Rebellato et al. 2023 — Inorganic contaminants in plant-based yogurts (Brazil retail)

Rebellato and colleagues at the Institute of Food Technology (ITAL, São Paulo) and Adolfo Lutz Institute measured 11 inorganic elements (Al, Cr, Co, Ni, As, Mo, Cd, Sb, Ba, Hg, Pb) in 43 plant-based yogurt sample-lots and 1 cow-milk natural yogurt sample-lot purchased from retail establishments in Campinas (São Paulo, Brazil) between August and October 2022. Quantification was by inductively coupled plasma mass spectrometry (ICP-MS) on an iCAP RQ instrument following ultrasound-assisted nitric acid / hydrogen peroxide digestion at 80 °C for 35 minutes. Element-detection patterns differed sharply between plant- and animal-based yogurts: in plant-based samples Al, Cr, Co, Ni, Mo, Ba and Pb were quantified at appreciable concentrations across multiple samples, with Cd detected in 2 sample-lots and As detected in 3 sample-lots; Sb and Hg were below the limit of quantification (LOQ) in all 44 sample-lots. In the cow-milk yogurt comparator, only Mo (72.54 µg/kg) and Ba (160.76 µg/kg) were above LOQ. Estimated dietary intake values from a single unit of the highest-concentration plant-based yogurt remained below available PTWI, PTMI, and BMDL reference values for adults but reached 46.7% of the EFSA tolerable daily intake (TDI) for nickel in children, rising to 93.4% of the Ni TDI at 2 units/day.

Key numbers

All concentrations reported in micrograms per kilogram (µg/kg) wet weight of the product as commercialized, per the source’s Sampling and Sample Preparation sections (approximately 0.5 g of sample digested in 4 mL HNO3 + 2 mL H2O2, then ultrasonic bath 80 °C / 35 min, diluted to 20 mL with ultrapure water, 0.45 µm PTFE filtration; n = 3 independent analytical replicates per sample-lot). Values are reported as mean ± standard deviation. Letters within the same sample column denote significant differences across lots of that sample (one-way ANOVA + Tukey’s test, p < 0.05). “n.d.” denotes not detected (< LOQ). LOQ = 200 µg/kg for Al; 4 µg/kg for Cr, Co, Ni, As, Mo, Cd, Sb, Ba, Hg, Pb. LOD = 119 µg/kg for Al; 2 µg/kg for the other elements. Sample identifiers (A–R) and lot numbers (1–3) together uniquely disambiguate every replicate in the source’s Table 3; the source’s per-sample brand codes (described in Verification notes) are not surfaced on this page.

Limits of detection and quantification (Section 3.1, source p. 5)

Analytical curves linear with r² > 0.99 for all elements. Recovery ranged 80 to 110% against certified reference materials (INCT-TL-1 Tea leaves for Al/Ni/Cd/Sb/Ba/Pb; ERM-DB 151 skimmed milk powder for Mo/Hg/Pb; Tort-3 Lobster Hepatopancreas for Co/As). Precision (CV across 7 independent repetitions of the yogurt sample) ranged 6 to 15%, meeting INMETRO specifications.

Plant-based yogurt concentrations by element (Table 3, source pp. 6–7)

Selected per-element ranges across the 43 plant-based sample-lots (excluding < LOQ entries):

The narrative ranges given in the abstract (Al < LOQ–9019.05; Cr < LOQ–88.14; Co < LOQ–40.56; Ni 31.71–700.46; As < LOQ–10.61; Mo < LOQ–355.70; Cd < LOQ–4.37; Sb < LOQ; Ba < LOQ–1505.71; Hg < LOQ; Pb < LOQ–21.58 µg/kg) match the Table 3 detected-sample extremes above.

Cow-milk yogurt comparator (sample R, lot 1; Table 3, p. 7)

Only Mo and Ba were quantified above LOQ in the cow-milk yogurt; all nine other elements were < LOQ.

