The extraction protocols were based on available literature for apple (Malus domestica) (11–14), carrot (Daucus carota) (15–18), and soybean (Glycine max) (19–21). We tested different tissue processing methods and various extraction buffers (Table 1), explained below.

For apples, small apple pieces (with peel, freshly bought from the store or stored at 4°C for 4 days, Jonagold variety) were brought into solution and mixed with extraction buffers #1, #2, and #3, and for some with the addition of protease inhibitors (Supplementary Figure S1A). The first method for carrots was to cut them into small pieces and grind them with a coffee grinder. The second method was to put small carrot pieces into a solution and blend them with a mixer. The third method was to snap freeze the carrot pieces with liquid nitrogen and crush them into powder with a pestle and mortar (Supplementary Figure S1B). Extraction buffers #1, #2, and #3 were used, with the addition of protease inhibitors. For soybeans, sample processing was done by mixing dry soybeans into powder and bringing them into solution (method 1, Supplementary Figure S1C) or by soaking dry soybeans in the extraction buffer and mixing (method 2, Supplementary Figure S1D), both with extraction buffer #1 and #2.

Protein concentration was determined via the Bradford (B6916, Sigma-Aldrich, Saint Louis, Missouri, United States) and bicinchoninic acid (BCA) assay (23227, Thermo Scientific, Waltham, Massachusetts, United States). Four replicate measures were performed, outliers were removed, and the average was used to determine the protein concentration (µg/ml). Protein concentration was used for normalization in further experiments. Aliquots were stored at −20 °C until further use.

Protein extract separation was done under reducing conditions using the MES SDS running buffer (NuPage, Invitrogen, Thermo Fisher Scientific). Separation was performed on a 10% Bis-Tris gel (NuPAGE, Invitrogen) at 160 V for 1 h. For protein visualization, a silver staining (Pierce Silver Stain kit, Thermo Fisher Scientific) was performed according to the manufacturer's protocol. The molecular weight (MW) of the proteins was estimated by comparison with an established protein marker (SeeBlue Plus2 Prestained Standard, Invitrogen).

Following SDS-PAGE, proteins were transferred by electrophoretic transfer at 30 V for 1 h20 onto a polyvinylidene difluoride (PVDF) membrane (Immobilon-P, Merck, Darmstadt, Germany). Transfer buffer (Table 2) was used to facilitate protein blotting. Remaining free binding spots were blocked with PBST-5% non-fat dry milk (NFDM) for 1 h at RT. After washing 4 times (5 min each) with PBST, the membrane was incubated with patient serum (1/8 dilution) overnight at 4 °C. Next, after washing again 4 times (5 min each) with PBST, the membrane was incubated with mouse anti-human IgE (GTX27382, GeneTex, Irvine, California, United States) (1/1,000 dilution) for 1 h at RT. Followed by again washing 4 times (5 min each) with PBST, and incubation with secondary horseradish peroxidase (HRP)-labeled goat anti-mouse IgG (1/20,000 dilution) for 2 h at RT. Finally, after the last wash (4 times, 5 min each, with PBST), protein bands were detected with chemiluminescence (Western Lightning Plus ECL-kit, Perkin Elmer, Waltham, Massachusetts, United States) and visualized with ImageQuant LAS 500 (GE Healthcare Life Sciences, Chicago, Illinois, United States) or the Odyssey XF Imager (LICORbio, Lincoln, Nebraska, United States).

The basophil activation test (BAT) was performed according to our in-house protocol as described previously (22). Peripheral blood was collected in Lithium Heparin tubes (BD Vacutainer). Fresh blood samples were stimulated with various concentrations of our protein extracts diluted in basophil stimulation buffer (BSB; 20 mM Hepes, 133 mM NaCl, 5 mM KCl, 7.5 mM CaCl2, 3.5 mM MgCl2, 1 mg/ml HAS, 0.5 mM Glucose, pH 7.4, supplemented with 60 ng/ml IL-3) for 25 min at 37 °C. For the boiled conditions, samples were put in boiling water for 30 min before basophil stimulation. Positive controls included anti-human IgE (50 ng/ml in BSB, Sigma-Aldrich, Saint Louis, Missouri, United States) and N-Formylmethionine-leucyl-phenylalanine (fMLP, 40 nM in BSB, Sigma-Aldrich). BSB was used as a negative control. Cells were stained with Phycoerythrin (PE) anti-human CD123 (Clone 6H6, 5 µg/ml), Alexa Fluor (AF) 647 anti-human HLA-DR (Clone L243, 5 µg/ml), Fluorescein isothiocyanate (FITC) anti-human CD63 (Clone H5C6, 20 µg/ml) (all BioLegend, San Diego, California, United States). Basophils were gated as CD123⁺/HLA-DR⁻ cells, and CD63 expression was used as a measure for basophil degranulation or activation. Samples were acquired using the LSR Fortessa flow cytometer (BD Biosciences, Franklin Lakes, New Jersey, United States) and analyzed with FlowJo v10.8.1 (FlowJo LLC, BD Biosciences).

