Measures to improve wine malolactic fermentation.

This review focuses on the considerable amount of research that has been directed towards the improvement of efficiency and reliability of malolactic fermentation (MLF), which is important in winemaking. From this large body of work, it is clear that reliable MLF is essential for process efficiency and prevention of spoilage in the final product. Impediments to successful MLF in wine, the impact of grape and wine ecology and how this may affect MLF outcome are discussed. Further focus is given to how MLF success may be enhanced, via alternative inoculation strategies, MLF progress sensing technologies and the use of different bacterial species. An update of how this information may be used to enhance and improve sensory outcomes through metabolite production during MLF and suggestions for future research priorities for the field are also provided.

Keywords: Oenococcus oeni; Lactobacillus; Malolactic fermentation; Wine

During winemaking, the initial conversion of grape must to wine is an alcoholic fermentation (AF) carried out by one or more strains of yeast, typically Saccharomyces cerevisiae. After the alcoholic or primary fermentation, a secondary fermentation known as malolactic fermentation (MLF) is often undertaken, depending on the style of wine that the winemaker seeks to achieve. Malic acid is one of the predominant organic acid in grapes, occurring in amounts of the order of 3 g L−1 (Palma and Barroso 2002). This compound can therefore contribute to the acidity, pH and mouthfeel of wine and can be a nutrient for several spoilage organisms. Accordingly, a malolactic fermentation (MLF) by which l-malic acid is decarboxylated to l-lactic acid serves several purposes: to reduce the harsh acidity of malic acid, give a concomitant modest increase in pH as well as increase wine microbial stability. The former of these outcomes are of greatest importance in sparkling and some white wines, whereas most red wines undergo MLF primarily for stability. Subsequent effects due to MLF also include impacts on both aroma and visual profile (Burns and Osborne 2013; Sumby et al. 2010).

S. cerevisiae has long been recognised as a poor metaboliser of extracellular malic acid, due to a lack of a mediated transport system, low substrate affinity for l-malic acid and the mitochondrial location of the malic enzyme MAE1, which catalyses the oxidative decarboxylation of malate to pyruvate (Boles et al. 1998). More recent reports have suggested that some strains of S. cerevisiae are able to consume up to 20% of the malic acid during fermentation (Redzepovic et al. 2003). Blazquez Rojas et al. (2012) reported that two S. cerevisiae strains (AWRI 1493 and AWRI 1554) could consume malic acid, but the quantum is not clear since the starting amount was not reported. However, even with strains capable of consuming l-malic acid during a typical ferment (3 g L−1), there would still be more than 1-2 g L−1 remaining.

MLF is typically conducted by lactic acid bacteria (LAB), particularly Oenococcus oeni, a Gram-positive microorganism that is thought to have evolved to exclusively survive in the fermented beverage environment (Campbell-Sills et al. 2017a). O. oeni usually grows more slowly than other LAB, but ultimately triumphs in wine owing to its greater tolerance to the combination of ethanol and acid (Lonvaud-Funel 1999). Despite this advantage, MLF can be uncertain and protracted thereby lengthening processing time and reducing winery throughput. All the while, wines are left with minimal protection from the preservative SO2 so as to encourage MLF, increasing the risk of wine oxidation or contamination with spoilage organisms. Several non-Saccharomyces species are also capable of metabolising l-malic acid by converting it into ethanol through malo-ethanolic deacidification (Benito et al. 2015; Volschenk et al. 2003). They could therefore be an alternative to traditional MLF. However, the sensory effect of using non-Saccharomyces yeasts instead of LAB to conduct MLF is unclear and needs to be studied further. Hence, the use of alternative yeast species is not the focus of this review.

While most reports relate to wine made from grapes, MLF may be conducted in other beverages and foods, such as pineapple juice, cider, durian pulp fermentation, cherry wine and many others. Nevertheless, while the information discussed here will be relevant to other fermented foods, this review focuses on what is known about the impediments to successful MLF in wine. Greater process reliability as well as an ability to tailor wine composition are attributes keenly sought by the wine industry in any new measures to improve MLF. To this end, the selection or generation of more robust strains, identification of alternate LAB, use of LAB-yeast co-inoculation instead of sequential inoculation strategies, alternate biomass introduction/removal methods, improvement of monitoring and a greater attention to the sensory influences of these organisms are all approaches by which researchers have sought to meet this demand. A summary of key recent findings under several of these strategies is included here, along with a discussion of future opportunities. For details of specific strain improvement methodologies, the reader is referred our earlier review (Betteridge et al. 2015).

