2.1. Identities and Concentrations of Substances Found in the PET Containers
Only low molecular weight compounds (up to approximately 200 g/mol) absorbed into the bottle wall of non-food PET containers were found in the present study. Notably, molecular weights up to 250 g/mol represent the crucial range for contaminants in PET that are relevant for migration and potential exposure and therefore for safety evaluation [
15]. This is due to the very low diffusivity of larger molecules because of their relatively high activation energies for diffusion [
15,
16,
17,
18]. Consequently, larger molecules rarely or even never penetrate into the PET matrix during the first-use life and therefore do not build up a significant potential in and remigration from the polymer matrix. Therefore, substances from non-food products with molecular weights above 200 g/mol were not found in the PET containers. In addition, migration modelling based on realistic diffusion coefficients [
18] confirms that the EFSA safety evaluation criterion which is set at a migration limit of 0.1 µg/L would generally be fulfilled for substances with molecular weights above 200 g/mol for long-term migration test conditions. Even after long storage periods of one year at room temperature, this migration limit cannot be exceeded for such substances, thus they are not relevant for safety evaluation.
The 36 investigated PET containers were grouped into six categories according to prior fillings, namely (i) dishwashing detergents, (ii) antifreeze and dish-rinsing agents, (iii) mouthwash products, (iv) kitchen and bathroom sanitary cleaning products, (v) shampoos, and (vi) shower gels and liquid soaps. An overview of the qualitative and (semi)quantitative data of these 36 PET samples is given in
Table 1,
Table 2,
Table 3,
Table 4,
Table 5 and
Table 6. It should be noted that the identities of the bottles could not be clarified in all cases, hence the presence of several unknowns in the tables. In a way, this situation mirrors real-life scenarios, which will always include a certain number of unknowns. Data presented in
Table 1,
Table 2,
Table 3,
Table 4,
Table 5 and
Table 6 include the concentration values for individual samples, as well as the mean value for the whole group. A calculation of mean values for the whole group is in our opinion meaningful because PET recollection systems will always contain a mixture of all these samples and never one sample type only. Following this logic, calculation of overall mean values from the group means would be the next relevant step to predict realistic contamination levels in a fully mixed recollection system. Such overall mean values would form the right basis for further exposure and safety assessment considerations for real-life scenarios. However, for numerous substances this would lead to “dilutions” of concentrations in the full mixture. To avoid this, our approach is to undertake a safety evaluation based on typical concentrations and related ranges found in the six groups. To include also considerations on the potential toxicological concern of the substances identified,
Table 1,
Table 2,
Table 3,
Table 4,
Table 5 and
Table 6 report the Cramer classification results for the substances, including the alert indicators for genotoxic (gtc) and non-genotoxic (ngtc) carcinogenicity.
The first group (
Table 1) includes six bottles used for dishwashing products. Only three substances, namely ethanol, phenoxy methanol and phenoxy ethanol, were found in individual bottles at elevated concentrations of 380 mg/kg, 13 mg/kg and 8.4 mg/kg, respectively. The concentrations of all other compounds found in the individual PET bottles within this category were below or far below 10 mg/kg. With the exception of ethylene glycol, which exhibited a low concentration distribution of between 2 and 5 mg/kg in the individual samples, the average concentration values for the mixture of the whole group ranged below or far below 1 mg/kg. Two unknown compounds were found, with group mean concentrations of 0.33 and 0.42 mg/kg. Furthermore, in a few cases wrong assignments by Toxtree were obtained (see footnotes in
Table 1), e.g., benzene, a known human carcinogen, were allocated negative gtc and ngtc assignments.
The second category (
Table 2) contains four PET bottles originally filled with antifreeze detergents and dishwashing rinsing agents. As expected, the most abundant compound found in these bottles was ethanol, which was present in all four bottles at high levels and exhibited a maximum concentration of 940 mg/kg. High concentrations were also found for other compounds, such as 2-butanone with a concentration of up to 45 mg/kg (sample 8), and concentrations of 23 and 16 mg/kg for two unknown compounds (unknown 3 and unknown 4). The other compounds found in this category, including two further unknowns, were all below 8 mg/kg in the individual bottles. The group means were correspondingly lower, with levels of between 0.5 and 5 mg/kg when excluding ethanol and 2-butanone.
