Large volume headspace GC/MS analysis for the identification of volatile compounds relating to seafood decomposition

Abstract Decomposition in seafood products in the United States is monitored by the Food and Drug Administration (FDA) laboratories using sensory testing, which requires highly trained analysts. A large‐volume headspace (LVHS) gas chromatography/mass spectrometry (GC/MS) method was developed to generate analytical results that can be directly compared to sensory evaluation. Headspace vapor was withdrawn from a 1‐L vial containing 50 g seafood sample using a large volume headspace autosampler. Various volatile compounds were collected simultaneously. Analytes were preconcentrated by a capillary column trapping system and then sent through a cryo‐focuser mounted onto the GC injector. A selected ion monitoring (SIM) MS acquisition method was used to selectively monitor 38 compounds of interest. Samples of red snapper, croaker, weakfish, mahi‐mahi, black tiger shrimp, yellowfin tuna, and sockeye salmon that have been assessed and scored by an FDA National Seafood Sensory Expert (NSSE) were used for method performance evaluation. Characteristic compounds potentially associated with seafood quality deterioration for each seafood species were identified by quantitative analysis using pooled matrix‐matched calibrations and two‐sample t‐test statistical analysis. Classification of fresh and decomposed samples was visualized on the analysis of variance (ANOVA)–principal component analysis (PCA) score plots. The results determined that the LVHS‐GC/MS technique appeared promising as a screening tool to identify compounds representative of sensory analysis.

1-L vial containing 50 g seafood sample using a large volume headspace autosampler.
Various volatile compounds were collected simultaneously. Analytes were preconcentrated by a capillary column trapping system and then sent through a cryo-focuser mounted onto the GC injector. A selected ion monitoring (SIM) MS acquisition method was used to selectively monitor 38 compounds of interest. Samples of red snapper, croaker, weakfish, mahi-mahi, black tiger shrimp, yellowfin tuna, and sockeye salmon that have been assessed and scored by an FDA National Seafood Sensory Expert (NSSE) were used for method performance evaluation. Characteristic compounds potentially associated with seafood quality deterioration for each seafood species were identified by quantitative analysis using pooled matrix-matched calibrations and twosample t-test statistical analysis. Classification of fresh and decomposed samples was visualized on the analysis of variance (ANOVA)-principal component analysis (PCA) score plots. The results determined that the LVHS-GC/MS technique appeared promising as a screening tool to identify compounds representative of sensory analysis.

K E Y W O R D S
large volume headspace GC/MS, pooled calibration curves, seafood, volatile compounds Chemical indices of decomposition can provide a significant support mechanism to sensory findings in some seafood products.
The FDA has established criteria for analyzing histamine (AOAC Official Method, 1987) and indole (AOAC Official Method, 1982) in seafood to support regulatory action in the absence of sensory evidence in some cases. Other potential chemical indicators of decomposition in specific seafood products may also have utility in determining the decomposed state of seafood (Boziaris & Parlapani, 2017;Joffraud et al., 2001). Several instrumental methods for analyzing chemical indices of seafood freshness have been proposed (Bai et al., 2019;Chan et al., 2006;Duflos et al., 2005;Self et al., 2018). However, these methods always use only a small portion (e.g., 2-10 g) of a whole fillet, which may not accurately represent an entire fish sample. For instance, decomposition caused by oxidation usually occurs in the fish abdomen while bacterial growth would be the main source of deterioration in fish back meat (Khoshnoudi-Nia & Moosavi-Nasab, 2019; Wang et al., 2020). Thus, sampling bias may occur when a very small amount of sample is collected from a single region of a fish filet. In addition, advanced chromatography-mass spectrometry techniques often require sample pretreatment, such as extraction or derivatization, to achieve higher sensitivity. But sample pretreatment does not always preserve the original proportion and integrity of volatile compounds, so measurement bias may occur when evaluating the agreement between instrumental analysis and sensory testing. Furthermore, previous studies tend to focus on one or two seafood species and use laboratory-based samples that were prepared in-house under controlled conditions. For these reasons, the present work was undertaken to develop a large volume headspace (LVHS)-gas chromatography/mass spectrometry (GC/MS) method, as a nonsensory analytical technique, for the analysis of seafood decomposition in a way that can be directly compared to FDA NSSE sensory evaluation scores. By using seven types of FDA NSSE graded wild-caught seafood products at known organoleptic states, the reliability of LVHS-GC/MS was evaluated.
Potential marker compounds indicative of decomposition for each seafood species were identified. Analytical challenges encountered in headspace analysis of volatiles are also discussed.

