Measurement of Diacetyl and Related Compounds in Coffee Roasteries and Breweries

Abstract

α-Diketones such as diacetyl (2,3-butanedione) and 2,3-pentanedione are generated during the roasting and fermentation of foods and are also used as flavoring compounds. Exposure to these compounds has been associated with obliterative bronchiolitis in workers. We report indoor air concentrations of diacetyl and 2,3-pentanedione, as well as acetoin (3-hydroxy-2-butanone), in several small coffee roasteries and breweries using standard integrated air sampling sorbent tubes followed by gas chromatography tandem mass spectrometry as well as the first use of on-site continuous real-time proton-transfer reaction time-of-flight mass spectrometry (PTR-ToF-MS). Diacetyl and 2,3-pentanedione were detected in most of the sorbent samples at concentrations between 0.02 and 8 ppb_v, and in general were higher in coffee roasteries compared with breweries. Three integrated air samples, all from the barista area at one facility, exceeded the NIOSH recommended exposure limit (REL) of 5 ppb_v for diacetyl. 2,3-Pentanedione concentrations in these three samples were greater than 50% of its REL, but did not exceed it. Acetoin, a precursor to diacetyl, was also detected at concentrations between 0.03 and 5 ppb_v in most sorbent tube samples, with concentrations generally higher in breweries. PTR-ToF-MS measurements exhibited similar trends and provided continuous real-time volatile organic compound data that showed episodic excursions with peak concentrations of diacetyl and 2,3-pentanedione between 15 and 20 ppb_v. Examination of the time series data identified specific activities associated with peak diketone emissions, including transfer of freshly roasted coffee beans to the cooling tray, or the opening of a brew kettle. Additional indoor air quality parameters including CO₂, NO₂, and PM_2.5 were also assessed on-site. Airway inflammation was assessed in 19 workers before and after each work shift using online measurements of fractional exhaled nitric oxide (F_ENO). The pre-shift mean F_ENO was 3.7 (95% confidence interval: −3.6, 11.0) ppb_v higher and the post-shift F_ENO was 7.1 (−1.9, 16.1) ppb_v higher for workers at coffee roasteries compared with breweries. The cross-shift change in F_ENO was 3.4 (−2.8, 9.6) ppb_v higher for workers at coffee roasteries compared with breweries. However, none of these differences were statistically significant, and the cross-shift change in F_ENO was not statistically different from zero for either group of workers. The findings from this pilot study demonstrate that α-diketones and related compounds are present in the indoor air of both breweries and coffee roasteries and may exceed health protective guidelines in coffee roasteries. Additional studies are required to fully characterize worker exposures in these settings and to identify specific work activities and processes associated with high exposures. Engineering controls, including targeted exhaust ventilation and the use of low-cost sensors, are recommended as an approach to protect workers from exposure to hazardous levels of α-diketones.

What’s Important About This Paper?

α-Diketones such as diacetyl (2,3-butanedione) and 2,3-pentanedione are generated during the roasting and fermentation of foods and are also used as flavoring compounds. Exposure to these compounds has been associated with obliterative bronchiolitis in popcorn and coffee manufacturing workers. This study found concentration excursions were higher in coffee facilities compared with breweries. Within the coffee facilities, high concentrations were associated with roasting and grinding activities and were episodic, suggesting that strategic engineering controls can limit personal exposures to diacetyl and related compounds.

Introduction

Diacetyl and related α-diketones, such as 2,3-pentanedione, are added to food products as natural and artificial flavoring agents that impart a buttery taste and aroma (NIOSH, 2016). They are also present as natural byproducts in some fermented food products, including cheese, yogurt, miso, and beer (Ott et al., 1999, Frank et al., 2006, Giri et al., 2010, NIOSH, 2016), and in roasted food products such as coffee (Duling et al., 2016). Inhalation exposure to diacetyl and 2,3-pentanedione is associated with respiratory disease including obliterative bronchiolitis (Bailey et al., 2015), a severe, irreversible lung disease characterized by fixed airway obstruction.

In the early 2000s, cases of obliterative bronchiolitis were identified in diacetyl-exposed workers from a diacetyl manufacturing plant, several microwave popcorn manufacturing facilities (NIOSH, 2016), and in coffee processing facilities (Bailey et al., 2015, NIOSH, 2016, Harvey et al., 2021). Table 1 summarizes the 8-h time-weighted average (TWA) and short-term exposure limit (STEL) from the American Conference of Governmental Industrial Hygienists (ACGIH), the European Commission (EC), and the US National Institute for Occupational Safety and Health (NIOSH). The US Occupational Safety and Health Administration (OSHA) does not currently have a permissible exposure limit for these compounds. Although much research has focused on diacetyl-exposed workers in the food and flavor industry (Kreiss et al., 2002, Parmet and Von Essen, 2002, Akpinar-Elci et al., 2004, Kanwal et al., 2006), fewer studies have focused on small coffee roasteries and breweries. There are several recent publications that demonstrate coffee worker exposures to diacetyl and pentanedione in exceedance of the NIOSH RELs. In general, coffee grinding activities have been observed to give rise to higher concentrations than roasting. For example, the TWA 8-h diacetyl concentrations associated with coffee roasting and grinding were reported as 6.4 and 9.4 ppb_v, respectively (McCoy et al., 2017). The results of several studies are summarized in recent reviews of the literature (LeBouf et al., 2020, Echt et al., 2021).

