Hydrogen Sulfide Formation in Canned Wines: Variation Among Can Sources

Background and goals Wines packaged in aluminum beverage cans


Introduction
The global market for canned wines in 2021 was $235.7 million USD (in revenue) and is expected to rise to $571.8 million by 2028 (Grand View Research 2021), nearly a 30-fold increase from 2014 ($2 million in revenue) (Weed 2019).The expected increase in popularity of canned wines is attributed to several reasons including their convenience, lightweight design, ruggedness, established recycling channels, and acceptance at venues where glass is prohibited such as stadiums, festivals, and pools (Jacoby 2019, Ruggeri et al. 2022, US EPA 2023).
As with other types of wine packaging, wine producers are concerned with potential effects of aluminum beverage cans on wine quality.Aluminum (Al 0 ) readily corrodes in the presence of oxygen or water to generate a passive layer of oxidized aluminum (Al 3+ , e.g., Al 2 O 3 ).In the absence of a coating, and at the low pH typical of wines and other beverages (pH <4), this passive oxide layer would dissolve (Robertson 2013), leading to an increase in dissolved Al 3+ and eventually package leaking and loss of the hermetic seal (Robertson 2013).To mitigate corrosion, a thin polymeric coating (also called a "liner", "varnish", or "lacquer") is applied to the inside of aluminum beverage cans (Robertson 2013).Until recently, nearly all beverage can liners were bisphenol A (BPA)-based epoxies, the "gold-standard" liner material since the 1950s because of their inertness, low cost, and excellent barrier properties (LaKind 2013, Geueke 2016).However, because of concerns regarding BPA's role as a potential endocrine disruptor, its use has been curtailed in recent years, either through explicit bans (as occurred in France in 2015) (Geueke 2016) or by requirements that the presence of BPA must be declared on the label (as is the case in California since 2016).Alternative beverage can liners (also referred to as "bisphenol A-non-intent," or "BPA-NI,") have become more widespread in recent years, including those based on older materials such as acrylic (LaKind 2013), and newer epoxies based on monomers with lower endocrine activity than BPA (e.g., tetramethyl bisphenol F) (Szafran et al. 2017).
Despite the presence of a liner, there is good evidence that canned wine may react with an aluminum beverage can to generate hydrogen sulfide (H 2 S, "rotten egg" aroma, ~1 µg/L odor threshold) (Allison et al. 2021).Recent work demonstrated that glass packaged commercial wines developed only trace levels of H 2 S (<6 µg/L, and typically undetectable) after eight months, but that the same wines could produce high levels of H 2 S (>1000 µg/L) when stored in acrylic lined cans (Montgomery et al. 2023).Suprathreshold H 2 S concentrations (up to ~50 µg/L) were also observed in BPA epoxy and BPA-NI epoxy cans (Allison andSacks 2021, Montgomery et al. 2023).Across wines, the best predictor of H 2 S accumulation during can storage was the initial concentration of molecular SO 2 ; similar results were observed under accelerated aging conditions using lined aluminum coupons.Free SO 2 and pH were also well-correlated with H 2 S production, but other factors (i.e., total SO 2 , chloride, copper, and alcohol) were poorly correlated with H 2 S production (Montgomery et al. 2023).Based on these results, H 2 S formation was hypothesized to occur through the following reaction (Allison et al. 2021, Montgomery et al. 2023): Because wines that produced high H 2 S also showed signs of liner degradation, it was speculated that SO 2 either directly damaged the liner or else was able to diffuse through the liner to generate H 2 S gas and induce delamination.
Earlier work evaluated liners sourced from a single can manufacturer; variation in H 2 S formation among can manufacturers with similar liner chemistries was not considered (Montgomery et al. 2023).We hypothesized that H 2 S formation during long-term storage in canned wines would vary among can manufacturers even for the same liner type.We further hypothesized that variation in H 2 S among can manufacturers for the same liner type could be explained by differences in the can composition and the consistency of the liner application, which could be evaluated by employing a range of tools to characterize can and liner properties.
Nitrogen liquid and gas (N 2 , Ultra High Purity) cylinders were supplied by Airgas USA LLC.A 500 mL liquid nitrogen (LN 2 ) sprayer was obtained from US Solid.Headspace vials (30 mm × 60 mm, 27 mL), 20 mm butyl rubber septa, 20 mm tear-away crimp seals, and a 20 mm hand crimper were all obtained from Supelco (product codes 27298, Z166065, 27016, and 33280-U, respectively).A 5 to 25 mL bottle-top dispenser was obtained from VWR (product code 82017-768).A coated, clear 1000 mL glass bottle with septa port was obtained from Ankom Technology.A Surebonder (FPC Corporation) electric glue skillet and ethyl vinyl acetate (EVA) hot glue pellets (B-2001; Surebonder -FPC Corp.) were obtained through Amazon.