Efficient control of MLF requires an extensive knowledge of the response of LAB to the stressful conditions found in wine. The ability of a LAB to undergo MLF is influenced by many factors including pH, temperature, wine inhibitor content (e.g. ethanol, SO2, medium chain fatty acids (MCFAs)), nutrient limitation, other potential as yet unknown factors, the yeast strains carrying out AF and interactions with the indigenous microflora of the fermentation (Cinquanta et al. 2018; Guzzon et al. 2009; Liu et al. 2017b). Survival in wine under such multi-stress conditions requires the maintenance of the functionality of the cell membrane, in order to control ion permeability and regulate solute and nutrient exchange between the cell and the external medium. Ethanol is considered to be the main stressor in wine because it can injure cell membrane integrity and impact cell viability. Ethanol tolerance is widely reported to be strain specific and the ethanol stress response is complex and well studied (reviewed by Bonomo et al. 2018).

The second most important stressor in wine is low pH. Most wines have a pH ranging from 3.8 to 3.2, with wines at the higher values being more prone to microbial spoilage as well as biogenic amine formation (Cinquanta et al. 2018). The physiological effect of pH on LAB is also well defined. RNA sequencing revealed the differential expression of several genes related to the metabolism of amino acids, carbohydrates, membrane transport and energy metabolism as part of the genetic response of O. oeni strain SD-2a to low pH (3.0 vs. 4.8; Liu et al. 2017a). More recently, Cinquanta et al. (2018) studied the effect of pH in two Italian wines (Falanghina and Tintilia) inoculated with O. oeni, Lb. plantarum or a 50:50 mix of both. The duration of MLF was influenced by the pH and the LAB strain used, with both O. oeni and Lb. plantarum completing at pH 3.8, neither completing MLF at pH 3.2 and Lb. plantarum failing to complete fermentation in Falanghia wine at pH 3.4. Interestingly, an evaluation of the capacity of a new Lb. plantarum V22 starter culture to complete MLF at the laboratory and semi-industrial scale revealed that bacterial survival was related more to pH evolution during MLF than the initial pH of the must (Lerena et al. 2016). In all cases, the musts showed initial pH values over 3.6, but the pH showed dynamic behaviour, changing as MLF progressed. In fermentations in which pH increased over time, Lb. plantarum V22 successfully metabolised most malic acid originally present in the must. By contrast, when the pH decreased over time, bacterial counts declined accordingly, as did the rate of malic acid consumption. This decrease in pH may be due to increased acetic acid production (if the Lb. plantarum strain was facultatively heterofermentative) or due to interactions with the yeast strain (which may produce various organic acids during fermentation (Henick-Kling 1993)). Such changes in must/wine pH and their impact on MLF kinetics should be studied further to allow better tailoring of strain performance and wine conditions.

Where bacterial growth or MLF are initiated despite high ethanol content, or unfavourable pH values, the risk of stuck MLF remains. Key contributors are high total SO2, lack of nutrients or phage infection, but there may be other as yet unknown factors such as specific inhibitory yeast metabolites. Although SO2 produced by yeast during AF will exist in the bound form immediately after fermentation (mostly to acetaldehyde), the total SO2 is still inhibitory to MLF because bacteria can metabolise the acetaldehyde fraction, releasing a proportion of SO2 (Osborne et al. 2000; Wang et al. 2018). A lack of nutrients can also be problematic and can be mitigated by the addition of commercially available substitutes; however, this should be used with caution as it is not advisable to leave nutrients in the wine for other spoilage microorganisms to utilise.

An additional impediment to successful MLF is the effect of phage on the LAB strain conducting MLF. There has been reinvigorated interest in this over the last couple of years and it has been reported that the low pH and high ethanol conditions found in wine can affect the lytic activity of phage (Costantini et al. 2017; Henick-Kling et al. 1986). This effect could be due to the modification of the bacterial cell surface induced by stress conditions. However, as phage have been isolated from wines having difficult MLF, other factors such as sensitivity of phage to ethanol, pH, and SO2 may also play a role in O. oeni resistance. Phage can potentially lyse bacterial cells at the start of AF and thereby interfere with MLF. Interestingly, SO2 is reported to have an antiviral activity (Henick-Kling et al. 1986; Philippe et al. 2017) and O. oeni strains that are more resistant to SO2 may be further benefited by protection from phage attack with small SO2 additions (5 g hL−1).

