Are the elemental fingerprints of organic and conventional food different? ED-XRF as screening technique.

Research has been conducted the last years to assess whether organically grown food is chemically different from produce of conventional agriculture and which markers are appropriate to discriminate between them. Most articles focus on one single food commodity, produced under strict controlled organic farming conditions, leaving open the question whether the difference would be seen when applied to the same commodity under different growing conditions. In this work 118 organic and 151 conventional samples of commercially available paprika powder, cinnamon, coffee, tea, chocolate, rice, wheat flour, cane sugar, coconut water, honey and bovine milk were characterised for their elemental composition using energy dispersive X-ray fluorescence. Resulting profiles were analysed using univariate and multivariate statistical techniques. Organic samples of a given commodity clustered together and were separated from their conventional counterparts. Differences in the elemental composition of food, could be used to develop statistical models for verifying the agronomical production system.

Introduction

Over the last years pan class="Chemical">consumer demand for organic food increased remarkably. The higher price consumers have to pay for organic products makes them an attractive target for fraudsters.

Regulation (EU) 2018/848 of the European Parliament and of the pan class="Chemical">Council, (Regulation (EU), 2018/84), makes provisions about production and labelling of organic food within the European Union, and refers to controls to be carried out to ensure that a product marked with the organic EU-logo respects the rules for organic agriculture. Current control systems refer to surveillance of production, preparation and distribution of the final organic product, but no reference is made in that legislation to analytical controls based on monitoring of some selected marker compounds. Nevertheless, as recently reviewed by Dias de Lima & Barbosa (Dias de Lima and Barbosa, 2019), many works have been published on the authentication of food grown under organic and conventional farming systems, most of them involving some form of chemometrics and data mining algorithms.

Analysis of pesticides and antibiotics, the use of which is forbidden in organic agriculture, is a way to differentiate organic and conventional food, and has been recently applied to different matrices such as honey (Lazarus et al., 2021) and spices (Maestroni et al., 2021). However, this approach requires different analytical methods for different food commodities because the list of forbidden substances varies from one commodity to another. Most important is the fact that fraudsters go frequently ahead of legislation, using new pesticides and antibiotics for which no limits exist, and which could eventually be more harmful than those known. The mentioned problems can be avoided analysing substances inherent to food, rather than limiting the analysis to forbidden compounds.

The approaches used so far for authentication purical">poses can be divided in targeted and untargeted analysis. Targeted analysis addresses the determination of specific substances in food and are mostly represented by analysis of stable isotopes by Isotope Ratio Mass Spectrometry (IRMS) (Liu et al., 2020), by multi-elemental analysis frequently carried out with ICP-based methods (Junior et al., 2018) and by chromatographic methods for the determination of metabolites (Eibler et al., 2018), or by combination of several of them. Untargeted analysis aims at analysing, in a holistic approach, all substances, some known and some unknown, in a certain sample. Untargeted analysis has become very popular in some analytical areas such as metabolomics, and is frequently carried out using different spectroscopic techniques such as Nuclear Magnetic Resonance (NMR) (Consonni et al., 2018; Mazzei and Piccolo, 2018) and Mid- and Near-Infra Red Spectroscopy (MIR and NIR, respectively) (Liu et al., 2018; Bizerra Brito et al., 2015), and chromatography-based methods. The advantages and disadvantages of non-targeted methods for food control have been discussed in detail by Esslinger, Riedl, & Fauhl-Hassek (Esslinger et al., 2014). The main advantage is that non-targeted analysis provide a holistic view of food composition that can highlight both, fraud related issues and food safety problems. The main disadvantage is related to the lack of standardisation, both from the chemical and statistical point of view, which can jeopardise the reproducibility of the results.

Elemental analysis has been used alone or in combination with other approaches to discriminate organic from conventional food in a range of food commodities such as, rice (Liu et al., 2020), cane sugar (Barbosa et al., 2015), eggs (Giannenas et al., 2009) and chocolate (Junior et al., 2018). It is not our purpose to make a thorough review of all works published in this field, since in depth reviews on the use of elemental profiles to classify organic food can be found elsewhere (Laursen, K.H., Schjoerring, J.K., Kelly, S.D. & Husted, S., (2014); Dias de Lima and Barbosa, 2019), including discussions about advantages and disadvantages of the different analytical approaches.

Interestingly, the use of X-Ray Fluorescence is only cited in one of the two mentioned reviews, as having been used for the classification of organic wheat and apples. However, Energy Dispersive-X Ray Fluorescence (ED-XRF) can be particularly suitable for its use in control laboratories, because it allows multi-elemental analysis with hardly any sample treatment. Milling of solid samples and preparation of pellets, is all that is required. Liquid samples such as honey, can be directly analysed without any pre-treatment at all. Hand-held and bench-top ED-XRF instruments, which can be used in situ, are available at a much lower investment costs than ICP-based instruments. The main drawback of ED-XRF is that its limits of quantification (LOQs) are in general higher than those of ICP-MS and of other ICP-based techniques. They can be too high for the analysis of some trace elements such as Cd, Pb or As. On the other hand, analysis of Cl and Br that can be cumbersome when sample digestion is required (Pereira, Enders, Iop, Mello & Flores, 2018), is straightforward by ED-XRF even in samples with low elemental content such as honey (Fiamegos et al., 2020).

Studies published so far focused mostly on one or two food commodities cultivated in a specific geographical zone, belonging to one single botanical variety and if possible cultivated by one single producer, to avoid the variability introduced by sources other than the organic/conventional nature of the product (Liu et al., 2020; Junior et al., 2018; Mazzei and Piccolo, 2018; Barbosa et al., 2016). In other works commercially available samples purchased at retailers were used to reflect the situation of the market (Liu et al., 2018; Consonni et al., 2018; Eibler et al., 2018).

In this work, eleven different food pan class="Chemical">commodities (paprika powder, cinnamon, coffee, chocolate, tea, wheat flour, rice, cane sugar, honey, cow milk and coconut water) were studied and only commercially available samples were analysed. Samples of different brands were purchased at retailers in different Member States of the EU, covering different geographical origins, cultivars, manufacturing processes, transport and storage conditions, to include as many sources of variation as possible.

The aim of this work was to elucidate if despite all the mentioned sources of variability, organic food has an elemental profile different from that of conventional agriculture. Mass fractions of Mg, P, Cl, S, K, Ca, Cr, Mn, Ni, Cu, Zn, Br, Rb, Sr and Ba were determined by ED-XRF, and evaluated by multivariate analysis to combine and concentrate the information extracted from all the evaluated variables (elements) in some few components. The study did not try to establish predictive classification rules for the individual commodities but to demonstrate empirically that organic farming leaves an imprint on the product’s elemental profiles.

Experimental

Samples

In order to include as much variability as possible in both the organic and conventional products, all samples included in the study are commercially available, purchased at retailers in different countries and covering as many different brands as possible. Samples were purchased during the period 2016–2020 to include different production campaigns and manufacturing lots in case products from the same brand were analysed more than once. All information available on the labels of the analysed samples is included in Table 1.

Table 1

Label information about the analysed samples and mass fractions (expressed in mg kg−1) obtained by ED-XRF.