Commercial wines and initial chemical analysis
Five commercial wines were generously donated by an industry cooperator in 20-L high-density polyethylene KeyKegs: Pinot grigio (PG), Sauvignon blanc (SB), French rosé (FR), sparkling rosé (RB), and sparkling white (WB).Details on the wine style, vintage, and basic wine chemistry are provided in Supplemental Table 1.Basic wine chemistry was determined by established methods at the Cornell Craft Beverage Analytical Laboratory (Geneva, NY).Briefly, alcohol by volume (ABV) was analyzed using a Foss OenoFoss, free SO 2 analysis was carried out by flow injection analysis on a Foss FIAstar 5000 Analyzer, titratable acidity (TA) was measured by titration with 0.1 N sodium hydroxide to pH 8.2 endpoint with a Metrohm 862 Compact titrator and a Hanna Instruments HI901W automatic titrator, and pH was measured on a Fisher Scientific Accumet Excel XL25 dual-channel pH/ion meter.
Molecular SO 2 was calculated from Equation 2 using acid dissociation constant values (pK a ) adjusted for alcohol as described by Coelho et al. (2015): The aluminum content in the wines was determined by a local facility (USDA-ARS Holley Center) using a Thermo Scientific iCAP 6500 series system for inductively coupled plasma-atomic emission spectroscopy; the protocol is described elsewhere (Zhou et al. 2016).Initial H 2 S was measured by gas detection tubes, as described below.

Canning procedure
Cans were filled with wine directly from the KeyKegs using the supplied KeyKeg manual pump.After filling, a few drops of LN 2 were added to sparge the headspace of O 2 , and the can was immediately topped with a lid and seamed on a manual double seamer (MK16 seamer; Oktober Design).Seam quality was validated using a standard industry protocol consisting of measuring the seam thicknesses at four different points (first operation, second operation, cover hook, body hook) at three locations around the seam (Oktober Design 2024).Prior to the canning experiments, total package oxygen (TPO, or the sum of liquid and headspace O 2 , normalized against volume) was measured as described elsewhere using model wine (Montgomery et al. 2023).For TPO measurements, O 2 in the can headspace and liquid were equilibrated by gently agitating for one hour.The can was then opened, and dissolved O 2 in the liquid was measured by a Fibox 3 LCD trace O 2 meter fitted with a DP-PSt6 O 2 dipping probe (PreSens).The headspace O 2 content could then be calculated from the headspace volume and literature value for O 2 volatility; the TPO was calculated based on the sum of headspace and dissolved O 2 , and was determined to be <1.5 mg O 2 /L for the tested cans.

H 2 S in canned wines over long-term storage as a function of pH and SO 2
To evaluate the effects of pH, free SO 2 , and molecular SO 2 on H 2 S production in canned wine, five wines (PG, SB, FR, RB, and WB) were prepared in three groups prior to canning: 1. Low molecular SO 2 , low free SO 2 -Control group, no adjustments 2. Low molecular SO 2 , high free SO 2 -KMBS added, pH adjusted to 3.65 to 3.80 with K 2 CO 3 3. High molecular SO 2 , high free SO 2 -KMBS added, no pH adjustment Following adjustments, the low and high free SO 2 values ranged from 15 to 20 and 40 to 50 mg/L, respectively.The low and high molecular SO 2 values ranged from 0.6 to 1.1 and 1.5 to 2.5 mg/L, respectively.The pH values ranged from 3.11 to 3.37 for the wines without pH adjustment, and from 3.65 to 3.80 for the wines with K 2 CO 3 added.The pH was adjusted in Group 2 so the free SO 2 was similar to that of Group 3.
Wines were canned as described above in one of three can types: BPA epoxy (Company X), BPA-NI epoxy (Company Y), and BPA-NI epoxy (Company Z), then stored at 20°C in an upright position away from sun and light, until analysis at four and eight months of storage.Three can replicates were prepared for each wine (n = 5), composition (n = 3), can type (n = 3), and time point (n = 2), for a total of 270 individual can samples.The H 2 S in each sample was measured at the appropriate time point.

Accelerated aging -preparation of aluminum coupons and testing protocol
Preparation of coupons was based on an approach described and validated elsewhere (Montgomery et al. 2023).Can tops and bottoms were removed with a Gryphon C-40 band saw (Gryphon Corp.), then cut vertically with scissors to open the body (Supplemental Figure 1).Rectangular coupons (1 cm × 4 cm total) were cut by stainless steel shears from the middle of the side of the can body, and two 1 cm × 2 cm coupons were prepared from the headspace region of the can body (Supplemental Figure 2A to 2C) for each accelerated aging trial to maintain a constant surface-tovolume ratio in the 27 mL vial.The bare, uncoated edges were then sealed with EVA hot melt glue.
Accelerated aging trials were performed as described by Montgomery et al. (2023).Bottled or kegged wine was transferred into an ethanol-sanitized 20-L plastic water cooler, previously sanitized by a 70% ethanol rinse.Prior to filling, wines were nitrogen-purged until dissolved O 2 was <0.1 mg/L by PreSens Fibox 3 LCD trace O 2 meter with DP-PSt6 O 2 dipping probe.During vial filling, nitrogen was used to backfill the cooler to limit O 2 pickup.For each accelerated test, a 27-mL glass crimp-top vial was purged with two to three drops of LN 2 before rapidly adding deoxygenated wine (25 mL) and a coated, edge-sealed aluminum coupon.A butyl rubber septum was then placed on top of the vial, but not sealed, to allow excess LN 2 to dissipate (10 to 15 sec), after which the vials were sealed with a 20-mm aluminum metal crimp cap.Accelerated aging took place at 50°C prior to H 2 S measurement after three and 14 days.This protocol achieved O 2 pickup of <0.5 mg/L O 2 , as determined by the PreSens meter (Montgomery et al. 2023); measurements on model solutions indicated that there was a negligible amount of O 2 ingress over three and 14 days of storage in-vial.