Plant-based formulation patterns highlighted by the authors (Sections 3.2.1–3.2.10, pp. 8–12)

• Aluminum: highest in soy-chocolate formulations containing sugar syrup, vegetable fat, cocoa powder, and soy protein isolate (sample M lot 3, 9019.05 µg/kg) and in pea/soy-protein veg-protein strawberry formulations containing protein isolate and coconut milk (sample C lot 3, 5844.72 µg/kg). Two sample groups (F and O) were < LOQ for Al across all lots.
• Chromium: highest in coconut-cream-based yogurt-flavored coconut food (sample D lot 1, 88.14 µg/kg) and natural coconut cream (sample O lot 1, 55.52 µg/kg); both formulations share coconut-cream, modified starch, and additive composition.
• Cobalt: highest in soy-chocolate samples (sample M lots 1–3, 28.03–40.56 µg/kg) containing sugar syrup, vegetable fat, cocoa powder, and soy protein isolate; below LOQ in five sample groups (E, F, G, J, K).
• Nickel: highest in two coconut-yogurt formulations from the same brand source (samples Q and P, 700.46 and 491.08 µg/kg respectively), both containing coconut milk powder requiring reconstitution. The authors flag this as the most exposure-relevant finding given the EFSA TDI of 13 µg/kg bw per day for Ni and the small body weight assumed for the child consumer profile (15 kg).
• Arsenic (total): detected above LOQ only in samples L lot 3 (10.61 µg/kg, berry fermented coconut cream), M lot 1 (4.33 µg/kg, soy-chocolate), and P lot 1 (4.38 µg/kg, red-fruit coconut yogurt). Remaining 40 plant-based and 1 animal-based sample-lots were < LOQ. (Wiki Key-numbers table reports the floor as 4.33 ± 0.15 since only the M lot 1 entry carries a tabulated SD; the P lot 1 value is 4.38 ± 0.24.)
• Molybdenum: highest in veg-protein samples A, B, C (293.41 to 355.70 µg/kg across A, B, C lots 1–3) and soy-chocolate sample M (190.39 to 207.61 µg/kg across M lots 1–3). The source narrative (Section 3.2.6, p. 10) describes the combined A/B/C and M cohort range as 190.39 to 355.70 µg/kg. The authors attribute this to the shared pea and soy protein isolate ingredient base across these formulations.
• Cadmium: above LOQ in only two sample-lots — sample M lot 2 (4.20 µg/kg, soy-chocolate) and sample Q lot 1 (4.37 µg/kg, coconut yogurt).
• Antimony: all 44 sample-lots < LOQ.
• Barium: highest in berry fermented coconut cream (sample L lots 1–3, 1072.76 to 1505.71 µg/kg) and soy-chocolate (sample M lots 1–3, 527.17 to 580.98 µg/kg). Authors attribute the L pattern to red-fruit (strawberry, blackberry, blueberry) and chia ingredients, and the M pattern to sugar syrup, vegetable fat, and cocoa powder.
• Mercury (total): all 44 sample-lots < LOQ.
• Lead: detected above LOQ in samples D lot 3 (4.39 µg/kg), E lot 3 (5.74 µg/kg), J lot 1 (6.07 µg/kg), M lots 1 (19.85 µg/kg) and 2 (21.58 µg/kg), and P lot 1 (4.33 µg/kg). The remaining 38 sample-lots were < LOQ. Sample M (soy-chocolate) showed the highest Pb concentrations, attributed by the authors to its sugar syrup, vegetable fat, and cocoa powder ingredient profile.

Estimated dietary intake (EDI) calculated by the authors

EDI computed using a deterministic single-unit-per-day model with the maximum measured element concentration applied to the package mass (range 90 to 250 g across products), expressed as µg metal per kg body weight per day. Body weight assumptions: 15 kg child, 60 kg adult (Section 2.6, p. 5).

