. The ADI of 1.9 mg EDTA/kg bw recommended by the Joint FAO/WHO Expert Committee on Food Additives (JECFA) in 1973 for humans 31 is derived from the NOAEL for rats of 250 mg EDTA/kg bw from the study by Oser et al. 32 , where 250 mg EDTA/kg bw was the highest concentration used; in other studies doses up to 2500 mg/kg bw have shown no toxicity 5,33 . Diarrhea as an adverse event has not been described in doses below 250-1000 mg/kg bw 6,34,35 , which is about three-fold higher than the highest dose in our study. Although EDTA is capable of disrupting the intestinal barrier in healthy animals as demonstrated by our experiments (Fig. 3b,c), the magnitude of the effect is obviously insufficient to www.nature.com/scientificreports/ cause a detectible clinical phenotype. Only for Fe-EDTA, several studies in colitis models have demonstrated an increase in intestinal inflammation and colitis-associated carcinogenesis; but the effect was attributed to iron and not EDTA 8,36 . In line of our findings, Constante et al. noted increased intestinal inflammation in DSS-treated mice with Fe-EDTA but not other iron compounds, concluding that intestinal toxicity might be specific to Fe-EDTA 37 .
None of these studies have pointed to EDTA as the specific toxic moiety. However, while the compound seems safe in a healthy intestine, our results show that this is clearly not the case in the presence of gut inflammation. Unfortunately, there is no human data in the setting of IBD, infectious diarrhea or colorectal cancer, which, in the light of our study are specific risk populations. Our study has several limitations. It utilizes only animal and cell culture models, and no data are available on human exposure to EDTA. The disruption of epithelial barrier components in both colitis models is not as apparent as in the healthy mice or in cultured epithelial cell monolayers. This is most likely due to the time point of sacrifice that was rather late after the initial inflammatory stimulus (in order to better observe tumour development). It remains not entirely clear how intestinal inflammation enhances EDTA toxicity, as similar changes are observed in healthy mice after a single short-term EDTA exposure. We hypothesize that a healthy mucosal barrier is more resistant to EDTA, and that a disruption of the barrier components by inflammation exposes deeper and otherwise protected mucosal structures to EDTA and facilitates translocation of commensal intestinal bacteria. Disruption of the barrier components (mucin 38,39 , tight junctions 40 , desmosomes 41,42 , hemidesmosomes 43 ) by genetic defects or by immunological or chemical methods causes intestinal inflammation by itself, which makes it impossible to study these two factors independently in vivo. A direct effect of EDTA in promoting dysbiosis and therefore inflammation is also possible, as EDTA alters the stability of bacterial cell walls, delays microorganism growth and prevents adhesion by its sequestering action on divalent cations 44 . It remains unclear whether the increased colitis-associated carcinogenesis is solely secondary to inflammation or whether EDTA also has direct procarcinogenic properties. Scheers et al. demonstrated that Fe-EDTA may promote the proliferation of CaCo-2 and Hutu-80 cancer cells by activating Erk via increased levels of amphiregulin and EGFR but not via the Wnt pathway 45 . Studies have also proposed that EDTA may disrupt DNA stability by interfering with DNA-bound proteins by chelating Zn 2+ , possibly also Ca 2+ and Mg 2+6 ; the functional relevance of such chelation remains unknown. It remains unexplored whether the chelating action of EDTA can influence the intracellular redox balance, thus promoting direct DNA damage.
Nevertheless, the relevance of our findings remains high. EDTA is widely used and very stable. It is detected in most large rivers and even found in the drinking water in concentrations up to 30 µg/l 4 . The gastrointestinal (GI) tract would be inevitably exposed to EDTA especially by water together with its use in foods, pharmaceutics, cosmetics and household chemicals. Because of differences in local regulations and practices, the extent of exposure to EDTA is likely to vary from country to country. According to our results, due to the presumed EDTA's toxicity in a specific population (IBD), a lowering and re-determining of the ADI is warranted. Also, specific recommendations in individuals with GI diseases such as IBD, irritable bowel syndrome, GI cancer or infectious diarrhea should be issued. It would be important to address the exposure to EDTA in healthy individuals and in patients with the above mentioned conditions in further studies, although studying dietary factors and linking them to a disease phenotype is notoriously problematic due to inconsistent exposure over time, ethical issues preventing a randomized design, difficult data collection and large number of subjects required for a cohort study. This study also highlights the shortcoming of the way food additive testing is done only in healthy animals. Other food additives and dietary agents have shown relevant intestinal toxicity in the presence of intestinal inflammation that was not apparent in healthy animals, such as emulsifiers 38 , TiO 2 46 , or most recently polyunsaturated fatty acids 47 . Within the healthy human population itself, disruption of GI barrier function is common and linked to numerous GI conditions as mentioned above. We propose to remove EDTA from ingested substances and to include intestinal inflammatory models in future safety testing.