Soybean plants (Glycine max, commercial variety “Tasso”) were grown in a commercially available seedling substrate (DCM product code 1004525, <1 kg NPK/m³), characterized by a low nutrient content to minimize salt stress and promote root development in early growth stages. The plants were cultivated under controlled greenhouse conditions and were subjected to different fertilization treatments (NPSOY) starting 2 weeks after sowing. A total of four conditions were established, with eight plants per condition. The unfertilized control condition (1) only received tap water, the moderately fertilized condition (2) was treated with tomato fertilizer, the highly fertilized condition (3) was treated with tomato fertilizer with additional NH₄NO₃, and the very highly fertilized condition (4) was additionally treated with extra NPK (nitrogen, phosphorus, potassium) granules (Supplementary Table S1). These conditions corresponded to estimated nitrogen input levels of 0 kg N/ha (Condition 1), 25 kg N/ha (Condition 2), 50 kg N/ha (Condition 3), and 75 kg N/ha (Condition 4), respectively. For reference, typical nitrogen application rates in soybean cultivation range between 25 and 50 kg N/ha under field conditions. Beans were harvested, and a selection of 6 samples (3 from unfertilized and 3 from fertilized plants) was extracted and normalized for protein concentration (Bradford). This selection was based on sample volume and contained representative samples from the unfertilized an fertilized conditions. This selection (n = 6) was further used to determine the allergenicity in birch pollen-allergic adult patients suffering from PFAS (n = 5) via basophil activation testing.

Birch pollen-allergic adult patients suffering from PFAS were included in our analysis. Patients were recruited during routine consultations at the Allergy Department (UZ Leuven). Ethical approval for this study was provided by the Ethical Committee Research of UZ/KU Leuven (S65184), and informed consent was obtained from all patients. Allergies were confirmed via routine skin prick testing and allergen-specific IgE measurements (Supplementary Table S2). Additionally, a validated questionnaire was used for the diagnosis of PFAS (23). Whole blood samples were collected for basophil activation testing, and serum for further immunoreactivity testing. Two subjects were included for the optimization experiments (NP-opt), and five patients were included for the soybean greenhouse experiment (Supplementary Table S2). A negative control, not suffering from any allergies, was included to control for background reactivity of our extracts (Supplementary Figure S2).

The study involving human participants was conducted in accordance with the Declaration of Helsinki. The study was reviewed and approved by the Ethical Committee Research of UZ/KU Leuven (protocol code S65184 and date of approval: 25/03/2022). The participants provided their written informed consent to participate in this study.

No statistical testing was performed for the optimization experiments. For the greenhouse pilot experiment, normality of the data was evaluated using the Shapiro–Wilk test. Differences between multiple groups were detected via one-way ANOVA, differences between two groups via (un)paired t-testing or its non-parametric alternative.

The protein extraction protocols for apples were compared for three different extraction buffers and the use of protease inhibitors. Only two out of the five extracts yielded a protein concentration just above the detection limit (Figure 1A, APL2 and APL4). APL4 (106 µg/ml) and APL5 (<100 µg/ml) were used for further optimization as these conditions were most consistently supported in literature. SDS-PAGE analysis (Figure 1B) revealed a limited protein profile including a protein band around 17 kDa, corresponding to the Bet v 1 homologue Mal d 1, a band around 28 kDa, although slightly different, possibly corresponding to the 23 kDa Thaumatin-like protein Mal d 2, and a band around 9 kDa corresponding to the nsLTP protein Mal d 3. Immunoblot analysis with serum of a BP-allergic individual suffering from oral allergy symptoms when eating raw apples was performed (Figure 1C). However, no IgE-reactivity was detected towards the expected Mal d 1 protein. Basophil activation testing (Figures 1D,E) revealed similar reactivity towards both apple extracts, and boiling diminished the reactivity more strongly for the APL5 extract, suggesting a relevant contribution of heat-labile proteins (e.g., Mal d 1).

