*Review* **Fermentation of Organic Residues to Beneficial Chemicals: A Review of Medium-Chain Fatty Acid Production**

#### **Panagiota Stamatopoulou, Juliet Malkowski, Leandro Conrado, Kennedy Brown and Matthew Scarborough \***

Department of Civil and Environmental Engineering, University of Vermont, Burlington, VT 05405, USA; Panagiota.Stamatopoulou@uvm.edu (P.S.); Juliet.Malkowski@uvm.edu (J.M.);

Leandro.Fernandes@uvm.edu (L.C.); Kennedy.Brown@uvm.edu (K.B.)

**\*** Correspondence: Matthew.Scarborough@uvm.edu and mscarbor@uvm.edu

Received: 14 October 2020; Accepted: 23 November 2020; Published: 28 November 2020

**Abstract:** Medium-chain fatty acids (MCFAs) have a variety of uses in the production of industrial chemicals, food, and personal care products. These compounds are often produced through palm refining, but recent work has demonstrated that MCFAs can also be produced through the fermentation of complex organic substrates, including organic waste streams. While "chain elongation" offers a renewable platform for producing MCFAs, there are several limitations that need to be addressed before full-scale implementation becomes widespread. Here, we review the history of work on MCFA production by both pure and mixed cultures of fermenting organisms, and the unique metabolic features that lead to MCFA production. We also offer approaches to address the remaining challenges and increase MCFA production from renewable feedstocks.

**Keywords:** chain elongation; carboxylate platform; medium-chain fatty acids; carboxylic acids; mixed culture fermentation

#### **1. Introduction**

Microbial fermentation processes play an important role in food production, nutrient cycling, and pollution remediation. For more than 9000 years, humans have relied on microbial fermentation to produce and protect food [1], and fermentation processes have been used to convert wastes into valuable products since at least the 13th century [2]. One wide-spread application of fermentation processes to recover valuable products from wastes is anaerobic digestion. Anaerobic digestion is used to produce biogas (a mixture of carbon dioxide, methane, moisture, and other trace gasses) from organic wastes, including sewage sludge, agricultural and food production residues, and high-strength industrial wastes. There are estimated to be more than 20,000 large-scale anaerobic digestion facilities across the globe [3]. These facilities largely combust biogas to generate heat and electrical power, but biogas can also be converted to renewable natural gas to augment fossil-fuel derived natural gas. Anaerobic digestion to produce biogas is a well-studied process and has been widely implemented, but there is interest in producing valuable chemicals other than biogas from these renewable feedstocks.

Medium-chain fatty acids (MCFAs) are monocarboxylic acids containing 6 to 12 carbon atoms (Table 1). Currently, MCFAs are largely produced as a byproduct of palm refining. Palm kernels contain a relatively high content of octanoic acid (aka, caprylic acid, C8), decanoic acid (aka, capric acid, C10) and dodecanoic acid (aka, lauric acid, C12) [4]. MCFAs have also gained attention as dietary supplements, especially as medium-chain triglycerides (MCTs; three MCFAs attached to a glycerol backbone) [5]. MCTs are typically produced from the hydrolysis and separation of fatty acids from kernel or palm oil and re-esterification to a glycerol backbone, and contain 50–80% C8 [6]. MCTs are used as a supplement in livestock feed [7–9] and are marketed under many tradenames for human consumption (e.g., MCT Oil, Brain Octane Oil). In addition to their role as a nutritional supplement, MCFAs are also used for the production of personal care products, pharmaceuticals, dyes, and antimicrobials [10]. MCFAs have also been proposed as a precursor for liquid transportation fuels [11,12].


**Table 1.** Properties of short- and medium-chain fatty acids.

<sup>1</sup> Solubility in water obtained from PubChem [13] at 25 ◦C unless otherwise noted. <sup>2</sup> Electron density is presented as the chemical oxygen demand (COD). This value is the theoretical mass of O2 required to completely oxidize a compound. <sup>3</sup> Solubility for octanoic acid was obtained by averaging the solubilities at 20 and 30 ◦C. <sup>4</sup> Solubility for nonanoic acid is reported for 20 ◦C.

MCFA production has been demonstrated by pure-cultures and mixed microbial communities (microbiomes) sourced from a variety of inocula. In bioreactor experiments, with either pure cultures or mixed communities, a variety of end products are typically formed, including short-chain fatty acids (SCFAs) and MCFAs. Studies of pure cultures provide valuable insights into MCFA production, and many novel MCFA-producing organisms have been described in recent years. Microbial communities are believed to be advantageous for MCFA production from complex feedstocks due to their ability to consume a wide range of substrates and to respond to perturbations due to changes in feedstock characteristics. Since the mid 2010 s, the targeted production of MCFAs over short-chain fatty acids (SCFAs) has been termed "chain elongation," and the technology has previously been reviewed by Angenent et al. [10] and others [14,15]. However, there have been many recent developments in MCFA production that warrant further review. Therefore, this review summarizes the current body of knowledge using both pure cultures and mixed microbial communities. Metabolic features, energetics, and thermodynamics that may drive MCFA production are then considered. Lastly, future work is proposed to improve our understanding of MCFA production and to elucidate tools to improve MCFA production from complex feedstocks.

#### **2. Medium-Chain Fatty Acid Production by Bacterial Isolates**

MCFAs have been known as a potential fermentation product for over 100 years. Work with bacterial isolates producing MCFAs has provided insights into the metabolism of MCFA production and the phylogenetic diversity of MCFA-producing organisms. Here, we summarize the chronological body of work on MCFA production leading to the 13 isolated bacterial strains currently known to produce MCFAs. We also discuss the variety of substrates that can be used and the natural environments from which these bacteria were isolated. Characteristics of the known MCFA-producing bacterial isolates are summarized in Table 2.

*Rhodospirillum rubrum* [16] was the first MCFA-producing bacterium to be isolated (originally named *Spirillum rubrum* in 1887 [17,18]). During dark fermentation, *R. rubrum* consumes glucose, fructose, sucrose and carbon monoxide, and produces C6 [10,19]. In the 1940s, *Clostridium kluyveri* was isolated from canal mud. *C. kluyveri* is an anaerobic, non-saccharolytic bacterium that produces C6 and C8, using ethanol, n-propanol and pyruvate as electron donors, and C2, C3 and C4 as electron

acceptors [14,20–22]. *C. kluyveri* is one of the most well-studied MCFA-producing species and a complete genome for this organism became available in 2008 [23]. The proposed metabolism of MCFA production is largely based on experiments performed with *C. kluveryi*, and supporting genomic evidence [22–26] and is discussed later in this review. Shortly after the isolation of *C. kluyveri* in 1938, *Eubacterium limosum* was isolated from pea-blanching wastes, and converted methanol and C2 to C4 and C6 [27,28].

Shortly after the isolation of *C. kluveryi* and *E. limosum,* Prevot and Taffanel isolated *Ramibacteriurn alactolyticum* [29]. In 1967, this organism was shown to produce carboxylic acid products, including C6 and C8, from sugars and lactate [30]. In 1973, the name was changed to *Eubacterium alactolyticum* [31], and 26 years later, the organism was renamed again as *Pseudoramibacter alactolyticus* [29]. The complete genome of *P. alactolyticus* was sequenced in 2011 as part of the Human Microbiome Project [32]. In 1951, a third MCFA-producing microbe was isolated from the rumen of a sheep [33]. *Rumen organism LC*, as it was initially named, used sugars and lactate to produce even- and odd-chain carboxylic acids. In 1958, the organism was renamed *Peptostreptococcus elsdenii* [34], and in 1971 Rogosa reclassified the isolate as the genus *Megasphaera* [35] and named the type species *Megasphaera elsdenii*. *M. elsdenii* is a well-studied organism that has been used for the production of MCFAs from a variety of complex and simple substrates [36–43]. The complete genome of the *M. elsdenii* type strain was made available in 2011 [44], and three additional strain genomes have been sequenced since then [45,46]. In 2020, a metabolic reconstruction and model of *M. elsdenii* was published [47].

Several additional MCFA-producing strains have been described since the early 2000s. In 2003, *Eubacterium pyruvativorans* was isolated from a sheep rumen [48]. *E. pyruvativorans* converts C3 and C4 to C5 and C6, using methanol and pyruvate as electron donors [48]. *Ruminococcaceae* bacterium CPB6, isolated from a fermentation pit in 2015, produces C6 from lactate and has been used to convert wastes from liquor-making to MCFAs [49,50]. The genome of the *Ruminococcaceae* bacterium CPB6 was published in 2017 [51]. *Caproiciproducens galactitolivorans* (originally *Clostridium* sp. BS-1 [52,53]) was isolated from an anaerobic digester and described in 2015. *C. galactitolvorans* produces C6 from galactitol and its genome was sequenced in 2019 [54,55]. In 2017 a new species of *Megasphaera* was described. *Megasphaera hexanoica* converts sugars, including fructose and mannose, to even- and odd-chain fatty acids, including C6. *M. hexanoica* was isolated from a cow rumen, and follow-up work has investigated conditions to increase MCFA production [56,57].

In the first nine months of 2020, five additional MCFA-producing bacteria have been isolated. *Caproicibacter fermentans* is a C6-producing bacterium isolated from a methanogenic bioreactor [58]. *Clostridium* sp. BL-3, BL-4 and BL-6 convert lactate to C6. One of the isolates, BL-4, also produces iso-butyrate. [59,60]. *Caproiciproducens* sp. 7D4C2 was recently isolated from an MCFA-producing bioreactor and produces C6 from sugars [61]. Further, though not isolated, two *Candidatus* species of bacteria, *Candidatus Weimeria bifida* and *Candidatus Pseudoramibacter fermentans*, were described in 2020 and metabolic reconstructions were proposed based on metagenomic and metatranscriptomic analyses [62].