H 2 S production under accelerated conditions: can source, can liner type, and within-can location
To evaluate variation in H 2 S production across commercial cans, 10 can types (three BPA epoxy, five BPA-NI epoxy, two acrylic) were sourced from a total of five commercial suppliers (designated V, W, X, Y, and Z).The cans were coded with letters V to Z, signifying the can producer, followed by 1 (BPA epoxy), 2 (BPA-NI epoxy), or 3 (acrylic), to signify the liner type.For two of the can sources, multiple production batches (n = 2 or 3) were tested to evaluate batch-to-batch variation.For example, X1 and X1-2 indicate two batches of BPA epoxy cans from manufacturer X. Y2-2 was observed to have incomplete liner coverage based on visual inspection.Y2-3 was the same liner material as Y2, but had a thinner layer of liner material applied.A high molecular SO 2 commercial 2020 German Riesling (pH 3.1, molecular SO 2 = 2.56 mg/L, free SO 2 = 43 mg/L, ABV = 9.3%) was used for accelerated aging trials.Accelerated aging trials took place for both three or 14 days, because earlier work had demonstrated that the average of three day and 14 day trials was most predictive of long term aging, and all treatments (can type × location in can × timepoint) were prepared in triplicate.

H 2 S production by immersed and nonimmersed regions of aluminum
To evaluate if immersed and non-immersed regions of aluminum produced similar H 2 S concentrations, a modified accelerated aging experiment was developed.Coupons were prepared from BPA-NI epoxy coated Al 3004 sheets as described above.For each accelerated aging test, coated coupons were inserted in one of four orientations (Supplemental Figure 3): 1.One coupon (4 cm × 1 cm) was fully submerged, and a second coupon (1 cm × 1 cm) was bonded to hot melt glue on the underside of the vial septa in the vial headspace.2. A single coupon (5 cm × 1 cm) was partially submerged, with ~1 cm 2 exposed to the vapor phase.3.One coupon (4 cm × 1 cm) was fully submerged, and a second coupon (1 cm × 1 cm) was also fully submerged.4. A single coupon (1 cm × 1 cm) was glued to the underside of the vial septa in the vial headspace, with ~1 cm 2 exposed to the vapor phase.

H 2 S quantification
H 2 S was measured by colorimetric gas detection tubes (GDT) attached to a commercial aeration-oxidation (A-O) apparatus, described in more detail elsewhere (Allison et al. 2021).An H 2 S selective GDT (4LT and 4LL tubes; Gastec International) was inserted between the receiver flask of an A-O apparatus (GW Kent, Inc.) and the vacuum source.Aspiration of a wine sample resulted in staining of the GDT tube via reaction of H 2 S with a metal salt; the stain length was proportional to the original H 2 S concentration.Interferences from SO 2 were prevented by inserting an SO 2 selective GDT (Gastec 5L) between the H 2 S GDT and A-O unit.The method detection limit was previously reported to be ~1 µg/L (Allison et al. 2021).

Characterization of polymeric liner and aluminum interior surface of cans
The interiors of the cans used in the long-term storage trials (X1, Y2, and Z2) were evaluated both before and after storage by several optical and spectroscopic techniques.For some techniques, unlined cans provided by the manufacturers were also analyzed as described below.To prepare the flat sections of cans for evaluation, cans were cut across the body with a Gryphon C-40 bandsaw (Gryphon Corp.), then cut vertically with scissors to open the body, as described for coupon preparation.For post-storage cans, a seam teardown tool (Oktober Design) was used to first remove the can lid.Samples were obtained from five different vertical locations along each can, as shown in Figure 1.Coupons were 0.25 cm × 0.25 cm, except for coupons from the can headspace, where smaller coupons (~0.25 cm × 0.1 cm) were used because larger, flat samples could not be prepared.Elemental composition of unused bare aluminum cans and pre-and post-storage lined cans were determined by x-ray fluorescence (XRF) (Bruker Tracer III-SD).Samples (1 cm × 1 cm) were prepared (Figure 1, Location 4) and run in triplicate, with the voltage set to 40 kV, current set to 40 µA, and a 60 sec dwell time.The portable device was mounted in the upward facing tabletop geometry and samples were placed on the measurement window with the inner face of the can sample pointing toward the window for data collection.XRF peak areas were fit using PyMCA (Solé et al. 2007).
Liner composition was characterized by Fourier transform infrared-attenuated total reflectance (FTIR-ATR) on a Bruker Vertex V80V Vacuum FTIR system (Cornell Center for Materials Research [CCMR], Ithaca, NY) in a nitrogen atmosphere.For FTIR analyses, coupons (0.5 cm × 0.5 cm) were prepared in triplicate from the middle of the can body (Figure 1, Location 4).Spectra were collected from 4000 to 700 cm -1 (Sultanova et al. 2019).
For profiling liner thickness and aluminum uniformity, laser-scanning profilometry was performed at the CCMR using a Keyence VK-X260 laser-scanning profilometer.Each sample was analyzed at 408 nm using the surface profile feature to study the aluminum surface uniformity and the thin film feature to study liner thickness and uniformity.For the thin film analysis, a refractive index of 1.7 was used for BPA and BPA-NI epoxy liners based on the literature value for polycarbonate (Sultanova et al. 2019) in the range of 435 to 1052 nm and extrapolating to 408 nm.The measurement size for a sample was roughly 285 μm × 210 μm.
Liner thickness in the body and bottom (dome) of the can was also evaluated by optical interferometry (SpecMetrix ACS-10 Model, Sensory Analytics LLC) (Komaragiri and Telep 2017).A broad range of wavelengths (700 to 1400 nm) was used and a refractive index of 1.55 was chosen for analysis.Three cans of each type were sampled for analysis.
To characterize liner integrity (e.g., the presence of pores), electrochemical impedance spectroscopy (EIS) was performed on cans in triplicate using a PalmSens3 potentiostat and PSTrace 5.9 software.Intact, unused lined cans were evaluated in place of coupons using protocols described elsewhere (Esteves et al. 2014, Daroonparvar et al. 2021).The setup is shown in Supplemental Figure 4 and is similar to aluminum can EIS measurements described by Grandle and Taylor (1994).Briefly, can bottom exteriors were sanded with 1000 grit sandpaper (Dura-Gold), washed with 70% ethanol, and dried to ensure a clean aluminum surface contact with the working electrode.The cans were filled with sodium chloride (NaCl) electrolyte solution (35 g/L), and a counter electrode (flat cell steel) and reference electrode (silver/silver chloride, potassium chloride as reference electrolyte) were inserted into the electrolyte solution and placed inside a Faraday cage to avoid external electromagnetic interferences.For the working electrode, electrical contact was made with the bottom of the can.The open circuit potential (OCP) was allowed to stabilize (drift <0.01 ∆mV/sec) before measurements were taken; the parameters used for OCP and EIS are shown in Supplemental Tables 2 and 3.For metal exposure measurements, a WACO Enamel Rater III was obtained, and a 10 g/L NaCl electrolyte solution was used to evaluate the three can types from the long-term aging study.A constant electrolyte fill level was maintained across all the cans, and the stainless steel electrode was placed into the can for measurement.The industry standard test was used, which consists of applying 6.3 V for 4 sec before measuring the current.
Liner adhesion quality was evaluated by an industry standard method, ASTM D3359-17 (Test Method B), using Scotch Bi-Directional Filament Tape 8959 (180° peel strength of 11 N/cm) on unused cans (ASTM 2022): six parallel horizontal and six parallel vertical cuts are made on the interior of the can, and pressure-sensitive tape is applied to that area.After 90 sec, the tape is removed and the percentage of the area with removed liner is determined.
The presence of uncoated regions on cans was evaluated qualitatively by ASBC Can Method 8 (ASBC 2011).An aqueous solution of hydrochloric acid (0.027 N) and copper sulfate pentahydrate (100 g/L) was added to cans (n = 3 replicates) and stored at room temperature for 45 min.Uncoated regions could be detected by the presence of Cu deposits, as described (ASBC 2011).