| Sample information
Seafood samples used for method development, optimization, and evaluation were fresh and unprocessed Atlantic salmon, tilapia, and cod purchased from local supermarkets in the Washington D.C. area.
Samples were stored at −60°C upon arrival at the laboratory.
Seafood samples for investigation were provided by the FDA Pacific Northwest Laboratory in Washington (Self et al., 2018).
Frozen, unprocessed fillet portions of seven seafood species (red snapper, croaker, weakfish, mahi-mahi, black tiger shrimp, yellowfin tuna, and sockeye salmon) were collected from Guyana (red snapper, croaker, and weakfish), Ecuador (mahi-mahi), Vietnam (black tiger shrimp and yellowfin tuna), and Alaska (sockeye salmon). Each portion was individually evaluated by FDA NSSE near the catch locations, vacuum packed, flash-frozen, and transported to FDA.
A sensory score on a 100-point scale was given to each portion by FDA NSSE, where "0" represents the best quality and "100" represents the worst quality. Scores between 0 and 50 were passing (nondecomposed) and scores between 51 and 100 were failing (decomposed). In the current study, samples with sensory scores between 15 and 25 were considered to be "fresh"; samples with sensory scores higher than 70 were considered to be "decomposed." Sensory scores and odor characteristics are described in Table 1.

| Sample preparation
All seafood samples were individually vacuum sealed and kept frozen at −60°C prior to use. In the Figure S1 shows that a typical fish filet in this study included both back and abdomen regions. Once a deep-frozen sample was slightly thawed, inedible skin or shell was removed. The edible parts were placed in a Robot Coupe R401B single-speed food processor and ground for 1 min while still mostly frozen. Aliquots of the ground seafood (50.0 ± 0.1 g) and drying agent Na 2 SO 4 (35.0 ± 0.1 g) were placed in a 1-L glass headspace vial. Figure

| Headspace sampling
Prior to headspace sampling, each 1-L headspace vial was incubated at 30°C for 30 min with agitation. After incubation, an aliquot of headspace vapor (50 ml) was withdrawn directly from the 1-L vial through the Silonite™ Male Micro-QT™ valve mounted on the top of the vial cap ( Figure S1C).
Collected headspace vapor went through a capillary column trapping system (CTS) to eliminate water and air. Vapor sample size was then reduced from 50 ml to 1 µl. Concentrated analytes went through a cryo-focuser mounted onto GC injector.
Liquid nitrogen focused the analytes onto the GC column (Wilson et al., 2012;Wylie, 1986), which significantly improved peak shape and resolution.
The entire headspace sampling process was automatically performed using a 7650HS-CTS analyzer (Entech Instruments).
Optimization of experimental parameters can be found in the   Figures S2-S5 and Table S1 and S2. This headspace autosampler system has options for static sampling, CTS trapping, cryo-focusing, and injection. Key parameters were provided in Table S2.