MCFA-producing isolates have been derived from a variety of natural sources, from canal mud to rumens to the human body. This wide range of isolation sources suggests that MCFA production may have yet undiscovered roles within natural ecosystems. With the exception of *R. rubrum*, all of the MCFA-producing isolates belong to the Firmicutes phylum, and there are currently six genera of MCFA-producing bacteria within the Firmicutes: *Clostridium*, *Pseudoramibacter*, *Megasphaera*, *Eubacterium*, *Caproicproducens*, and *Caproicbacter.* It is expected that many more MCFA-producing bacteria will be isolated in coming years, and studies with these organisms and their genomes are expected to provide new insights into MCFA production.


**Table 2.** Isolated bacterial strains known to produce MCFAs.

#### **3. MCFA Production form Complex Feedstocks with Microbial Communities**

While microbial isolates are important for understanding MCFA production, the application of large-scale MCFA production is expected to rely on microbial communities using "open culture" processes [10]. Several renewable substrates have been tested for MCFA production, including the organic fraction of municipal solid waste (OFMSW), food waste (FW), lignocellulosic biomass and residues of lignocellulosic biomass refining (LCB), waste from grain ethanol production, brewing distilling, and wine making (EBDWW), dairy processing wastes (DPW) and manure. In some processes, pretreatments have been used to condition the feedstock, including homogenization and microbial pre-treatment (Figure 1). Homogenization typically includes the addition of water to convert high-solid feedstocks into pumpable liquids. Microbial hydrolysis and fermentation (a.k.a, acid digestion) rely on anaerobic microbes to break down complex organic matter and to ferment a portion of the resulting hydrolysis products to acids and alcohols. Both liquid- and solid-state processes (generating an acid-rich leachate) have been used. Alternatively, feedstocks can be fed directly to a single reactor that performs hydrolysis, fermentation, and MCFA production. Sometimes, a supplemental electron donor (such as ethanol) is added. The MCFA-production step has been tested using both suspended-growth and attached-growth processes. Further, up-flow granular sludge processes have been used. The advantage of attached growth and granular processes is that they decouple the solids retention time from the hydraulic retention time, allowing the growth of slow-growing microbes without the need for high hydraulic retention times. If included, the extraction of MCFAs takes place either subsequent to MCFA production or as part of the same reactor system. When implemented on a reactor recirculation

stream this process is referred to as pertraction. An advantage of pertraction is that it prevents the accumulation of MCFAs within the fermentation broth. While most tested extraction processes rely on liquid–liquid separation with organic solvents, electrolytic extraction has also been proposed [64]. Typical processes for MCFA production are depicted in Figure 1. The selected processes largely depend on the complexity of the feedstock and the availability of electron donors to drive MCFA production. Here, we provide a review of past research on converting complex organic substrates to MCFAs based on the type of feedstock (Table 3). Past studies have investigated different feedstocks, reactor configurations, hydraulic retention times (HRTs), and inoculum sources. We summarize these studies based on net conversion of chemical oxygen demand (COD) provided to the reactor, including both COD in the feedstock and COD added as a supplemental electron donor.

**Figure 1.** Processes for converting organic wastes to MCFAs. Several processes have been implemented for converting organic waste streams to MCFAs.



*Processes* **2020**, *8*, 1571


**Table 3.** *Cont.*

#### *Processes* **2020**, *8*, 1571


**Table 3.** *Cont.*

#### *3.1. Organic Fraction of Municipal Solid Waste (OFMSW)*

Organic fraction of municipal solid waste (OFMSW) consists of yard waste, food waste, and other organic waste materials. As such, it can contain a mixture of plant- and animal-based products that vary widely in composition. High fractions of yard waste can increase the cellulose, hemicellulose, and lignin content, while high amounts of meat wastes can increase the protein content. OFMSW is collected as a solid material and may require dilution and homogenization prior to processing, especially if the food waste is treated in liquid-based reactors. An alternative leach bed-based approach has also been proposed. In these configurations, leachate is circulated through the solid waste to stimulate fermentation and capture fermentation products [66]. The highest productivities and COD conversion were obtained in work using an up-flow fixed-bed reactor supplemented with ethanol, relying on a two-stage process in which OFMSW was pre-fermented prior to MCFA-production [65]. The lower productivity of MCFAs in other OFMSW experiments [66] may be partly contributed to by a high fraction of yard waste in the feedstock. Cellulose, hemicellulose, and lignin polymers in yard waste are particularly recalcitrant [83]. While OFMSW is an abundant feedstock, it contains recalcitrant materials that may necessitate pre-treatment (e.g., hydrolysis) to achieve high conversion efficiencies of COD to MCFAs.

#### *3.2. Food Waste (FW)*

Food waste (FW) (a component of OFMSW) is another abundant organic waste that has been tested as a feedstock for MCFA production. As with OFMSW, FW characteristics can also vary significantly, with vegetables and starch-based products providing carbohydrates of various monomeric and polymeric compositions and meats providing high quantities of protein and fat. The pre-processing of food waste can include homogenization and pre-fermentation. De-packaging of food waste is an additional concern [84]. Work published in 2018 by Nzeteu et al. [67] demonstrated the use of a leach bed reactor with liquid recirculation, and achieved a maximum conversion of 19% of the influent COD in food waste to C6. Base on the characterization of reactor influent and effluent, they showed that 90% of hemicellulose, 60% of proteins, 50% of cellulose, and 20% of fats were degraded. In addition to the sustained demonstration with the leachate reactor, Nzeteu and colleagues tested the ability to produce additional MCFAs from the leachate bed effluent through the addition of either ethanol, H2, or both ethanol and H2 as supplemental electron donors in batch experiments. They found that lactate in the leachate reactor effluent was converted to C6 when H2 or a combination of H2 and ethanol was added. Another process for processing food waste is a two-stage configuration in which food waste is first pre-treated in a hydrolysis and fermentation bioreactor. Roghair and colleagues [68] used a batch hydrolysis and fermentation reactor operated at a pH of 5.5 and retention time of 18 days to acidify food waste prior to feeding to an MCFA-producing bioreactor. In total, 28% of the COD was converted to MCFAs with a 4 day retention time, but decreasing the retention time to 1 day reduced MCFA production [68]. Reddy et al. performed batch tests with and without bioaugmentation with *C. kluyveri* and found that the augmentation of *C. kluyveri* improved MCFA production [69]. Contreras-Davila et al. enriched a bioreactor that produced lactate as an intermediate to MCFAs, and the microbial community was dominated by species related to the genera *Lactobacillus* and *Caproicproducens* [70]. While lacking the high lignocellulosic content of OFMSW, the production of MCFAs from food waste still has challenges, including the de-packaging of abundant food wastes (such as those obtained from a grocery store). High protein content is also a concern due to potential ammonia toxicity as amino acids are degraded.

#### *3.3. Lignocellulosic Biomass (LCB)*

LCB feedstocks include the non-grain parts of crops (such as corn-stover), non-food plants (such as switchgrass), and paper. Lomkar, Fu and Holtzapple showed the production of C6 from a mixture of chicken manure and shredded paper by adding supplemental ethanol in batch experiments [71]. Scarborough, et al. demonstrated the production of C6 and C8 from the stillage (the material remaining

after distillation) of a switchgrass biorefining process [72]. The lignocellulosic stillage was rich in xylose and complex carbohydrates, and lactate accumulated as an intermediate upon spike-feeding the reactor with stillage. Subsequent multi-omic studies suggested that two organisms, Ca. *Weimeria bifida* and Ca. *Pseudoramibacter fermentans*, expressed genes for producing MCFAs [62,85]. Metabolic modeling, however, suggested that all of the MCFAs were produced by Ca. *Weimeria bifida* from xylose, while Ca. *Pseudoramibacter bifida* was predicted to produce C4 [86]. Other studies have explored the use of grass directly [73,74]. Weimer, Nerdahl and Brandl achieved a high conversion of COD (35%) with milled grass and ethanol using a microbiome from cow rumen fluid augmented with *C*, *kluyveri* [73]. Khor et al. used a two-stage process in which grass was first fermented to lactate using a leach bed reactor [74]. Leach bed fermenter effluent was converted to C6 in a microbial community enriched in *Clostridium-related* organisms. MCFAs were extracted via electrolytic extraction and Kolbe electrolysis was used to convert the enriched caproate stream into dodecane, which can be used as a liquid transportation fuel [74]. MCFA production from corn silage (partly fermented corn stover commonly used as animal feed) has also been demonstrated [75]. Corn stover hydrolysate (produced from the chemical depolymerization and enzymatic hydrolysis of corn stover) has also been converted to MCFAs using a pure culture of *M. elsdenii* [37]. The high COD and carbon content of LCB feedstocks make them attractive for MCFA production. Further, integrating MCFA production into next generation biorefineries may provide additional sources of revenue.