Statistical analysis and software
Statistical analysis was done via JMP Pro 16 and JMP Pro 17 (SAS Institute, Inc.).Analysis of variance (ANOVA; α = 0.05) was used to evaluate the effects of storage time, liner, and wine composition on H 2 S production.A p value of <0.05 was used to determine significant differences among treatment groups.

H 2 S in canned wines with varying liner sources and composition over long-term storage
Using a wine adjusted to varying pH and SO 2 concentrations, previous work demonstrated that H 2 S formation in the presence of aluminum coupons was higher for acrylic liners than for epoxy liners, and that H 2 S was best correlated with molecular SO 2 rather than free SO 2 or pH.However, this earlier work was performed on a single wine, single liner type, and was not validated with long-term storage conditions (Montgomery et al. 2023).
To confirm that molecular SO 2 , but not free SO 2 , best predicts H 2 S production across a broader range of cans and wine sources, five commercial wines were adjusted with K 2 CO 3 and SO 2 to yield three treatments per wine: I. Low pH, low free SO 2 , low molecular SO 2 II.High pH, high free SO 2 , low molecular SO 2 III.Low pH, high free SO 2 , high molecular SO 2 Other hypothetical combinations, e.g., low pH, high free SO 2 , and low molecular SO 2 , could not be generated because molecular SO 2 is dependent on the free SO 2 and hydrogen ion concentration ([H + ]).However, these treatments were sufficient for decoupling the relative importance of these three factors.The "high" molecular SO 2 range was 1.6 to 2.6 mg/L.Although this range is high compared to typical molecular SO 2 recommendations to prevent microbial spoilage (0.5 to 0.8 mg/L) (Zoecklein et al. 1999), the molecular SO 2 values in this paper were calculated using ethanol-adjusted pK a values, and "conventional" molecular SO 2 values based on the pK a of water will underestimate molecular SO 2 by 25 to 50% (Coelho et al. 2015).Thus, this range is equivalent to 0.8 to 2.0 mg/L molecular SO 2 based on conventional calculations.
The unadjusted Treatment I wines contained undetectable H 2 S except for trace levels (<3 µg/L) in the RB wine.H 2 S was not measured in the wines immediately following pH and/or SO 2 adjustment (Treatments II and III), but was assumed to be unchanged.Wines were then stored in cans from three different manufacturers (X1 = BPA epoxy, Y2 and Z2 = BPA-NI epoxy) and H 2 S was measured after four and eight months of storage (Figure 2).In the unadjusted wines (Treatment I), average H 2 S was below sensory threshold (<10 µg/L) in all five wines at both four and eight month time points.Interestingly, no correlation was observed between molecular SO 2 and H 2 S concentration in the adjusted wines, in contrast to an earlier report (Montgomery et al. 2023).Potentially, this is because the range of molecular SO 2 values in the current study (0.59 to 1.07 mg/L) was considerably smaller than the range of the earlier work (0.13 to 2.4 mg/L).
H 2 S was notably higher in Treatment III than in Treatment II or Treatment I (ANOVA, p < 0.05).For Treatment III, the average H 2 S formed after four months was 28.7 µg/L, with some samples producing >100 µg/L H 2 S (Figure 2, top).By comparison, H 2 S formation in Treatment I and II wines averaged 2.2 µg/L (Figure 2, top).Similar differences were observed after eight months of product storage (Figure 2, bottom), which also showed that higher H 2 S formation is formed in the presence of high molecular SO 2 (Treatment III), and not by free SO 2 alone (Treatment II), in agreement with previous observations under accelerated aging conditions with a single wine and liner source (Montgomery et al. 2023).All H 2 S concentrations-particularly for Treatment III-were higher than those observed in bottled wine (≤6 µg/L) as well as the H 2 S sensory threshold (~1 µg/L) (Allison et al. 2021).Previously, it was speculated that the neutral molecular SO 2 would be able to diffuse through the nonpolar liners to react with the metal surface.However, as molecular SO 2 is a component of free SO 2 and its proportional contribution will be favored at low pH, it is thus not possible to distinguish the effect of molecular SO 2 from the interaction of [free SO 2 ] × [H + ].No significant correlation was observed between H 2 S accumulation and any of the other wine parameters (alcohol, TA, dissolved aluminum, residual sugar) reported in Supplemental Table 1 (one-way ANOVA, p > 0.05 for all tests).