| Interference removal
Excessive water in the headspace above seafood samples was reduced using Na 2 SO 4 anhydrous powder to prevent chromatographic distortion and mass spectrometric interference. In addition to moisture, interferences due to the prevalence of volatile organic compounds in the laboratory can be problematic in headspace analysis.
Reagents, glassware, and other sampling hardware may also yield artifacts to sample analysis. Thus, steps were taken to reduce interfering compounds.
In this study, all glass vials were baked at 150°C overnight before use. Lids, valves, and O-rings were stored in a vacuum oven.
A SIM MS acquisition method was used to acquire signals at only the selected mass fragments in a certain time segment. Moreover, baseline correction, a chemometric data preprocessing method, was used to further remove interfering and irrelevant signals from analytical signals (Wang et al., 2014). A blank (i.e., headspace vial only containing the drying agent Na 2 SO 4 ) was run in the same manner as seafood samples. A total of 30 blanks were obtained. Twoway (chromatographic and mass spectrometric) baseline correction was performed with an in-house algorithm after data normalization (Wang et al., 2013).

| Calibration curves
A calibration mixture stock solution was prepared with 38 chemical standards listed in Table 2. Calibration solutions were made by serial dilutions of the stock solution with pentane to create multiple concentration levels. The internal standard solution was 4-heptanone (653 μg/kg in pentane).
During this research, two types of calibration curves were prepared. The first type of calibration curve was made by adding 10 µl of each calibration solution and 10 µl of internal standard solution onto a piece of blank Whatman ® qualitative filter paper (Grade 1, 85 mm circle) and the second type was pooled matrix-matched calibration curves generated using 50 g of ground tilapia (along with 35 g of Na 2 SO 4 powder) spiked with 10 µl of each calibration solution.
Filter paper was initially used as a blank to make calibration standards because nonpolar chemical standards are insoluble in water and to avoid potential interferences from solvents. Only a very small amount of calibration solution (10 µl) was spiked into each calibration standard for the purpose of minimizing solvent (pentane) vapor in the headspace glass vial. Details of concentrations of individual analytes in each calibration level can be found in Table S3. Precision and accuracy of filter paper-based calibration curves were assessed using three biological replicates of blank (filter paper) spikes.
For the pooled matrix-matched calibration curves, tilapia was chosen to be the matrix because its fat content was between the low and high range of other types of seafood to be analyzed (Genualdi et al., 2013). Five sets of calibration curves were individually made on five different days. Concentration levels varied by analyte (Table 2). A regression line of best fit was generated from the pooled data of all calibrations combined so day-to-day variation and matrix effects were factored into the curves. Peak area measurement, generation of calibration curves, and concentration calculation were performed by MS Quantitative Analysis, version 10.2 (Agilent Technologies). Calibration curves shown in Figure S6 were reconstructed using Excel, version 2102 (Microsoft). The established pooled matrix-matched calibration curves were assessed using three biological replicates of matrix spikes. A different matrix (cod) was used to evaluate the pooled matrix-matched calibration curve made with tilapia. Cod as a very lean fish is low in fat and carbohydrates.
Blank fresh cod typically had a small amount, if any, of the analytes detectable prior to spiking. Matrix spikes were prepared with 50 g of cod (along with 35 g of Na 2 SO 4 powder) spiked with 10 µl of a spike solution. Blank subtraction was performed prior to quantification.
Raw GC/MS data sets were acquired as network common data form (netCDF) files. With an in-house algorithm, netCDF files were read into MATLAB. The data sets were binned by retention time from 5.8 min to 21.0 min with a 0.01 min increment and binned by massto-charge ratio from m/z 22 to m/z 175 with a m/z 0.1 increment.
Data sets were normalized to unit vector length to reduce random errors, such as slightly varying amounts of samples in different injections. After baseline correction, the separability of data clusters was visualized on ANOVA-PCA score plots.
Concentrations of target compounds in fresh and decomposed seafood samples were subject to two-sample t-tests to determine potential marker compounds indicative of decomposition for each species. A two-sample t-test was performed using the ttest2 function in MATLAB, which returned the values of average, standard deviation (STD), and h. The h value is a test decision for the null hypothesis that two groups of data come from independent random samples from normal distributions with equal means. The result h is 1 if the test rejects the null hypothesis at the 5% significance level, and 0 otherwise. Thus, a compound with a h value of 1 was considered to be a potential marker compound.