#### *3.4. Ethanol, Brewery, Distillery, and Winery Waste (EBDWW)*

EBDWW includes high-strength organic streams containing stillage (material remaining after distillation), yeast biomass (trub, dregs, or lees), and spent grains. EBDWW can contain high amounts of ethanol, which is a well-studied electron donor for MCFA production [10]. Multiple studies have investigated MCFA production from beer and/or stillage from grain-based ethanol biorefineries [53,64,75,76]. Andersen et al. produced C6 and C8 from beer and stillage derived from wheat-based ethanol production. The production of C10 was also reported, and the microbial community was enriched in *Lactobacillus* and *Clostridium*-related organisms [53]. While ethanol was provided in the feedstock (targeting an influent concentration of 6 g/L under steady-state conditions), lactate was also predicted to be an important electron donor produced as an intermediate from carbohydrates [56]. Urban et al. converted 58% of the COD in corn beer to MCFAs, extracted the MCFAs with pertraction, and converted MCFAs to alkanes using Kolbe electrolysis [75]. Similar to work by Khor et al. [74], Urban et al. demonstrated the complete conversion of a complex organic material to alkanes with MCFAs as an intermediate. Carvajal-Arroyo et al. implemented a granular sludge process to convert 29% of COD in stillage to MCFAs [76]. The resulting granules were enriched in *Ruminococcaceae*-related organisms. Andersen et al. converted stillage to a mixture of VFAs—including a minor amount of C6—in a bioreactor integrated with electrolytic extraction. Electrolytic extraction has the advantage of removing the MCFA products while also supplying H2 as an electron donor and consuming H+ to reduce or eliminate the need for the addition of pH control chemicals [64]. Wu et al. used an up-flow granular sludge reactor to convert 77% of the COD in liquor-making wastewater to MCFAs (77%). Zhu et al. used a recently isolated MCFA-producing strain (*Ruminococcaceae* CPB6) and converted 51% of COD in lactate-rich liquor-making wastewater to MCFAs [50]. Kucek et al. also achieved high conversion efficiencies from wine lees (a waste stream from wine production containing yeast biomass) using a bioreactor equipped with a pertraction system. In summary, EBDWW feedstocks are attractive due to their high organic content and availability of electron donors, including ethanol and lactate, which can drive MCFA production from the carbohydrates remaining in the waste stream.

#### *3.5. Dairy Processing Wastewater (DPW)*

Cheese, milk and yogurt processing produce low-value streams that are rich in organic materials. DPW can be rich in simple substrates, such as lactose, that have the potential for many biorefining applications. Whey, a dairy-derived protein, is also abundant. Three known studies have investigated

the use of DPW for MCFA production. Using acid whey from yogurt production, Xu et al. used a two-stage, temperature-phased process to produce MCFAs from acid whey. The MCFA production reactor was coupled to a pertraction system, and a sample from the first stage fermentation was diluted to maintain target organic loading rates. This system achieved a 61% conversion of COD and the highest production rates of MCFAs reported in the literature thus far [79]. The high production rates are likely due to an easily-degradable substrate, pre-fermentation to produce lactic acid, the use of an up-flow filtration reactor to decouple solids retention time from the hydraulic retention time, and the removal of MCFAs with pertraction to prevent product inhibition. In another study, acid whey from the production of quark (a curd cheese product) was used as a feedstock to produce C6 with an up-flow sludge-blanket reactor [80]. As with other studies, the microbial community was enriched in *Ruminococcaceae*-related organisms [80]. In a third study, the feedstock was created by mixing cheese whey with water [81]. This study provided a robust analysis of different operational parameters, including the HRT and the organic loading rate. The biofilm generated in the up-flow reactor was enriched in microbial communities, including organisms related to *Lactobacillus*, *Clostridium*, and *Ruminococcaceae* [81]. DPW are an attractive substrate due to their high content of simple substrates, including sugars (such as lactose) and lactate.

#### *3.6. Manure*

To our knowledge, there has been only one published study investigating manure as a substrate for MCFA production. Zhang et al. used a two-stage batch system to convert swine manure to MCFAs [82]. The swine manure contained 37% solids, and the solids were comprised of 25% hemicellulose, 20% protein, 18% cellulose, and 6.5% fat. Diluted manure was used as the feedstock, and ethanol was added as a supplemental electron donor. A COD conversion of 28% was achieved, but this required a long duration of 76 days. Still, this study demonstrates that manure may be a suitable substrate for MCFA production.

#### *3.7. Ideal Feedstocks and Processes for MCFA Production*

Past work shows that MCFAs can be produced from a wide variety of feedstocks (Table 3). Supplemental electron donors, such as ethanol, can be added to potentially increase MCFA yield, but the addition of ethanol results in significant economic and environmental costs [87]. Feedstocks containing easily degradable materials—such as DPW, containing high amounts of sugar—may not require supplementation with an electron donor. Further, some feedstocks—such as EBDWW—may be rich in electron donor materials, such as lactate or ethanol. A variety of reactor configurations have also been tested. Pretreatment with microbial hydrolysis and acidification may improve MCFA yields for some substrates, and the real-time removal of products via pertraction can greatly increase production rates. However, there is still much work required to optimize reactor systems, based on cost and productivity, for MCFA production.

#### **4. Metabolic Features of Medium-Chain Fatty Acid Production**

The primary route for MCFA production is believed to be reverse β-oxidation. β-oxidation is a route of fatty acid catabolism, and metabolic engineering to reverse this pathway in model organisms has been suggested [88]. Certain fermenting organisms, however, have long been known to use reverse β-oxidation as a strategy for redox balancing. Here, we focus on the natural metabolism of MCFA-producing bacteria described through various phenotypic and genomic studies.

#### *4.1. Reverse* β*-Oxidation*

MCFAs are proposed to be produced by reverse β-oxidation (Figure 2) [10,89]. The cycle is initiated by an acyl-CoA C-acyl transferase (E.C. 2.3.1.16, E.C. 2.3.1.9) that combines acetyl-CoA with an acyl-CoA (e.g., acetyl-CoA, propionyl-CoA, butyryl-CoA, valeryl-CoA, hexanoyl-CoA). In the second step, the 3-hydroxyacl-CoA is reduced via 3-hydroxyacyl-CoA dehydrogenase (E.C. 1.1.1.157, E.C. 1.1.1.35) with electrons from NADH. In the third step, enoyl-CoA hydratase (aka, crotonase; E.C. 4.2.1.55, E.C. 4.2.1.17) creates a double bond between the second and third carbons creating

an enoyl-CoA. In the fourth step, an electron-bifurcating acyl-CoA dehydrogenase (E.C. 1.3.1.109) oxidizes two molecules of NADH, with one electron pair being transferred to the enoyl-CoA and one being transferred to ferredoxin, producing reduced ferredoxin (FdRED) and an acyl-CoA [90]. The electron-bifurcating acyl-CoA dehydrogenase complex relies on two electron transfer flavoproteins (EtfA and EtfB) to mediate the transfer of electrons [91]. In total, reverse β-oxidation increases the length of an acyl-CoA by two carbons, while oxidizing three molecules of NADH to NAD<sup>+</sup> and reducing one molecule of oxidized ferredoxin (FdOX) to FdRED.

**Figure 2.** Reverse β-oxidation with an electron-bifurcating acyl-CoA dehydrogenase.

#### *4.2. Terminating Reverse* β*-Oxidation*

Reverse β-oxidation produces a CoA-ligated carboxylate (Figure 2). The terminal step of reverse β-oxidation is the cleaving of the carboxylate from CoA. Acyl-CoA hydrolases (E.C. 3.1.2.20) use water to cleave the CoA from the acid (Figure 3). A CoA transferase (E.C. 2.8.3.8) transfers a CoA from one acid to another. The replacement of the CoA with phosphate acyl transferases (e.g., phosphate butyryl transferase, E.C. 2.3.1.19) and subsequent ADP phosphorylation via a kinase (e.g., butyrate kinase, E.C. 2.7.2.7) have also been proposed [92].

**Figure 3.** Terminal enzymes for hydrolyzing the CoA-ligated products of reverse β-oxidation to release a carboxylate and CoA.

#### *4.3. Energy-Conservation Coupled with Reverse* β*-Oxidation*

Reverse β-oxidation does not directly produce ATP. Instead, organisms are predicted to use one or more enzymes to couple highly exergonic redox reactions to the generation of an ion motive force (IMF). The IMF can be used to generate ATP via a membrane-bound ATP synthase. One example is the *Rhodobacter* nitrogen fixation (RNF) complex (aka, ion-translocating ferredoxin NAD<sup>+</sup> oxidoreductase, E.C. 7.2.1.2) (Figure 4). The RNF complex couples the reduction of NAD<sup>+</sup> by FdRED with the translocation of proton or sodium ions across the cell membrane to generate IMF. This enzyme complex has been found in many fermenting bacteria, as well as methanogenic archaea [93–96]. The catalyzed reaction is reversible and the RNF complex plays a role in maintaining ratios of intracellular NADH and FdRED [97]. Another example is a membrane-bound energy-conserving hydrogenase (ECH) that couples the reduction of protons by FdRED with IMF generation. Regardless of the route of IMF generation, FdRED (a highly electronegative electron carrier) is needed to generate IMF. Therefore, the electron-bifurcating acyl-CoA dehydrogenase (Figure 2) is key to producing energy from reverse β-oxidation.

**Figure 4.** Membrane-bound enzymes for energy conservation in MCFA-producing bacteria. The RNF complex couples the reduction of NAD+ with FdRED with ion translocation to generate an ion motive force. Similarly, energy-conserving hydrogenase couples the reduction of protons with FdRED to ion translocation. The ion motive force is consumed by ATP synthase to produce ATP.

#### *4.4. Additional Electron Bifurcating Enzymes for Energy Conservation and Redox Management*

Other enzymes are proposed to conserve energy by producing or conserving FdRED to use for IMF generation [98]. One example is NAD(P)+ transhydrogenase (Nfn) (Figure 5), which transfers electrons from NADH and FdRED to NADPH and can be used in reverse to produce FdRED and NADH from NADPH [26]. Another example is HydABC (Figure 5), which produces H2 through the bifurcation of electrons from FdRED and NADH [99]. While Nfn produces FdRED directly, HydABC decreases the amount of FdRED needed for H<sup>+</sup> reduction compared to ferredoxin hydrogenase (Hyd1, E.C. 1.12.7.2) (Figure 5). Alternatively, HydABC may be used in reverse for the production of FdRED and NADH from H2.