Table 1.

Recommended time-weighted and short-term exposure limits for diacetyl and 2,3-pentanedione

	ACGIH		EC		NIOSH
	TWA 8 h	STEL 15 min	TWA 8 h	STEL 15 min	TWA 8 h	STEL 15 min
Diacetyl	10 ppbv	20 ppbv	20 ppbv	100 ppbv	5 ppbv	25 ppbv
2,3-Pentandione	—	—	—	—	9.3 ppbv	31 ppbv

Both integrated and continuous (direct reading) air sampling methods are common for measuring indoor and outdoor air. The two methods can provide complementary data sets (Davey et al., 2020), both of which have unique advantages and disadvantages. For example, integrated sampling that uses a sorbent tube combined with GC–MS analysis (LeBouf and Simmons, 2017) is both sensitive and selective providing quantitative measurements of individual volatile organic compounds (VOCs) on discrete samples. Continuous monitoring of VOCs using proton-transfer reaction time-of-flight mass spectrometry (PTR-ToF-MS) (Warneke et al., 2011, Yuan et al., 2017) provides sensitive (ppt_v) and mass selective measurements with frequencies as high as 1 Hz. These time-resolved measurements provide unique insights on the impact of specific activities and/or engineering control systems that mitigate exposure. However, PTR-ToF-MS methods are generally less selective than gas chromatography tandem mass spectrometry methods, as structural isomers will appear at the same mass/charge ratio and are therefore not distinguished.

The objective of this pilot study is to use PTR-ToF-MS to continuously measure the concentrations of α-diketones and a known diacetyl precursor (acetoin) in several small coffee and beer processing facilities. These compounds were simultaneously measured over the course of the workday to identify short-term concentration excursions and their relationship to specific activities. Simultaneously, we collected integrated samples over an intermittent 8-h period as well as a continuous 1-h period on sorbent tubes at several locations within these facilities for off-line analysis following OSHA methodology (OSHA 1013 [2008], OSHA 1016 [2010]). We also present data for auxiliary indoor air quality including methane, carbon dioxide, nitrogen oxides, and fine particulate matter (PM_2.5). Finally, workers were recruited to participate in exhaled breath monitoring for fractional exhaled nitric oxide (F_ENO), which is a marker of airway inflammation.

Methods

Method of workplace recruitment

Several craft coffee roasting and brewing business owners on Vancouver Island, BC, Canada were approached in person, via email, or through phone contacts. Participating facilities were selected based on location, size of operation, and interest level of owners/operators. Two coffee roasters (CR#1 and CR#2) and two breweries (BR#1 and BR#2) were selected to participate in the study. Details about the facilities (size, layout, sampling strategy, and number of employees) can be found in Supplementary Table S1 and Supplementary Figs S1–S8.

Study subjects

Workers were recruited from each of the facilities to participate in exhaled breath monitoring for F_ENO. Several days prior to the dates of sample collection, a one-page summary document describing the research study was sent to facility managers, who forwarded this information to their workers. On the morning of sample collection, workers who wished to participate in the exhaled breath sampling met with the research study coordinator and completed a consent form to participate in the study. The study procedures were reviewed and approved by the University of Washington Institutional review board (IRB ID: STUDY00002320). Between one and seven workers from each facility participated in the exhaled breath monitoring. F_ENO measurements were obtained from each subject immediately before they started the work-shift, and again at the end of the work shift. For two of the facilities, F_ENO measurements were obtained before and after two work shifts, whereas at the other two facilities only one pair of F_ENO measurements was obtained per worker.

Exhaled breath monitoring

Airway inflammation was assessed by measurements of F_ENO using a portable chemiluminescence analyzer (NIOX VERO; Morrisville, NC), according to manufacturer instructions. Quality control procedures were completed as specified by the manufacturer, which included (i) verification of instrument zero response consisting of a NO free gas sample automatically generated from ambient air and (ii) daily measurement of a positive control from one member of the research team with a stable F_ENO value within the normal biological range. In addition, subjects were instructed to avoid strenuous exercise or eating nitrate rich foods during the 3 h prior to F_ENO measurement.

Air sampling

We used two complimentary methods to measure diacetyl, 2,3-pentanedione, 2,3-hexanedione, 2,3-heptanedione, and acetoin in indoor air. The first method utilized sorbent tube samplers with subsequent GC–MS/MS analysis to measure time-integrated air concentrations. The second method employed was PTR-ToF-MS, which allows for direct continuous measurement for compounds of interest.