Conclusion
We demonstrate a previously unrecognized intestinal toxicity of EDTA, a chelator used as a food additive and in pharmaceuticals among numerous other applications. EDTA salts induce massive intestinal inflammation and increased colorectal carcinogenesis in biologically different animal and cell culture models of inflammatory bowel disease at doses that are comparable to human use. We propose that the disruption of the epithelial barrier function may be the mechanism of the observed effect. Interestingly, this toxicity is not evident in healthy animals and therefore has been missed by regulatory safety testing. We therefore suggest the inclusion of intestinal inflammatory models in safety testing procedures for food additives as a strategy to detect otherwise unrecognizable toxicity in the intestine as a primary organ of exposure. www.nature.com/scientificreports/ no experimental diets were administered. Animals received food and tap water (unless otherwise mentioned) ad libitum. The experimental diets were commenced as specified for every experiment. The composition of the diets is summarized in Supplementary Tables S1 and S4. The researchers involved were not blinded to the group allocation. The 173 mg EDTA/kg bw dose was chosen as the NOAEL dose for Fe-EDTA according to EFSA 5 . The 21 mg EDTA/kg bw dose was chosen to represent the ADI in humans and is derived according to the formula ADI = NOAEL × 0.01 × 12.3, where 0.01 is the overall default uncertainty factor to account for inter-and intraspecies variability as recommended by EFSA 48 , and 12.3 is the body surface area-based metabolic weight conversion factor between mice and humans 49 . The conversion between mg/kg bw and mg/kg chow was performed using the quotient 0.149, also recommended by EFSA 48 . The ADI for humans as set by JEFCA is 1.9 mg EDTA/kg bw (or 2.5 mg Ca-EDTA/kg bw), corresponding to 23.37 mg EDTA/kg bw for mice and is slightly higher than the one we used (21 mg/kg bw) 31 .

Methods
For the AOM/DSS model 15,16 , azoxymethane (Sigma Aldrich, St. Louis, MO, USA) was administered at a concentration of 7.5 mg/kg bw intraperitoneally at a time point specified for every experiment. Dextran sodium sulfate (MP Biomedicals, Eschwege, Germany) was given with the drinking water at a concentration of 1.5% weight/volume for five days. The timepoint of each DSS cycle is denoted at the description of the corresponding experiments. During a subsequent recovery phase, the animals received tap water.
For the IL10 −/− model 17 , colitis induction and synchronization was done using piroxicam (Sigma Aldrich) at 200 ppm with the chow 50 for one cycle of 8 days (Fig. 1) or two cycles of five days with a recovery phase of 4 days between cycles (Fig. 2).
Throughout the animal experiments, the mice were weighed at least once per week, and their stool was examined for consistency and the presence of overt or occult blood (using a guaiac test, Haemoccult, Beckman Coulter). These variables were used to obtain a clinical disease activity score (DAI) 51 (Supplementary Table S2).
At the end of the experiment, the mice were sacrificed by cervical dislocation after anesthesia with xylazine and ketamine.
Histology and immunohistochemistry. The intestines were collected, flushed and prepared using the Swiss roll technique 52 and subsequently fixed in phosphate buffered formalin 10%. A subset of the intestines was not flushed but fixed including the content using the mucus preserving methacarn solution (methanol 60%, chloroform 30%, glacial acetic acid 10%). Five µm cuts were prepared after standard dehydration and paraffin embedding procedures. For the examination of colitis activity and presence of tumours, hematoxylin and eosin stains were performed using a standard method. The intestines were then examined under a light microscope (Olympus BX51 microscope with an Olympus DP73 microscope camera, Tokyo, Japan). A histological colitis activity index (HAI) 53,54 (Supplementary Table S3) was obtained. The intestinal tissue was also examined for the presence of tumours, and the area of each tumour was measured using the CellSens Dimension Version 1.17 software (Olympus, https ://www.olymp us-lifes cienc e.com/de/softw are/cells ens/). The presence of invasiveness (i.e. breakthrough through the mucularis mucosae) was noted; a dysplasia grading was not performed. The examiners (TA and AC) were not blinded for the group allocation. The following variables were calculated: total tumour burden (i.e. sum of all tumour areas per mouse), mean tumour size, tumour multiplicity (number of tumours per mouse) and invasive tumour multiplicity.
The intactness of the mucus layer was evaluated using methacarn-fixed intestines. Periodic acid Schiff 's base (PAS) mucus stain was performed using a standard method, and the intactness of the layer was examined and graded 0%-100% under a light microscope.
The intactness of intercellular contacts was examined by immunohistochemistry, which was performed using a standard method. The antibodies used are listed in Supplementary Table S5. The intensity of the stain was graded as 0 (no expression), 0.5 (weak expression) and 1 (strong expression), multiplied with the estimated relative area with the corresponding staining intensity, thus resulting in an immunoreactivity score ranging from 0 to 100%. The examiner (GD) was blinded for the group allocation by opacifying the slide labeling.