For the Ni EDI specifically, the authors note that consumption of 2 units/day of the highest-Ni plant-based yogurt (sample Q, 700.46 µg/kg) corresponds to 93.4% of the EFSA TDI of 13 µg/kg bw per day for the 15 kg child profile, and call for caution at that consumption rate. For Mo, the authors compute an adequate-intake equivalent of 88.9 µg/day from sample B (355.70 µg/kg × package mass), exceeding the EFSA AI of 65 µg/day for adolescents/adults and recommending moderate consumption. For all other elements, single-unit and multi-unit consumption stay well below the relevant reference value.

Methods (brief)

Analytical method: inductively coupled plasma mass spectrometry on a Thermo Scientific iCAP RQ ICP-MS (Bremen, Germany), RF power 1550 W, He collision-gas mode at 5.00 mL/min, micromist nebulizer at 0.98 L/min, dwell time 0.3 s (0.02 s for internal standard), monitored isotopes ²⁷Al, ⁵³Cr, ⁵⁹Co, ⁶⁰Ni, ⁷⁵As, ⁹⁷Mo, ¹¹¹Cd, ¹²³Sb, ¹³⁷Ba, ²⁰²Hg, ²⁰⁸Pb, internal standards ⁴⁵Sc, ⁷²Ge, ¹¹⁵In, ¹⁰³Rh, ²⁰⁹Bi, ¹⁹⁵Pt at 50 µg/L. No chemical speciation was performed — reported As and Hg values are total arsenic and total mercury; Cr is total chromium with no hexavalent speciation.

Sample preparation: ultrasound-assisted acid digestion adapted from Fioravanti et al. 2020. Approximately 0.5 g of sample weighed into a 50 mL graduated tube; 4 mL HNO3 (sub-boiling distilled) and 2 mL 30% H2O2 added; closed tube held overnight (~17 h); then heated in Easy 180H ultrasonic bath (Elma, Germany) at 80 °C for 35 min; cooled to room temperature; volume adjusted to 20 mL with ultrapure water (Gehaka reverse-osmosis system, > 18.2 MΩ·cm); filtered through 0.45 µm PTFE filter (Agilent Technologies, Tokyo, Japan). All mineralizations performed in triplicate including analytical blank.

Analytical control: certified reference materials (INCT-TL-1 Tea leaves, Institute of Nuclear Chemistry and Technology, Warszawa, Poland; ERM-DB 151 skimmed milk powder, Joint Research Center, Geel, Belgium; Tort-3 Lobster Hepatopancreas, National Research Council, Ottawa, Canada). CRM coverage by element: tea leaves for Al, Ni, Cd, Sb, Ba, Pb; skimmed milk powder for Mo, Hg, Pb; Lobster Hepatopancreas for Co and As. A previously analyzed plant-based yogurt sample (sample F, lot 1) was used for method-validation precision repeatability. Standards: multi-element 100 mg/L for Al/Cr/Co/Ni/As/Mo/Cd/Ba/Pb; Sb 1000 mg/L; Hg 100 mg/L; all from Specsol (Quimlab, Jacareí, Brazil). Calibration curves five-point, 0.1 to 100 µg/L for non-Al elements and 5 to 100 µg/L for Al. Internal standard solutions (Sc, Ge, In, Rh, Bi, Pt at 1000 mg/L stock, Fluka, Steinheim, Germany) diluted to 50 µg/L working concentration.

Statistical analysis: results reported as mean ± standard deviation of n = 3 independent analytical replicates per sample-lot. One-way ANOVA with Tukey’s post-hoc test (p < 0.05) for among-lot comparisons within each sample, using Statistic 7.0 (StatSoft, Tulsa, OK, USA). Multivariate Principal Component Analysis (PCA) using Pirouette 3.11 (Infometrix Inc., Bothell, WA, USA) on a 44 × 9 matrix (samples × elements, excluding Sb and Hg which were < LOQ throughout).