ID	BIO/NON BIO	Category	Origin	Continent	Mg	P	Cl	S	K	Ca	Cr	Mn	Fe	Ni	Cu	Zn	Br	Rb	Sr	Ba
LoQ (mg/kg)					1450	171.0	78.00	700.0	566.0	118.4	1.89	2.55	4.60	0.16	1.20	5.80	1.70	4.20	1.19	2.40
U(%)(k = 2)					13.0	6.0	2.0	10.0	3.0	3.5	20.0	11.0	6.5	25.0	10.5	6.5	22.0	5.0	8.0	18.0
Paprika powder
Organic
PAPR0012	BIO	NA	Spain	Europe	1766	3205	5963	3515	35,230	2560		12.85	195.6	0.68	9.29	34.33	10.02	9.49	20.66	4.43
PAPR0014	BIO	Sweet	Spain	Europe	1700	3387	3037	3082	35,781	2795		16.19	129.0	0.17	7.26	30.69	1.91	14.68	4.17	3.89
PAPR0016	BIO	Sweet smoked	Spain	Europe	1500	3696	7984	3795	36,228	4199		14.23	241.4	0.75	9.20	37.59	16.82	6.16	24.83	3.98
PAPR0046	BIO	Sweet	Spain	Europe	1740	3956	7197	4076	39,316	1939		13.18	146.6	0.57	10.62	38.60	16.18	5.11	24.77
PAPR0052	BIO	NA	Spain	Europe	1851	3794	8792	3789	38,837	2888		17.36	171.7	0.74	10.46	34.99	31.82	5.00	31.06	3.75
PAPR0054	BIO	NA	Spain	Europe	1877	2989	6794	3745	36,116	1688		11.81	107.5	1.08	9.90	30.57	12.41	8.24	21.12	2.56
PAPR0056	BIO	Sweet	Spain	Europe	1987	3431	5936	3169	37,365	1952		12.39	148.1	0.60	9.71	31.02	15.73	6.08	18.54	3.07
PAPR0057	BIO	Sweet	Spain	Europe		3485	3248	2879	34,365	2236		18.32	107.3	0.21	6.61	31.38	4.48	17.85	6.58	3.19
PAPR0013	BIO	Sweet	NA	NA	2222	2725	6331	2875	31,814	3447		13.95	81.43	0.39	7.13	19.66	27.28	5.18	45.81	4.90
PAPR0015	BIO	NA	NA	Europe	1645	3753	6624	3849	39,020	1789		12.12	143.3	0.86	10.24	32.67	18.90	4.26	27.59
PAPR0029	BIO	NA	NA	NA		3579	7941	3847	36,823	2725		13.12	201.4	0.63	10.08	37.39	17.05	4.85	22.12	3.27
PAPR0031	BIO	NA	Hungary	Europe		3290	3237	3509	30,172	2300		9.63	135.8	2.11	8.47	24.67	5.19	18.29	6.55	3.14
PAPR0051	BIO	Sweet	na	Europe	2097	3578	5804	3505	38,838	2371		17.28	176.2	1.03	9.99	28.24	12.68	7.50	25.74	3.41
				Mean(mg kg−1)	1839	3451	6068	3510	36,146	2530		14.03	152.7	0.75	9.15	31.68	14.65	8.67	21.50	3.60
				Median(mg kg−1)	1808	3485	6331	3515	36,228	2371		13.18	146.6	0.68	9.71	31.38	15.73	6.16	22.12	3.41
				Std(mg kg−1)	217.0	340.4	1871	394.1	2789	705.3		2.56	43.94	0.49	1.36	5.34	8.52	5.00	11.24	0.67
				Std (% of the median)	12.0	9.8	29.6	11.2	7.7	29.7		19.4	30.0	71.9	14.0	17.0	54.2	81.2	50.8	19.7
Coventional
PAPR0008	NON BIO	Sweet	Spain	Europe		2610	3046	2126	21,257	2167		14.90	328.0	0.61	6.96	16.72	4.07	15.96	16.24	6.50
PAPR0009	NON BIO	Sweet	Spain	Europe		3819	5092	4257	28,718	1821		18.28	181.8	0.24	8.97	29.59	6.73	14.33	10.57	4.28
PAPR0011	NON BIO	Sweet	Spain	Europe		3116	3575	2658	26,194	1839		14.98	363.2	0.64	7.38	19.07	2.18	16.08	16.88	5.50
PAPR0022	NON BIO	Sweet	Spain	Europe		2786	5308	2831	31,346	2551		35.84	760.4	1.12	9.92	20.64		21.72	25.78	13.70
PAPR0023	NON BIO	Sweet smoked	Spain	Europe		3255	4774	3093	31,919	2336		16.81	438.4	0.83	9.46	20.12		26.84	21.70	7.26
PAPR0025	NON BIO	Sweet	Spain	Europe		2868	4481	2876	29,152	2282		16.98	388.7	0.95	8.39	20.76	7.32	16.37	20.72	7.59
PAPR0053	NON BIO	Sweet	Spain	Europe		3592	3555	2477	22,034	2591		17.74	291.3	0.78	8.90	21.07	3.45	14.74	15.69	5.22
PAPR0004	NON BIO	Sweet	NA	NA		2399	2791	2018	21,456	2199		14.48	282.5	0.45	6.87	16.53	2.02	15.22	16.06	5.42
PAPR0017	NON BIO	Sweet	NA	NA		2738	5250	2969	26,504	4145		19.35	418.9	1.86	8.05	22.21	2.12	23.03	25.48	8.29
PAPR0018	NON BIO	Sweet	NA	NA		2754	2334	2084	24,184	1534		22.46	278.2	0.40	7.70	22.60	4.65	19.42	9.19	7.17
PAPR0019	NON BIO	Sweet	NA	NA		2902	3085	2151	19,225	2350		15.35	304.1	0.50	10.22	22.80		18.34	17.52	6.52
PAPR0020	NON BIO	NA	NA	NA		2732	5541	2954	31,855	4049		25.22	600.5	2.63	8.25	21.68	4.63	20.35	24.02	9.56
PAPR0021	NON BIO	NA	USA	America		3496	5204	3356	34,915	2436		17.58	437.5	0.76	9.39	20.09		26.97	24.21	7.32
PAPR0024	NON BIO	Sweet	Hungary-Spain	Europe		3339	3916	3263	28,944	3272		16.57	273.6	0.74	10.23	21.96	4.32	11.14	15.23	5.03
PAPR0026	NON BIO	NA	NA/USA	NA/USA		2885	4289	2582	27,050	1604		14.97	270.2	0.57	6.56	23.85	26.61	9.64	9.30	4.43
PAPR0027	NON BIO	NA	NA	NA		2832	4091	2413	24,333	2509		15.97	394.4	0.71	8.71	20.60		20.61	18.44	6.50
PAPR0028	NON BIO	Sweet	NA	NA		1896	4365	2008	25,415	1739		11.27	251.9	0.60	8.96	19.48	6.87	22.82	17.68	5.40
PAPR0030	NON BIO	NA	Hungary-Kalocsa	Europe		3077	3383	2970	28,099	2026		13.96	138.4	1.15	10.72	20.21	15.69	8.35	8.28	3.45
PAPR0032	NON BIO	Sweet	NA	Europe		3340	4051	2806	28,341	2710		20.74	531.2	0.82	9.74	22.51	2.55	18.99	20.74	7.87
PAPR0033	NON BIO	Hot	NA	Europe		2961	3566	2506	26,103	1905		15.59	308.9	0.62	8.53	20.90	2.91	17.66	15.01	5.43
PAPR0034	NON BIO	Sweet	NA	Europe		3320	4313	3125	29,330	1903		14.77	275.9	1.36	9.63	24.11	4.76	28.19	14.63	4.45
PAPR0035	NON BIO	Hot	NA	Europe	1559	3048	5043	3028	33,773	2701		17.29	500.7	0.62	9.16	19.96	2.80	29.69	25.93	8.52
PAPR0036	NON BIO	NA	NA	NA	1617	2776	6109	3239	32,690	4917		20.64	679.0	2.74	8.37	18.62	1.86	18.45	37.26	10.42
PAPR0037	NON BIO	Sweet	NA	NA	1877	2602	6136	2849	29,784	4063		16.75	408.7	1.50	7.49	19.00	3.20	19.76	25.27	7.66
PAPR0038	NON BIO	Sweet	NA	Europe		3565	4720	2972	30,876	3364		26.05	512.3	1.16	9.80	21.02		21.07	22.80	8.86
PAPR0039	NON BIO	Sweet	NA	Europe		3547	4987	2896	29,426	3817		30.86	630.0	1.25	9.82	21.60	2.30	17.94	24.63	10.28
PAPR0040	NON BIO	Sweet	NA	Europe		3420	4183	2736	28,064	1939		18.60	415.8	0.84	10.29	20.81		25.04	19.05	6.64
PAPR0041	NON BIO	Smoked	NA	Europe	1589	3095	4250	2870	26,382	3273		21.52	355.9	0.72	8.10	19.42	7.88	23.87	19.92	5.50
PAPR0042	NON BIO	Smoked	Bulgaria	Europe	1615	3699	4511	2959	33,640	1722		14.97	320.3	0.68	9.64	23.37		21.47	15.29	4.65
PAPR0043	NON BIO	Sweet	Hungary	Europe		3443	3898	3230	28,930	2328		16.59	217.8	1.15	9.08	22.06	17.67	14.71	12.89	4.37
PAPR0044	NON BIO	Smoked	NA	Europe	1694	3198	5292	3015	35,374	2392		20.62	525.0	1.45	9.81	41.07	1.77	22.84	18.37	7.13
PAPR0047	NON BIO	Sweet	Israel	Asia		2732	4711	2361	27,968	1879		10.70	237.0	0.39	6.50	14.51	1.70	19.51	20.70	5.28
PAPR0048	NON BIO	Smoked	NA	NA		3172	6251	2793	33,820	2677		13.92	327.2	0.67	8.22	18.70	2.75	30.57	24.15	5.55
PAPR0049	NON BIO	NA	NA/USA	NA/USA	1904	3781	5326	3282	34,447	2207		16.39	374.7	1.07	9.79	20.13	1.98	30.13	22.89	5.02
				Mean(mg kg−1)		3092	4487	2833	28,700	2617		18.38	388.9	0.98	8.89	21.34	4.80	20.45	19.63	6.78
				Median(mg kg−1)		3105	4496	2886	28,937	2371		16.89	369.0	0.80	9.02	20.70	3.33	20.06	19.49	6.51
				Std(mg kg−1)		436.1	975.4	467.3	4224	832.4		5.26	145.4	0.57	1.10	4.43	4.12	5.44	5.84	2.18
				Std (% of the median)		14.0	21.7	16.2	14.6	35.1		31.1	39.4	71.6	12.2	21.4	123.8	27.1	30.0	33.4
Coffee
Organic
C0023	BIO	Unknown	Indian	Asia	2025	1608	324.9	1626	21,088	1425		33.50	38.33	0.63	11.54	8.49		21.40	5.70	4.52
C0024	BIO	Arabica	Mexico	America	2117	1482	249.7	1678	19,486	1520		30.60	35.62		11.19	6.76		32.60	7.82	6.03
C0025	BIO	Unknown	Bolivia	America	1646	1493	307.9	1583	19,836	1427		31.02	39.47	0.43	10.75	8.19		20.30	4.87	3.61
C0026	BIO	Arabica	Guatemala	America	1894	1459	315.5	1575	19,704	1706		35.69	32.86	0.46	11.33	8.88		29.90	3.94	3.11
C0027	BIO	Arabica	Republic of Congo	Africa	2259	1543	386.7	1567	20,792	1165		21.11	60.20	0.21	10.72	7.80		52.06	9.17	6.02
C0028	BIO	Arabica	Ethiopie	Africa	1639	1738	298.3	1584	20,279	1233		16.25	37.09		10.31	6.42		22.62	4.52	4.40