Measurement of fecal EDTA content. Stool samples from a subset of animals were collected immediately post mortem, snap frozen in liquid nitrogen and stored at − 80 °C. The stool samples were dried at 105 °C and homogenized mechanically, by sonification and shaking for 1 h in a Fe-complexing solution (20 mg Fe(III) sulfate diluted to 100 ml with a 1.5 mM sulfuric acid solution; all reagents from Sigma Aldrich). The centrifuged supernatant was subjected to high performance liquid chromatography using a HPLC-System 1260 with diode array detector (Agilent Technologies, Santa Clara, CA, USA) and a Hamilton PRP-X100 anion exchange column (10 µm particle size, Sigma Aldrich). The main peak of EDTA was measured at 300 nm using diode array detection and G2170AA Rev. B.04.03 software (Agilent, https ://www.agile nt.com). The analyses were performed at the Chemcon laboratory (Vienna, Austria). Paracellular permeability assay. T84 cells (source: ATCC, Manassas, VA, USA) were grown in Dulbecco's modified Eagle's medium-F12 (Gibco, Thermo Fisher Scientific, Waltham, MA, USA) supplemented with 10% fetal bovine serum (Biochrom, Berin, Germany). 1 × 10 5 cells were plated on 24-well polystyrene transwells (0.4-μm pore size; Costar, Corning, NY, USA) for 14 days to form intact monolayers as described previously 23 . To mimic inflammation, the monolayers were pretreated with tumour necrosis factor alpha (TNFα, 50 ng/ml, Miltenyi Biotec, Bergisch Gladbach, Germany) and interferon gamma (IFNγ, 50 ng/ml, eBioscience, Thermo Fischer Scientific) for 1 h and then with EDTA compounds (Fe-EDTA at 4 mM, Ca-EDTA at 4 mM or Na-EDTA at 0.625 mM, corresponding to 1168 mg EDTA/l or 182.5 mg EDTA/l) for 3 h. A lower dose of Na-EDTA compared to the other compounds had to be used, as higher concentrations led to an immediate detachment of the cells. 10-kDa fluorescein isothiocyanate-labeled dextran (FITC-dextran, 1 mg/ml, Sigma Aldrich) was then Confocal laser endomicroscopy. Healthy 6 weeks old C57BL/6 male mice (n = 3 per group) were used.
One day prior to the experiment, a bowel preparation solution (2.6 g NaCl, 13.5 g glucose, 1.5 g KCl, 2.9 g trisodium citrate, 34.5 g polyethylenglycol 35000 (all from Sigma Aldrich), distilled water to a total volume of 1 l) was given instead of drinking water. The animals were anesthetized using ketamin 100 mg/kg and xylazine 12 mg/ kg intraperitoneally. After removing any stool rests in the sigmoid colon by flushing under visualization using a miniature sigmoidoscope (Karl Storz, Tuttlingen, Germany; airpump from Eheim, Deizisau, Germany), EDTA or saline were applied via the working channel of the endoscope. After administration of fluorescein (0.05 mg/g bw intraperitoneally, Fluorescite, Alcon Ophthalmika GmbH, Vienna, Austria), two rectal applications of 2 ml Na-EDTA (1 mM, corresponding to 292 mg EDTA/l) or saline for 10 min were given.  Fig. 1 (a pilot experiment) were based on the assumption of doubling the tumour multiplicity with the iron compounds compared to control, using a power of 0.8 and significance of 0.05. For the experiment depicted in Fig. 2, sample size calculations were based on the estimates for total tumour burden from the experiment from Fig. 1 using a power of 0.9 and a significance of 0.05. Bonferronicorrected t-tests were utilized for the calculation of the sample size. The other presented parameters were considered as secondary endpoints. Results from animals that died prematurely have not been used in the final analysis. The sample size for the further experiments was empirically set at 3, as they were considered pilot experiments. For all final analyses, non-parametric Kruskal-Wallis ANOVA with Dunn's post-hoc tests were used for comparisons of more than two groups, and Mann-Whitney U-test for comparisons between two groups. For the microbiome analysis, generalized UniFrac distances were visualized using non-metric multi-dimensional scaling (NMDS). Cluster significance was assessed using permutational multivariate analysis of variance. Testing for significant differences in diversity and bacterial abundances was performed using Kruskal-Wallis Rank Sum Test with Benjamin-Hochberg method for correction for multiple comparisons. All tests were performed as twotailed tests. A p-value below 0.05 was considered statistically significant for all tests. The datasets generated and analyzed during the current study are available from the corresponding author on reasonable request.
The study was carried out in compliance with the ARRIVE guidelines. www.nature.com/scientificreports/ Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creat iveco mmons .org/licen ses/by/4.0/.