Sorbent tube sampling method

Full-shift 8-h TWA samples, designed to represent work shift average exposure, were collected at two to four locations within each facility. The locations were selected based on proximity to different tasks or activities within the facility. In addition, several 45- to 51-min short-term exposure samples were collected at each facility in conjunction with specific activities. For 8-h TWA samples, the pump was programmed to sample intermittently for 1 min in every 5 min across 8 h, to avoid breakthrough of the analyte on the sorbent tubes. For the short-term samples, designed to capture specific work activities, the pump operated continuously. These area samples allow for comparison of the standard sorbent tube method with the PTR-ToF-MS but do not represent the personal exposure of each employee. As such, exposure limits provide a benchmark against which the area samples may be compared, but the area samples do not represent actual exposures. Sample volumes ranged from 7.6 to 8.8 l for the short-term samples and from 6.9 to 9.4 l for the TWA samples.

Sampling for the target analytes diacetyl, acetoin (3-hydroxy-2-butanone), 2,3-pentanedione, and 2,3-hexanedione was undertaken using two precleaned silica gel sorbent tubes (7 × 110-mm, 600 mg, Cat. No. 226-183, SKC Inc., Eighty Four, PA) connected in series, as specified in OSHA 1012 (2008) and OSHA 1016 (2010). Air was drawn through the tubes using battery powered personal sampling pumps (PCXR, SKC Inc.) connected to the sorbent tubes via an adjustable low flow tube holder (SKC, Inc.), at a nominal flow rate of either 100 ml min⁻¹ (TWA samples) or 200 ml min⁻¹ (short-term exposure samples). Flow rates were set and verified in the field using a calibrated rotameter (SKC Inc.). Sorbent tubes were protected from light by wrapping with aluminum foil. Immediately after collection, samples were stored on ice in the field and transferred to a refrigerator at the end of each workday.

Sorbent tube analysis method

Three α-diketones and acetoin were measured by a GC–MS/MS method, with some modifications (OSHA 1013, OSHA 1016, LeBouf and Simmons, 2017, Echt et al., 2021). Sorbent material was removed from the glass tubes then extracted in 2 ml of 5% methanol/acetone spiked with internal standards diacetyl-d⁶ and 2,3-pentanedione-1,1,1,4,4-d⁵. The GC–MS/MS system (Agilent 7890B/7000, Santa Clara, CA) used for analysis was operated in multiple reaction monitoring mode with precursor and product ion transitions appropriate for qualitative and quantitative determination (Supplementary Table S2). Samples were analyzed in two batches, each within 17 days of sample collection (OSHA 1013, OSHA 1016). Mid-range calibration verification standards were analyzed throughout each batch (>1:20 injections), with acceptable recoveries for all compounds (77–103%). Assay contamination was assessed using laboratory and field blanks (sorbent tubes through which no air was drawn). Low levels of all four analytes were detected in laboratory blanks, ranging between 0.001 ng per sample (2,3-hexanedione) to 1.75 ng per sample (diacetyl). Analyte levels in the field blank samples were not significantly different from the levels in the laboratory blanks. Therefore, sample concentrations were corrected by subtracting the average of the laboratory blanks. Minimum detection limits (MDL) ranged from 1 ng per sample (2,3-hexanedione) to 2 ng per sample (diacetyl) and determined as the greater of the lowest calibration standard (after blank subtraction) or the laboratory blank. This MDL is equivalent to an air concentration of approximately 0.02–0.06 ppb_v assuming a 10-l air sample.

Real-time measurements

The Mobile Mass Spectrometer Lab (MMSL) housed in a research purposed Mercedes-Benz Sprinter van was equipped with a proton-transfer reaction time-of-flight mass spectrometer (PTR-ToF-MS, Ionicon PTR-TOF1000, IONICON Analytik, Innsbruck, Austria), an NO_x analyzer (model 42i, ThermoFisher), and a fast greenhouse gas analyzer (FGGA 30r, Los Gatos) for CO₂, CH₄, and H₂O was parked with the engine off within 3 m of an outside entryway to each sampling site and plugged in to building power. These instruments operate continuously and provide parallel time series data collected at 0.1–1 Hz. Fluorinated ethylene propylene (Cole Parmer) tubing with ¼” outer diameter (OD) was used for NO_x and greenhouse gas analysis. Sample lines were typically between 10 and 30 m in length depending on the proximity to the designated sampling point. The NO_x analyzer was calibrated by the vendor (CD Nova Ltd) immediately prior to deployment. The greenhouse gas analyzer was not calibrated immediately prior to deployment but reported ambient outdoor concentrations of 410–420 and 1.8–1.9 ppm_v for CO₂ and CH₄, respectively, consistent with government agency reported values for these well-mixed gases. An Optical Particle Sizer (model 3330, TSI Inc., Shoreview, MN) was employed inside each facility to count and size particles between 0.3 and 10 µm at 1-s intervals. PM_2.5 was calculated by summing particle size bins up to 2.5-µm diameter and reported in µg m⁻³ using vendor algorithm. Ambient outdoor conditions (relative humidity, pressure, and temperature) were recorded with an on-board weather station (Model 597, MetOne Instruments Inc., Grants Pass, OR) during sampling trips.