**Figure 5.** Additional enzymes for redox balancing in MCFA-producing bacteria. NfnAB couples NADH and FdRED oxidation with NADP<sup>+</sup> reduction. HydABC relies on electrons from both NADH and FdRED to reduce protons. A ferredoxin hydrogenase uses only FdRED for hydrogen production.

#### *4.5. Hypothesized Routes for MCFA Production from Ethanol with and without Acetate as an Electron Acceptor*

To illustrate how bacteria tie MCFA production to energy conservation, we considered the well-studied example of ethanol and C2 conversion to MCFAs [23,89,98]. Ethanol conversion with reverse β-oxidation is well described in *C. Kluyveri*, and the metabolism has been reconstructed for C4 and C6 production from the co-utilization of ethanol and acetate [10,23]. One example of this overall transformation is the conversion of 1 mol C2 and 2 mols ethanol to C6 (Figure 6). In our example, we assumed C2 is incorporated via acetyl-CoA CoA transferase (Figure 3). Under this condition, C6 production results in the production of 1 mol of ATP (0.5 mol ATP per mol ethanol). Under this condition, no H2 is predicted to be produced. The ratio of ethanol-to-C2, however, is not fixed and H2 is predicted as a product when the molar ratio of ethanol-to-C2 is greater than 2:1 (Figures S1–S13).

**Figure 6.** An example model of C6 production from ethanol (2 mols) and acetate (1 mol) without production of H2.

While ethanol and acetate can be used as co-substrates, acetate is not required as an electron acceptor for converting ethanol to MCFAs [100]. When 2 mols of ethanol are converted to acetyl-CoA, a total of 4 mols NADH are predicted to be produced (Figure 7). One cycle of reverse β-oxidation consumes 3 mols NADH, leaving 1 mol NADH and 1 mol FdRED. There are multiple potential ways to maintain redox balance, depending on the hydrogenase present. With ECH, FdRED could reduce H<sup>+</sup> to H2 while producing IMF (Figure 7). IMF can then be consumed by the RNF complex to transfer electrons from the remaining NADH to FdRED, in order to recycle NAD<sup>+</sup>. FdRED can also be used via ECH to generate IMF that drives ATP synthase. In this model (Figure 7A), 0.5 mol ATP are predicted (0.25 mol ATP per mol of ethanol). If 3 moles of ethanol are consumed, then six NADH are produced (Figure 7B). Two cycles of reverse β-oxidation would consume all six NADH and produce 2 mols FdRED. Then, the 2 mols FdRED could be used to generate IMF with ECH, resulting in the net production of 1 mol ATP (0.333 mol ATP per mol of ethanol consumed). Thus, the production of C6 is predicted to increase the ATP yield by 33% per mol ethanol compared to C4 production. These two scenarios

illustrate how MCFA production can increase ATP production compared to the production of C4, but the overall ATP production is lower than when acetate is used as an electron acceptor (Figure 6).

**Figure 7.** Example models of C4 production (**A**) and C6 production (**B**) when ethanol is used as a sole substrate without acetate as an electron acceptor. The conversion of C6 is predicted to increase the ATP yield per mol of ethanol by 33%.

#### *4.6. Thermodynamic and Energetic Drivers of MCFA Production*

The energetic drivers of MCFA production have been explored [101], but recent discoveries of novel energy-conserving mechanisms in anaerobic bacteria suggest that energetic drivers are not fully elucidated [90]. To illustrate the potential energetic drivers of MCFA-production, we considered the conversion of ethanol to MCFAs. All free energies of formation were assumed based on previously described methods [85]. When consuming ethanol as a sole substrate, H2 is predicted to be produced for all scenarios, and it is shown that the amount of H2 produced decreases with the length of the product (Equations (1)–(4)). The reaction depicted in Equation (3) was also described in Angenent et al. [10].

$$1\,\mathrm{C}\_{2}\mathrm{H}\_{6}\mathrm{O}^{\cdot} + 1\,\mathrm{H}\_{2}\mathrm{O}^{\cdot} \rightarrow 1\,\mathrm{C}\_{2}\mathrm{H}\_{3}\mathrm{O}\_{2}^{-\cdot} + 2\,\mathrm{H}\_{2} + 1\,\mathrm{H}^{+} \quad \Delta\mathrm{G}\mathrm{O}^{\cdot} = \,\mathrm{+9.65\,\mathrm{kJ}\,\mathrm{mol}^{-1}\,\mathrm{ethanol}} \tag{1}$$

$$1\,\mathrm{C}\_{2}\mathrm{H}\_{6}\mathrm{O} \rightarrow \frac{1}{2}\,\mathrm{C}\_{4}\mathrm{H}\mathrm{7O}\_{2}^{-1} + 1\,\mathrm{H}\_{2} + \frac{1}{2}\,\mathrm{H}^{+} \qquad \Delta\mathrm{G}\mathrm{0'} = -14.5\,\mathrm{kJ}\,\mathrm{mol}^{-1}\,\mathrm{ethanol} \tag{2}$$

$$1\ \mathrm{C}\_{2}H\_{6}\mathrm{O}\_{2} \rightarrow \frac{1}{3}\ \mathrm{C}\_{6}H\_{11}\mathrm{O}\_{2}^{-} + \frac{1}{3}\ \mathrm{H}\_{2}\mathrm{O} + \frac{2}{3}\ \mathrm{H}\_{2} + \frac{1}{3}\mathrm{H}^{+} \quad \Delta\mathrm{G}0' \ = \text{--} 22.6\ \mathrm{kJ}\ \mathrm{mol}^{-1}\ \text{ethanol} \tag{3}$$

$$1\ \mathrm{C}\_{2}\mathrm{H}\_{6}\mathrm{O}\_{2} \rightarrow \frac{1}{4}\ \mathrm{C}\_{8}\mathrm{H}\_{15}\mathrm{O}\_{2}^{-} + \frac{1}{2}\ \mathrm{H}\_{2}\mathrm{O} + \frac{1}{2}\ \mathrm{H}\_{2} + \frac{1}{4}\ \mathrm{H}^{+} \quad \Delta\mathrm{G}\mathrm{O}^{\prime} = -27.4\ \mathrm{kJ}\,\mathrm{mol}^{-1}\ \text{ethanol} \qquad (4)$$

The co-consumption of ethanol and acetate reduces the need for H2 production (Equations (5)–(8)) as acetate can accept electrons that would otherwise be used to produce H2. When consuming 1 mol C2 per mol of product (Equations (5)–(8)), no H2 is predicted. The ratios of ethanol-to-C2, however, are not fixed and the utilization of 3 mols ethanol to 2 mols acetate has been described previously for *C. kluyveri* [10]. Many ethanol-to-C2 ratios have been tested, and high ratios of ethanol to acetate have been shown to favor MCFA production [100].

$$1\,\mathrm{C}\_{2}H\_{6}O + 1\,\mathrm{C}\_{2}H\_{3}O\_{2}^{-} \rightarrow 1\,\mathrm{C}\_{4}H\_{7}O\_{2}^{-} + 1\,H\_{2}O \quad \Delta G0' = -38.6\,\mathrm{kJ}\,\mathrm{mol}^{-1}\,\mathrm{ethanol} \tag{5}$$

$$1\,\mathrm{C}\_{2}H\_{6}O + \frac{1}{2}\mathrm{C}\_{2}H\_{3}O\_{2}^{-} \rightarrow \frac{1}{2}\,\mathrm{C}\_{6}H\_{11}O\_{2}^{-} + 1\,H\_{2}O \quad \Delta\mathrm{G}\vartheta' = -38.7\,\mathrm{kJ}\,\mathrm{mol}^{-1}\,\mathrm{ethanol} \tag{6}$$

$$1\ \mathrm{C}\_{2}\mathrm{H}\_{6}\mathrm{O} + \frac{1}{3}\mathrm{C}\_{2}\mathrm{H}\_{3}\mathrm{O}\_{2}^{-} \rightarrow \frac{1}{3}\mathrm{C}\_{8}\mathrm{H}\_{15}\mathrm{O}\_{2}^{-} + 1\ \mathrm{H}\_{2}\mathrm{O} \quad \Delta\mathrm{G}\mathrm{O}' = -39.7\ \mathrm{kJ}\,\mathrm{mol}^{-1}\,\mathrm{ethanol} \tag{7}$$

$$\text{Al }\mathrm{C}\_{2}\mathrm{H}\_{6}\mathrm{O}^{\cdot} + \frac{2}{3}\mathrm{C}\_{2}\mathrm{H}\_{5}\mathrm{O}\_{2}^{\cdot -} \rightarrow \frac{5}{6}\mathrm{C}\_{4}\mathrm{H}\_{7}\mathrm{O}\_{2}^{\cdot -} + \frac{2}{3}\mathrm{H}\_{2}\mathrm{O}^{\cdot} + \frac{1}{3}\mathrm{H}\_{2} + \frac{1}{6}\mathrm{H}^{+} \quad \Lambda\mathrm{G}\mathrm{O}^{\cdot} = -30.6\,\mathrm{kJ}\,\mathrm{mol}^{-1}\,\mathrm{ethanol} \tag{8}$$

Metabolic reconstructions of ethanol consumption predict that hydrogen-producing enzymes play a role in determining the most energetically favorable products (Table 4). If C2 is produced from ethanol, no net ATP production is predicted with Hyd1. With ECH, up to 1.0 mol ATP per mol ethanol is predicted, but the amount of free energy needed to generate this much ATP is only feasible at H2 partial pressures below 10−<sup>5</sup> atm, assuming a minimum quantum of –50 kJ per mol ATP [102] (Figure 7). With Hyd1, longer products are predicted to increase the ATP yield. With ECH, shorter products increase the ATP yield, but these scenarios all rely on low H2 partial pressures for the predicted ATP production to be feasible (Figure 7). The co-utilization of C2 can obviate the need for H2 production, and the predicted ATP yield (0.5 mol ATP per mol ethanol) when no H2 is produced is higher than all scenarios using ethanol as a sole substrate and Hyd1 as the hydrogenase. If Hyd1 is the only hydrogenase available, co-utilization of C2 with ethanol is always predicted to produce more ATP per mol ethanol than consumption of ethanol as a sole substrate, and all products result in the same ATP yield (0.5 mol ATP per mol ethanol). However, if ECH is available, the ATP yield is higher if ethanol is the sole substrate, but the overall free energy release is lower per mol ATP produced. This suggests that ethanol consumers with ECH may only co-utilize acetate under high H2 partial pressures.