Finally, it is possible that strain-specific differences in mleA (malolactic enzyme) expression and l-malic acid consumption are due to the individual strain's ability to adapt to increased ethanol concentrations. Miller et al. (2011) reported that while malic acid and low pH increased mle expression in Lb. plantarum, increased ethanol concentration reduced mle expression. Accordingly, very low levels of malic acid could also be a reason MLF does not proceed to completion. Certainly, ethanol has been reported on numerous occasions to be an inhibitor of MLF (Gockowiak and Henschke 2008; Vailiant et al. 2008), thus a decrease in mleA expression could be the basis for MLF failure. More recently, Betteridge et al. (2018) reported that O. oeni strain A90, derived from directed evolution experiments to be better adapted to high ethanol, showed an initial drop in mleA expression (1 h of ethanol exposure), but ultimately had higher mleA expression than the parent (SB3) after 24 h in high ethanol conditions. This isolate also consumed l-malic acid faster than the parent in the presence of ethanol.

Although these impediments are well known, they provide ready targets for strain optimisation or culture/fermentation management approaches to help improve the success of MLF. The challenge still remains significant given the diversity of MLF conditions and practices as well as the microflora involved.

A diverse community of microorganisms is present on grapes and therefore transfers into winemaking with the possibility of influencing wine processing and sensory properties. Of these, LAB not only contribute to decarboxylation of malic acid but also produce other benefits such as the liberation of aroma precursors and positive enzymatic activities (proteases, lipases, esterases, tannases, glycosidases; Grimaldi et al. 2005a, 2005b; Matthews et al. 2006). In attempting to define the microbiology of grapes and wine, the use of culture-based or culture-independent methods to reveal these microorganisms determines apparent population complexity, with the latter generally uncovering more species, some in very small numbers.

In a review by Barata et al. (2012), over 50 microbial species were identified on grape berries. Of these, the number of LAB (mostly Lactobacillus spp. and Pediococcus spp.) was limited to a few species totalling only of the order of 102 cfu g−1 on sound grapes. LAB species isolated from grapes include Lb. plantarum, Lb. casei and Lb. hilgardii. Grape must contains a greater diversity of species and in addition to the former includes Lb. brevis, Pediococcus damnosus, P. parvulus, P. pentosaceus, Leuconostoc mesenteroides and O. oeni (Davis et al. 1985, 1986, 1988; Miranda-Castilleja et al. 2016). LAB densities in crushed grapes are about 102 cfu mL−1 to 104 cfu mL−1, depending on climatic conditions towards the end of grape maturation, which is inversely correlated with must acidity (Lonvaud-Funel 1999). The frequency of detection on grapes specifically of O. oeni is much lower and requires adequate methods to promote the development and identification of minority populations (Franquès et al. 2017; Renouf et al. 2005, 2007). The general consensus is that O. oeni cannot be detected on grapes and they are often not detected by culture until the end of alcoholic fermentation (AF) and during MLF (Ultee et al. 2013). One exception to this (Franquès et al. 2017) involved an extended 15 days of semi-selective cultivation in MRS medium supplemented with l-malic acid, fructose, nystatin, sodium azide, l-cysteine and tomato juice, implying that the low numbers make the selection method critical when trying to isolate O. oeni present prior to MLF.

There is limited information regarding the succession of LAB in fermentations where the traditional approach of inoculating for MLF after the primary fermentation is used. Further in this context, the interactions of the inoculum with the indigenous microflora are poorly defined. By comparison, it is known that when LAB are introduced earlier in the process and co-inoculated with yeast, there is often a drop in cfu (of between 101 and 104 cfu mL−1) during the first 24 h followed by rapid growth towards numbers required to carry out MLF (Ong 2010; Fig. 1). This ability of inoculated strains to implant and recover in the juice/wine is critical to MLF success. During fermentations that are not inoculated, population profiles vary depending on SO2 additions prior to crushing, but the general trend remains the same, i.e. O. oeni and some species of Lactobacillus can survive during AF, whereas Pediococcus and other LAB gradually decline (Fig. 1). Thus, during the first days of AF, the number of LAB usually increases to near 104 cfu mL−1 and then decreases to around 102 cfu mL−1 at the end of AF, due mainly to competition from yeast and sensitivity to SO2 and ethanol. After AF, LAB (almost exclusively O. oeni, but some Lactobacillus sp. may be present) increase and MLF begins when cells reach 106 cfu mL−1 (Fig. 1). O. oeni is largely responsible for MLF given it is the species best adapted to wine.Inoculation strategies and an amalgamation of the growth of the two main LAB species present during wine fermentation (un-inoculated). Data taken from Bokulich et al. (2016); Davis et al. (1986); Marzano et al. (2016); Ong (2010); Wibowo et al. (1985). Other LAB species would be expected to reach a higher cell number if the pH is above 3.5 (Davis et al. 1986). Chains of purple cocci—Oenococcus; chains of pink rods—Lactobacillus