C0029	BIO	Arabica	NA	Non European	1920	1335	287.0	1627	19,788	1650		26.22	35.15		9.92	7.10		18.67	7.52	5.14
C0032	BIO	Arabica	Nicaragua	America	2036	1379	370.7	1718	21,194	1479		22.50	36.55		12.86	10.53		36.36	6.75	5.67
C0033	BIO	Arabica	Guatemala	America	1702	1341	262.0	1561	18,357	1624		40.84	35.36		9.82	7.37		35.78	3.00	3.94
C0039	BIO	Min 90 % Arabica	Tanzania, Peru, Honduras, Congo	Non European	1870	1457	326.9	1508	20,356	1380		36.96	35.86		11.41	8.51		54.12	10.39	6.39
C0041	BIO	Arabica/Robusta	Latin America	America	1715	1708	301.6	1504	20,191	1513		20.01	31.50	1.24	14.96	6.69		42.57	6.92	1.81
C0044	BIO	Arabica	Costa rica	America	1941	1341	231.8	1614	17,855	1114		26.54	36.01	0.17	14.26	7.70		75.35	6.86	3.99
C0045	BIO	Arabica	Bolivia	America	1725	1456	192.8	1486	17,454	1333		46.05	34.84	0.69	12.56	8.68		34.14	6.89	2.53
C0046	BIO	Arabica	Brazil	America	2057	1388	420.8	1669	19,587	1436		21.39	38.80	0.24	14.04	7.60		21.91	2.85	2.48
C0047	BIO	Arabica	Colombia	America	1587	1546	296.7	1425	19,109	1344		40.28	31.10	0.24	12.01	7.82		30.89	18.34	11.57
C0048	BIO	Arabica	Costa rica	America	2251	1284	299.7	1495	18,206	1335		36.40	37.40	0.27	11.99	7.38		30.69	12.67	6.42
C0049	BIO	Arabica	Colombia	America	1648	1526	244.7	1403	17,104	1379		22.11	30.30		10.94	7.52		11.10	5.33	4.42
				Mean(mg kg−1)	1884	1476	301.0	1566	19,434	1415		29.85	36.85	0.46	11.80	7.85		33.56	7.27	4.83
				Median(mg kg−1)	1894	1459	299.7	1575	19,704	1425		30.60	35.86	0.35	11.41	7.70		30.89	6.86	4.42
				Std(mg kg−1)	217.3	128.1	57.07	87.24	1245	161.3		8.67	6.57	0.33	1.51	0.99		15.69	3.83	2.24
				Std (% of the median)	11.5	8.8	19.0	5.5	6.3	11.3		28.3	18.3	93.8	13.2	12.9		50.8	55.8	50.7
Conventional
C0005	NON BIO	Arabica	Ethiopia	Africa		1507	406.5	1470	20,854	1420		15.79	39.48		10.50	6.12	1.74	14.70	2.60	2.40
C0006	NON BIO	Robusta	Uganda	Africa	1900	2037	408.2	1718	24,654	1468		16.41	44.86	0.44	14.36	8.30		40.09	6.91	5.88
C0007	NON BIO	Arabica	South/Central America	America	1858	1585	472.3	1760	22,707	1409		38.71	43.99	0.21	12.04	8.43		29.39	10.89	4.72
C0009	NON BIO	Arabica/Robusta	South America/ Brazil & Guatemala	America	1932	1741	674.6	1874	24,536	1458		20.58	46.54	0.28	12.37	8.23		35.39	16.65	4.38
C0010	NON BIO	Arabica	Indian Malabar/ Latin America	America/Asia	1792	1534	414.8	1660	21,783	1355		36.11	39.06	0.40	11.73	7.95		37.07	13.81	6.01
C0011	NON BIO	Arabica/Robusta	South America/ East Africa	America	2065	1636	443.3	1739	21,811	1335		36.62	40.19	0.55	12.28	8.01		30.12	6.86	4.53
C0013	NON BIO	Arabica	Guatemala/ Antigua	America	1667	1379	344.0	1601	21,463	1748		26.74	44.66	LoQ	11.02	7.34		61.88	8.76	5.48
C0014	NON BIO	Arabica	Indonesia-Sulawesi	Asia	1965	1524	328.8	1560	21,897	1386		25.80	37.90	<LoQ	10.53	9.05		83.62	7.39	5.15
C0016	NON BIO	Unknown	NA	NA	2081	1692	498.5	1734	24,132	1469		33.50	48.49	0.67	11.47	8.70	10.13	44.40	6.04	5.47
C0017	NON BIO	Arabica	Costa Rica	America	2341	1393	366.9	1651	21,307	1365		30.81	46.26	0.19	10.94	7.49		32.05	12.20	6.01
C0018	NON BIO	Unknown	NA	NA	2352	1642	393.0	1732	21,799	1516		46.19	45.37	0.55	12.54	8.42		31.30	6.71	5.18
C0019	NON BIO	Unknown	NA	NA	2229	1593	470.5	1743	22,374	1574		37.60	47.68	0.37	11.80	7.89		29.65	5.37	4.66
C0020	NON BIO	Arabica	Columbia	America	1966	1510	358.0	1668	20,573	1410		54.18	36.85	0.49	10.99	7.82		23.81	9.56	7.78
C0021	NON BIO	arabica	Brazil	America	2256	1507	491.0	1707	22,373	1461		35.18	39.53	0.23	11.93	8.30		30.92	5.08	4.03
C0022	NON BIO	Arabica	Nicaragua	America	2037	1503	342.7	1630	21,236	1447		22.21	37.26		11.93	10.28		32.46	5.84	6.11
C0034	NON BIO	Arabica	Columbia	America	2452	1611	342.2	1612	20,643	1340		38.49	37.81	0.34	12.42	8.11		35.65	15.97	7.82
C0035	NON BIO	Arabica	Ethiopia	Africa	2330	1774	314.0	1591	20,815	1228		17.45	34.48	0.32	10.01	6.59		29.10	4.29	4.22
C0036	NON BIO	Arabica	Indonesia	Asia	1805	1743	380.9	1653	22,726	1411		33.40	37.74	0.17	10.63	8.89		65.62	7.47	5.74
C0037	NON BIO	Arabica	Nicaragua	America	2135	1587	431.4	1732	22,684	1455		33.75	37.36		11.34	8.77		25.06	5.38	6.15
C0038	NON BIO	Arabica/Robusta	India	Asia	1595	1862	456.4	1744	23,840	1290		26.92	40.02	1.89	11.84	8.44		27.06	3.24	3.95
C0040	NON BIO	Arabica	NA	NA	1859	1444	357.5	1523	19,489	1390		41.50	39.74	0.18	11.40	8.63		29.50	9.85	5.58
C0001	NON BIO	Unknown	Colombia	America	1803	1479	364.1	1510	19,998	1372		44.09	38.47	0.80	13.85	7.93		14.62	10.36	9.77
C0002	NON BIO	Arabica	Brazil	America	2060	1428	434.3	1659	22,399	1456		29.98	42.73	0.33	14.05	7.23		26.24	4.73	3.48
C0003	NON BIO	Arabica/Robusta	South India	Asia	1985	1627	348.4	1575	22,033	1277		31.40	40.71	1.15	12.74	6.95		31.25	4.11	4.95
C0004	NON BIO	Arabica	Colombia	America	1790	1411	295.9	1493	19,511	1402		34.81	37.43	0.34	13.74	7.26		36.42	16.38	8.98
C0008	NON BIO	Arabica	South America/ East Africa	America/Africa	1682	1502	307.1	1509	20,425	1280		26.16	38.09	0.28	14.01	7.55		30.88	7.62	5.75
C0012	NON BIO	Arabica	Brazil	America	2697	1269	433.4	1543	19,927	1178		25.58	48.45	0.36	9.92	6.89		25.60	1.81
C0015	NON BIO	Arabica	Ethiopia-Sidamo	Africa	1924	1643	332.8	1580	20,845	1396		17.15	48.35		9.67	21.13		26.58	4.72	5.26
C0043	NON BIO	Arabica	Nicaragua	America	1651	1488	441.9	1597	20,047	1175		24.33	30.20		10.21	7.46		13.08	5.91	4.77
				Mean(mg kg−1)	2007	1574	401.8	1640	21,686	1395		31.08	41.02	0.48	11.80	8.42		33.57	7.81	5.51
				Median(mg kg−1)	1965	1534	393.0	1651	21,783	1402		31.40	39.74	0.35	11.80	8.01		30.88	6.86	5.37
				Std(mg kg−1)	266.3	157.8	77.72	97.65	1427	115.4		9.42	4.63	0.39	1.31	2.58		14.81	4.05	1.58
				Std (% of the median)	13.6	10.3	19.8	5.9	6.5	8.2		30.0	11.7	111.4	11.1	32.3		48.0	59.1	29.4
Tea
Organic
TEA0008	BIO		China	Asia	1560	2492	493.0	2925	17,758	4778		1345	112.1	4.07	9.30	24.77		79.38	14.50	26.07
TEA0009	BIO		Sri Lanka	Asia		2768	941.9	2943	20,576	4570		638.2	96.52	3.78	15.08	26.68	1.71	49.80	32.68	41.26
TEA0010	BIO		India	Asia	1928	2699	505.8	2949	19,550	4801		412.4	136.5	3.44	17.22	35.74	1.92	64.02	21.85	24.46
TEA0011	BIO		NA	Non EU	1923	3368	1532	2623	22,058	3413		809.4	126.9	5.90	15.05	29.14		52.56	6.83	11.13
TEA0012	BIO		Sri Lanka	Asia	2549	2501	968.5	2765	20,295	4213		616.3	111.6	5.40	17.42	25.84	3.14	65.95	26.36	49.40
TEA0013	BIO		China	Asia	1744	2863	624.7	2393	19,694	3547		675.7	76.93	3.06	11.80	26.74		40.78	7.63	12.33
TEA0014	BIO		China	Asia		2181	345.0	2466	12,005	3894		1279	108.3	4.16	17.20	24.10	3.36	101.04	15.66	21.13
TEA0015	BIO		India	Asia	1815	2924	623.1	2598	20,424	3665		408.3	193.9	6.01	21.77	37.28		70.18	19.46	24.33
TEA0016	BIO		India	Asia	1521	3610	753.3	2951	20,937	3521		415.5	129.2	6.02	13.83	37.64		63.03	17.22	24.31
TEA0017	BIO		Vietnam	Asia		2955	624.8	2823	22,245	3573		557.6	118.5	3.61	18.75	43.41	2.57	348.60	16.87	14.56
TEA0018	BIO		India	Asia	2143	2873	744.9	3826	21,871	4698		561.9	224.0	6.32	14.46	28.84		101.06	20.44	38.38
TEA0019	BIO		NA	Asia	2041	2135	952.5	2434	17,413	4460		567.8	111.4	5.32	17.01	23.22	3.77	41.20	30.83	51.35
TEA0020	BIO		Sri Lanka	Asia	1925	2399	1250	2847	20,292	4545		584.3	90.21	4.27	15.92	26.46	3.15	43.59	20.46	32.21
				Mean(mg kg−1)	1915	2751	796.8	2811	19,625	4129		682.4	125.9	4.72	15.75	29.99	2.80	86.25	19.29	28.53