**Table 4.** Predicted ATP production and free energy release for the conversion of ethanol to acetate, butyrate, hexanoate, and octanoate with either a ferredoxin hydrogenase (Hyd1) or energy-conserving hydrogenase (ECH).

<sup>1</sup> The ATP predicted to be produced from substrate-level phosphorylation (SLP). <sup>2</sup> The ATP predicted to be produced via ATP synthase (ATPS).

To further illustrate the potential importance of H2 and hydrogenase enzymes, we performed a thermodynamic analysis to assess the feasibility of a subset of predictions (Table 4) with varying conditions of H2 partial pressure. We considered either Hyd1 or ECH as the sole enzyme for hydrogen production. All products are predicted to result in thermodynamically favorable reactions up to a partial pressure of 10−<sup>1</sup> atm (Figure 8A). At H2 partial pressures greater than 10−<sup>1</sup> atm, the production of acetate as the sole product is predicted to be infeasible (**Δ**G0 > 0). Because the overall reaction is the same regardless of the type of hydrogenase assumed, there is only one line per product shown in Figure 8A. We then incorporated the predicted ATP yields from our metabolic reconstructions (Figure 8B). With ECH, the predicted free energy release per mol ATP is greater than −50 kJ at increasing H2 partial pressures for each potential product (C2 = 10−<sup>5</sup> atm; C4 = 10−<sup>4</sup> atm; C6 = 10−<sup>3</sup> atm; C8 = 10−<sup>1</sup> atm). This suggests that H2 partial pressure may be a control to drive the production of MCFAs if ECH is used as a hydrogenase. With Hyd1, the production of acetate from ethanol is predicted to produce a net of 0 ATP in order to maintain redox balance (Figure S1). ATP predictions from all other products are feasible up to a H2 partial pressure of 3 atm, indicating that H2 partial pressure may not be an appropriate control to drive MCFA production if Hyd1 is the hydrogenase.

**Figure 8.** *Cont.*

**Figure 8.** Free energy release for different products from ethanol under a range of H2 partial pressures. Plots show the overall free energy release (**A**) and the free energy release per mol of ATP (**B**). Areas shaded in gray are considered infeasible and the dashed lines represent scenarios with Hyd1 as a hydrogenase instead of ECH.

#### **5. Path Forward**

Taken together, the metabolic reconstructions and thermodynamic analysis suggests that (1) the most energetically advantageous products depend on enzymes other than those directly performing reverse β-oxidation, and (2) the infinite number of combinations of substrate ratios and product formation make simulating MCFA production a challenge. Still, as we learn more about the behavior of microbes producing MCFAs and how these microbes interact with complex microbial communities, we should improve our ability to design, diagnose, and control MCFA-producing bioreactors. We have identified specific knowledge gaps below.

#### *5.1. Controlling Product Length*

Controlling the terminal product length remains a challenge. While we have illustrated that some conditions should favor the production of MCFAs over C2 and C4, almost all studies to date create mixtures of end products, varying in length from C2 to C6 or C8 (Table 3). The highest specificities of C6 and C8 production have resulted from the incorporation of pertraction systems [103]. This real-time removal of C6 and C8 is thought to increase their production in part because the protonated forms of MCFAs are toxic [104–106]. Thus, their production may be limited by their toxic effects. While there has been recent work on the toxic effects of these compounds in MCFA-producing bioreactors [107], an improved understanding of the toxicity of these products is needed. Quantifying these effects, for instance via Monod-based inhibition terms, would be particularly valuable for modeling the impacts of MCFA toxicity and the favorable production of specific carbon chain lengths across various systems.

Enzyme specificity is also expected to control the length of the terminal product. The enzymes directly involved in reverse β-oxidation (Figure 2) and the termination of reverse β-oxidation (Figure 3) are all likely to play a role in dictating the predominant product length. Individual enzymes of the fatty acid biosynthesis (Fab) pathway are well characterized, and metabolic engineering approaches have been used to produce specific chain lengths [108,109]. Additional kinetic studies of reverse β-oxidation enzymes can inform modeling approaches, and structural studies could enable metabolic engineering to improve enzyme specificity towards target products. Genetically tractable MCFA-producers, however, are only beginning to emerge [36]. As metabolic engineering tools continue to improve and additional MCFA-producing isolates become available, isolate-based studies should elucidate strategies to improve the control of target chain lengths.

#### *5.2. Microbial Community Assembly, Function, and Resiliency*

The many microbial community-based studies for MCFA production have elucidated common functional guilds that become enriched in these communities, while phylogenetically distinct, functional roles tend to be conserved. As an example, several reactors fed carbohydrates have been enriched in *Lactobacillus* species and lactate-consuming MCFA producers. A framework for these functional guilds was recently proposed [86].

Further, these microbial communities have demonstrated resilience (sustained production of MCFAs) when operated as open cultures, often fed with non-sterile feedstocks. This resilience may be due to the presence of MCFAs, which select for only organisms with some tolerance to these products. In addition, although these communities contain organisms that rely on oxygen-sensitive ferredoxin [110], they survive in conditions with feedstocks that likely contain small amounts of oxygen. Thus, we predict that facultative anaerobes (such as *Lactobacillus*-related organisms) may play a role in quenching oxygen. When separating stages of fermentation (e.g., two-stage processes, Figure 1) or implementing pertraction, it is important to consider the traits that may make these microbial communities resilient.

#### *5.3. Product Uses*

MCFA production from renewable feedstocks may allow for reduced dependence on palm refining and fossil fuels. While MCFAs currently have high market prices, it is expected that the widespread production of MCFAs would reduce their value. Thus, more work on refining MCFAs to other products is needed. Kolbe electrolysis has already been demonstrated as a potential option to convert MCFAs to alkanes for use as a liquid transportation fuel. Another option may be to purify MCFAs and use them as a feed supplement. MCFAs are expected to have many of the same potential benefits as MCTs and may have potential as a dietary supplement for both humans and animals. More work on the role of MCFAs as a platform chemical is needed, and this work should consider the potential societal and environmental benefits along with the economic benefits. Thus, emphasis on life cycle assessment of MCFA-production is needed, and the study performed by Chen et al. provides a good example [87]. If implemented with sustainable approaches, MCFA production from organic wastes will be a step towards local circular bioeconomies which are critical for addressing persistent environmental and social challenges [111].

**Supplementary Materials:** The following are available online at http://www.mdpi.com/2227-9717/8/12/1571/s1, Figure S1: Proposed model for conversion of ethanol to acetate with Hyd1, Figure S2: Proposed model for conversion of ethanol to butyrate with Hyd1, Figure S3: Proposed model for conversion of ethanol to hexanoate with Hyd1, Figure S4: Proposed model for conversion of ethanol to octanoate with Hyd1, Figure S5: Proposed model for conversion of ethanol to acetate with ECH, Figure S6: Proposed model for conversion of ethanol to butyrate with ECH, Figure S7: Proposed model for conversion of ethanol to hexanoate with ECH, Figure S8: Proposed model for conversion of ethanol to octanoate with ECH, Figure S9: Proposed model for conversion of ethanol and acetate to butyrate, Figure S10: Proposed model for conversion of ethanol and acetate to hexanoate, Figure S11: Proposed model for conversion of ethanol and acetate to octanoate, Figure S12: Proposed model for conversion of ethanol and acetate to butyrate with an ethanol-to-acetate ratio of 3:2 and Hyd1, Figure S13: Proposed model for conversion of ethanol and acetate to butyrate with an ethanol-to-acetate ratio of 3:2 and ECH.

**Author Contributions:** Conceptualization, M.S. and P.S.; review of literature—P.S., J.M., and L.C.; writing—original draft preparation, M.S. and P.S.; writing—review and editing, P.S., J.M., L.C., K.B. and M.S.; supervision, M.S. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Acknowledgments:** We thank the College of Engineering and Mathematical Sciences at the University of Vermont for supporting M.S. with start-up funding and J.M. and K.B. with a summer research experience for undergraduates.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## *Review* **Quantification of Volatile Compounds in Wines by HS-SPME-GC/MS: Critical Issues and Use of Multivariate Statistics in Method Optimization**

**Sandra Pati 1, Maria Tufariello 2,\*, Pasquale Crupi 3,4, Antonio Coletta 3, Francesco Grieco 2,\* and Ilario Losito <sup>5</sup>**


**Abstract:** The aim of this review is to explore and discuss the two main aspects related to a HeadSpace Solid Phase Micro-Extraction Gas-Chromatography/Mass-Spectrometry (HS-SPME-GC/MS) quantitative analysis of volatile compounds in wines, both being fundamental to obtain reliable data. In the first section, recent advances in the use of multivariate optimization approaches during the method development step are described with a special focus on factorial designs and response surface methodologies. In the second section, critical aspects related to quantification methods are discussed. Indeed, matrix effects induced by the complexity of the volatile profile and of the non-volatile matrix of wines, potentially differing between diverse wines in a remarkable extent, often require severe assumptions if a reliable quantification is desired. Several approaches offering different levels of data reliability including internal standards, model wine calibration, a stable isotope dilution analysis, matrix-matched calibration and standard addition methods are reported in the literature and are discussed in depth here.