| Identification of volatile compounds in seafoods
Seafood is diverse in many ways. The present study included seven seafood species caught from various habitats. To discover marker compounds relating to quality deterioration in seafood, initially a nontargeted fingerprinting method using GC/MS data sets collected in the SCAN mode was attempted (data not shown). However, a high number of unspecified data points from interfering compounds limited the possibility of making one group of samples strikingly different from other groups. Thus, a targeted method was developed herein. Targeted analysis often has a higher selectivity and sensitivity than nontargeted analysis. Moreover, the reliability of a targeted method can be validated using chemical standards, as analytical targets have been predefined (Ballin & Laursen, 2019 The percent accuracy was the calculated concentration of each compound divided by its expected concentration in cod spike, expressed in units of percent. c %RSD stands for percent relative standard deviation.

TA B L E 2 (Continued)
primary goal of the current study is to establish and evaluate a new large volume headspace sampling method for detecting these target compounds in seafood and identify compounds that have significant differences in concentration between fresh and decomposed seafood.
Even in vacuum packed seafood during storage, spoilage bacteria still could generate volatile compounds, such as 2-methyl-1-butanol and 2-butanone (Jørgensen & Henrik, 2001). Some low molecular weight compounds are primary and secondary lipid oxidation products with strong olfactory attributes, imparting the characteristic odor of rancid fish oil (Kulås et al., 2002), including 2,4-octadiene,  (Duflos et al., 2005;Shimoda et al., 1996). showed that a prolonged incubation period did not promote the mass transfer across the phase boundary ( Figure S2) but extended the sample run time (Table S1). Agitation during incubation proved to be advantageous to help less volatile compounds diffuse in the headspace phase and be effectively withdrawn from the vial ( Figure S3).

| Large volume headspace sampling optimization
In many cases, the addition of salts or solvents to sample matrix may decrease the partition coefficients and allow more compounds to pass into the headspace phase. Three different types of matrix modifier were tested during method development: saturated sodium chloride (NaCl) water, 10% potassium hydroxide (KOH) water, and Na 2 SO 4 anhydrous powder. The addition of NaCl solution to aqueous samples did not result in an observable difference for the current study of seafood. Adding 10% KOH to ground seafood, in an attempt to breakdown fat and lipids, was not successful in this study because saponification did not occur at the current incubation tem-  (Table S1).
It should be noted that the large volume headspace sampling method is different from traditional headspace techniques, such as solid-phase microextraction (SPME) and purge-and-trap. There are several advantages of the large volume headspace sampling method, including (1) reduced sampling bias. Since a 1-L headspace vial has a higher capacity than a regular headspace vial (6-27 ml), different regions of a whole fillet were tested in a single analysis. Also, (2) the overall pattern of VVOCs and VOCs in seafood headspace was not altered during sample preparation because a minimal sample pretreatment protocol was used. Moreover, (3) a variety of aroma compounds were recovered at the same distribution as experienced by the sensory analyst. As the headspace vapor collection process was nonselective, compounds generated from different sources were extracted in the same manner.