				Median(mg kg−1)	1924	2768	744.9	2823	20,295	4213		584.3	112.1	4.27	15.92	26.74	3.14	64.02	19.46	24.46
				Std(mg kg−1)	296.5	426.1	328.5	367.4	2715	538.8		301.3	40.70	1.14	3.13	6.36	0.76	81.34	7.69	13.24
				Std (% of the median)	15.4	15.4	44.1	13.0	13.4	12.8		51.6	36.3	26.7	19.6	23.8	24.4	127.1	39.5	54.1
Conventional
TEA0001	NON BIO		Sri Lanka	Asia	2635	2842	952.1	2927	20,752	5165		1203	279.3	3.32	11.70	26.88	4.84	70.31	42.63	40.93
TEA0002	NON BIO		NA	NA	2098	2665	1149	2991	23,936	4856		760.7	163.2	5.36	13.40	28.42	100.6	65.55	14.82	33.80
TEA0003	NON BIO		NA	NA	2223	2018	1455	2577	18,283	6251		1176	423.3	5.77	9.90	27.55	6.18	50.62	31.95	37.52
TEA0004	NON BIO		India & Sri Lanka	Asia		2829	918.9	2944	21,450	4375		912	279.2	3.98	13.97	27.40	2.72	83.33	36.50	29.47
TEA0005	NON BIO		NA	NA	2437	2166	988.5	2682	18,499	5631		912.8	259.8	4.50	15.77	28.04	2.92	69.27	28.04	32.56
TEA0006	NON BIO		Kenya	Africa		2548	992.5	2753	20,297	4990		1663	276.8	4.74	10.19	27.32	5.51	116.02	43.06	33.15
TEA0007	NON BIO		Sri Lanka	Asia	1919	2504	1320	2826	19,488	4944		757.7	138.3	3.90	15.28	30.52	2.63	38.56	21.63	28.67
				Mean(mg kg−1)	2262	2510	1111	2814	20,386	5173		1055	260.0	4.51	12.89	28.02	4.13	70.52	31.23	33.73
				Median(mg kg−1)	2223	2548	992.5	2826	20,297	4990		912.8	276.8	4.50	13.40	27.55	3.88	69.27	31.95	33.15
				Std(mg kg−1)	281.1	315.6	206.0	152.0	1944	604.6		321.9	92.98	0.86	2.35	1.21	1.57	24.76	10.57	4.31
				Std (% of the median)	12.6	12.4	20.8	5.4	9.6	12.1		35.3	33.6	19.1	17.5	4.4	40.4	35.7	33.1	13.0
				Mean(mg kg−1)
Rice
Organic
RICE0002	BIO	Basmati	NA	NA		1162	430.4	1374	1039.8			8.47	7.50		1.67	16.00
RICE0015	BIO	Basmati	NA	NA		1178	240.7	1612	795.4			8.75	6.63	0.22	1.83	18.39	3.14			7.66
RICE0016	BIO	Basmati	India	Asia		976.3	407.2	1412	934.7			8.91	5.62		1.58	15.15
RICE0026	BIO	Basmati	NA	Non EU		1433	455.6	1370	1129.1			9.94	5.81		1.24	15.77				6.95
RICE0030	BIO	Basmati	NA (Himalayas)	NA		1277	174.0	1485	781.7			8.70		0.19	1.60	17.69		4.54
RICE0047	BIO	Basmati	NA	NA		1716	491.5	1194	1415.5			10.12	12.28		1.46	16.71
RICE0049	BIO	Basmati	Pakistan	Asia		1047	403.5	1405	1160.9			8.55			1.28	15.17
RICE0054	BIO	Basmati	India	Asia		1277	249.8	1511	797.3			8.79	4.87	0.22	2.41	18.13
RICE0055	BIO	Basmati	Pakistan	Asia		1017	290.5	1234	1213.7			5.57		0.40	1.38	11.31	15.35
				Mean(mg kg−1)		1232	349.2	1400	1029.8			8.64	7.12		1.61	16.04
				Median(mg kg−1)		1178	403.5	1405	1039.8			8.75	6.22		1.58	16.00
				Std(mg kg−1)		232.5	112.0	130.5	220.6			1.30	2.68		0.36	2.15
				Std (% of the median)		19.7	27.7	9.3	21.2			14.8	43.2		22.5	13.5
Conventional
RICE0004	NON BIO	Basmati	India-Pakistan	Asia		1678	216.7	1471	1164	268.7		7.49	14.02		2.01	21.20	44.12			2.46
RICE0029	NON BIO	Basmati	NA	Non EU		1149	535.8	1509	924.8			8.48			2.06	16.79				2.49
RICE0031	NON BIO	Basmati	India-Pakistan	Non EU		1847	507.6	1058	1580	254.2		10.61	9.37		1.44	16.29				33.27
RICE0033	NON BIO	Basmati	NA	NA		1054	516.2	1441	989.5			7.79	7.58	0.20	1.91	16.74
RICE0042	NON BIO	Basmati	NA	NA		1302	255.9	1350	958.1			7.87			1.72	15.11	41.10			2.66
RICE0043	NON BIO	Basmati	India	Asia		1452	463.6	1389	1186			9.77	6.01		1.29	16.06
RICE0044	NON BIO	Basmati	India and Pakistan	Asia		1639	324.6	1358	1185			8.14	5.63		1.60	14.95	4.39			8.66
RICE0048	NON BIO	Basmati	India and Pakistan	NA		1375	331.6	1365	1027	184.1		9.88	6.53		1.30	14.81
				Mean(mg kg−1)		1437	394.0	1367	1127			8.75	8.19		1.67	16.49				9.91
				Median(mg kg−1)		1413	397.6	1377	1096			8.31	7.06		1.66	16.18				2.66
				Std(mg kg−1)		272.3	126.5	137.9	211.1			1.17	3.16		0.31	2.06				13.33
				Std (% of the median)		19.3	31.8	10.0	19.3			14.0	44.7		18.6	12.7				501.0
Wheat flour
Organic
Flour002	BIO	White flour Stone milled (Triticum aestivium)	Spain	EU		1356	828.0	1528	2105	323.1		11.15	17.16		2.02	10.23	9.81		2.00
Flour003	BIO	White flour (Triticum aestivium)	NA	EU/NON EU		1163	599.5	1373	1595	208.2		6.53	13.23		1.62	7.71	2.00
Flour004	BIO	White flour Stone milled (Triticum aestivium)	NA	EU		1235	734.8	1795	1549	265.3		9.56	17.39		1.65	8.20	4.70
Flour005	BIO	White flour Stone milled (Triticum durum)	NA	EU/NON EU		2926	658.8	1721	3941	449.7		15.49	72.39		4.16	44.16			2.54	4.35
Flour012	BIO	White flour (Triticum durum)	NA	EU/NON EU		1432	772.7	1812	1958	1528		9.34	26.73		2.24	13.37	4.78		4.01
Flour014	BIO	White flour (Triticum aestivium)	NA	EU		953.7	543.3	1418	1536	175.6		3.22	12.26				2.62			32.94
Flour016	BIO	White flour (Triticum aestivium)	NA	EU		2169	626.5	1289	2520	241.7		18.91	22.37		2.60	20.43				2.62
Flour019	BIO	White flour Stone milled (Triticum aestivium)	France	EU		1668	611.8	1277	1909	219.0		15.86	18.52		2.59	14.80	4.00
				Mean(mg kg−1)		1613	671.9	1527	2139	426.3		11.26	25.01		2.41	16.99	4.65
				Median(mg kg−1)		1394	642.7	1473	1934	253.5		10.36	17.96		2.24	13.37	4.35
				Std(mg kg−1)		645.3	97.18	222.1	800.6	453.2		5.23	19.70		0.87	12.76	2.76
				Std (% of the median)		46.3	15.1	15.1	41.4	178.8		50.5	109.7		38.7	95.4	63.6
Coventional
Flour007	NON BIO	White flour (Triticum aestivium)	NA	NA		1189	687.9	1338	1870	229.3		6.56	12.00		1.45	7.93
Flour008	NON BIO	White flour (Triticum durum)	Spain	Europe		1509	688.5	1532	1964	240.5		10.29	15.96		1.75	10.06
Flour009	NON BIO	White flour (Triticum durum)	Portugal	Europe		1169	792.9	1654	1706	240.2		5.73	14.43		1.35	7.35
Flour011	NON BIO	White flour (Triticum durum)	UK	Europe		1275	690.7	1702	1841	1064		6.58	17.66		1.49	8.51			1.72	3.35
Flour013	NON BIO	White flour (Triticum aestivium)	France	Europe		1408	657.2	1281	1559	227.7		12.14	18.17		2.01	9.48	2.33		303.76
Flour017	NON BIO	White flour (Triticum aestivium)	Poland	Europe		945.5	744.3	1376	1334	189.0		4.94	9.83			5.83
Flour018	NON BIO	White flour (Triticum aestivium)	Poland	Europe		1051	682.6	1409	1386	190.6		6.30	11.22		1.12	6.51				6.26
Flour020	NON BIO	White flour Stone milled (Triticum aestivium)	Spain	Europe		1454	780.0	1600	2181	270.9		17.09	19.88		2.49	13.33	2.50		2.35	2.87
				Mean(mg kg−1)		1250	715.5	1486	1730	331.6		8.70	14.89		1.67	8.63
				Median(mg kg−1)		1232	689.6	1470	1773	234.7		6.57	15.20		1.49	8.22
				Std(mg kg−1)		199.0	50.0905	156.7	291.4	297.3		4.18	3.63		0.46	2.37
				Std (% of the median)		16.1	7.3	10.7	16.4	126.6		63.7	23.9		31.0	28.8
Sugar
Organic
Sugar004	BIO	Cane	Antilles	America									5.65
Sugar005	BIO	Cane	Antilles	America						140.4			5.21
Sugar012	BIO	Cane	NA	NON EU			146.8			394.7			7.04
Sugar013	BIO	Cane (Rapadura)	NA	NON EU			3227		7719	1666			48.24					13.09	6.57	3.75
Sugar014	BIO	Cane	NA	NON EU					655.3	276.8			14.74
Sugar019	BIO	Cane	NA	NON EU			365.0		1061	229.4			9.71
				Mean(mg kg−1)			1246.3		3145	541.4			15.10
				Median(mg kg−1)			365.0		1061.1	276.8			8.38
				Std(mg kg−1)			1719		3966	635.2			16.61
				Std (% of the median)			470.9		373.8	229.5			198.3
Conventional
Sugar002	NON BIO	Cane	NA	NA						155.6
Sugar006	NON BIO	Cane	NA	NA			176.5			364.7
Sugar020	NON BIO	Cane (Panela)	Latin America	America			1468		3169	1154		3.74	17.67						2.38
Sugar021	NON BIO	Cane	NA	NA
				Mean(mg kg−1)			822.4			558.2
				Median(mg kg−1)			822.4			364.7
				Std(mg kg−1)			913.4			526.7
				Std (% of the median)			111.1			144.4