EDI calculation: deterministic model from Kroes et al. 2002 using maximum measured concentration per element multiplied by package mass (range 90 to 250 g across the product set, single unit/day consumption), divided by reference body weight (15 kg child, 60 kg adult). Reference values: JECFA PTWI for Al (2 mg/kg bw/week) and Cd (PTMI 25 µg/kg bw/month); EFSA TDI for Ni (13 µg/kg bw/day) and BMDL0.1 for Pb (12 µg/kg bw/day); JECFA BMDL0.5 for As (3.0 µg/kg bw/day); EFSA AI for Mo (65 µg/day adult; 20 µg/day child 4–6 y; 10 µg/day child 7–11 m); WHO adequate-characterization value for Ba (20 µg/kg bw/day).

Limitations explicitly named by the authors: at the time of study, no Brazilian or international regulation established maximum tolerable limits for inorganic contaminants in plant-based foods (the Brazilian Resolution 722/22 and Normative Instruction IN 160/22 cover general food categories but not plant-based dairy alternatives); large variation observed across lots of the same brand-product within the August–October 2022 sampling window; no chemical speciation performed; no formal hypothesis testing across brands or formulations beyond per-sample-lot ANOVA. The Conflicts of Interest statement declares none.

Implications

Certification: This is a Brazilian retail-market survey covering plant-based yogurt — a product category for which no harmonized regulatory maximum limits currently exist (the authors specifically note this gap with respect to Brazilian Resolution 722/22 and Normative Instruction IN 160/22). It contributes direct occurrence evidence for Al, Cr, Co, Ni, Mo, Cd, Ba, and Pb in coconut-cream-based, coconut-pulp-based fermented, pea/soy-protein-isolate veg-protein, and soy-chocolate plant-yogurt formulations. The two most exposure-relevant findings are (a) coconut-milk-powder reconstituted yogurt at 491–700 µg/kg Ni, which translates to 46.7% of the EFSA TDI for nickel in a 15 kg child at one unit/day and 93.4% at two units/day, and (b) soy-chocolate formulations at 9019 µg/kg Al and 19.85–21.58 µg/kg Pb. The single cow-milk-yogurt comparator was < LOQ for all heavy metals except Mo and Ba. Per-brand replication is small (1 to 3 lots per product across 5 brands and 17 flavors); these values should be treated as range-of-product-evidence rather than population estimators.

App: Route to coconut, soy, soy-protein-isolate, and cocoa ingredient pages, and to aluminum, chromium, cobalt, nickel, arsenic-total, molybdenum, cadmium, barium, and lead metal pages. The matrices array includes yogurt to surface this paper to category-level synthesis once a plant-yogurt product slug is locked. Downstream synthesis on this source should preserve formulation-level distinctions (coconut-cream vs coconut-milk-powder vs pea/soy-protein-isolate vs soy-chocolate) because the contamination profile differs systematically across these formulation classes.

Courses: Useful for illustrating (a) the regulatory gap between rapidly growing plant-based-dairy product categories and existing inorganic-contaminant maximum tolerable limits, (b) the role of formulation choices (cocoa powder, sugar syrup, vegetable fat, soy protein isolate, coconut milk powder) in driving heavy-metal contamination signal beyond the underlying ingredient matrix, (c) the differential element-detection patterns between plant- and animal-based yogurts (Al/Cr/Co/Ni/Mo/Ba/Pb in plant products versus only Mo and Ba in cow-milk yogurt), and (d) the practical application of the deterministic single-unit-per-day EDI model against PTWI, TDI, PTMI, BMDL, and AI reference frameworks in a single product-category survey.