The third group (
Table 3) includes six bottles filled in the first use with mouthwash solutions. The presence of flavour compounds in these bottles is quite expected. The most abundant compounds detected were the two cis-trans isomers of anethole, with the highest concentration for the trans isomer of 28 mg/kg in sample 16. Besides two unknowns (7 and 8) found at low levels of below 1 mg/kg, two gtc alerts were obtained from Toxtree for the substances carvone and eugenol, which can be considered as wrong assignments.
Seven bottles were investigated in the category of sanitary cleaning products (
Table 4). These bottles contained a variety of different compounds, including seven unknowns. Two samples contained three substances at higher levels: in sample 17, both anethole isomers were found at 31 mg/kg and 44 mg/kg, respectively, and in sample 23 the solvent 2-butanone was quantified at 120 mg/kg. For most of the substances, however, the concentrations were below or far below 8 mg/kg in the individual bottles and the mean values were correspondingly lower. It is noteworthy that the four aldehydes hexanal to nonanal were incorrectly assigned to gtc by Toxtree, and that the substance 2-ethylacrolein, found at 0.6 mg/kg in sample 22 and with a group average level at 0.14 mg/kg, is reported to be genotoxic in vitro (AMES test).
In category 5 (shampoo bottles,
Table 5) with six different bottles, only one sample contained ethanol, with a concentration of about 1100 mg/kg. All other compounds found in this category, including two unknowns, were below 9 mg/kg and exhibited mean values in most cases of below 2 mg/kg.
In the last category (
Table 6), seven bottles originally filled with shower gels or liquid soap were analysed. Again, ethanol was found in three of the seven bottles at high concentrations of up to 400 mg/kg. The other compounds, including six unknowns, were determined at concentrations below or far below 13 mg/kg in the individual samples, with correspondingly much lower mean values. Benzaldehyde, found at 0.9 mg/kg in sample 33, received an incorrect gtc assignment by Toxtree.
2.2. Comparison with Published Data on Contaminant Levels in Post-Consumer PET
Quantitative data on contaminants in post-consumer PET recyclates are rare in the scientific literature. Huber and Franz investigated 22 post-consumer PET flake samples [
20] and found that the flavour compound limonene, a ubiquitous substance present in many products, was detected in 15 samples at concentrations of up to 3 mg/kg in the polymer. In one sample, phthalate esters were determined, albeit without full identification and quantification. In a subsequent more comprehensive study, Franz and Welle [
21] investigated about 150 post-consumer PET flake samples from food bottles originating from 14 European PET recyclers. The aim was to determine “fingerprints” from headspace gas chromatography/flame ionization detection (GC/FID) analyses to screen for quality and potential migrants. In addition, acetaldehyde and limonene were quantified. The results showed concentrations of between 1.5 mg/kg and 11 mg/kg for limonene and 15 to 40 mg/kg for acetaldehyde, which was quite different from virgin PET (around 1 mg/kg). Since a fraction of the samples were also retrieved from green dot collections, these samples most likely did contain PET containers from non-food applications. Other substances, including those potentially related to non-food fillings, were not targeted in their study. Nevertheless, the GC/FID fingerprints revealed the presence of such substances, albeit at peak intensities that were comparable to or lower than limonene. This indicated that other substances were present at the same concentration range as limonene, i.e., up to approximately 10 mg/kg.
In the USA, Bayer investigated PET materials (flakes superficially cleaned by a commercial wash process) from five different post-consumer PET feed streams for bottle-to-bottle recycling processes [
22] taken from deposit and curbside collections. Whereas the investigated deposit collections contained nearly 100% food containers, the amount of non-food containers in curbside collections ranged from 0.04% to 6%. From one of these curbside collections a fraction of non-food use containers was sorted out for this study. Bayer identified and semi-quantified 121 compounds absorbed in PET. The total concentration of these compounds was approximately 28 mg/kg in deposit bottles (nearly 100% PET bottles previously used for food applications) and 39 mg/kg in PET bottles used for non-food fillings, respectively. The key substances found in both fractions were, from a concentration standpoint, the flavour compounds hexanal, benzaldehyde, limonene, methyl salicylate, and carvacrol. These data for PET containers from non-food applications are in good agreement with the concentrations found in the present study with the exception of ethanol and 2-butanone, which were not found by Bayer. The most likely reason for this absence is that Bayer analysed the post-consumer PET samples after a commercial washing process, which included a hot washing step for about 10 min at temperatures of about 85 °C with caustic soda and detergents followed by a drying step, which significantly reduces the concentrations of volatile solvents like ethanol. Published studies involving surrogate contamination have shown that simple washing and drying reduce the concentrations of volatile contaminants in PET by at least 98%, and even non-volatile contaminants by 79% [
23,
24].