| PCA analysis with data generated by LVHS-GC/MS
Seafood freshness assessment using LVHS-GC/MS chemical profiles was directly compared with sensory scores obtained by FDA NSSE. Complex GC/MS data sets were interpreted using a chemometric data analysis strategy. PCA converted GC/MS data sets to a lower-dimensional space without any awareness of the class labels.
The ANOVA-PCA score plots were used to visualize the separation of data clusters. As shown in Figure 2, for red snapper, croaker, weakfish, mahi-mahi, black tiger shrimp, and yellowfin tuna, fresh and decomposed samples were completely separated. Sample classification based on LVHS-GC/MS analysis agreed well with FDA NSSE sensory analysis. For sockeye salmons, 95% confidence intervals of two data clusters overlapped to some extent, indicating that profiling volatile compounds of interest may not be sufficient to accurately assess the freshness of sockeye salmon.
The major obstacle to accurate classification of fresh and decomposed seafood samples using instrumental analysis was the large variability among multiple samples of the same type. Essentially, in the present study, each sample was representative of an individual fish so biological replicates represented different fish with a sensory score that fell within a certain range. Therefore, the data variability was greater than if one fillet was homogenized and separated into multiple samples for replicates.
The large data variance was observed on PCA score plots as wide 95% confidence intervals. In PCA, a primary set of data sets can be represented by a subset of independent principal components (PCs) (Gniazdowski, 2017) such that new data sets typically contain fewer variables than the original ones. Ideally, a small number of principal components can contain as much information as the full set of primary variables. For example, Figure 2a shows that the percentage of variance explained by the first and the second PCs was 61% and 18%, respectively. Their cumulative percentage of variance was 79%. It means that the first two principal components carryover 79% of the information contained in primary variables. It should be noted that PCA was used to observe the clustering of LVHS-GC/MS data sets. Cohesion and separation of data clusters can be improved if a higher number of the extracted PCs were used (e.g., four PCs).
But choosing a subset of principal components seems slightly out of scope and will not be discussed herein. The large data variance implied that marker compounds could potentially be better characterized by quantitative analysis.

| Quantitative analysis with LVHS-GC/MS
Acquiring quantitative data can be challenging with GC headspace To confirm the LVHS-GC/MS procedure is suitable for quantitative analysis of volatiles, filter paper-based calibration curves were assessed using three biological replicates of blank (filter paper) spikes. Precision in this study was determined as the repeatability relative standard deviation (%RSD) of three spikes. The percent accuracy was the calculated concentration of each compound divided by its expected concentration in the spike, expressed in units of percent. Due to common coelution and retention time shift of peaks, a qualifier ion has been chosen for each compound in addition to a target ion. The extracted ion chromatogram for the target ion was used for the quantitation, while the qualifier ion was used to facilitate distinguishing this compound from any others with similar retention times.
Partition coefficients of analytes in filter paper-based calibration standards were free of matrix effects. Looking at just the peak area responses, under the same experimental conditions, %RSD of three blank spikes ranged from 2% to 22%, with 26 of 37 compounds having %RSDs <10%; and accuracy ranged from 78% to 112%, with 24 of 37 compounds having accuracy in the range of 90% and 110%. Quantitation of acetoin was not achieved in this study because extraction followed by direct injection GC/MS (Pinu & Villas-Boas, 2017) or derivatization followed by headspace GC/MS (Tian et al., 2009) is usually necessary for sensitive detection of acetoin.
Result shows that the precision and accuracy of filter paper spikes were satisfactory (Table S3).
However, it was found that target compounds in cod spikes had extremely low peak area responses, indicating that the matrix af-  Figure 3 as an example. Curves for the rest of the analytes can be found in Figure S6.
Precision and accuracy of the pooled matrix-matched calibration curves were evaluated using three replicates of spiked cod.
Depending on the analyte, the concentration in the spiked cod was at least two times higher than the lowest detectable concentration in blank cod. Blank subtraction was conducted prior to quantification. As shown in Quantitative data of marker compounds for seven seafood species are provided in Table S4. Two observed sources of data variability in the preset study were (1) biological variation caused by the fact that each biological replicate represented an individual wild-caught seafood sample, and (2) noncontrollable variation, as it is impossible to strictly control the VVOC and VOC emissions from 50 g of ground sample into a 1-L headspace. Even with this variability, differences in headspace concentrations of certain compounds caused by decomposition were still noticeable.