The long-term aging study also allowed for comparison of H 2 S production across can liners for different wines.Results for Treatment III wines stored in the three liners are shown in Figure 3. Z2 cans produced lower H 2 S (average = 5.9 µg/L ± 2.1) than the same wines stored in X1 or Y2 cans (average = 36.6ug/L ± 3.2), as shown in Figure 3 (two-way ANOVA, p < 0.05).These differences are notable because Y2 and Z2 reportedly use the same liner material (Valspar V70) and differ only in their manufacturers.Furthermore, Z2 produced less H 2 S than X1, even though the latter used BPA epoxy, a liner typically considered to be the "gold standard" for beverage cans.Additionally, Can Z2 wines also had significantly lower variance than either Can X1 or Can Y2 wines (Levene's test, p < 0.05).Previous work on canned wines also reported high variation in H 2 S production among can replicates (relative standard deviation >50%); all cans in the previous study were from the same can manufacturer, and it was not possible to determine if variation arose from sample preparation, analytical characterization, or can-to-can differences (Montgomery et al. 2023).Other authors have noted that can-to-can variability may be a considerable source of variation in beverage can corrosion, potentially due to variation in liner thickness or cure quality (Grandle and Taylor 1997, Folle et al. 2008, Soares et al. 2019).This current work suggests this may also be important for explaining variation in H 2 S production; the specific role of liner thickness is discussed in greater detail later.
X1 cans produced the most H 2 S in the low SO 2 (Treatment I) control wines (ANOVA, p = 0.029; data not shown), and the amount of H 2 S in the Treatment I control group rose for each liner, from four to eight months (Figure 2).

H 2 S production from can headspace versus body
Previous work demonstrated that cans with higher amounts of H 2 S formation following long-term storage also had greater visible corrosion (Montgomery et al. 2023).Visible corrosion was not scored in the current study.However, we observed that the location of visible corrosion varied among can sources.Can X1 had visible corrosion mostly in the body of the can, with lesser amounts in the neck region of the can (not shown), but Y2 had considerable corrosion in the upper neck region (Supplemental Figure 5B).
To evaluate if the neck and body regions of a can had different susceptibility to corrosion, coupons were created from the headspace region of the neck (Figure 1, Locations 1 and 2) and the side walls of the body (Figure 1, Location 4) and were incubated in a commercial Riesling under accelerated conditions (three days at 50°C).The resulting H 2 S concentrations are shown in Figure 4. H 2 S production was highest for the headspace neck region of Y2, approximately three-fold higher than the body region, which agrees with the observed differences in visible corrosion.However, because the surface area of the body is ten-fold greater than the neck, these differences are not likely to explain differences in overall performance among can manufacturers.
To determine if H 2 S production was greater in non-immersed regions, an accelerated aging trial was performed with the location of aluminum coupon varied.Negligible H 2 S production was observed when aluminum was present only in the headspace, indicating that H 2 S generation (Equation 1) requires contact with the wine.Additionally, no increased visible corrosion or H 2 S production was observed when the coupon was fully immersed (Treatment III) versus partially immersed (Treatments I and II), as shown in Figure 5. Thus, the visible corrosion observed for Can Y2 is likely because the coating in this region is providing a less effective barrier (see Figure 4), and not because corrosion is accelerated in non-immersed regions, as compared to immersed regions.

H 2 S production during accelerated aging of lined aluminum coupons from multiple manufacturers
To further characterize the variation in H 2 S production among can manufacturers for similar liner types, accelerated aging testing was conducted with coupons produced from cans used in the long-term study discussed above (i.e., X1, Y2, and Z2), along with seven other liners, for a total of 10 liner treatments.For these experiments, a single, commercially available German Riesling with high molecular SO 2 was used.
H 2 S production under accelerated conditions across liners is shown (Figure 6).In agreement with previous work (Montgomery et al. 2023), acrylic liners generated much higher amounts of H 2 S (up to 100 µg/L) than the other liner types, with the exception of Y2-2 (Figure 6; Tukey test, p < 0.05).As with long-term aging, differences in H 2 S formation were observed among can manufacturers even for the same liner type.For example, Z2 produced less H 2 S than all three batches of Y2 (ANOVA, p < 0.05), and Z1 produced less H 2 S than X1 (ANOVA, p < 0.05).Similarly, there were significant differences in H 2 S production among acrylic coatings from different producers (Y3, W3), as well as significantly higher variation in H 2 S (Levene's test, p < 0.05).Finally, batch-to-batch variation was observed among Y2 cans, with approximately ten-fold higher H 2 S production in the visually defective batch (Y2-2) than in either the original batch (Y2) or thinner liner cans (Y2-3).The Y2-2 cans had clear visible defects in the moat of the can (see Supplemental Figure 6), which likely accounted for their poor performance.