PHOTO (COLOR)

Early, culture-based studies on this topic reported the succession of bacterial populations through stages of wine fermentation, thereby offering novel insights into these microbial communities. More recently conducted surveys of this type have been completed with metagenomic approaches, the richness of data attesting to the superiority of next generation sequencing (NGS) over classical methods (David et al. 2014). Such genomic methods for determining the grape and wine microbiome (reviewed by Morgan et al. 2017) have in fact led to an exponential increase in information on species abundance both before and during un-inoculated fermentation (Bokulich et al. 2016; David et al. 2014; Marzano et al. 2016; Piao et al. 2015; Portillo and Mas 2016). For example, Pinto et al. (2015) used high-throughput sequencing to fully characterise both eukaryotic and prokaryotic communities in samples collected from six Portuguese wine regions and reported a clear relationship between the microbial community and fermentation stage. As expected, the biodiversity decreased for both prokaryotic and eukaryotic communities as the selectivity of the environment increased with progression of fermentation. LAB were detected at low abundances and O. oeni was not detected. Amongst the LAB, high numbers of Lactobacillus (Lactobacillaceae), Leuconostoc (Leuconostocaceae), Lactococcus and Streptococcus (Streptococcaceae) were present. In a survey of over 200 commercial wine fermentations, the presence of multiple species of yeast and bacteria throughout the ferment was observed, although only reported as relative abundance (Bokulich et al. 2016). Therefore, as the researchers increase the use of genomics to study wine fermentations, an opportunity arises from the deposition of raw data into publicly accessible databases to allow comparisons between studies.

Actual numbers aside, greater insights into bacteria in fermentations are being reported. Portillo and Mas (2016) showed that acetic acid bacteria and LAB are more abundant than previously thought in a Grenache wine fermentation study, with similar results arising from low-sulphured or unsulphured wine fermentations (Bokulich et al. 2015). Additionally, NGS analysis has confirmed that bacteria not previously described in this context may also be present during the process (Godálová et al. 2016). Results such as the above are likely to vary depending on SO2 additions. As expected, SO2 affects microbial diversity in a dose-dependent manner, with 25 mg L−1 being cited by Bokulich et al. (2015) as the ideal concentration to achieve microbial stability when wine pH is sufficiently low. These other species may as yet be implicated in successful/unsuccessful fermentation outcomes and further study as to their impact on both AF and MLF is needed. For example, some Lactobacillus sp. have been implicated in causing stuck AF (Bokulich et al. 2016), and different strains of O. oeni have been shown to interact in either a negative or positive way depending on the strains tested (Brandam et al. 2016). Additionally, Ramakrishnan et al. (2016) reported the ability of LAB strains to induce a metabolism-modifying prion [GAR+] in S. cerevisiae, hampering early yeast dominance in the fermentation and delaying the rapid depletion of amino acids by the yeast thereby enabling proliferation of bacteria present in the juice.

It is therefore possible that the inherent microbial diversity could lead to an increased risk of incomplete fermentation. Fermentations are a complex environment where yeast and LAB can interact either by cell-to-cell contact or by production of molecules that influence the survival and activity of other cells in that environment (Liu et al. 2017b). Study of these interactions is providing fascinating insights. Yeast-LAB interactions, including effects on wine, have been reviewed recently by Liu et al. (2017b). Although this review only includes studies up to 2014, it highlights the importance of understanding the impact of microbial diversity on MLF outcomes. Many factors can influence compatibility between yeast and LAB and therefore alter wine attributes. Problematic MLF can arise due to production of inhibitory compounds including medium-chain fatty acids, ethanol and SO2 and depletion of essential nutrients prior to MLF (Bisson 1999; Capucho and San Romão 1994; Comitini et al. 2005; Fleet 2003; Fourcassie et al. 1992; Gobert et al. 2017; Guilloux-Benatier et al. 1998; Knoll et al. 2008; Lonvaud-Funel 1995; Osborne and Edwards 2006; Ultee et al. 2013). However, yeast production of such compounds does not fully explain problematic MLF due to the complexity of yeast-LAB interactions and apparent strain- and cultivar-specific compatibility (Nehme et al. 2008; Tristezza et al. 2016). Additionally while there is potential for cell-to-cell contact to also play a distinct role in MLF success or failure (Ramakrishnan et al. 2016), there is currently no identification of specific mechanisms underlying physical interactions and their subsequent effect on MLF outcomes. Consequently, the link between yeast-LAB interactions, compatibility and sensory outcomes remains ill-defined.