**Keywords:** wine; HS-SPME-GC/MS; multivariate statistical analysis; calibration

#### **1. Introduction**

A wine aroma profile (and thereby a volatile profile) is one of the most important quality criteria affecting wine acceptability by consumers [1]. Its characterization is very complex because volatile molecules usually belong to different classes such as alcohols, esters, aldehydes, acids, terpenes, phenols and lactones [2] with a wide range of polarity and concentrations. Furthermore, the non-volatile wine matrix affects the partitioning of aroma compounds between the matrix and the gas phase depending on their specific chemical properties and their interactions with aroma compounds [3].

An effective extraction from the aqueous wine matrix is essential for the accurate qualitative and quantitative analysis of wine volatiles. Therefore, various extraction and preconcentration approaches have been developed followed by separation and identification usually by mono-dimensional or bi-dimensional gas chromatography coupled to mass spectrometry (GC/MS, GC X GC/MS). Liquid-liquid [4], simultaneous distillation [5], solid phase [6], solid phase micro- [7], supercritical fluid [8] and stir bar sorptive extraction [9], among others, have been largely described for the analysis of wine volatile compounds [10].

195

**Citation:** Pati, S.; Tufariello, M.; Crupi, P.; Coletta, A.; Grieco, F.; Losito, I. Quantification of Volatile Compounds in Wines by HS-SPME-GC/MS: Critical Issues and Use of Multivariate Statistics in Method Optimization. *Processes* **2021**, *9*, 662. https://doi.org/10.3390/ pr9040662

Academic Editor: José Manuel Moreno-Rojas

Received: 18 March 2021 Accepted: 9 April 2021 Published: 9 April 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

On the other hand, headspace solid phase microextraction (HS-SPME), firstly described by Arthur and Pawliszyn [11], is considered to be one of the most effective techniques employed [10]. As HS-SPME is based on the partitioning of the analyte between the extracting phase immobilized on a fused silica fiber (as in all SPME approaches) and the headspace of the wine, it exploits a favorable transfer of volatile compounds into the latter. As with all SPME approaches, it combines sampling, analyte isolation and enrichment into one step and is suitable for full automation; moreover, as heat can be exploited for the desorption of the volatile analytes from the extracting phase, it does not require solvents either for the extraction or for the analyte desorption stage.

As with other SPME approaches, HS-SPME is usually a non-exhaustive extraction technique in which only a small percentage of the analyte is removed from the sample matrix. Both equilibrium and pre-equilibrium approaches can be employed. In the first case, enough time is available during the extraction step to reach the partitioning equilibrium between the sample matrix and the extraction phase and convection conditions do not affect the amount of analyte extracted. Under these conditions, if the sample volume (Vs) is much larger than the product between the volume of the extracting phase on the fiber (Vf) and the partition coefficient of the analyte between the fiber coating and the sample headspace (Kfs) it can be easily demonstrated that the amount of analyte extracted is proportional to the sample concentration but it does not depend on the sample volume. Indeed, the following equation is valid:

$$\mathfrak{n} = \mathsf{K}\_{\mathfrak{fs}} \mathsf{V}\_{\mathfrak{f}} \mathsf{C}\_{\mathfrak{f}}$$

where n is the mass of an analyte absorbed on or partitioned into the coating and C0 is the initial concentration of the analyte in the sample [12].

In the second case, a short pre-equilibrium extraction is carried out and, if the convection is constant, the amount of extracted analyte is proportional to the extraction time. Even though the sensitivity is usually lower than that achievable with equilibrium extraction, the extraction time is shorter, thus reducing the overall analysis time. On the other hand, a non-equilibrium extraction requires a strict control of extraction conditions (e.g., convection, if present, time and temperature) in order to guarantee an acceptable reproducibility for quantitative analysis.

In all cases, the development of a quantitative SPME method requires a careful optimization of parameters affecting the extraction efficiency such as the type and volume of the extracting phase, the agitation method, the extraction time and sample modification (pH, ionic strength, organic solvent content) and, especially when volatile analytes are involved, temperature. Moreover, as volatile compounds absorbed on/partitioned into the fiber coating are usually thermally desorbed by exposing the fiber to the heated space inside a GC injector before being separated into a GC column and detected, the desorption conditions of analytes (temperature, transport gas flow, eventual presence and characteristics of a liner into the GC injector) need to be optimized too.

The use of multivariate statistics approaches, in particular those pertaining to the design of experiments (DoE) instead of the classical univariate (one-at-a-time, OVAT) approach, has become increasingly relevant in recent years [13–15]. The increasing relevance of HS-SPME in wine analysis applications is illustrated by the linear increase in the number of publications on the topic occurred since the introduction of SPME (Figure 1).

**Figure 1.** The number of papers related to Solid Phase Micro-Extraction (SPME) applications to wine.

In particular, the histogram in Figure 1 shows the number of papers related to SPME applications to wine based on a search on the Scopus database between 1998 and 2020 using the keywords "HS-SPME/wine/volatile" and excluding reviews, conference and data papers and book chapters. Despite the numerous reports describing SPME applications [2,16,17], papers concerning the critical issues of SPME-GC/MS quantification methods for wine volatiles are still limited and the number of rigorous quantitative applications reported in the literature is still low. This evidence can be considered a remarkable lack as divergent results concerning matrix effects [18,19] and standardization methods [20,21] have been reported in the literature for the quantification of wine volatiles.

Starting from the considerations made so far, this paper aims to review two relevant aspects related to the application of HS-SPME-GC/MS to the quantitative analysis of volatile compounds in wines: 1) recent advances in the use of optimization approaches in the method development stage; 2) critical aspects related to quantification methods so far reported. Indeed, both of these aspects are expected to contribute significantly to a successful application of HS-SPME-GC/MS methods pursuing the assessment of the safety, quality, authenticity and/or traceability of wines through the profile of their volatile compounds.

#### **2. Optimization of HS-SPME Parameters with Multivariate Statistical Analysis**

The success of HS-SPME depends on the analyte distribution between the matrix and the headspace, which is influenced in turn by a few parameters including the vapor pressure of volatiles, matrix ionic strength, stirring and equilibration time and temperature. The extraction time and temperature, the fiber sample distance and the nature and thickness of the fiber coating also play important roles in terms of extraction yield [15,22,23]. In light of this, statistical approaches able to optimize many factors simultaneously should be adopted to select the best conditions for the achievement of reliable and reproducible results. Multivariate statistics techniques can be successfully exploited for this purpose [24–27] (Figure 2).

**Figure 2.** Optimization of Headspace Solid Phase Microxtraction (HS-SPME) parameters with a multivariate statistical analysis.

As a general approach, optimization procedures start with the definition of a measurable quantity, *Y*, as a reference parameter for the goodness/quality of the experiment itself. In most cases, the goodness of the experiment and the value of the *Y* parameter are positively correlated. Once *Y* is identified, factors *Xi* (experimental parameters) with *i* ranging from 1 to *k* that are expected to affect the results of the experiment must be selected. In determining the best procedure for optimizing the factors, however, it must be taken into account that: (1) a range of admissible values has to be determined for each factor; (2) it is usually not possible to experimentally test all of the possible values of a single factor in its range nor all of the possible combinations of different factors.

Once the variability ranges have been established for the factors, optimization based on multivariate statistics can be achieved through two fundamental approaches related to the DoE: two-level factorial designs and response surface methodologies (RSM) [28–30] with the latter including, among others, central composite (CCD) [25,31], Doehlert matrix (DM) [32,33], three-level factorial (3K) and Box Behnken (BBD) designs [27].

Factorial designs are a class of experimental designs offering a relatively large amount of useful information from a small number of experiments; for this reason, they are often exploited for a screening of factors. When the number of experiments that can be carried out is limited, factorial designs offer an efficient way of obtaining the maximum information from these experiments. In this type of approach, a number of levels (different values of each factor within its variability range) is established. A measurement is made for each possible combination of the levels of each of the *k* factors. The results are compared and the combination that maximizes the *Y* parameter is determined by developing a relatively simple model based on the responses. In the two-level full factorial design, two levels per factor are established; the number of combinations and therefore the number of measurements to be made is thus 2k. In a more generic case, if a number N of levels per factor is set, the number of measurements to be taken is Nk [34–37].

This approach is quite different from the classical univariate approach (OVAT), which is based on the evaluation of the effect of one variable at a time with all other variables kept constant; thus, it is unable to provide information on the interactions among factors. It is worth noting that several univariate optimization studies have been conducted to determine the optimal conditions for the volatile extraction of wine by HS-SPME [15]. However, factorial designs may lead to similar accomplishments with a reduced number of experiments. Carrillo et al. [38] employed a full factorial design (24) to establish the relative influence of four factors (type of fiber, temperature, pre-incubation time and sodium chloride addition) and their interactions on the chromatographic responses of extracted compounds in order to develop a method for the determination of oak-derived volatile compounds in wine. Robotti et al. [39] used an experimental design for the optimization of

the SPME procedure according to both the maximum signal intensity and repeatability. In a study on Maresco sparkling wine volatiles, Tufariello et al. [21] focused on the equilibration of parameters and extraction times through a full two-factor three-level design with the purpose of optimizing the overall time of analysis and eventually finding an interaction factor between the two parameters affecting the analytical response.