Polymeric liner and aluminum surface characterization
Earlier work demonstrated that liner type critically affects H 2 S production during canned wine storage, with wines stored in the presence of acrylic liners forming greater than ten-fold more H 2 S than epoxy liners (Montgomery et al. 2023).However, both long-term and accelerated aging results of the current work showed that comparable variation can occur among cans with the same liner types sourced from different manufacturers.
For the three can sources used in the long-term study (X1, Y2, and Z2), variation in performance (both average H 2 S formation and can-to-can variation) was hypothesized to have arisen from one, or a combination, of aluminum alloy composition, polymeric liner composition, liner degree of cure, and liner thickness and uniformity.

Aluminum alloy composition
The aluminum alloy compositions of cans from BPA-NI epoxy producers Y and Z were determined by XRF spectrometry and are shown in Supplemental Figure 7.These two cans were selected because they used the same liner chemistry.The results are semiquantitative, because calibration curves were not run.The only element to show variation >20% among can sources was chromium (Cr), which was approximately three-fold higher in Z. Considering the low typical concentrations of Cr in the Al 3004 alloy (<0.05%) (United Aluminum 2023), variation in this element appeared unlikely to explain differences in H 2 S formation.Other transition metals detected in the alloys, especially Cu (in its soluble Cu(II) form), are well-known to form nonvolatile complexes or polysulfides following reaction with H 2 S, and could potentially limit accumulation of H 2 S if solubilized (Kreitman et al. 2019).However, variation in Cu and other elements (beyond Cr) were minor (<20%) and unlikely to explain the observed differences in H 2 S accumulation.

Liner composition by FTIR-ATR
Liners of cans used in the long-term storage experiment were characterized by FTIR, and spectra are shown in Supplemental Figure 8. Minor differences were seen between BPA epoxy (Can X) and the two BPA-NI epoxy (Cans Y2 and Z2) cans, but the liner used in Y2 and Z2 cans showed no visible differences.This latter result was expected because literature from the producers indicated both liners were Valspar V70 (tetramethyl bisphenol F) based coatings.Significant peaks were observed at ~1725, 1510, 1210, 1140, and 1030 cm -1 .