It is also difficult to predict MLF progression based on yeast-LAB presence alone since compatibility depends on the combination of yeast and LAB used, the timing of LAB inoculation and the grape cultivar (Comitini and Ciani 2007; Delaquis et al. 2000; Maicas et al. 1999; Mendoza et al. 2010; Ugliano and Moio 2005). Thus, despite many studies revealing the importance of yeast-LAB combinations on MLF outcomes for wines, in-depth knowledge offering predictability of robust MLF based on yeast, LAB and grape variety selections is lacking. The interactions between Saccharomyces, non-Saccharomyces and LAB are complex and will take time to be delineated. With the rise in popularity of indigenous (un-inoculated) fermentations, the interactions of the wine microbial community as a whole is becoming more important and merits further study. More data on how microbes including Saccharomyces and non-Saccharomyces yeast and LAB interact with each other and (in the case of indigenous fermentations) other microbes will enable winemakers to develop winemaking practices (e.g. amount of SO2 on grapes pre-ferment) that encourage selection of appropriate microbial communities for improved and more successful fermentations.

MLF will often occur during the typical succession that transpires in most wine fermentations. However, MLF by indigenous strains may be slow or incomplete, and it is often more expedient to inoculate with an MLF starter culture. Multiple inoculation strategies can be used, but there are two main ones, sequential and co-inoculation (reviewed in Bartowsky et al. 2015 and Sumby et al. 2014). In the context of MLF, sequential inoculation refers to the practice of allowing alcoholic fermentation (AF) to complete before addition of LAB to initiate MLF. Co-inoculation for MLF involves the addition of an LAB culture early in the primary fermentation, often 24 h or more post-yeast inoculation (Fig. 1). Where LAB starter cultures are used, there are also a number of formulations and application methods. For example, starter cultures that have received proprietary treatment to prepare the cell membrane to cope with the stresses found in wine can be added directly to the fermentation. Alternatively, a freeze-dried starter (e.g. O. oeni) is reactivated and adapted (with or without the addition of a specific activator) followed by acclimatisation to the wine. A third method involves propagation and adaptation in the winery, as is often done for low pH and acidic sparkling or white wines. In this case, the bacteria are propagated under progressively increasing stress over several days. Regardless of the inoculation strategy chosen, timing of MLF induction can have a significant effect on chemical and sensory properties of wine and a greater effect on sensory properties in comparison to yeast treatment alone (du Plessis et al. 2017).

The traditional MLF management practice adopted by most wineries is sequential inoculation whereby LAB are inoculated following AF completion. However, co-inoculation of yeast and LAB is gaining in popularity because it can help secure and improve MLF. It can shorten the time between AF and MLF, thereby reducing the risk of microbial spoilage (Guzzon et al. 2013; Lasik-Kurdyś et al. 2017) and having multiple positive effects on wine composition (summarised by Sumby et al. 2014). Thus, the spoilage yeast Brettanomyces bruxellensis can be found at most stages of the fermentation but is particularly problematic post AF, during MLF and wine barrel ageing. Brettanomyces produces organoleptically unpleasant volatile phenols. Periods of wine processing with low levels of SO2, such as occur when seeking to encourage MLF, can favour the growth of Brettanomyces. More rapid completion of MLF so that protective levels of SO2 can be established is therefore highly desirable. Alternatively, the isolation or development of LAB that can grow and function against a background of at least moderate amounts of SO2 may represent a strain development target worthy of pursuit.

Another opportunity exists around reducing/eliminating SO2 additions during winemaking by identifying novel biological alternatives such as the potential of indigenous mixed cultures in the control of B. bruxellensis (Berbegal et al. 2017). Different strains of S. cerevisiae, non-Saccharomyces yeasts and O. oeni were co-inoculated under multiple strategies. With regard to the interaction between S. cerevisiae and O. oeni co-inoculated into spiked red must (B. bruxellensis added at 1 × 103 cfu mL−1), the results showed a decrease in 4-ethyl guaiacol and 4-ethyl phenol to below their sensory perception threshold at 21 days after commencement of AF compared to fermentation with the S. cerevisiae-only control (Berbegal et al. 2017). Interestingly, the B. bruxellensis population was also reduced in cell density in the presence of O. oeni. What remains to be tested is whether this difference persists over time. LAB show great potential to be used as either whole-cell or cell-free additions during winemaking, and this is an area requiring further study. For a review on how LAB are currently used as antifungal additions in other food industries, please see Arena et al. (2018).