VOCs were continuously sampled using perfluoroalkoxy alkane tubing (¼” OD) insulated with foam pipe wrap and connected to the PTR-ToF-MS. A dedicated pump (Model DOA-P704-AA, GAST Manufacturing, Inc.) was used to pull sample air into the vehicle (~20 l min⁻¹), which was subsampled by the PTR-ToF-MS at 0.75 l min⁻¹ resulting in a sample transport time of approximately 7–13 s, depending on the location. PTR-ToF-MS measurements were made at 1 Hz and minimum detection limits (MDLs; ppb_v) were defined as the concentration, which had an average signal three times greater than the standard deviation of a blank, after background subtraction. The MDL for diacetyl was 0.3 ppb_v and for 2,3-pentanedione was 0.6 ppb_v. Although some progress has been made by others to differentiate isobars/isomers with PTR-ToF-MS (Shen et al., 2012), no effort was made to do so here. As such, time-resolved VOC concentrations reported should be considered semiquantitative due to potential interference from compounds with similar or identical m/z. The mass resolution (m/z/Δm/z) ranged from 1500 to 2000 at m/z 59. Calibrations are shown in Supplementary Fig. S9, and a list of the target analytes and potential isobars/isomers is presented in Supplementary Table S3.

A direct comparison between PTR-ToF-MS and GC–MS/MS proved challenging due to differences in the sampling methods, the time scale of the measurements themselves, and difficulties in constructing a well-mixed air standard. The relative standard deviation for the replicate constructed air samples were 4.3 and 20% for the PTR-ToF-MS and GC–MS/MS analysis, respectively (Supplementary Table S4). There is good correlation between the two sets of measurements, with a R² of 0.88 for diacetyl and a R² of 0.84 for pentanedione for colocated samples across all facilities (Supplementary Fig. S10). In general, the PTR-ToF-MS values were higher (~30%) than the colocated sorbent tube concentrations in the field measurements. This positive bias may have resulted from the presence of VOCs with the same m/z as the target analytes.

Results

The results of a total of eight on-site sampling days at coffee and brewery locations revealed indoor air concentrations of α-diketones above the MDLs. Observations were made over three consecutive days at one each of the coffee and brewery locations to capture the full range of operational activities. Diketone levels were observed to be strongly dependent on sampling location and specific activities within the facilities.

Sorbent tube measurements

The 8-h sorbent tube measurements are summarized in Table 2, which indicates that diacetyl and 2,3-pentanedione were detected in most of the sorbent samples at concentrations between <0.04 and 8 ppb_v. Acetoin was also detected at concentrations between 0.03 and 5 ppb_v in most sorbent tube samples. Figure 1 is a box plot summarizing the sorbent tube results for both the TWA samples and the short-term samples, across the four facilities surveyed in this study. Overall, the TWA levels of these compounds were well below occupational exposure guidance values in these facilities, although diacetyl and 2,3-pentandione were observed to be elevated at one of the coffee roasters. Specifically, over the 3-day sampling campaign, the average concentrations near the barista bar at CR#1 were 7.1 and 7.0 ppb_v for diacetyl and 2,3-pentanedione, respectively. A detailed summary of the sorbent tube results is tabulated in Supplementary Tables S5 and S6.

Table 2.

Summary sorbent tube measures of diacetyl and related compounds in breweries and coffee roasteries (concentrations in ppb_v)

8-h TWA	Diacetyl	2,3-Pentanedione	Acetoin	2,3-Hexanedione
Coffee roasteries (# samples = 15)
Fraction > MDL	15/15	15/15	15/15	0/15
Mediana	0.87	0.63	0.13	—
Mean (SD)	1.06 (2.69)b	0.79 (2.72)b	0.23 (1.64)	—
Range	0.19–7.88	0.14–8.01	0.04–4.62	—
Breweries (# samples = 12)
Fraction > MDL	10/12	5/12	12/12	0/12
Mediana	0.28	0.07	0.36	—
Mean (SD)	0.23 (0.28)b	0.08 (0.03)b	0.34 (0.25)	—
Range	<0.07–0.99	<0.04–0.14	0.06–0.97	—

^aOnly samples > MDL included in calculation of mean, median, and SD.

^bMeans differ significantly (P = 0.05) for brewery versus coffee roaster (based on independent samples t-test).

Figure 1.

Comparison of (a) diacetyl, (b) 2,3-pentanedione, and (c) acetoin concentrations across the four facilities surveyed. 2,3-Hexanedione was only detected in three samples (all at facility CR#1) and hence is not included in these plots. 2,3-Pentanedione was not detected at facility BR#2. Data <LOD treated as missing. On each box, the central mark indicates the median, and the bottom and top edges of the box indicate the 25th and 75th percentiles. Observations above the 95th percentile are plotted using the ‘o’ symbol. Note that the y-axis is truncated to permit visual comparison across facilities: three diacetyl samples were above 2 ppb_v and are not included here (see Supplementary Table S5 for all data).