Begley et al. [
10] investigated the sorption kinetics of hexachloro cyclohexane (lindane) from a commercial medical shampoo into PET materials, one being amorphous sheet material and the other material from PET bottle wall at 20 °C and 40 °C. The amount absorbed into PET from the shampoo after 231 days was 28 mg/kg in the sheet material at 20 °C and at the same level in the bottle material at 40 °C. However, the concentration in the sheet material at 40 °C reached a surprisingly high level of 745 mg/kg. This can be explained by the fact that the shampoo contained 7% acetone, a very aggressive solvent that causes swelling of the amorphous PET sheet material. As a consequence, swelling of the PET matrix triggers higher sorption of the shampoo ingredient (the formulation contained 1% lindane) under the exaggerated sorption conditions at 40 °C. In our opinion, this high value cannot be considered as a potentially realistic contamination level. Firstly, PET containers and amorphous sheet material are not equivalent, and the latter are typically used for non-food products. Secondly, such a long-term contact at 40 °C is not a realistic storage scenario. Furthermore, from analysis of selected non-food PET bottles obtained directly from collection bins for recycling plastic bottles in the USA, the authors [
10] found 130 mg/kg to 204 mg/kg of methyl salicylate in a antiseptic mouthwash bottle, 1 mg/kg of the antibacterial agent 5-chloro-2-(2,4-dichlorophenoxy)-phenol (trade name Triclosan) in a PET bottle used for hand soap, as well as 1.6 mg/kg of the flavour compound limonene in a paint remover bottle. The values for Triclosan and limonene are in very good agreement with our findings; however, we found methyl salicylate only in one mouthwash bottle and at a level of more than 100 times lower (
Table 3). The reason for this may be due to a strong interaction of ethanol with PET, which enhances the uptake of methyl salicylate into the polymer. Unfortunately, the concentration of ethanol was not determined in the aforementioned study, although a level of more than 1000 mg/kg would be expected in view of the reported parameters. In our case, ethanol was not detected (sample 13). This is likely due to the fact that modern packaging design aims to achieve better compatibility between fillings and packaging materials, with the effect that sorption of product ingredients will be minimized.
Franz et al. [
13] published data from a Europe-wide screening of washed post-consumer PET flakes. Overall, 689 different post-consumer PET flake samples were collected from twelve European countries and were analysed regarding for post-consumer contaminants. Since one sample consisted of 10–15 flakes, each originating from a different bottle, it was concluded that some 7000–10,000 bottles were included in this screening project. In contrast to Bayer [
22], the authors did not have the information needed to distinguish between post-consumer PET bottles from food contact applications and non-food containers, thus could not determine the fraction of non-food containers, if any. However, it is most likely that also small amounts of PET containers from non-food applications were present among the investigated samples, depending on the recollection system. The proportion of such PET containers, however, was not available. As a result, mean concentrations for the most prominent substance limonene (flavour in soft drinks) and acetaldehyde (PET typical degradation substance) in rPET flakes were reported to be 2.9 mg/kg and 18.6 mg/kg, respectively, with maximum concentrations of 20 mg/kg and 86 mg/kg, respectively. In addition, several substances untypical of PET, such as phthalates, adipates and erucamide, were found at levels from 0.05 mg/kg to 0.2 mg/kg, with a maximum concentration of 0.5 mg/kg for the substance dioctyl phthalate. Due to the sample characteristics mentioned above (10-15 flakes per sample measured), this translates to approximately a tenfold higher concentration in the particular flake and therefore in the associated bottle, i.e., to a concentration range from 0.5 mg/kg up to 5 mg/kg. It cannot be established whether or not these substances originated from non-food fillings or from carry-over from other sources, such as labels, sleeves, cups, coatings, etc., but it provides an indication that these values are in congruence with the measured values of this study. In any case, due to the fact that the investigated post-consumer PET flakes were drawn from the market, they most probably also contained some material from non-food PET containers.
Dutra et al. investigates also recycled FET samples. Several post-consumer substances were identified in post-consumer rPET flakes and pellets samples. The list of identified substances is in agreement with our findings. However, no quantitative data are given in this study [
25].