The protein extraction protocol for carrots was compared for three different extraction buffers and the use of protease inhibitors (Figure 2A). All extracts yielded a similar protein concentration between 200 and 400 µg/ml. The CAR5 extract, made by just blending carrots in PBS, yielded the highest protein concentration of 509 µg/ml. SDS-PAGE analysis (Figure 2B) revealed a diverse protein profile, including a protein band around 17 kDa, corresponding to the Bet v 1 homologue Dau c 1, a band around 28 kDa, although slightly different, possibly corresponding to the 33 kDa Isoflavone reductase-like protein Dau c 5. In some extracts (i.e., CAR1, CAR5, CAR7, CAR9, and CAR10), a band around 14 kDa is visible, which corresponds to the profilin protein Dau c 4. Immunoblot analysis (Figure 2C) with serum of a BP-allergic individual suffering from oral allergy symptoms when eating raw carrots revealed IgE-reactivity towards the expected Dau c 1 protein, which was diminished after prolonged boiling of the extract. However, reactivity was rather limited, and also targeted other proteins with no known allergenic relevance. Basophil activation testing (Figures 2D,E) revealed limited variability between the different carrot extracts. Boiling diminished the reactivity of CAR9 and CAR10 and completely reduced CAR11 reactivity, again suggesting a relevant contribution of heat-labile proteins (e.g., Dau c 1).

The protein extraction protocol for soybeans was compared for two processing methods and two extraction buffers (Figure 3A). Dry processing of the soybeans in a coffee grinder till soy flour was practically the most optimal method and yielded a higher protein concentration compared to the wet processing method. SDS-PAGE analysis (Figure 3B) revealed an extensive protein profile including a protein band around 17 kDa, corresponding to the Bet v 1 homologue Gly m 4. Additional proteins include Gly m 1 or Gly m 2, around 7 or 8 kDa, the profilin protein Gly m 3, around 14 kDa, the Gly m 5 and Gly m 6 subunits between 20 and 65 kDa, Gly m 7, around 76.2 kDa, and Gly m 8, around 28 kDa. Immunoblot analysis (Figure 3C) with serum of a BP-allergic individual suffering from oral allergy symptoms when eating soybeans was performed. However, only limited IgE-reactivity was detected towards the expected Gly m 4 protein, but this reactivity was diminished after prolonged boiling of the extract. Basophil activation testing (Figures 3D,E) revealed similar reactivity towards the three tested soybean extracts, and boiling diminished the reactivity of all three extracts, again suggesting a high contribution of heat-labile proteins (e.g., Gly m 4).

Soybean plants were cultivated under different fertilization treatments. Soybeans were collected, and we saw no differences in the amount of harvested soybeans per condition (Supplementary Figure S3). A selection of six beans (including 3 from the unfertilized control condition and 3 from the fertilized conditions) was used for further testing. We observed a visual difference in the beans between the two fertilization conditions. Beans from the unfertilized (U) condition were bigger, while beans from the fertilized (F) condition appeared smaller and shriveled (Figure 4A). The weight per bean was significantly lower for the beans from the fertilized conditions (Figure 4C, p = 0.0013). Soybean extracts were made, and total soluble protein concentration was determined. Protein concentration (Bradford) was not significantly different between the two conditions, but trended towards higher values in the fertilized beans, despite their smaller size (Figure 4D, p = 0.1211). On SDS-PAGE, we observed a difference in the protein profile of our soybean extracts (Figure 4B), where the unfertilized condition showed some additional protein bands, around 30 kDa and 8 kDa, compared to the fertilized condition. Five patients were included for basophil activation testing. We measured lower basophil reactivity or sensitivity towards the beans from the fertilized conditions (Figures 4F: AUC p = 0.049, and 4G: EC50 p = 0.012).