PTR-ToF-MS measurements

Time series data collected at CR#1 shows diacetyl and 2,3-pentanedione concentrations were highly correlated (R = 0.84, Supplementary Fig. S10b), with each other over the 9-h observation period (Fig. 2). The top panel (Fig. 2a) identifies three broad activity categories as roasting, cleaning, and grinding. During roasting, numerous concentration excursions are evident which generally correspond to when the freshly roasted beans were dumped from the roaster onto cooling trays. During cleaning, the roaster, exhaust vent, and surrounding area were cleaned with brushes or brooms. During grinding, roasted beans from storage containers were ground and packaged in conventional coffee bags with valves. The air inside the facility was not well mixed, and the sampling line for continuous measurements was in the center of the facility closer to the roasting than the grinding operations (Supplementary Fig. S2). Figure 2b is an expanded view of the time series during the highest concentrations observed between 10 and 11 am, during which time operators processed five batches of espresso. Activities during this time included the use of a non-ventilated mini-roaster, roasting and cooling of coffee beans, and grinding of roasted beans.

Figure 2.

Time series of diacetyl and 2,3-pentanedione concentrations measured at CR#1 on 20 February 2018. Top panel is the time series for the whole day, and gray shading indicates data shown in the bottom panel. The green dashed line displays the NIOSH time-weighted average REL for diacetyl at 5 ppb_v. Bottom panel shows a 1-h period when five batches of espresso beans were processed.

An example of PTR-ToF-MS time series for diacetyl from a brewery (BR#1) is shown in Fig. 3. Broad activity categories are identified as follows: lautering (separating wort from residual mashed grain), boiling, and transfer of wort (the liquid extracted from mashing the grain). One concentration excursion was observed during lautering when the mash tun was temporarily opened (Fig. 3,ii). The highest concentration excursions were observed during boiling when the kettle lid was opened for a significant amount of time (~1 h) to prevent overboiling (Fig. 3,iv). During transfer, the wort was cooled and pumped to large fermenting tanks. No significant concentration excursions were observed during this stage of the process.

Figure 3.

Time series of diacetyl concentration during brewing at BR #1 on 26 February 2018. The green dashed line displays the NIOSH time-weighted average REL for diacetyl at 5 ppb_v.

Continuous α-diketone measurements revealed median concentrations for diacetyl and 2,3-pentanedione were typically below 2 ppb_v (Supplementary Table S5) and summarized in Fig. 4. However, it is clear in the time-resolved PTR-ToF-MS data that concentration excursions up to 15–20 ppb_v can occur in indoor air and that these elevated levels were strongly corelated with specific activities inside the facilities. Diacetyl and pentanedione had measured concentration excursions above the MDL at each facility on each sampling day.

Figure 4.

PTR-ToF-MS results for diacetyl, 2,3-pentanedione, 2,3-hexanedione, and acetoin concentrations at breweries and coffee roasteries. On each box, the central mark indicates the median, and the bottom and top edges of the box indicate the 25th and 75th percentiles. Observations above the 95th percentile are plotted using the ‘o’ symbol.

Three full workdays (~8 h day⁻¹) of sampling were collected for CR#1, and one full workday of sampling was collected at CR#2. On the first day of sampling at CR#1 (CR#1a), there was no coffee roasting taking place. Both the second (CR#1b) and third (CR#1c) days of sampling had active roasting for approximately 5.5 h (typically 7:30 am–1 pm). The maximum and median concentrations for diacetyl and pentanedione measured at CR#1 were higher on the days of active roasting. The median concentration for diacetyl on roasting days at CR#1 (1.30 ppb_v) was twice that of CR#1 on a non-roasting day (0.67 ppb_v). Furthermore, the number of observations with diacetyl greater than 5 ppb_v was substantially higher on days when roasting occurred. Very similar observations are made for pentanedione with median concentrations of 0.45 ppb_v on the non-roasting day and 1.24 ppb_v on days when roasting occurred. During active coffee roasting, the median concentration for acetoin at CR#1b was 0.10 ppb_v. The second coffee roastery (CR#2) consisted of a single roaster in a building with a high ceiling and modern ventilation. The main activity was roasting and the median diacetyl and pentanedione concentrations were 0.58 ppb_v and 0.81 ppb_v, respectively.

Three full workdays (~8 h day⁻¹) of sampling were collected for BR#1, and one full workday of sampling was collected at BR#2. The brewing activity at BR#1 was similar on day 1 (BR#1a) and day 3 (BR#1c), with median concentrations for diacetyl of 0.98 and 0.69 ppb_v, respectively. On day 2 (BR#1b), the main activity was canning with minimal brewing activity. The median diacetyl concentration was at or below our detection limit of 0.3 ppb_v on that day. The second brewery (BR#2) was a larger facility (see Supplementary Table S1 and Supplementary Fig. S5), which did not require opening of the kettles to prevent over-boil as in BR#1. The median diacetyl and pentanedione concentrations were at or below our detection limits at this facility. Hexanediones were generally observed to be below 2 ppb_v, except at the BR#2, where several observations between 2 and 4 ppb_v were made. Heptanediones were not generally observed in indoor air across this sampling campaign. The maximum and median acetoin concentrations were higher at breweries than coffee roasting facilities, as shown in Figure 4.