As a conclusion here it can be stated that both our data and the reported literature data, indicate that levels of contamination from non-food applications can be found from the sub-mg/kg range up to around 30 mg/kg when we exclude solvents of a very volatile character, such as ethanol and 2-butanone. Concerning our data presented here, by far the highest fraction is found in the range below 10 mg/kg for individual bottles, and hence below 1 mg/kg for the group mean values.
2.3. Estimation of the Fraction of Non-Food Products PET in the Recycling Feed Stream
The estimation of the fraction of non-food products PET in the recycling feed stream is a very difficult undertaking. Many producers of super-clean recycled PET purchase washed PET flakes produced from ground PET bottles by recycling companies that only run washing lines but not super-clean recycling processes. Use of such PET flakes as input materials make it nearly impossible to determine the amount of the PET fraction previously used for non-food applications, since PET flakes from food and non-food applications are indistinguishable. Considering the step prior to grounding into flakes, i.e., the recollected bottles compressed into compact bales, in principle it is possible to distinguish between the two, but due to the high compression of the baled bottles the fraction of non-food bottle might be nevertheless difficult to be determined. However, since quantitative data about the amount of non-food containers in the input stream and their contamination levels are required for the evaluation of compliance with EFSA criteria, recyclers need to find a way to perform an analytical sorting. Such proprietary data are not usually available for the different recollections systems and could only be estimated by the particular recycler based on opening the bales and counting the different bottles. However, such a laborious and time-consuming procedure will result only in a snapshot of the actual input material and not substantiate a general picture. Furthermore, the fraction of non-food PET will not be consistent in different feed streams. Due to the different recollection systems in the European member states, it is quite obvious that the fraction non-food PET containers might be very different, whereby the percentage may range from virtually zero for deposit bottles systems up to a higher percentage for curbside systems and more general plastics recollections. Supporting data are very rare in the scientific literature. Begley et al. [
10] mention that 20% of the PET containers foreseen for PET recycling in the USA might be non-food product bottles. However, this seems to be a rough estimate from the recollection figures from 1999 and not necessarily current for the situation today.
In the following, an attempt is made to approximately estimate the maximum amount of PET containers from non-food applications based on the available market data. The PET resin market can be divided into two main application fields: (i) the fibre applications and (ii) the packaging applications, e.g., PET containers and sheet material. In 2008, the worldwide market size was 45 million tonnes of PET resins, from which 63% were for fibre and yarn applications that are not collected in packaging recollection systems and therefore can be excluded from these considerations. From the non-fibre PET resins, 35% were used for bottling of carbonated soft drinks, 25% for mineral water, and 17% for other beverages, which amounts to 77% used overall for the beverage bottles market. Furthermore, 9% of the PET resins were used for other food packaging applications, 10% for films and sheet, and 4% for other (non-food) packaging application, which amounts to 23% not used for the beverage bottle market. We consider (i) that the input materials for PET bottle-to-bottle recycling processes are typically PET containers and (ii) that sheet/film materials are typically not collected or separated during recycling (in the first washing step) so that 10% sheet material does not enter the recycling process. We also exclude (iii) the 9% fraction for other food from recycling and assume (iv) that the full 4% fraction for other packaging goes into the category non-food products packed in PET containers. We assume conservatively that this 4% fraction enters the recycling feed stream together with the 77% beverage bottle material, and that both fractions together make up 100% of the recycling feedstock. This recycling feed stream scenario would then contain 5% non-food packaging and 95% beverage bottles. This, of course, would be an average scenario of all recollection systems and could vary, as already stated and within certain limits, from one recollection system to the next. In conclusion, scenarios from very low percentages (<1%), as most likely originating from deposit systems, up to percentages of 5% or even somewhat higher, as possibly originating from less controllable curbside recollections, are estimated. However, a percentage of 20%, as mentioned above, appears unlikely from the reported market data but may occur sporadically under specific local recollection conditions.