In total, the study includes 118 orpan class="Chemical">ganic products (17 coffee, 5 coconut water, 10 dark chocolate, 8 wheat flour, 20 honey, 7 bovine milk, 13 paprika powder, 10 cinnamon, 9 basmati rice, 6 cane sugar and 13 tea), and 151 conventional products (29 coffee, 6 coconut water, 13 dark chocolate, 8 wheat flour, 20 honey, 8 bovine milk, 34 paprika powder, 14 cinnamon, 8 basmati rice, 4 cane sugar and 7 tea).

Reagents and standards

The trueness of the method used in the anpan class="Chemical">alysis of liquid samples was evaluated with a multi-elemental stock solution containing Ag, Al, B, Ba, Bi, Ca, Cd, Co, Cr, Cu, Fe, Ga, In, K, Li, Mg, Mn, Na, Ni, Pb, Sr, Tl and Zn at 1000 mg kg−1 each (Merck), and with individual solutions of Cl, P and Rb 1000 mg kg−1 (Merck), that were diluted with deionised water up to a final concentration of 10 mg kg−1. Two chocolate certified reference materials (CRMs), ERM-BD512 (European Commission, JRC) and SRM 2384 (US NIST) were also used to evaluate the trueness of the method when applied to chocolate and honey. Chocolate and honey have a similar density at 40 °C where the ED-XRF measurements took place. The tobacco CRM PVTL-6 (with certified values for all the elements included in the study), regularly used to check the accuracy of the results, was from the Institute of Nuclear Chemistry and Technology, Warsaw, Poland.

CRMs and repan class="Chemical">ference materials (RM) were used in the analysis of solid samples for calibration purposes and to test the accuracy of the results. The CRMs used are listed elsewhere (Fiamegos and de la Calle Guntiñas, 2018).

Blank mepan class="Chemical">asurements in the analysis of liquid (bovine milk and coconut water) and semi-liquid (honey and dark chocolate) samples, were carried out with deionised water obtained with a Milli-Q Plus system (> 18.3 MΩ) (Millipore, Billerica, MA, USA).

CEREOX® wax and the reference samical">ple FLX-S13 were opan class="Chemical">btained from Fluxana GmbH, Bedburg-Hau, Germany.

Instrumentation and sample preparation

An Epsilon 5 ED-XRF spectrometer (PANpan class="Chemical">alytical, Almelo, The Netherlands) was used to determine the elemental mass fractions of Mg, P, Cl, S, K, Ca, Cr, Mn, Ni, Cu, Zn, Br, Rb, Sr and Ba. Unfortunately, nitrogen cannot be analysed by ED-XRF because this technique does not allow the accurate determination of light elements. The first element that can be accurately analysed with the Epsilon 5 instrument is Mg although with high limits of quantification. Solid samples were analysed with a method whose validation and performance characteristics (including uncertainty and limit of quantification) are described elsewhere (Fiamegos et al., 2018) and included in Table 1. Liquid and semi-liquid sample analysis were carried out with the Auto Quantify application of the ε5 software (PANalytical), under a helium atmosphere and using holders for liquid samples, which were sealed with a 6 μm Mylar film giving rise to blank values for P and Ca. De-ionised water was measured ten times, and the mass fractions obtained were recorded. No blank correction was applied to the data used for modelling purposes, because subtraction of a constant number from random variables does not affect co-variances, since the same constant is also subtracted from the mean.

The performance of the ED-XRF spectrometer was controlled and recalibrated every week with the reference sample FLX-S13 to correct for the normal drift. The tobacco CRM PVTL-6 was measured every week after measuring the FLX-S13 standard to check the accuracy of the measurement. No systematic bias was observed during the period of time in which the analyses were carried out.

Solid samples were milled in a Planetary MonoMill-Minimill II of Fritsch-PANalytical, (Almero, The Netherlands) to obtain a fine powder and 40 mm diameter pellets were done using aluminium cup dies and pressing the sample with a semi-automatic press of Hertzog Maschinenfabrik Gmbh (Osnabrück, Germany). For some matrices CEREOX® wax was added to the matrix and carefully mixed with a metal-free spatula before making the pellets, to avoid crumbling during measurements. The wax used had concentrations of the elements studied below the LOQ of the method. Table 2 summarises the conditions used for all the matrices included in this study.

Table 2

Sample treatment for all the matrices included in the study.

Matrix	Sample weight (g)	Milling	Milling speed (rpm)	Milling time (min)	Number of tungsten carbide balls	Wax	Wax weight (g)	Aluminium cap	Press(kilo Newton)	Pressing time (min)
Paprika powder	4	No	–	–	–	Yes	1	Yes	200	3
Cinnamon	6	No	–	–	–	No	–	Yes	250	3.5
Coffee	5	No	–	–	–	No	–	No	200	3.5
Tea	7	Yes	300	4	3	No	–	Yes	250	3.5
Rice	6	Yes	300	5	5	Yes	1	Yes	200	3.5
Wheat flour	6	No	–	–	–	Yes	1	Yes	250	3.5
Cane sugar	5	Yes	300	5	2	Yes	1	Yes	250	3.5

Sample treatment for all the matpan class="Species">rices included in the study.

Bovine milk wpan class="Chemical">as mixed shaking manually before transfer of an aliquot of 10 mL to the sample holder. Honey was homogenised with a metal-free spatula before transferring 10 mL to the sample holder. Chocolate samples were warmed in an oven for 70 °C until melted, then homogenised with a metal-free spatula and then a 10 mL aliquot was transferred to the sample holder.