| Determination of chemical markers for each species
The off-odors and off-flavors of unprocessed seafood are dependent on seafood species and origin (Whitfield, 1998). Statistical tests were performed to determine which target compound could be possible markers relating to decomposition of each species. For each target compound, its concentrations in multiple biological replicates of fresh and decomposed samples were provided in Table S4. Twosample t-test determined whether the "fresh" and the "decomposed" populations were statistically different from each other. The requirements of two-sample t-test were met, since (1) data in each group were obtained via a random sample from the population, (2) data values within a group were independent, (3) the measurements were continuous, and (4) variances for the two groups were assumed to be equal. The target compounds with significant differences in concentration between fresh and decomposed samples are displayed in Figure 4 for each seafood type and by compound class.
Our results show that 2-methylbutanal and 3-methylbutanal had large percent differences in concentration between fresh and decomposed samples in all species except yellowfin tuna and pentanal was an important marker for yellowfin tuna, weakfish, and croaker. Table S4 shows that different aldehydes were positively correlated with the spoilage of different species (h = 1). Monitoring aldehydes could be useful for assessing the freshness of seafood in the future. Especially, 3-methylbutanal, an intermediate in the catabolism of leucine (Cserháti & Forgács, 2003), was the only significant marker compound found in all seven seafood species. This short-chain branched aldehyde is an important flavor compound in various foods. Its biochemical conversion routes have been thoroughly reviewed by other researchers (Smit et al., 2009). One possibility is that during the degradation of unprocessed seafood, leucine is liberated from protein by extracellular and intracellular proteolysis; then, multiple transaminases and leucine dehydrogenase catalyzed the conversion of leucine toward α-keto isocaproic acid; lastly, the branched-chain α-ketoacid dehydrogenase complex catalyzed the decarboxylation of α-keto isocaproic acid to form 3-methylbutanal. Another option for heat-treated seafood products is that 3-methylbutanal can be produced by nonenzymic Strecker degradation. An ideal marker would be one of microbial degradation. Since 3-methylbutanal is a chemical indicator of protein degradation and can also be produced as a storage odor (Duflos et al., 2005;Jónsdóttir et al., 2008;Smit et al., 2009), further studies are needed to determine how effective of a marker it would be for seafood decomposition.

| Ketones
Ketones are another class of volatiles closely related to distinct flavors of seafood. They can be produced by lipid peroxidation.
The primary products of lipid oxidation, such as conjugated dienes and hydroperoxides, may be indicators of the initial stage of oxidative degradation . In the current study, however, 2,4-octadiene was not found to be a chemical indicator for any species. In addition, ketones can be produced by the metabolism of sugars (Dillon, 2014). We found that 2,3-butanedione (diacetyl), a reactive diketone produced from pyruvate, was strongly positively correlated with the decomposition of croaker and mahi-mahi, and its concentration in mahi-mahi was significantly higher than in other species. The reason for this difference is not clear. It is likely that diacetyl production is increased with aeration (Dillon, 2014) and mahi-mahi are mostly found in the surface water. Moreover, mono-ketones could be gradually derived from hydroperoxides through the splitting of fatty acid chains . It was found that 2-heptanone and 3-hexanone were not possible chemical indicators of decomposition for any species, while others (2-butanone, 2-nonanone, 2-pentanone, 2-undecanone, 3-pentanone, and 3-methyl-2-butanone) could be chemical indicators for certain species. The pathways leading to their formation have been reviewed previously (Kawai & Sakaguchi, 1996;Schulz et al., 2020).

| Alcohols
Alcohols have been widely detected along with aldehydes and ketones in various seafood by other researchers (Bai et al., 2019;Duflos et al., 2005;Fratini et al., 2012;Joffraud et al., 2001;Jørgensen & Henrik, 2001). In addition to being a potential chemical indicator of decomposition, delicate flavor of very fresh fish could also be contributed to alcohols (Alasalvar et al., 2005;Lindsay, 1994). Our results suggest that 2-methyl-1-butanol could be a chemical indicator of decomposed yellowfin tuna; 2-methyl-1-propanol could be a chemical indicator of decomposed mahi-mahi, black tiger shrimp, and sockeye salmon; (E)-2-penten-1-ol could be a chemical indicator of decomposed sockeye salmon; and (Z)-2-penten-1-ol could be a chemical indicator of decomposed red snapper, weakfish, and croaker.