2.4. Evaluating the Impact of the Contamination Levels in Non-Food PET Applications on the Safety of Recycled PET Bottles for Food Contact
Regarding the impact of the contamination levels in non-food containers on the safety of rPET in direct food contact, the question of interest is: Which concentrations of which substances from non-food PET containers can be established in beverages filled in PET bottles made from rPET based on usual super-clean recycling processes? Or, in other words: Which amounts of such non-food substances would remain in the rPET bottle walls after the reprocessing conditions and migrate from there into the beverages, typically after one year at room temperature (25 °C), as defined by EFSA as typical storage conditions for mineral water bottles? Further: How toxicologically relevant is the potential exposure of the consumer to these concentrations in beverages? An answer to these questions can be most likely only given by designing a conservative scenario of the situation in combination with mathematical modelling of the underlying diffusion and migration processes and assisted by the so-called Threshold of Toxicological Concern (TTC) concept [
26]. We therefore assume the following scenario: A 5% fraction of non-food PET is processed together with 95% post-consumer PET bottle material into new beverage bottles in a typical super-clean recycling process. The contaminants in the non-food PET after surficial washing and rinsing, which correspond to usual commercially washed rPET flakes used as input for super-clean processes, are represented by the substances listed in
Table 1,
Table 2,
Table 3,
Table 4,
Table 5 and
Table 6. The cleaning efficiency of the super-clean process is assumed to be 90% for any contaminant, which is the typically lowest efficiency of such processes, usually ranging between 99% and 90%, depending on the substance; cleaning efficiencies of only 90% can be considered as worst-case for PET super-clean recycling processes. The rPET produced in this manner is then manufactured into beverage bottles (for simplicity reasons) with a surface-to-volume ratio of 6 dm
2 per litre beverage. Migration into the beverage is modelled after contact conditions of one year at 25 °C (room temperature). Two generic migration curves derived from these assumptions are presented in
Figure 1. The figure depicts the migration into the beverage as a function of the molecular weight of contaminants using two different prediction models for the diffusion coefficients of potential migrants. These generic curves are based on the assumption of a starting contamination level of 10 mg/kg in the non-food PET, which is reduced by a factor of 20 due to the assumption of only a 5% fraction of non-food PET. This results in a concentration of 0.5 mg/kg. The concentration is further decreased by 90% due to the cleaning efficiency of the super-clean process, leading to a contamination level of 0.05 mg/kg in the final rPET bottle. The migration is then modelled using this 0.05 mg/kg as the initial concentration in the polymer, the so-called c
P,0 value, for the migration model for the above-mentioned contact conditions of one year at 25 °C.
Of the two prediction models depicted in
Figure 1, the rather conservative
Piringer A
P model, which is described in [
27,
28,
29], is generally accepted and applied in many EFSA evaluations of recycling processes. The other model uses the
Welle equation [
9], which gives a more realistic estimate of the diffusion in the PET polymer. From these two generic curves, using 5% non-food PET and 10 mg/kg input contamination level in non-food PET as defined generic starting points, the concentrations of other substances in the beverage can be easily derived for any other non-food PET contamination level and/or percentage in the feed stream because of the linear relationship of the migration value of c
P,0 and the percentage used. When we increase the fraction of non-food PET up to 20%, these values increase each by a factor of four compared to a non-food container level of 5%. On the other hand, when we decrease the concentration in the non-food PET from 10 mg/kg to 0.5 mg/kg, the migration decreases by a factor of 20. A combination of both will lead to a reduction by a factor of five.
As an example: for the substance carvone (molecular weight 150 g/mol), if present at 10 mg/kg in the non-food feed stream, migrations of 0.04 µg/kg and 0.001 µg/kg can be derived from the generic curves of the
Piringer and the
Welle model, respectively. When considering the value measured in sample 16, which is 2.4 mg/kg (see
Table 3), the related migration value would be roughly a factor 4.17 lower, i.e., 0.009 µg/kg and 0.00024 µg/kg, respectively. Based on the mean group value of samples 11–16, however, the migration would be a factor 25 lower, i.e., 0.0016 µg/kg and 0.00004 µg/kg, respectively. The curves in
Figure 1 allow a very quick and straightforward safety evaluation for all substances identified with no genotoxic alert, as listed in
Table 1,
Table 2,
Table 3,
Table 4,
Table 5 and
Table 6. With the exception of ethanol, migration will in all cases be lower than 1 µg/kg and, in most cases, even far (one to several orders of magnitude) below 1 µg/kg. Therefore, regardless of the Cramer class assignment of a compound, either I, II or III, it can be considered to be safe. As mentioned above, ethanol and other small and very volatile molecules such as 2-butanone are determined in higher concentrations as 10 mg/kg. These very volatile substances will be removed by the super-clean process at cleaning efficiencies equal to or higher than 99% and are therefore not further considered here.