Data treatment and multivariate analysis

Elemental mass fractions as obtained by the ε5 software for solid samples were used without any correction because no systematic bias was observed during the validation of the method for any of the elements included in the study (Fiamegos et al., 2018). The Auto Quantify application of ε5 system was developed by the manufacturer of the ED-XRF for the analysis of fuel, and is based on the use of fundamental parameters, a mathematical algorithm that incorporates the effect of the matrix in the results. The liquid and semi-liquid matrices included in this study are different than fuel and most results were affected by a constant bias of 50 % as calculated in the analysis of the multi-elemental standard solution and the two chocolate CRMs used to evaluate the trueness of the results.

The software Statistica (TIBpan class="Chemical">CO, Version 13.5.0.17) was used to carry out univariate t-tests (95 % confidence interval), to highlight significant differences in elemental mass fractions between the organic and conventional products for the different food commodities.

Multivariate analyses were carried out with the software SIMCA Version 15.0.2 (Satorius Stedim Biotech AS, Malmö, Sweden) (Eriksson et al., 2013). Principal Component Analysis (PCA) (unsupervised) and Partial Least Square-Discriminant Analysis (PLS-DA) (supervised) were the techniques used to concentrate the information provided by the 16 variables (mass fractions of 16 elements) into fewer components. Both techniques helped to visualise whether the various food commodities studied, form clusters based on their organic/conventional nature. Mass fractions as obtained, without further corrections for bias or blank, were used in multivariate analysis, including those that were below the LOQ.

The number of principal components were always kept to three to avoid overfitting. Permutations tests were run for the PLS-DA models to evaluate if the classification of the samples in the two groups, organic or conventional, is significantly better than any other random classification in two other arbitrary groups (Golland et al., 2005; Mielke and Berry, 2001). Permutations where chosen because the number of observations (samples) was frequently low, which could introduce some problems in the use of cross-validation (Westerhuis et al., 2008).

Bi-plots were pan class="Chemical">constructed to visualise which elements where present in higher concentrations in the organic and conventional groups, respectively, for the different food commodities.

Results and discussion

Table 1 summarises the mass fractions obtained for the different elements analysed in all solid samples included in this study together with the mean, the median and the standard deviation. The median was used as robust measure of location to eliminate the influence of outliers. In general, the good agreement between mean and median, taking into consideration their associated expanded uncertainties (95 % CI), indicate that the results obtained were normally distributed. Table 1 does not include the results obtained for liquid (honey and milk) and semi-liquid (chocolate) samples, because they were obtained with the Auto Quantify application of the ε5 software, developed for the analyses of fuel. The analysis of the CRMs SRM 2384 and ERM-BD512 revealed a systematic bias of 50 % for most elements in the liquid or semi-liquid products. Chromium mass fractions are not included in Table 1 because they were frequently around the detection limit of the method; however, as some clear patterns were observed in some food commodities, Cr was used for modelling purposes for those foods.

Producers are required to follow strict rules, including controls and administrative formalities, to obtain a certificate of conformity, which allows them to label their products as being “organic”. During the conversion period from conventional to organic farming, producers have to follow the organic production rules but are not entitled to sell their products as being organic. In addition, at times of surplus supply, organic produce may end up as conventional product on the market. With the approach followed in this work, i.e. purchasing the test items at retailers, those products would be considered conventional, while having the chemical composition of organic food and would then be projected into the organic products by multivariate analysis. Unfortunately, for products sampled at retail there is no way to differentiate real false positives and products that result from organic farming but are, for whatever reason, not labelled as “organic”.

The elements that differentiate orpan class="Chemical">ganic and conventional products, vary from one food commodity to another and for that reason separate data evaluation studies were conducted for each type of food included in this survey.

Paprika powder

Paprika powder belongs to the group of spices, frequently affected by fraudulent activities. The group was well represented in this study and encompassed 34 conventional and 13 organic samples. Most of the samples were sweet paprika but also two hot and six smoked paprikas (1 organic and 5 conventional), were analysed. For thirteen samples, no information whether they were sweet or hot was available on the label. PCA score plots constructed for the organic and conventional products, respectively, demonstrated that hot and smoked paprika are randomly distributed among the remaining samples in their groups, and were not flagged as outliers (neither by the Hotelling’s test nor by the Mahalanobis distance). Hence, hot and smoked paprika were retained in the study. Also samples with no indication about sweet or hot were kept.

According to the Student’s t-test (95 % CI), pan class="Chemical">Mg, P, Cl, S, K, Cr, Mn, Fe, Zn, Br, Rb and Ba mass fractions were significantly different in organic and conventional products, while those of Ca, Ni, Cu and Sr were not. Significantly different elements were used as variables to construct the PCA score plot, Fig. 1a, that shows two well-defined clusters based on the organic/conventional nature of the product. According to the bi-plot, Fig. 1b, organic paprika powders are richer in Mg, K, S, P, Cl, Zn and Br, while conventional products are richer in Fe, Mn, Cr, Ba and Rb.

Fig. 1

a) PCA score plot with two principal components of organic (BIO) and conventional (NON BIO) paprika powders, R2X[Cum] = 0.772 and b) Bi-plot organic and conventional paprika powders. Ellipse: Hotelling’s T2 (95 %).

One conventional paprika powder is clearly projected among the organic products in the PCA score plot. As discussed earlier, this sample could indeed be mislabelled or a product without organic label but actually grown and processed under organic conditions.

Cinnamon

Another spice for which preliminary results were obtained was cinnamon. Unfortunately, the elemental composition of cinnamon is different in the two most widely commercialised botanical varieties of cinnamon, cassia (Cinnamomum cassia) and Ceylon cinnamon (Cinnamomum zeylanicum). After separation of the available samples into cassia and Ceylon cinnamon and into organic and conventional within each one of the two species, the number of samples in each class was reduced and due to the decreased statistical power the obtained results could only be considered as indicative. Nevertheless, the PLS-DA score plot, shows that two clusters are formed based on the organic and conventional origin of the samples (Supplementary S1 a and b). PLS-DA, which is a supervised technique, was more powerful in separating the two groups than PCA as it maximises the distance between and minimise the distance within groups, whereas the main use of the unsupervised PCA is to reduce the dimensionality of the data space to enable its visualisation.

Coffee

The initial repository of coffee samples consisted of 46 samples (17 organic and 29 conventional) mostly of American (Mexico, Honduras, Guatemala, Nicaragua, Costa Rica, Colombia, Peru, Bolivia and Brazil), African (Ethiopia, Uganda, Tanzania and Republic of Congo) and Asian (India and Indonesia) origin; no information about geographical origin was provided for five samples. Most samples were 100 % Arabica (Coffea arabica), one sample was 100 % Robusta (Coffea canephora syn. Coffea robusta), and five were mixtures of Arabica and Robusta; no information about botanical variety was provided for six samples. Decaffeinated samples were not included in the study.

Initially, pan class="Chemical">all samples were taken for statistical evaluation. According to the Student’s t-test, P, Cl, S, K and Fe mass fractions were significantly higher in the conventional coffees than in the organic ones. Nitrogen, P and K need to be added yearly to soils in coffee plantations, as mentioned in a FAO report (Winston et al., 2005). Potassium is used in high quantities to increase the yields in coffee cultivation, frequently in the form of KCl (Cuzato Manuso et al., 2014), what would explain the higher contents of those two elements in conventional coffee. Fe is also in the list of nutrients required for coffee cultivation (Winston et al., 2005).

All elements, with the exception of Br (quantifiable in only one sample) were used for modelling purposes. The PCA score plot, Fig. 2a, shows that organic samples tend to cluster together although several conventional coffees were projected among the organic ones.

Fig. 2

a) PCA score plot of organic (BIO) and conventional (NON BIO) coffees, R2X[Cum] = 0.513, b) PLS-DA score plot of organic and conventional 100 % Arabica coffees from three different continents, R2X[Cum] = 0.422, R2Y[Cum] = 0.562, c) PLS-DA score plot of Arabica American coffees, R2X[Cum] = 0.446, R2Y[Cum] = 0.743, d) PCA score plot of organic and conventional chocolates, R2X[Cum] = 0.773, e) PCA score plot of organic and conventional chocolates with 70 to 75 % cocoa, R2X[Cum] = 0.703, f) PCA score plot of organic and conventional teas, R2X[Cum] = 0.802. Ellipse: Hotelling’s T2 (95 %).

a) PCA score plot of orpan class="Chemical">ganic (BIO) and conventional (NON BIO) coffees, R2X[Cum] = 0.513, b) PLS-DA score plot of organic and conventional 100 % Arabica coffees from three different continents, R2X[Cum] = 0.422, R2Y[Cum] = 0.562, c) PLS-DA score plot of Arabica American coffees, R2X[Cum] = 0.446, R2Y[Cum] = 0.743, d) PCA score plot of organic and conventional chocolates, R2X[Cum] = 0.773, e) PCA score plot of organic and conventional chocolates with 70 to 75 % cocoa, R2X[Cum] = 0.703, f) PCA score plot of organic and conventional teas, R2X[Cum] = 0.802. Ellipse: Hotelling’s T2 (95 %).

To reduce the intrinsic variability of the coffee data set possibly resulting from the two plant species (C. arabica and C. robusta), only 100 % Arabica coffees were selected (13 organic and 20 conventional) for further modelling purposes. Different cultivars such as Caturra-catuai, Java, Pink Bourbon, Castillo, etc, exist within the Arabica variety. Nevertheless, information about cultivar was only available for a reduced number of organic samples bought in specialised shops, and it was not taken into consideration in the construction of the models. Chlorine and K were present in concentrations significantly higher in conventional than in organic coffees. Fig. 2b, shows the PLS-DA score plot of the 33 Arabica coffees, where two clear clusters organic/conventional, can be observed.