Additional indoor air quality measurements

Continuous measurements of NO, NO₂, CH₄, CO₂, and PM_2.5 were made at all locations over the 8-day sampling period. Summary boxplots are presented in Fig. 5 and representative time series are presented in Supplementary S11, S14, S15, S17, and S18. Elevated concentrations NO, NO₂, and CH₄ were observed during roasting at CR#1 (CR#1b and CR#1c) but remain well-below occupational exposure limits. Elevated PM_2.5 was observed at CR#1 during roasting, consistent with visually observable smoke released when roasted beans were dumped onto the cooling tray. At CR#1, the 8-h TWA PM_2.5 concentrations, measured on two separate roasting days, were 30 and 31 µg m⁻³. An example time series showing coincident periodic concentration fluctuations for PM_2.5 and diacetyl is shown in Supplementary Fig. S11. At the brewery BR#1, elevated CO₂ concentrations (median = 432 ppm_v, max = 4614 ppm_v) were observed during the day when canning was the main activity (BR#1b). At the brewery BR#2, elevated CO₂ concentrations were also observed (median = 448 ppm_v, max = 6866 ppm_v) but could not be attributed to a specific activity. While individual observations occurred above 5000 ppm_v, none of the CO₂ measurements exceeded TWA or STEL exposure limits.

Figure 5.

Comparison of NO, NO₂, CH₄, CO₂, and PM_2.5 concentrations between breweries and coffee roasteries. On each box, the central mark indicates the median and the bottom and top edges of the box indicate the 25th and 75th percentiles. Observations above the 95th percentile are plotted using the ‘o’ symbol. Concentration excursions that extend beyond the y-axis limits for PM_2.5 are indicated with an asterisk and the maximum concentration is reported.

F_ENO measurements

We measured pre-shift and post-shift F_ENO in the workers from the breweries and coffee roasteries. These results are summarized in Table 3 and Fig. 6. Pre-shift F_ENO levels ranged between 8 and 38 ppb_v, while post-shift levels ranged between 7 and 51 ppb_v. We used linear mixed models to compare pre-shift F_ENO, post-shift F_ENO and cross-shift change in F_ENO between breweries and coffee roasteries, controlling for individual facilities as a random effect. The pre-shift mean F_ENO was 3.7 ppb_v higher (95% confidence interval: −3.6, 11.0) and the post-shift F_ENO was 7.1 ppb_v higher (−1.9, 16.1) for workers at coffee roasteries compared with breweries. The cross-shift change in F_ENO was 3.4 ppb_v higher (−2.8, 9.6) for workers at coffee roasteries compared with breweries. Based on the likelihood ratio test, none of these differences were statistically significant, and the cross-shift change in F_ENO was not statistically different from zero for either group of workers.

Table 3.

Summary measures of F_ENO in breweries and coffee roasteries (concentrations in ppb_v)

	Pre-shift	Post-shift	Cross-shift change
Breweries (# subjects = 12, # sample pairs = 16)
Median	23	12	7.7
Mean (SD)	22 (11)	22 (12)	0.1 (7.7)
Range	8 to 38	7 to 43	−21 to 10
Coffee roasteries (# subjects = 7, # sample pairs = 10)
Median	24	28	9.0
Mean (SD)	26 (8.0)	29 (12)	3.5 (9.0)
Range	17 to 37	12 to 51	−10 to 19

SD, standard deviation.

Figure 6.

Comparison of pre-shift F_ENO, post-shift F_ENO and cross-shift change in F_ENO between breweries and coffee roasteries. On each box, the central mark indicates the median, and the bottom and top edges of the box indicate the 25th and 75th percentiles.

Discussion

Acetoin, diacetyl, and 2,3-pentanedione were measured at concentrations ranging from <0.04 to 8 ppb_v using time-integrated sorbent sampling techniques. Concentrations varied depending on sampling locations within the facility and type of activities taking place. In general, diacetyl and 2,3-pentadione were higher in coffee roasteries compared with breweries. Acetoin was generally higher in breweries. Three samples, all from the barista area at one coffee roastery, exceeded the NIOSH recommended exposure limit (REL) of 5 ppb_v for diacetyl. 2,3-Pentanedione concentrations in these three samples were greater than 50% of its REL but did not exceed it.