A more complex approach to the DoE sometimes following the application of a factorial design is represented by the response surface methodology (RSM) [29,40,41]. In this case, the experiment is designed to provide a more refined model of the response surface, estimating interactions and even quadratic effects related to factors. RSM designs are used to find optimal process settings (in many cases as an improvement of those found using the factorial design), to troubleshoot process problems and weak points and make a product or process more robust against external and non-controllable influences. In this case, given *k* factors, a quadratic model of response *Y* is commonly adopted including N terms: a constant term, *k* independent linear terms and *k* independent quadratic terms (i.e., one for each factor in both cases) and *k*(*k*-1)/2 mixed quadratic terms that take into account the interactions between all possible couples of different factors. A minimum of N measurements based on different combinations of experimental factor values is thus required to obtain the parameters of the model through the resolution of a linear system of N equations in N unknowns. Under these conditions, however, no estimate of the parameter variability (and then of its significance) can be obtained in this case unless a number of experiments greater than that of the parameters is performed, e.g., by replicating experiments for specific combinations of factors. A multivariate linear regression can be adopted to find the values of parameters (and obtain estimates of their variability) in this case.

A fundamental aspect to guarantee a good robustness of the results especially when the number of measurements has to be minimized is the choice of the combinations of the *k* factor values. This is the aspect that most differentiates RSM approaches actually reported in the literature for the optimization of the extraction conditions for wine volatile compounds. Arcanjo et al. [31] reported the application of RSM based on a rotatable central composite design (RCCD) to improve the conditions for extracting volatiles from Isabel wine. In this work, the RCCD was used to evaluate the effects of three independent parameters: temperature, equilibration and extraction time on three different responses, namely, the total number of peaks detected upon a chromatographic separation, the total area under the detected peaks and the area of the peak of isoamyl acetate, a typical volatile compound present in wine headspace, to define the best conditions for a high extractability of volatiles. A central composite design (CCD) based on a factorial design plus eight axial points plus five replicates in the center of the design was chosen by Barros et al. [42] to develop a new automatic HS-SPME method for the determination of volatiles in white wines produced from several grape cultivars. The Box Behnken design (BBD) was applied in research for characterizing aroma-active monoterpenes in berries to determine optimum levels of three variables influencing the terpene recovery on SPME fiber: extraction temperature, extraction time and equilibrium time and consequently to display their linear and quadratic effects [29]. Recently, the DoE approach consisting of a definitive screening design (DSD) and latent variable modelling has been reported for the optimization of a method based on the HS-SPME-GC/MS analysis of volatile fatty acids in wine [14]. A DSD is a particular class of three levels of screening designs and is capable of providing estimates of the main effects that are unbiased or unconfounded regarding all second-order interactions and among themselves. The original version of the DSD was restricted to quantitative factors [43] but a more recent evolution extended the application to incorporate two-level categorical factors [44]. These designs require a low number of runs, one more than twice the number of factors. For more than six factors, they also have the potential capability to estimate the full quadratic model.

#### **3. Matrix Effects and Calibration Approaches**

Wine volatile composition (influencing the headspace-extracting phase partition coefficient), the non-volatile wine matrix (influencing the liquid-headspace partition coefficient) along with ethanol levels (influencing both coefficients) were assessed as the factors mainly responsible for matrix effects [3,19,45]. Indeed, SPME is extremely sensitive to any experimental parameter that may affect the liquid-headspace and the headspace-extracting phase distribution coefficients.

Rocha et al. [45] investigated the effect of variations in the concentration of each analyzed standard on the other components of a model wine in order to clarify the extent of the changes in the concentration of one matrix component on the headspace equilibrium and, consequently, in the SPME absorption of the other liquid matrix components. The authors showed that the increase of the concentration of one compound results in a decrease of the absorption of all of the others and, as a consequence, of their relative response factors. Compounds with higher relative response factors were less influenced by the matrix composition. Thus, quantification by SPME was shown to be highly dependent on the matrix composition.

Rodríguez-Bencomo et al. [3] investigated the effects of five representative matrices, namely, a young white wine, a young red wine, an oak-aged red wine, a sparkling wine and a sweet, aged wine after deodorization, compared with a control model wine with no matrix effect. The authors found that the wine non-volatile fraction (composed of monosaccharides and disaccharides, amino acids, polyphenols and proteins and other high molecular weight compounds) strongly affected the volatility of odorous molecules through two different effects: retention and salting-out effects. In the first case, the volatile compound was retained by the non-volatile matrix (particularly by mannoproteins); thus, a decrease in the amount of aroma in the headspace was observed; in the second case there was the opposite situation, i.e., a few soluble non-volatile compounds (monosaccharides, disaccharides and amino acids) could bind water molecules thus decreasing the amount of free water in the matrix and increasing the concentration of the target compound, which in turn favored its volatilization. The behavior of the investigated matrices was quite different. Indeed, the non-volatile matrix composition is not the same in all wines; as an example, a red wine is significantly richer than a white wine in terms of polyphenols and tannins with effects on analyte solubility and on the interactions of the analyte matrix components.

Ferreira et al. [19] adopted three non-volatile matrices obtained from a white wine, a young red wine and an aged red wine, respectively, after complete dealcoholization and de-aromatization to investigate matrix effects in an SPME wine analysis. The authors revealed strong effects due to ethanol to the content of other major volatiles and to the composition of the non-volatile matrix depending on both the wine matrix composition and the compound nature. In particular, the authors showed that the effect of ethanol, which has potential effects on analyte solubility and on fiber extraction, due to potential competition, was higher than 20% of the total variability in 20 out of 61 investigated compounds. In general, the higher the ethanol level, the smaller the peak area. The effect of the levels of major volatile compounds accounted for more than 20% of the total variability in 19 cases and in a few, the effect of this factor became dominant. In most cases, the higher the level of major volatile compounds, the lower the signal. The authors attributed this behavior likely to competitive effects in the fiber exerted by some major compounds. The non-volatile matrix exerted an effect higher than 20% in 21 cases and in 11 cases became the dominant factor. In this case, the effects did not follow a well-defined trend, suggesting that intermolecular interactions occurred between a few analytes and matrix components.

In order to quantify volatile compounds in wine while keeping matrix effects into account, several approaches have been proposed including internal standards, model wine calibration, a stable isotope dilution analysis (SIDA), matrix-matched calibrations and standard addition methods (Table 1).


**Table 1.** Advantages and disadvantages of SPME calibration methods.

The use of an internal standard (IS) added to the matrix to be analyzed before extraction can compensate for matrix effects, losses of analytes during extraction and irreproducibility in instrumental responses but cannot fully avoid them; thus, quantification mistakes may occur. The IS must not be present (at least at detectable levels) in the samples, it should be similar in analytical behavior to the target analytes and it should be well-resolved from the latter. The compounds 4-methyl-2-pentanol and 2-octanol are the most used internal standards for volatile wine analysis [46,47]. An internal standard can also be employed for the equilibrium in fiber standardization. In this case, it is loaded on to the fiber coating prior to the extraction step instead of being spiked into the sample. The technique was developed for automated sampling from milliliter quantities of liquids in vials and used for the analysis of benzene, toluene, ethylbenzene, xylenes [48] and two carbamate pesticides [49] in a wine sample. However, no application has been yet reported in the literature for the determination of a wide range of volatile compounds in wine.

In general, quantitation with an IS is based on the calculation of the ratio between analytical responses (usually chromatographic peak areas) obtained for the analyte and the IS that is subsequently multiplied by the IS concentration. The analyte concentration is thus expressed as IS equivalents. By using this equation, the response factor (i.e., the ratio between a signal produced by an analyte and the quantity of the analyte producing the signal) of all analytes is supposed to be equal to the one of the IS. This is a quite drastic assumption when HS-SPME is adopted because it implies not only that analytes and the IS have the same instrumental response but also that they exhibit a similar behavior in terms of transfer into the headspace and of yield of extraction and then of desorption from the SPME coating. In other terms, they are expected to be influenced equivalently by matrix effects.

As it cannot be usually expected that the IS behaves the same as all of the analytes and, indeed, this requisite is generally not verified, quantification based on an IS is considered a semi-quantitative method. However, this approach is simple, versatile and can be successfully used when just a comparison of the number of individual compounds among samples is desired [50,51]. As the conversion of the analyte/IS response ratio into the analyte concentration can be misleading, many authors have expressed their results as a chromatographic peak area percentage for each volatile in order to compare it among the studied treatments [52,53]. In this case, an equal matrix effect on all analytes is also supposed to occur and the single peak area/total peak area ratio allows just a rough correction.

In model wine calibration, commercial standards are dissolved in ethanol and successively diluted in a wine model solution (e.g., 12% ethanol and 5 g/L of tartaric acid at pH 3.2 or more complex ethanolic solutions) to prepare the different levels of a calibration

curve for each compound. Volatile compounds are analyzed at each calibration level using the same method as that adopted for real wine samples and linear regression is applied to calculate the respective concentration [54]. In this case, a different response of analytes due to their different nature can be inferred; however, no matrix effect can be considered. The internal standard can be added to the model wine calibration solutions and real samples to compensate for the losses of analytes during extraction, irreproducibility in the instrumental response and matrix effects. However, this approach is reliable only if the matrix effect is assumed to be equal for the IS and for all of the analytes. In this case, a calibration plot is developed by calculating the ratio of the peak area of the analyte to the IS one for calibration solutions containing different analyte concentrations with a fixed IS concentration. This ratio is subsequently used to calibrate the sample [55,56]. In order to improve the accuracy of results, the use of multiple standards has been also proposed [57].