A special group of contaminants in non-food containers are substances with genotoxic and carcinogenic potential, and these require a separate discussion, as follows: In
Table 1,
Table 2,
Table 3,
Table 4,
Table 5 and
Table 6, several substances are listed for which the Toxtree software assigns a genotoxic or non-genotoxic carcinogenicity (gtc or ngtc) alert. Examples include aliphatic aldehydes, benzaldehyde, eugenol and others. These assignments are evidently incorrect and are not further discussed here. Two substances remain, namely benzene, listed in
Table 1, which is a known human carcinogen (but not gtc or ngtc assigned by Toxtree), and 2-ethylacrolein, which is known to be mutagenic in vitro [
19]. Both substances were found at group means levels of 0.15 mg/kg and 0.14 mg/kg, respectively. Benzene and ethylacrolein have similar molecular weights of 78.1 g/mol and 84.1 g/mol, respectively. In terms of migration, these values would correspond to concentrations in a beverage of 0.00107 µg/kg (benzene) and 0.00101 µg/kg (2-ethylacrolein) for the
Piringer prediction model and 0.000338 µg/kg (benzene) and 0.000240 µg/kg (2-ethylacrolein) for the
Welle prediction model. These values were calculated for an input fraction of 5% non-food PET and a cleaning efficiency of 90% for the super-clean process.
The more conservative value (
Piringer) corresponds to a dietary exposure of benzene of 0.000161 µg per kg body weight (bw) per day for an infant drinking 0.75 L of water, which is the crucial scenario applied by EFSA [
4,
5]. This is a factor 15.5 lower than the EFSA “benchmark” criterion from TTC concept of 0.0025 µg per kg bw per day applied to any potential genotoxic contaminants migrating from rPET food contact materials (FCMs). From the more realistic model (
Welle), a dietary exposure of benzene of 0.0000507 µg per kg bw per day follows, which is a factor 49.3 lower than the EFSA benchmark value.
It is interesting to note that a recent publication [
30] reported migration values for benzene from rPET bottles into mineral water of 0.03 µg/L to 0.44 µg/L. The level of concern for public health was evaluated based on the margin of exposure (MOE) concept [
31,
32]. The reported MOE values ranged from 3 × 10
5–8 × 10
6 and, being considerably higher than 10
4, were considered to be of low concern. In our case, the modelled migration values are at least one order of magnitude lower, which increases the MOE correspondingly. In conclusion, the expectable level of concern for benzene and 2-ethylacrolein can be considered at most as extremely low.
The final question to be discussed and answered is how to deal with the unknowns, which in principle could consist partly or even totally of genotoxic substances? The “sanitary cleaning products” group (see
Table 4) exhibited the highest number of unidentified substances (unknowns 9–15), amounting to seven compounds with group mean levels below of 1 mg/kg and with estimated molecular weights of between 130 and 160 g/mol. When assuming that all of these seven substances are genotoxic, the following simple consideration can be made: Summing up the concentrations of the seven substances gives 1.82 mg/kg, which we round up to 2 mg/kg. We simplify the approach by reducing these seven substances to only one, with a mean molecular weight of 145 g/mol and a concentration of 2 mg/kg. A migration (
Piringer) of 0.0418 µg/kg can be derived from the generic migration curve in
Figure 1 (
Piringer), which yields 0.00835 µg/kg (when divided by 5; c
P,0 is only 2 mg/kg). From this, the exposure for an infant drinking 0.75 L of water would be 0.00125 µg per kg bw per day, which is a factor 2 below the EFSA “benchmark” criterion; by comparison, the
Welle prediction model yields an exposure of 0.0000360 µg per kg bw per day, which is a factor 69.4 lower. From these results, even assuming that all unknowns are genotoxic (which can be reasonably assumed not to be true), the expectable level of concern, if any, can be considered very low.
As mentioned above, Begley et al. [
10] debated a potential fraction of 20% PET non-food product bottles in the recycling feed stream, which according to our approach based on reported market data appears to be unlikely. Nevertheless, when assuming 20% non-food PET, all safety margins derived above for the 5% non-food fraction would decrease by a factor of 4 due to the linear proportionality of the calculated migration (
Figure 1) with the initial content of contaminants in the PET material. For instance, the safety factor of 69.4 derived above from a genotoxicity assumption for all seven unknowns would decrease by a factor of 4, resulting in a fourfold higher exposure that is still lower than the EFSA “benchmark” criterion by a factor 17.4. This indicates that even a 20% non-food PET fraction would not pose a significant level of concern for the health of consumers.