The time-resolved PTR-ToF-MS measurements followed a similar trend to the sorbent tube data and revealed peak concentrations of diacetyl and 2,3-pentanedione between 15 and 20 ppb_v. Examination of the time series data identified specific activities that were associated with peak diketone and acetoin emissions, including transfer of freshly roasted coffee beans to the cooling tray, or opening of a brew kettle at a craft brewery. In general, diacetyl concentrations were the highest of the four diketones measured here, with concentrations decreasing for the higher homologs. The background indoor concentration was less than 2 ppb_v for diacetyl and less than 1 ppb_v for all other diketones monitored at both coffee and craft brewing sites during periods with little or no roasting or brewing activities. For example, Supplementary Fig. S16 shows diketone concentrations below 1 ppb_v over the course of a non-roasting workday except for a short transient excursion during the brief transfer of several kilograms of roasted coffee beans. This observation is supported by the low concentrations of diketones reported for integrated sorbent tube samplers. Similar observations were made on Day 2 at BR#1 where ambient background diketone levels of less than 1 ppb_v (Fig. 4) were observed. However, during active roasting and brewing days, we observe significant concentration excursions ranging from 5 to 20 ppb_v for diacetyl and 3 to 12 ppb_v for 2,3-pentanedione. The ratio of diacetyl to 2,3-pentandione is roughly 1.6–1.7 in the coffee roasting facility. Very little increase in the concentration of the hexane and heptane diones was observed. In one of the breweries, concentrations of diacetyl were observed to increase to 6–20 ppb_v associated with opening a kettle lid when boiling wort (Fig. 3). Similar concentration time series were observed for 2,3-pentanedione, 2,3-hexanedione, and acetoin, although at considerably lower concentrations. The ratio of diacetyl to 2,3-pentandione is roughly 7–8 in the brewing facility.

The highest diacetyl concentrations were measured at CR#1 and BR#1, with the highest measured concentration below the 15-min REL of 25 ppb_v. At CR#1, concentrations up to 21 ppb_v of diacetyl, 14 ppb_v of 2,3-pentanedione, and 1.2 ppb_v of 2,3-hexanedione were measured. These excursions were short duration (<5 min) with concentrations approaching background levels after each batch was released from the cooling tray (Fig. 2 and Supplementary Fig. S11). Concentration excursions were also measured at the brewery BR#1 on the first and third day of sampling when the brew master opened the mash kettle for process monitoring and to prevent overboiling (Fig. 3). The highest acetoin concentrations were recorded at BR#2 (Fig. 4), but were not associated with any specific activity, presumably because of the numerous concurrent processes occurring throughout the day.

On-site survey measurements in which the sampling inlet for the PTR-ToF-MS was moved to different locations revealed elevated diketone concentrations associated with roasted bean storage and grinding activities. For example, the time series data in Supplementary Fig. S12 illustrates diketone concentrations exceeding 2500 ppb_v (primarily 2,3-butanedione and 2,3-pentandione) in the headspace of bulk storage bins of roasted coffee beans. A similar reconnaissance survey of potential point sources at a craft brewing facility did not reveal notable ‘hot spots’ for the target analytes.

The diacetyl concentrations we measured at the coffee roasteries are similar to those of Echt et al. (2021) who observed some 8-h TWA exposures above the NIOSH REL of 5 ppb_v for workers involved in coffee bean roasting and grinding at a small-scale craft coffee facility. Pierce et al. (2015) reported personal diacetyl exposures in the range 13–16 ppb_v for a barista grinding and brewing coffee in a residential kitchen—a study designed to simulate potential exposures experienced by baristas and customers in a small commercial coffee shop. In contrast, higher diacetyl and 2,3-pentanedione concentrations (8–100 ppb_v) were reported by Bailey et al. (2015) and Duling et al. (2016) from a larger commercial coffee processing facility. A summary of diacetyl measurements made in coffee production facilities has recently been published (LeBouf et al., 2020).

There are limited data in the literature describing exposures to diacetyl and 2,3-pentanedione in association with brewing of beer. In the only report we found, Eversmeyer (2016) reported diacetyl levels < 0.003 ppb_v (TWA) and < 0.006 ppb_v (STEL) for two breweries. Our measurements are somewhat higher than those of Eversmeyer but still well below exposure guidelines and suggest that exposures to diacetyl and related compounds in breweries do not exceed health protective levels; however, short-term RELs may be exceeded in association with specific processes (e.g., opening the kettle when wort is boiled; see Fig. 3).

Continuous on-site monitoring

Auxiliary air quality parameters (NO, NO₂, CO₂, CH₄, and PM_2.5) were measured continuously over 8 days of an 8-h work shift to capture various activities at coffee roasting and craft brewing operations. Background levels were recorded before the start of the workday and found to be similar to ambient outdoor concentrations. Fine particulate matter (PM_2.5) was observed to be below 50 µg m⁻³ on non-roasting and brewing days. However, short-term excursions > 300 µg m⁻³ were observed when roasted coffee beans were transferred to the cooling trays at one of the roasting facilities. These excursions are highly episodic and observed to be strongly correlated with diketone concentration excursions (Supplementary Fig. S11). PM_2.5 was generally <75 µg m⁻³ at brewery facilities but was observed in the range of 100–400 µg m⁻³ for periods of several hours associated with the transfer of grains and other materials. Elevated levels of nitrogen oxides (NO and NO₂) were observed during roasting when natural gas furnaces were operational, either for heating coffee roasters or the mash tun in brewing. For example, NO concentrations ranging from 50 to 120 ppb_v and NO₂ levels ranging from 50 to 80 ppb_v were observed during roasting and dropped sharply to levels below 15 ppb_v once the roasting furnaces were turned off (Supplementary Fig. S14). Levels of NO and NO₂ at breweries were generally less than 40 ppb_v with some short excursions greater than 40 ppb_v (Supplementary Fig. S15). Measured concentrations of PM_2.5, NO₂, and NO were all well below OSHA permissible exposure limits. However, the 8-h TWA PM_2.5 concentrations measured on roasting days at the coffee facility CR#1 (up to ~30 µg m⁻³) do exceed the WHO 24-h PM_2.5 guideline of 15 µg m⁻³. Elevated levels of carbon dioxide (CO₂) and methane (CH₄) were observed during coffee roasting reaching levels of 700 ppm_v and 40 ppm_v, respectively. The concentration of both gases dropped to ambient levels once roasting furnaces were shut off (Supplementary Fig. S17). Concentrations of CO₂ at one brewery were between 500 and 1800 ppm_v and between 2 and 6 ppm_v for CH₄. Concentration spikes were largely associated with opening fermentation tanks and bottling activities. Carbon dioxide levels were generally higher at the larger brewery reaching levels of 5000 ppm_v (which is the 8-h TWA NIOSH REL) late in the afternoon (Supplementary Fig. S18). Given these observations and the emergence of low-cost gas sensors (especially for CO₂ and PM_2.5), we recommend that facility operators consider their use to inform ventilation requirements and protect worker safety.