Wiki pages this source may touch

• coconut
• soy
• soy-protein-isolate
• cocoa
• milk-and-dairy
• aluminum
• chromium
• cobalt
• nickel
• arsenic-total
• molybdenum
• cadmium
• barium
• lead

Verification notes

Brand and product identifiers: the source uses coded brand identifiers (IVV for 8 product samples A–H; IFR for 4 product samples I–L; IBT for 1 product sample M; IMD for 2 product samples N, O; IPV for 2 product samples P, Q; INB for the 1 cow-milk comparator sample R) rather than full commercial brand names. These coded designations appear to be brand abbreviations rather than the manufacturers’ full commercial names. Per Part 12 strict reading (locked 2026-05-17), brand codes are stripped from this page’s body, Key numbers, and tables — they are author-supplied brand identifiers, and attaching them to per-sample contamination values constitutes brand attribution. Sample letter (A–R) plus lot number (1–3) uniquely disambiguate every replicate in Table 3 without recourse to the brand code. Product-form descriptors (coconut-cream-based, coconut-milk-powder reconstituted, soy-chocolate, pea/soy-protein-isolate veg-protein, berry fermented coconut cream) carry the formulation distinctions the synthesis layer needs. The brand-code-to-sample-group mapping is recorded here in Verification notes for traceability only.

Speciation: no chemical speciation was performed by the authors. Arsenic is reported as total As; mercury as total Hg; chromium as total Cr (no hexavalent speciation). The metals: frontmatter therefore uses tAs, tHg, and Cr (not iAs, MeHg, Cr-VI). All 44 sample-lots were < LOQ for Hg and Sb.

Sample-lot replicate structure: the source reports 1 to 3 lots per sample-product across 17 sample-products and 1 cow-milk comparator. Triplicate analytical replicates per sample-lot. The Table 3 entries are mean ± SD across the 3 analytical replicates within each sample-lot. The “n = 43 plant-based” and “n = 1 animal-based” in the abstract refer to sample-lots (44 total in the lot count, since some sample-products have only 1 lot purchased). The 44 × 9 PCA matrix in Section 3.3 confirms this count.

Product taxonomy gap: no products/plant-yogurt.md, products/dairy-alternative-yogurt.md, or equivalent slug exists in the current product taxonomy snapshot. The products: frontmatter array is therefore [] and the matrices array carries yogurt to surface this paper to downstream category routing once a plant-yogurt product slug is locked by Karen’s Step 0 Lock workflow. The Wiki-pages-this-source-may-touch list routes by ingredient and metal pages only.

Ingredient taxonomy gap: pea protein isolate, modified starch, agar agar, potato starch, guar gum, xanthan gum, lactic acid bacteria, Streptococcus salivarius / Bifidobacterium animalis fermentation cultures, sugar syrup, and vegetable fat — all named in the source’s Table 2 ingredient lists — do not have dedicated ingredient slugs in the current taxonomy. The ingredients: frontmatter array routes only to the four ingredient slugs that exist in the current taxonomy and are clearly implicated in the contamination signal (coconut, soy, soy-protein-isolate, cocoa). Milk-and-dairy is included to route the cow-milk-yogurt comparator data.

Aluminum highest-concentration sample: 9019.05 ± 759.89 µg/kg in sample M (soy-chocolate, lot 3) per Table 3, p. 7. The Section 3.2.1 narrative reports this same value (p. 8) and identifies sample C lot 3 (veg-protein strawberry, 5844.72 µg/kg) as the second-highest. Both narrative and table agree.

Nickel highest-concentration samples: 700.46 ± 23.50 µg/kg in sample Q (coconut yogurt, lot 1) and 491.08 ± 40.44 µg/kg in sample P (red-fruit coconut yogurt, lot 1) per Table 3, p. 7. The Section 3.2.4 narrative (p. 9) reports the same two values and attributes them to coconut milk powder requiring reconstitution. Both narrative and table agree.

Lead highest-concentration samples: sample M (soy-chocolate) lots 1 and 2 at 19.85 ± 3.93 and 21.58 ± 5.39 µg/kg respectively per Table 3, p. 7. Section 3.2.10 narrative (p. 12) reports range “4.39 to 21.58 µg/kg” across samples D, E, J, M, P that exceeded LOQ, consistent with the per-sample-lot values 4.39 (D lot 3), 5.74 (E lot 3), 6.07 (J lot 1), 19.85 (M lot 1), 21.58 (M lot 2), 4.33 (P lot 1). Table and narrative agree.