Aluminum surface smoothness, liner thickness, and liner uniformity
Enamel rating, EIS, copper sulfate rinses, and liner adhesion tests were performed to evaluate liner integrity.Metal exposure analysis ("enamel rating") uses two electrodes: one on the can exterior and one immersed in an electrolyte solution contained within the can.Application of a direct current (DC) potential (6.3 V) results in an electrical current (in milliamps [mA]) proportional to the extent of exposed aluminum (Sencon 2019).Manufacturers that use enamel ratings typically have quality control cutoffs based on the corrosivity of the beverage; for beer, a noncorrosive beverage, the recommended cutoff is 75 mA (Fetters et al. 2004), but a lower cutoff (5 mA) is recommended for highly corrosive beverages like wine (personal communication with an anonymous industry member).In this study, enamel rating currents were not significantly different among the three groups of cans used in the long-term aging study, and all measured values were <4 mA (Supplemental Figure 9).We attempted to evaluate the enamel rating approach by testing the previously mentioned Y2-2 cans with obvious visual defects in liner coverage within the "moat" of the can (Supplemental Figure 6).Surprisingly, only two of the 15 cans failed the enamel rater test, as indicated by the presence of a short-circuit, with the others having ratings <1.5 mA.This may be due to the formation of a thicker, nonconductive Al 2 O 3 layer on the exposed areas between the time of can manufacturing and their laboratory evaluation.At the least, these results suggest that enamel rating may not be as useful for end users (e.g., wineries) as they would be for can manufacturers.
EIS is similar to enamel rating except that an alternating current (AC) potential is applied instead of DC potential, and the resulting data modeled as one of several possible equivalent circuits (Lazanas and Prodromidis 2023).In our work, the EIS data was modeled as a resistor-capacitor (RC) circuit, and the impedance at low frequency (0.05 Hz) was evaluated because it is reported to correlate well with poststorage visual can corrosion and coating performance in a model corrosive beverage (3.5% NaCl, adjusted to pH 3 Figure 6 Hydrogen sulfide (H 2 S) production for 10 different can types, as well as the underside of the can lid, after three days in accelerated aging conditions with a commercial German Riesling.Three technical replicates for each type of can (represented on the x-axis) were tested.Error bars represent one standard error.The cans were coded with letters V to Z, signifying the can manufacturer, followed by 1 (BPA epoxy), 2 (BPA-NI epoxy), or 3 (acrylic), to signify the liner type.For two of the can sources, multiple production batches (n = 2 or 3) were tested to evaluate batch-to-batch variation (e.g., Y2-2 indicates it is the second batch of BPA-NI epoxy cans from manufacturer Y).
with acetic acid), as reported in Grandle and Taylor (1994).Average impedance values at 0.05 Hz were ~10 MΩ, which is above the minimum value recommended to limit corrosion (McIntyre and Pham 1996).However, no significant differences in either average values or variance were observed among the three can producers (Supplemental Figure 10).This observation was surprising because several reports have suggested that impedance values can be related to long-term can performance (Grandle and Taylor 1994, Kern et al. 1999, Hollaender 1997).One possible explanation is that the wine (especially the SO 2 in wine) degrades the liner during long-term storage-this behavior was highly evident visually for acrylic liners in previous work, and could also occur for epoxy-type liners (Montgomery et al. 2023).
In the current study, EIS measurements were performed within one hour of filling, but other reports have suggested looking for changes in EIS data over a longer time course (out to 14 days) to observe evidence of liner degradation (Grandle and Taylor 1994, de Vooys et al. 2012, Lu et al. 2017).This was not attempted in the current work but would be appropriate for further study.
Treatment of unused cans from the long-term aging study with acidified copper sulfate resulted in no deposition of Cu onto the internal surface of the can or aluminum sulfate precipitation, indicating that there were no voids in the liner of sufficient size to be apparent to the naked eye, i.e., <0.05 mm (data not shown).To validate the test, the Y2-2 cans with visual defects were also evaluated, but no displacement reaction could be observed.As with the enamel rating tests, considerable time may have passed between can manufacturing and liner testing, allowing for formation of a protective oxide layer.Similarly, no material was removed by the ASTM D3359-17 adhesion test (data not shown).
Liner thickness values were determined by performing laser scanning profilometry and analyzing a uniformly coated subsection of the aluminum (Table 1).The Z2 cans had a significantly thicker liner (3.27 µm ± 0.37 µm) than the Y2 cans (2.96 µm ± 0.35 µm).The liner thickness of X1 was intermediary (3.01 µm ± 0.43 µm) and did not differ from Cans Y and Z (Student's t-test, Figure 7).The observation that Z2 cans had slightly (~10%) higher average liner thickness than the other cans could be a potential explanation for the better performance of Z2 cans.However, based on profilometry, these same Z2 cans also had an initially higher proportion of exposed aluminum and thin liner coverage (<0.5 µm), as shown in Table 1.Thus, these latter parameters are presumably not responsible for the lower levels of H 2 S observed in Z cans.Manufacturer Z's BPA epoxy and manufacturer V's BPA-NI epoxy, both slim cans, produced very little H 2 S in the accelerated aging protocol (Figure 6) and had significantly thicker liners (Figure 7), despite a larger internal surface area, than the rest of the 10 liners analyzed (Student's ttests, p < 0.05).
A regression of inverse liner thickness versus H 2 S produced is shown in Figure 8, with acrylic liners and the defective Y2-2 cans excluded.The rate of SO 2 permeation and thus, H 2 S formation, was assumed to be proportional to the inverse of the liner thickness, and significant correlation was observed between the inverse of liner thickness and H 2 S. Additionally, the two liners that produced the least amount of H 2 S (Figure 6, Y2-3 and Z1) had the lowest liner thickness standard deviations (Figure 7), suggesting that a lack of variation in liner thickness (and not just average liner thickness) was important for minimizing H 2 S. Interestingly, the cans with the thinnest liner, Y2-3, produced intermediate amounts of H 2 S. All the cans studied, except for Y2-3, were slated for "hard-to-hold," beverages such as kombuchas, sour beer, wine, energy drinks, and ready-to-drink beverages, meaning a greater amount of liner was applied to the can, and more strict quality control checks (Enamel Rater, manufacturer-specific tests) were met than for typical cans.The Y2-3 cans are a "soda-weight" or "beer-weight" can with a smaller amount of liner applied, and a lower cure temperature, than cans for "hard-to-hold" products.
Although Company Y's acrylic liner (Y3) was the thickest measured (Figure 7), high H 2 S was also observed for this liner.Previous work demonstrated that SO 2 will degrade acrylic liners, resulting in more exposed aluminum and likely explaining the much higher levels of H 2 S observed in acrylic-lined cans (Montgomery et al. 2023).
Liner thickness was also evaluated by interferometry.This technique reports average liner thickness values for multiple 1 to 2 mm 2 areas within a can, but finer resolution at the micron level (as was performed with profilometry) was not available.Summary statistics for the three can sources used in long-term studies are reported (Table 2).The order of thickness (Z>X>Y) for both average and minimum values in the can body was the same as the order measured with profilometry.
The liner thickness on the underside of can ends, which rarely show visual signs of corrosion, was measured by examining a flat portion.The average thickness of can ends was >8 µm, roughly three-fold thicker than the can bodies.
Table 1 Percentage of sample surface area at different liner thickness cutoffs.The percentages are averages of three technical replicates.Samples were taken from Locations 1 to 5 as shown in Figure 1.X1, Y2, Z2: X, Y, and Z signify the can manufacturer; 1 and 2 signify the liner type (BPA epoxy and BPA-NI epoxy, respectively).For two of the can sources, multiple production batches (n = 2 or 3) were tested to evaluate batch-to-batch variation (e.g., Y2-3 indicates the third batch of BPA-NI epoxy cans from manufacturer Y).Liner thicknesses were also determined at five different locations throughout the can body and headspace for the three cans used in long-term aging studies.Measurements were performed by laser-scanning profilometry (Figure 1).The area of thinnest liner coverage (~2.5 µm) was in the top of the headspace (Location 1) of the Y2 cans (Figure 9).As reported above, the headspace of the Y2 cans generated more H 2 S than any other can location, which further supports the hypothesis that thinner liners will typically have a shorter onset time before measurable H 2 S production occurs, assuming all other factors are the same.