When the approach of model wine calibration is used, the accuracy of the data can be established by the calculation of recovery R, which is obtained by comparing a wine sample spiked with known concentrations of the analytes (Cadded, generally one or two levels) and an unspiked wine extracted and analyzed by the same procedure. Concentrations of the spiked (Cspiked) and unspiked (Cunspiked) samples are then calculated using the calibration line equation and R is finally evaluated as: (Cspiked–Cunspiked)/Cadded). However, percent recoveries exceeding 100 are often found for several analytes thus suggesting the presence of enhancive matrix effects, which are not easily confirmable if the recovery is calculated at a single or at two concentration levels [21].

Furthermore, a complete set of the standards required for the calibration is generally not commercially available. Therefore, the concentrations of volatile compounds for which it is not possible to establish calibration curves were estimated by several authors on the basis of the equations of compounds with the same functional groups and/or similar numbers of C atoms [55]. However, the ideal internal standard is an isotopically labeled analogue (usually its polydeuterated analogue) of the analyte of interest. Indeed, calibration with isotope labelled standards can produce satisfactory results though the compounds are not available for all analytes of interest and thus should be synthesized. In this case, each analyte is assumed to experience the same matrix effects as its isotopically labelled homologue.

Sejer Pedersen et al. [58] described a new method for the analysis of geraniol, nerol, linalool and α-terpineol in wine based on the preparation of [2H7]-geraniol, [2H7]-nerol, [2H7]-linalool and [2H7]-α-terpineol for the use of internal standards followed by liquidliquid extraction and a GC/MS analysis. Standard solutions with unlabeled compounds and internal standards were prepared for calibration purposes. For comparison, they also used an HS-SPME-GC/MS method finding an adequate sensitivity and limits of quantitation down to 1 μg L–1 or below with, however, a worse signal-to-noise ratio especially at lower concentrations.

Siebert et al. [59] first proposed the use of polydeuterated internal standards for stable isotope dilution analyses of 31 wine fermentation products by HS-SPME-GC/MS. Nine of the labelled standards were commercially available while 22 were synthesized. Capone et al. [60] developed a SIDA method for the quantitative analysis of seven C6 compounds. The two methods above [58,59] were successfully used by Ugliano et al. [61] and Bindon et al. [62]. Furthermore, Bindon et al. quantified isobutyl methoxypyrazine (IBMP), isopropyl methoxypyrazine (IPMP) and sec-butyl methoxypyrazine (SBMP) using IBMP-d3 as an internal standard. More recently, Tomasino et al. [63] used the SIDA approach to quantify monoterpene isomers in forty-six Pinot Gris wines by HS-SPME multidimensional gas chromatography mass spectrometry. Dang et al. [64] quantified volatile phenols using relative peak areas, i.e., the ratio between the peak area of the analyte and the one of the normalizing standard (d3-4-methylguaiacol). Bee-DiGregorio et al. [65] used deuterated internal standards for the linalool and 3-isobutyl-2-methoxypyrazine determinations. The authors prepared calibration solutions by spiking unlabeled and deuterated internal standards into Chardonnay juice or Pinot Noir homogenate containing no detectable linalool

or 3-isobutyl-2-methoxypyrazine, respectively. However, although the SIDA has been shown to offer an effective strategy when a narrow range of analytes is investigated, this approach remains little used (just three out of twenty-five wine publications in 2020 used this method).

Among the further quantification methods cited before, matrix-matched calibration involves the preparation of several standard solutions in a sample matrix and then the extraction, desorption and analysis of volatile analytes under the same conditions are adopted for real samples so that a calibration line can be obtained in the presence of matrix effects. As usual, the calibration line equation can be subsequently used to estimate the concentrations of analytes in real samples once these are subjected to extraction and analysis and the corresponding analyte responses are obtained. Provided that standards are available for the analytes of interest, multi-analyte stock solutions can be prepared and then easily used for the preparation of matrix-matched multiple standards. Unfortunately, the matrix-matched calibration has a significant limitation as it requires a blank matrix, i.e., a matrix totally free from target analytes, to be available, which is a quite difficult requirement in the case of wine.

Liu et al. [66] firstly proposed the use of a volatile-free wine obtained by distillation for calibration purposes. The authors compared the chromatographic peak areas of 19 compounds, which were obtained using several model wine matrices including water, an 11.5% ethanol/water (*v/v*) solution, a concentrated synthetic wine, a 'volatile-free' wine and a real wine spiked with the same number of analytes. The peak areas of the analytes decreased significantly with the increase in the complexity of the matrix composition suggesting the presence of interactions between the interfering substances and the analytes. The slopes of the calibration lines (taking into account the peak area relative to an internal standard) obtained using the model wine matrices were completely out of the confidence interval of the slope values related to the volatile-free wine and the real wine for most of the volatile compounds. Conversely, the slopes of the linear models obtained with the volatile-free wine were within the confidence interval of the slopes obtained with the real wine for 16 out of the 19. A few other authors have used different deodorized and representative wines [3,19] according to the complexity of the matrix (e.g., young white wine, young red wine, oak-aged red wine), confirming the occurrence of interactions between the matrix and the analytes and the need to use a de-aromatized matrix for accurate calibrations. Furthermore, beside using a de-aromatized matrix, Ferreira et al. [19] proposed the employment of a large pool of internal standards and a calibration strategy based on the calculation of multivariate internal standards (MIS) as linear combinations of the normalized signals of multi-internal standards. However, in order to verify the occurrence of matrix effects when proposing other matrices for calibration, all of the authors cited so far [3,19,66] compared the results with a calibration in real wine. In other words, they used the technique of standard addition as a control because this calibration technique is considered the most reliable.

The technique of standard addition [67] is based on the addition of known concentrations of the analyte of interest to multiple aliquots of the sample, leaving the sample volume (and the proportion ethanol/water) virtually unchanged. Thus, no effect of ethanol can be observed as no variation of its concentration occurs. The sample as such and the spiked samples are subsequently analyzed and a plot of the analytical response versus the added concentration is constructed for each analyte. In this case, the intercept of the regression line on the axis reporting added concentrations provides an estimate of the concentration in the original sample. In the context of the HS-SPME-GC/MS analysis of wine volatiles, the standard addition method was adopted by Carrillo et al. [38] for the quantification of volatile oak compounds in aged red wines. In this case, a significant difference with the results provided by the internal standard method was found for most of the compounds [38]. The standard addition method was also used by Pizarro et al. [68] to quantify haloanisoles and volatile phenols in both white and red wines. The method was also adopted in conjunction with vacuum-assisted HS-SPME-GC/MS for the determination

of haloanisoles in red wines [69]. Recently, Tufariello et al. [21] compared a model wine calibration and the standard addition method for the quantification of volatiles in a white sparkling wine. As a result, they found that the quantification based on the model wine was not accurate for several analytes (with concentrations differing even by one order of magnitude) due to both suppressive and enhancive effects, depending on the analyte, on the final response exerted by the wine matrix. Therefore, although time-consuming, especially if a large number of samples has to be analyzed, the standard addition method should be the eligible method when accurate quantitative results for wine volatiles are highly desirable.

#### **4. Conclusions**

HS-SPME-GC/MS is a well-established technique for the analysis of volatile compounds in wine, offering undeniable and well-known advantages such as easiness of use, solventless analyte extraction and potentially high sensitivity. Unfortunately, the presence of many variables to optimize in the method development step may become a difficult challenge. However, the use of multivariate optimization approaches can be very useful to reduce the complexity of method optimization due to their capability to provide a lot of information when many factors are involved without requiring a high number of experiments. The quantification step of analytes is another critical point that needs to be carefully considered when performing an HS-SPME-GC/MS analysis of wine volatile compounds due to the potentially high relevance of matrix effects. Indeed, different quantification methods are usually adopted but none of them appear to be free from drawbacks such as failures in matrix effect reproduction (e.g., using model wines that cannot be considered reliable surrogates of real wines) or the need for time-consuming measurements (such as the standard addition method) or expensive compounds (such as isotopically labelled standards). The choice of the most appropriate quantification method should thus depend on a careful balance between the actual goal of the analysis (sometimes absolute concentrations of specific analytes implying more complex approaches are not strictly required) and practical aspects.

**Author Contributions:** Conceptualization, S.P. and I.L. review of literature—S.P., M.T., F.G., P.C., A.C.; writing—original draft preparation, M.T., S.P. and I.L.; writing—review and editing, S.P., M.T., F.G., P.C.; supervision, I.L. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** Data sharing not applicable.

**Acknowledgments:** This work was partially supported by the Apulia Region projects: "Innovazione nella tradizione: tecnologie innovative per esaltare le qualità dei vini autoctoni spumante della murgia barese-INVISPUBA" (P.S.R. Puglia 2014/2020 Misura 16.2); "Birra: dal campo al boccale - BEˆ2R" (P.S.R. Puglia 2014/2020 Misura 16.2). We would like to thank Domenico Genchi, Leone D'Amico, Vittorio Falco and Giovanni Colella of the Institute of Sciences of Food Production—CNR for their skilled technical support provided during the realization of this work.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


MDPI St. Alban-Anlage 66 4052 Basel Switzerland Tel. +41 61 683 77 34 Fax +41 61 302 89 18 www.mdpi.com

*Processes* Editorial Office E-mail: processes@mdpi.com www.mdpi.com/journal/processes

MDPI St. Alban-Anlage 66 4052 Basel Switzerland

Tel: +41 61 683 77 34 Fax: +41 61 302 89 18

www.mdpi.com