Additional field studies are required to strengthen the findings of this study, which was limited to a small number of facilities (2 roasteries and 2 breweries) that were also relatively small operations. Measurements were only made during one pilot study period, which does not account for potential seasonal variability. Furthermore, the highest diacetyl concentration recorded was at the barista station at CR#1, where only one sample was collected (n = 1). The relationship between PM2.5 and diacetyl in coffee roasting facilities requires further confirmation at additional sites.

Comparison of measured F_ENO levels to ATS guidelines

None of our study participants had elevated F_ENO, defined by ATS as F_ENO > 50 ppb_v (Dweik et al., 2011), and we did not observe any association between F_ENO levels and occupational exposure. In contrast, a recent study from Harvey et al. (2020) reported that 10% of coffee production workers had elevated F_ENO levels. Average diacetyl exposures measured in Harvey’s study were much greater than concentrations of diacetyl measured in our study. A cross-sectional study of workers in a Tanzanian coffee factory that processed and packed green coffee beans also found that workers had higher F_ENO levels than unexposed controls (Moen et al., 2012). A subsequent study reported coffee production workers had higher prevalence of chronic respiratory and asthma symptoms than workers in two control factories that produced beverages and fish products; F_ENO levels were marginally higher in coffee workers compared with controls (Sakwari et al., 2013).

Conclusion

In this pilot study in relatively small-scale coffee roasteries and breweries, we found that diacetyl and 2,3-pentanedione concentrations were generally below the NIOSH RELs. The only exception to this was the 8-h average concentrations measured near the barista bar where average air concentrations of 7.1 and 7.0 ppb_v for diacetyl and 2,3-pentanedione, respectively, were observed over the 3-day sampling campaign. Similarly, F_ENO levels were not elevated with respect to ATS guidelines, and no association between occupational exposures and changes in F_ENO was observed.

The continuous monitoring capability of the PTR-ToF-MS highlighted the episodic nature of diketone exposures in these facilities and aided in identifying specific tasks associated with these excursions. Elevated concentrations of diketones had strong association with roasting and grinding within coffee facilities. In brewing operations, the potential for exposure to diketones appears to be coincident with opening kettles during boiling of wort. This information can be used to inform when and where better ventilation and/or personal protective equipment might be warranted. Possible short-term exposure to high concentrations of diketones can also occur when employees enter confined coffee storage rooms and when opening storage bins of roasted coffee beans.

This report adds to the growing body of literature reporting diketone concentrations that approach or exceed RELs in coffee production facilities. We demonstrate the first use of on-site PTR-ToF-MS at coffee and brewing facilities to directly measure air concentrations of several diketone compounds of concern over a multi-day sampling campaign. The continuous VOC monitoring capabilities of PTR-ToF-MS are especially informative when surveying sites and relating short-term concentration excursions to specific activities. While fine particulate matter may also be an occupational health concern in its own right, its association with diacetyl concentrations in coffee roasting facilities suggests that it may be a suitable proxy for diketone exposure in this setting. The indoor installation of low-cost sensors for CO₂ and PM_2.5 could be used to monitor ventilation and provide useful operational information.

Funding

Funding and support for this project have been provided by the National Institute of Environmental Health Sciences (grant number P30ES007033). The content is solely the responsibility of the authors and does not necessarily represent the views of National Institutes of Health. The authors also gratefully acknowledge Vancouver Island University and the University of Victoria for ongoing support of the Applied Environmental Research Laboratories and Graduate Student Researchers. Support for LR and JD was also provided through the Natural Sciences and Engineering Research Council of Canada (NSERC Discovery Grant 2016-06454). The mobile mass spectrometry lab was made possible with funding from the Canada Foundation for Innovation (CFI32238) and the British Columbia Knowledge Development Fund.

Conflict of interest

The authors declare no conflict of interest relating to the material presented in this article. Its contents, including any opinions and/or conclusions expressed, are solely those of the authors.

Data availability

The data underlying this article will be shared on reasonable request to the corresponding author.