Conclusions
The molecular SO 2 fraction of sulfites was confirmed to be the best predictor of H 2 S formation during storage of canned wines for a range of wines and can liners.Considerable variation in H 2 S formation and visible corrosion occurs not only among different types of liner chemistry, but also for cans produced with the same liner material from different can manufacturers.Can-to-can variation was significantly higher for certain can manufacturers.Physical, optical, and mechanical tests of unused cans generally failed to predict performance during long-term storage, but differences in performance among cans were modestly correlated with differences in liner thickness.These observations, along with visible evidence of liner damage following storage, suggest that the chemical resistance of the liner to reactions with the wine (especially sulfites in wine), along with initial liner thickness, is critical to the stability of canned wines.The variation in liner performance identified in this work is likely to be relevant to other canned foods and beverages.Because the food or beverage may interact with the can liner, it is recommended to perform storage experiments with a can or coupon prior to evaluating the liner integrity.

Figure 1
Figure 1 Locations within can body sampled for liner and aluminum surface analysis: 1, top of neck, adjacent to seam; 2, tapered portion of can neck; 3, upper can body below the neck; 4, middle can body; and 5, lower can body.

Figure 2
Figure 2Hydrogen sulfide (H 2 S) production by treatment group and wine after four months (top) and eight months (bottom).Boxes represent range of H 2 S produced across three can liners, stored in duplicate (n = 6 points for each time point, wine, and treatment).↓ pH, ↓ fSO 2 , and ↓ mSO 2 , low pH, low free SO 2 , and low molecular SO 2 , respectively; ↑ pH, ↑ fSO 2 , and ↑ mSO 2 , high pH, high free SO 2 , and high molecular SO 2 , respectively.FR, French rosé; PG, Pinot grigio; RB, bubbly rosé; SB, Sauvignon blanc; WB, white bubbly.

Figure 3
Figure 3 Hydrogen sulfide (H 2 S) production by liner (X1, Y2, and Z2) for each of the five Treatment III (high molecular SO 2 ) wines after four months of storage.Error bars represent one standard error of three technical replicates.The asterisks indicate p < 0.05 (analysis of variance).X1, Y2, Z2: X, Y, and Z signify the can manufacturer; 1 and 2 signify the liner type (BPA epoxy and BPA-NI epoxy, respectively).For example, Y2 indicates a BPA-NI epoxy can from manufacturer Y.

Figure 4
Figure 4Hydrogen sulfide (H 2 S) production from can body (see Figure1, Location 4) and headspace (see Figure1, Locations 1 and 2) coupons from three different batches of cans (X1, Y2, and Z2), using the accelerated aging assay in triplicate (technical replicates), measuring at three and 14 days of storage.A negative control (no coupon) produced 0.5 µg/L H 2 S. Error bars represent one standard error.X1, Y2, Z2: X, Y, and Z signify the can manufacturer; 1 and 2 signify the liner type (BPA epoxy and BPA-NI epoxy, respectively).For example, Y2 indicates a BPA-NI epoxy can from manufacturer Y.

Figure 5
Figure 5 Dependence of hydrogen sulfide (H 2 S) formation on location of coated aluminum coupon (headspace [HS] versus immersed).Treatment 1, two coupons, one in headspace and one immersed; Treatment 2, one intact coupon, partially in headspace and partially immersed; Treatment 3, two coupons, both immersed; and Treatment 4, coupon in HS only.Three technical replicates were tested for each treatment.Error bars represent one standard error.A negative control group produced 0.5 µg/L H 2 S.

Figure 7
Figure 7Can body liner thickness from 10 different can types, as measured by laser-scanning profilometer (n = 3 cans per liner type).Error bars represent one standard deviation.The cans were coded with letters V to Z, signifying the can manufacturer, followed by 1 (BPA epoxy), 2 (BPA-NI epoxy), or 3 (acrylic), to signify the liner type.For two of the can sources, multiple production batches (n = 2 or 3) were tested to evaluate batch-tobatch variation (e.g., Y2-2 indicates it is the second batch of BPA-NI epoxy cans from manufacturer Y).

Figure 8
Figure 8 Hydrogen sulfide (H 2 S) formation after three days at 50°C versus inverse liner thickness for epoxy lined cans (BPA and BPA-NI).Each point represents the average H 2 S for a different liner (n = 8 technical replicates per liner).Error bars represent one standard error.

Figure 9
Figure 9 Liner thickness measurements by laser-scanning profilometry, measured across the cans (X1, Y2, and Z2).Three technical replicates were analyzed.Error bars represent one standard deviation.X1, Y2, Z2: X, Y, and Z signify the can manufacturer; 1 and 2 signify the liner type (BPA epoxy and BPA-NI epoxy, respectively).For example, Y2 indicates a BPA-NI epoxy can from manufacturer Y.

Table 2
Liner thicknesses (in µm) measured by laser interferometry for X1, Y2, and Z2 cans.Values are averages of multiple 1 to 2 mm 2 spots around the can interior.Four cans were analyzed at 24 spots in the can body and eight spots in the dome of each can.X1, Y2, Z2: X, Y, and Z signify the can manufacturer; 1 and 2 signify the liner type (BPA epoxy and BPA-NI epoxy, respectively).For example, Y2 indicates a BPA-NI epoxy can from manufacturer Y.