The 33 Arabica samples covered different geographical origins: America (Mexico, Guatemala, Nicaragua, Costa Rica, Colombia, Bolivia and Brazil), Africa (Ethiopia and Republic of Congo) and Asia (India and Indonesia) and mixtures of them. For some samples no information about geographical origin was available. Since geographical origin influences the elemental composition of food (Fiamegos et al., 2020), the study was limited to American coffees to further reduce the dispersion of the data not related to the organic/conventional character of the samples. American coffees were selected because they were better represented in the study (22 samples, of which 10 organic and 12 conventional). The PLS-DA score plot in Fig. 2c shows two distinct clusters even if North, Central and South American samples are included in the plot; two conventional samples, one with a “Fair Trade” label, are projected next to the organic samples, which could indicate that they have been produced by a regime similar to organic farming. According to the Student’s t-test Cl, K and Fe mass fractions are significantly higher in conventional than in organic American Arabica coffees.

Chocolate

Among all the food pan class="Chemical">commodities included in this study, the most processed one is chocolate and it is the only one containing several ingredients. The initial sample set included 10 organic and 15 conventional chocolates. Although all the chocolates analysed were defined as dark chocolate, information about cocoa content was not always available, and for some samples only the minimum cocoa content was given. Among all the chocolates for which information about the cocoa percentage was available, the lowest value was 55 % and the highest 80 %. The specific geographical origin of the cocoa used was not available for all samples, and for many the only reference to geographical origin was Non-EU. No information was available about the cocoa variety such as Forastero, Criollo and Trinitario, whose composition of volatile compounds has been demonstrated to be different (Acierno et al., 2016).

Initially, pan class="Chemical">all samples were kept for the statistical evaluation of results. Of the 16 elements analysed, only P, S, Ca, K, Mn, Fe, Ni, Zn, Rb and Sr mass fractions were used because the remaining elements were present at quantifiable amounts in less than 50 % of the samples, or in none of them. This was the case for Mg, which is a light element and ED-XRF is rather insensitive for them.

The results obtained were in good agreement with thopan class="Chemical">se found in cocoa samples using a portable ED-XRF instrument (Herreros-Chavez et al., 2019), and with other techniques as reviewed in the same paper.

The median of most elements is higher in conventional than in organic chocolate, probably reflecting the use of fertilisers, but only the mass fractions of three elements are significantly different. According to the Student’s t-test, the mass fraction of Ca is significantly higher in organic than in conventional chocolates, while the opposite applies to Fe and Rb. The largest difference was observed in the Fe content that was twice as high in conventional chocolates as in organic samples. The elemental composition of 36 chocolate samples characterised by ICP-MS, including 38 elements, has been previously used to classify organic chocolate by Junior et al. (Junior et al., 2018), who found higher concentrations of Fe, Zn and Mg in the conventional than in the organic chocolate.

Fig. 2d shows the PCA score plot pan class="Chemical">constructed with all chocolate samples. Organic and conventional samples are projected in two distinct clusters but two conventional samples, those with the highest cocoa content, 78 and 80 % respectively, overlapped with the organic chocolates. To exclude the possibility that the two clusters are not due to the organic/conventional nature of the chocolate but to their cocoa content, a new model was constructed using exclusively the chocolates with a cocoa concentration in the range 70–75 %, Fig. 2e, obtaining two perfectly separated clusters. In the group of chocolates with 70–75 % cocoa, the elements with significantly different mass fractions in organic and conventional samples are Fe, Mn and Rb. Due to the reduced number of samples, this information can only be considered as indicative.

Tea

Twenty tea samples, 13 organic and 7 pan class="Chemical">conventional, were included in the study, all of them black tea. Ten samples were made with tea from Sri Lanka, India and mixtures of them, three from China, one from Vietnam and one from Kenya. For the remaining 5 samples no information about their geographical origin was given.

Chlorine, Ca, Cr, Mn, Fe, Cu and Sr mass fractions were significantly different in the organic and conventional groups. Only the content of Cu was lower in the conventional than in the organic samples, while for the other elements the opposite was observed.

The mass fractions found were in good agreement with vpan class="Chemical">alues previously published (Matsuura et al., 2001; Kumar et al., 2005; Polechońska et al., 2015). The Br mass fraction of one conventional tea was extremely high, probably due to an interference or integration artefact. Since the values for the other elements in that sample did not show any special pattern, the Br content was not used for statistical or modelling purposes and it is not given in Table 1.

Only the mentioned seven elements were upan class="Chemical">sed to construct the PCA score plot shown in Fig. 2f, in which two clusters are observed for the organic and conventional teas. One organic tea was projected among the conventional ones. The three Chinese teas are slightly detached from the other samples within the organic cluster and the same applies to the Kenyan tea within the conventional tea group. This indicates that the differences between organic and conventional teas are larger than those due to their geographical origin.

Rice

Most studies published in the literature on elemental characterisation of rice are related to classification according to geographical origin, either at country (Kelly et al., 2002; Chung et al., 2018), or region level (Liu et al., 2019; Neves Lange et al., 2019). Classification of organic, green and conventional Chinese rice based on stable isotopes and elemental composition, using chemometrics has also been achieved (Liu et al., 2020); that study included only Japonica rice from one single Chinese province, all paddy fields belonging to the same company. Another study has also been published on discrimination of Brazilian organic rice by multi-elemental analysis combined with pattern recognition techniques (Barbosa et al., 2016).

In this study only Basmati rice was included, to eliminate variations due to botanical variety and to a certain extent to restrict the geographical origin. Basmati rice grows only in three regions of India and Pakistan (Delwiche, 2016). The study included 17 Basmati rice, 9 organic and 8 conventional according to their labels, cultivated in India and Pakistan. Some samples were a mixture of Indian and Pakistani rice. According to the Student’s t-test none of the elements measured were significantly different in the organic and conventional groups. Nevertheless, when the mass fractions of P, Cl, S, K, Cr, Mn, Fe, Cu and Zn were used in multivariate analysis, a separation into two clusters was achieved, as shown in the PLS-DA score plot, Fig. 3a. One organic sample clustered together with the conventional and one conventional projected together with the organic ones.

Fig. 3

a) PLS-DA score plot of organic (BIO) and conventional (NON BIO) Basmati rice, R2X[Cum] = 0.693, R2Y[Cum] = 0.526, b) PLS-DA score plot of organic (BIO) and conventional (NON BIO) wheat flour, R2X[Cum] = 0.784, R2Y[Cum] = 0.781, c) PLS-DA score plot of organic and conventional cane sugar, R2X[Cum] = 0.855, R2Y[Cum] = 0.809. Ellipse: Hotelling’s T2 (95 %).

Wheat flour

Wheat flour is a staple food that has received a lot of attention from consumers in recent years. Different types of wheat flour are commercially available, covering not only farming practices (organic and conventional), but also wheat varieties (Triticum durum and Triticum aestivum), production processes (whole meal flour, white flour, flour obtained with stone mills) and geographical origin. In this study, whole flour was excluded, but samples belonging to all the other mentioned categories were included in the sample set.

P, Cl, S, K, Ca, Mn, Fe, Cu, Zn and Br where present in at least half the samples of one of the two categories, organic/conventional, and their mass fractions were used for statistical studies. With the exception of Cl, the mass fractions of all elements were slightly higher in organic than in conventional flours. This finding is in agreement with a previous work (Wang et al., 2020), that compared the results of P, K, Mg, Mn, Zn, Cu and Al in organic and conventional wheat. Only the Br mass fraction was significantly higher in organic than in conventional flour (around four times higher). Fig. 3b shows the PLS-DA score plot obtained for all the analysed flours, where it can be seen that organic flours form a cluster separated from the conventional ones. The group of organic flours included T. aestivum and T. durum white flours and flours grounded with stone mills. Unfortunately, little information was available regarding the geographical origin of the analysed organic samples, known for only two (Spain and France) of the seven samples.

Cane sugar

Next to rice and wheat flour, cane sugar was added to the list of food commodities mostly composed of carbohydrates. Classification of 13 organic and 9 conventional cane sugars, all of them cultivated in the state of São Paulo, Brazil, has been achieved using 32 elements determined by ICP-MS (Barbosa et al., 2015). Cane sugar was selected to challenge the approach followed in this study, since the mass fractions of many of the elements analysed are below the quantification limit of the ED-XRF method used (Zdinikova and de la Calle, 2020), although still detectable.

This study included 10 samples, 6 organic and 4 pan class="Chemical">conventional, on which hardly any information about geographical origin was available. Two samples were produced in the Antilles and one in Latin America; the only information available for the remaining ones was “product of non-EU agriculture”. The elemental composition of cane sugar is strongly affected by the type of processing undergone. Non centrifugal cane sugar (NCS), also called rapadura or panela among other names, is obtained by evaporation of cane juice, while “brown sugar” is obtained mixing refined saccharose with molasses (Jaffe, 2015); NCS contains higher levels of Ca, Cl, Cr, Cu, I, Fe, Mg, Mn, P, K, Se, Na, and Zn, than brown sugars. In this study eight samples were brown sugars, and one organic and one conventional samples were labelled as “rapadura” and “panela”.