Element conversion / unit notes: all source values reported in µg/kg of product as commercialized (wet weight). No conversions were applied on this page. The JECFA Al PTWI reference value of “2 mg/kg bw” stated in the source narrative (Section 3.2.1, p. 8) is reproduced as the source states it; the value is the JECFA-established PTWI of 2 mg/kg bw per week per JECFA 2011 (source reference [35]). The EFSA Ni TDI of 13 µg/kg bw/day (Section 3.2.4, p. 9) reflects the EFSA CONTAM 2020 updated risk assessment (source reference [40]).

Cow-milk comparator: a single sample-lot (sample R, lot 1, “Natural yogurt”, described in Table 2 as “Whole milk and/or reconstituted whole milk, skim milk powder, and lactic acid starter”). Used by the authors solely as a reference for comparison with the plant-based samples. The single-sample replicate (n = 1 lot, n = 3 analytical replicates) does not support any population-level inference for cow-milk yogurt; downstream synthesis on the cow-milk-yogurt data point should treat this as a single-product anchor rather than as a sample-set.

License confirmed CC-BY 4.0 from source p. 1 (Copyright notice: ”© 2023 by the authors. Licensee MDPI, Basel, Switzerland.”). DOI 10.3390/ijerph20043707 resolves to Int. J. Environ. Res. Public Health 2023, 20, 3707; received 30 January 2023, revised 17 February 2023, accepted 17 February 2023, published 19 February 2023. Funding: FAPESP (São Paulo Research Foundation) grants 2022/07015-2 and 2017/50349-0 and CNPq grants 407080/2021-0 and 306054/2020-5. Conflicts of Interest: none declared. Data Availability: “available upon reasonable request.”

Audit subagent (2026-06-08, agent a12b08e73c3132bbc) flagged four findings. (1) ❌ Mo highest-sample misattribution: page reported “355.70 ± 3.70 (sample A, veg-protein cookies-and-cream, lot 2)” but Table 3 shows sample A lot 2 = 348.39 ± 8.47 and the 355.70 ± 3.66 value is in sample B (veg-protein peanut butter) lot 1. Verified against PDF Table 3, p. 6 — correct attribution is sample B lot 1, SD 3.66. Corrected. (2) ❌ Mo IVV-cohort range misattribution: page reported “highest in IVV-brand veg-protein samples A, B, C (190.39 to 355.70 µg/kg)” but Table 3 shows samples A/B/C lots span 293.41 (C lot 1) to 355.70 (B lot 1); the 190.39 µg/kg floor is sample M lot 1 (soy-chocolate), not A/B/C. The source narrative (p. 10, Section 3.2.6) describes the COMBINED A/B/C and M cohort as ranging 190.39 to 355.70 µg/kg. Corrected to show A/B/C range (293.41–355.70) and M range (190.39–207.61) separately, with a note on the source’s combined-cohort range. (3) ⚠️ Part 12 strict reading on brand codes (IVV/IFR/IBT/IMD/IPV/INB): audit argued sample letter + lot number already disambiguate replicates, so attaching brand codes to per-sample contamination values constitutes brand attribution. Verified the strict-reading argument is correct. Brand codes stripped from Key numbers, narrative, and per-element Verification-note tabulations; the brand-code-to-sample-group mapping is retained in a single traceability paragraph above. (4) ⚠️ Matrices vocabulary: audit flagged plant-based-drink and dairy-alternative as non-canonical (yogurt is the load-bearing matrix). Verified against the system-prompt matrices vocabulary — yogurt is the canonical entry; the other two are not. Removed; matrices now [yogurt]. All other audit checks (slug vocabulary for ingredient/metal wikilinks, speciation discipline tAs/tHg/Cr, ICP-MS methods fidelity, Part 2 wiki/HMTc firewall) returned ✅ clean. Verdict REVISE with the four corrections above applied.

Page history

The five most recent substantive edits to this page. The full version history lives in git; when DOI minting comes online (see schema docs), each entry below will also link to a version-pinned DataCite DOI.