| Sulfur compounds
Sulfur compounds (dimethyl sulfide, dimethyl disulfide, dimethyl trisulfide, and carbon disulfide) were found to be good indicators of decomposition for different seafood species. This result is in accordance with previous research (Bai et al., 2019;Duflos et al., 2005).
The generation of aliphatic sulfur-containing volatile compounds in seafood has been confirmed to be caused by enzymatic degradation of alk(en)yl cysteine sulfoxides (Varlet & Fernandez, 2010).
The development of sulfurous and putrid odors could be the turning point in the spoilage process (Whitfield, 1998), as it can move the overall seafood odor from desirable to rotten. Thus, it would be necessary to monitor the level of sulfur-containing volatile compounds to better control the quality of seafood. It is interesting that higher concentrations of dimethyl trisulfide (DMTS) and dimethyl disulfide (DMDS) were found in sockeye salmon than in other seafood species. The trend toward production of DMTS and DMDS is consistent with that reported for wild sea bream (Alasalvar et al., 2005), but the cause of accumulation in salmon is currently unknown.

| Other target compounds
Ethyl butyrate, an ester having a fruity flavor, was found to be indicative of the quality deterioration of croaker, mahi-mahi, black tiger shrimp, and sockeye salmon. It has a low odor threshold (1 ppb; Singh & Singh, 2008). Pseudomonas fragi may produce ethyl butyrate in fish during the early stages of spoilage (Wang et al., 2017).
Other compounds (alkane, chlorinated hydrocarbon, aromatic hydrocarbons, and saturated hydrocarbons) have been found by other researchers to exist in various seafood products, but their natural sources are poorly understood. Hsieh et al. suggested that benzene derivatives could be present due to environmental pollutants in crayfish (Tanchotikul & Hsieh, 1989). Alasalvar et al. (2005) determined that chloroform did not appear to contribute to the aroma of sea bream and may be an artifact. Heptane, toluene, chloroform, 1,1,3-trimethylcyclohexane, 1,2-dimethylcyclohexane, and methylcyclohexane were investigated in the present study. The results indicate that they were generally not reliable chemical indicators for fish spoilage, since their intensities were not strong enough for having sufficient statistical power. Previously, it has been reported that the occurrence of heptane in mackerel could be caused by spoilage (Duflos et al., 2005), however, heptane was not detected in any of the seven seafood species of interest in the current study.
It should be noted that the contribution of specific volatile compound to seafood's off-odor and off-flavor is a combined effect of its concentration in seafood and odor detection threshold (Tuckey et al., 2013). In this work, we established a new LVHS-GC/MS approach to detect the concentration change in various compounds in decomposed and fresh seafood, while their aroma properties have been identified previously by sensory panels. The change in concentration of important markers in decomposed and fresh samples for each species is illustrated in Figure 4. However, we could not determine if a change in the concentration of one specific compound would result in a change in sensory assessment. Extension of this work using the gas chromatography-olfactometry (GC-O) technique (Brattoli et al., 2013) will be implemented in the future to investigate specific effect of each target compound's concentration change on the sensory evaluation of seafood freshness.

| CON CLUS IONS
This study demonstrated the ability of a large volume headspace GC/ MS method for evaluating the quality of seven types of FDA NSSE seafood samples with verified sensory scores. With the completion of this initial study, the LVHS-GC/MS technique showed promising results as a screening tool for differentiating fresh and decomposed seafood samples. The entire process was fully automated and required minimal sample preparation.

ACK N OWLED G M ENTS
The authors would like to thank Dr. Betsy Yakes at FDA CFSAN for project coordination efforts. The authors appreciate the teamwork of the larger FDA Seafood Decomposition Workgroup, including those in the Office of Regulatory Affairs/Office of Regulatory Science, CFSAN/Office of Food Safety, and CFSAN/Office of Regulatory Science.

CO N FLI C T O F I NTE R E S T
The authors declare no conflict of interest.

E TH I C A L A PPROVA L S TATE M E NT
This study does not involve any human or animal testing.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are available on request from the corresponding author.

R E FE R E N C E S
Alasalvar, C., Taylor