Despite the reduced number of samical">ples, the high proical">portion of data where the mpan class="Chemical">ass fractions were below the respective LOQs, Table 1, and the inclusion of both NCS and brown sugars, cane sugars formed two distinct clusters based on their organic and conventional nature, Fig. 3c (constructed with all the elements indicated in Table 1 plus Si). As expected, the “rapadura” and “panela” samples were projected separated from the brown sugars, but always within their respective organic/conventional cluster.

Analysis of brown cane sugar for classification as organic or conventional would benefit from the use of techniques with lower LOQs, such as ICP-MS. Coconut sugar, contains elemental mass fractions significantly higher than cane sugar (Zdiniakova & de la Calle, 2020), hence ED-XRF could be used to characterise it as organic or conventional. So far, coconut sugar seems to be exclusively the product of organic agriculture and it was not possible to find any conventional samples on the market, to confirm this hypothesis. Keeping in mind the large differences in elemental content of cane and coconut sugars, different models must be constructed for each one of the two types of sugar.

Honey

It is known that the elemental profile of honey can be used for classification purposes according to botanical variety and geographical origin (Fiamegos & de la Calle, 2020; Ghidotti et al., 2021). These two sources of variability must be considered when studying the organic/conventional nature of honey. A set of 20 organic honeys was available for the study, including 3 chestnut (2 Spanish, 1 Italian), 3 eucalyptus (2 Spanish, 1 Italian), 2 lime (Romanian), 1 lemon (Italian), 3 orange (2 Spanish, 1 Mexican), 2 lavender (Spanish), 1 rosemary (Spanish), 2 sunflower (1 Spanish, 1 Romanian), 1 thyme (Spanish) and 2 robinia (1 Italian, 1 Romanian) honeys. Two reduce the variability due to botanical variety and geographical origin, a set of 20 conventional samples matching the botanical varieties represented in the organic group, and as much as possible also the geographical origin, were selected for the study. The only differences between the organic and conventional groups were that the conventional lemon and sunflower honeys were all Spanish.

The elements used for modelling purical">poses were P, Cl, K, Ca, Mn, Fe, Zn, Br and Rb. The remaining elements could not be quantified in any of the honey samples. According to the Student’s t-test, only P and Zn were significantly different in the organic and conventional groups, both being higher in conventional than in organic honeys.

When the 40 samples were included in the models, a trend to form separated clusters related to the organic/conventional nature of the honeys was observed. However, the PLS-DA model, Supplementary S1 c, shows that three organic samples are projected among the conventional ones and three conventional among the organic ones. In a second step, separated PLS-DA models were constructed for dark (chestnut, eucalyptus and lime) and light honeys (robinia, lemon, orange, lavender and rosemary), respectively. Sunflower and thyme honeys have on a general basis, mass fractions higher than the light honeys and lower than the dark ones, for this reason they were excluded in further studies.

The PLS-DA score plot constructed for dark honeys, Fig. 4 a, shows two separated clusters for organic and conventional samples, with only one organic sample projected among the conventional ones. Also in this case P and Zn were the only significantly different elements in the two groups. The separation is less clear in the case of light honeys, Supplementary S1 d; the PLS-DA score plot shows two separated clusters but the three organic orange samples were projected together with the conventional ones. Ideally, models should be constructed for honeys of one single botanical variety but unfortunately, this was not possible in this study due to the limited number of available samples.

Fig. 4

a) PLS-DA score plot of organic (BIO and conventional (NON BIO) dark honeys, R2X[Cum] = 0.563, R2Y[Cum] = 0.580, b) PLS-DA score plot of organic (BIO and conventional (NON BIO) Ultra-High-Temperature (UHT) full fat bovine milks, R2X[Cum] = 0.730, R2Y[Cum] = 0.689, c) PLS-DA score plot of organic (BIO and conventional (NON BIO) coconut water, R2X[Cum] = 0.840, R2Y[Cum] = 0.897. Ellipse: Hotelling’s T2 (95 %).

Bovine milk

Next to honey, bovine milk wpan class="Chemical">as introduced in this study as representative of food of animal origin. Elemental profiles have already been used to classify cow’s milk on the basis of geographical origin (Zain et al., 2016).

Several markers such as fatty acids and stable carbon isotope ratios, have been proposed to authenticate European organic milk, but none of them allowed the proper classification of Australian organic and conventional milks, and the question was raised whether they can be applied on milk from countries with warm and dry weather (Eibler et al., 2018). IR hand held devices have been applied to classify pasteurised retail milks in The Netherlands (Liu et al., 2018), and NMR to classify organic buffalo milk (Mazzei and Piccolo, 2018).

In this study P, Cl, S, K, Ca, Zn and Br were measured in 15 Ultra-High-Temperature (UHT) full fat cow milks (7 organic and 8 conventional) from Belgium, The Netherlands and Germany. None of the elements were present in significantly different amounts in the two populations according to the Student’s t-test, but with the exception of Zn (present in similar concentrations in the two groups), all mass fractions were higher in organic than in conventional milks. The PCA score plot, Supplementary S1 e, showed that the samples were projected in two clusters related to their organic/conventional character. The PLS-DA score plot, Fig. 4 b, shows a clear separation of the two clusters.

During the ED-XRF analysis, liquid milk samical">ples can clot decreasing the reproducibility of the results. Lyophilisation and further analysis of the obtained powder would be a pre-concentration technique that would eliminate the clotting problem. However, the introduction of a lyophilisation step would lessen the utility of ED-XRF as a fast screening tool.

Coconut water

Coconut water has attracted consumer’s attention as a healthy natural product low in carbohydrates. Analyses of coconut water to detect the fraudulent addition of exogenous sugars is broadly covered in the literature (Psomiadis et al., 2018; Richardson et al., 2019), but nothing has been published so far, to the best of our knowledge, on classification of organic coconut water. In this work 11 pasteurised coconut waters, all of them from different brands, were analysed. Unfortunately, information about the geographical origin of the product was available for only one sample (Brazil). Multivariate analyses were carried out with P, Cl, K, Ca, Mn, Br and Rb mass fractions. Fe and Zn could be quantified in less than half the samples and so their mass fractions were not used for modelling purposes. Only Cl and Br mass fractions were significantly different in the two groups, higher in conventional than in organic samples, according to the Student’s t-test. The PLS-DA score plot shows that also in this food commodity, organic and conventional samples form two separated clusters, Fig. 4 c.

Conclusions

The elemental composition of plants is influenced by various factors. Next to plant related factors such species, variety and physiological age, factors related to soil type and fertility, climate, and agronomic practices play a prominent role. Conventional and organic farming differ in the latter aspect as regards how fertilisation, pest and weed control is managed. Fertilisers allowed in organic farming are mainly animal manure and N-fixing plants (green manure) which has a profound effect on organic matter and the soil microbiome. Microbial communities govern to a certain extent nutrient transformation by enzymes release into soil and influence thereby availability and uptake of major and minor plant nutrients. Within the soil microbiome, arbuscular mycorrhizal fungi (AMF) play a key role in plant nutrition due to their capacity to improve plant mineral uptake (Smith and Smith, 2011; Rouphael et al., 2015). This could explain differences in mineral composition of plant food in dependence of the farming system since microbiological and enzymatic activities are higher in organically farmed soil (Jezierska-Tys et al., 2020).

The results discuspan class="Chemical">sed in this work demonstrate that despite of the type of food commodity and its geographical origin, the elemental profile of organic food differs to an extent from conventional food that the two groups can be separated by statistical techniques. This is true for food of vegetable and animal origin, for carbohydrate rich matrices (rice, flour) as much as fat rich matrices (chocolate). To the best of our knowledge, it is the first time that such a broad selection of food commodities have been included in a study on chemical composition of organic and conventional food. The study has been run under the worst case scenario circumstances, analysing commercially available samples from different regions, botanical varieties, industrial processes, transport and storage conditions. Despite all the mentioned sources of variability, organic food clusters together and separated from the conventional. No systematic trend has been observed that applies to all food commodities; for instance, Cl and Br mass fractions are higher in organic paprika powder than in conventional, while the opposite applies to coconut water.

To elucidate if organic food is hepan class="Chemical">althier than the conventional on the basis of its elemental content, goes beyond the purpose of this study. Further studies should be conducted, including aspects such as bioavailability, before conclusions on health benefits can be extracted.

The differences in elemental profiles could be exploited by control laboratories in the fight against fraud with the support of multivariate analysis since it was never the case that one single element could be used as marker for classification purposes. As with any other chemical marker, construction of databases with results obtained from samples representative of the product cross-section of particular interest in a certain market, is a must.

ED-XRF has demonstrated to be a suitable analytical technique for this type of studies, certainly as screening method, since one single calibration curve can be used in the analysis of all the different matrices, and because it does not include any sample treatment, other than milling and preparation of pellets. Some samples though, would benefit of the use of analytical techniques with lower LOQ’s, such ICP-based techniques.

Funding resources

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Declaration of Competing Interest

The authors declare no conflict of interest.

CRediT authorship contribution statement

Yiannis Fiamegos: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Validation, Writing - review & editing. Sergej Papoci: Formal analysis. Catalina Dumitrascu: Data curation, Formal analysis, Writing - review & editing. Michele Ghidotti: Data curation, Formal analysis, Writing - review & editing. Tereza Zdiniakova: Data curation, Formal analysis, Writing - review & editing. Franz Ulberth: Funding acquisition, Resources, Writing - review & editing. María Beatriz de la Calle Guntiñas: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Project administration, Resources, Supervision, Validation, Visualization, Writing - original draft, Writing - review & editing.