Optimal yeast-LAB combinations may differ for co-inoculation versus sequential MLF (du Plessis et al. 2017; Muñoz et al. 2014). For example, although non-Saccharomyces yeast strains had a beneficial effect on the progress of a sequential MLF, during co-inoculation some Candida zemplinina and Lachancea thermotolerans strains had a negative impact on LAB growth and MLF (du Plessis et al. 2017). There is as yet only limited information on how O. oeni competes in an indigenous fermentation with either non-Saccharomyces yeast or other LAB. Based on the ability of O. oeni to increase from being often undetectable to become the dominant species in wine, there are potentially some very interesting interactions occurring between O. oeni and the indigenous microflora throughout the winemaking process.

The optimal MLF inoculation strategy for each yeast strain or yeast combination to improve wine flavour and quality appears to be strain dependent, with variation in wine composition not always amounting to perceivable sensory differences. MLF co-inoculations often lead to modifications of both volatile and sensory outcomes when compared to sequential inoculations (Cañas et al. 2012; Delaquis et al. 2000; du Plessis et al. 2017; Versari et al. 2016). du Plessis et al. (2017) reported wine flavour profile modification was dependent on the non-Saccharomyces yeast strains and MLF strategies used. Wines that underwent co-inoculation MLF scored higher for certain sensory descriptors than wines that underwent sequential MLF; in particular, lower volatile acidity and higher glycerol concentrations. Despite this, some yeast combinations yielded better wines with sequential MLF (du Plessis et al. 2017). Increased information in this area will allow winemakers to make more informed decisions and provide further insight on how to improve MLF outcomes.

It is now well documented that there are strain-specific differences in response to the stressors found in wine. O. oeni strains have a compact genome of 1.8 Mb and several metabolic pathways related to growth in oenological environments. Its genome size most likely reflects a high level of organisation and simplicity that may be the basis for its adaptation to the wine environment (Mills et al. 2005; Sternes and Borneman 2016; Zé-Zé et al. 1998, 2000). Even so, there is still a need for improvement of O. oeni isolates that conduct MLF with numerous studies looking to improve MLF process via methods such as strain improvement, and by use of alternative strains or process improvements.

There are a number of LAB that have been utilised as MLF starters belonging to the species O. oeni, Lactobacillus plantarum, Lb. hilgardii, Lb. brevis, Lb. casei and Pediococcus sp. (Table 1). Each have demonstrated different properties, with significant strain dependence in such characteristics. O. oeni has been the most utilised because of its ability to survive in the harsh wine conditions of high ethanol and low pH. But in the last decade, increased research has been directed towards other LAB species that could provide novel attributes to wine. For example, Bou and Krieger (2012) described the use of LAB strains of the genera Lactobacillus and Pediococcus that were capable of initiating and completing MLF after direct introduction, without a prior acclimatisation step (reviewed by Bartowsky et al. 2015 and Sumby et al. 2014).

Further analysis of indigenous strains able to complete MLF under regional- or varietal-specific conditions has the potential to offer up new strains with increased genetic diversity and better adaptation to local conditions. Indigenous LAB isolated at the end of MLF from 16 different Chilean wineries were shown to be genetically different from commercial strains and lacked genes conferring detrimental properties as well as genes encoding enzymes linked to aroma compounds (Romero et al. 2018). One strain in particular, 139, had several promising oenological properties including glucosidase activities (Romero et al. 2018).

Some new strains of Lb. plantarum are thought to have a greater sensory impact on wines since they can produce enzymes such as β-glucosidases, proteases, esterases and decarboxylases with potential benefits for wine composition (du Toit et al. 2011; Matthews et al. 2004). It is anticipated that these activities will be reflected in the characteristics of the resulting wines with fruity characters being enhanced after MLF performed by these organisms. More recently, nine Pediococcus spp. isolated from commercial wines were studied for their impact on the chemistry, microbiology and sensory quality of Pinot Noir wine. The strains studied demonstrated a range in production of the important flavour compound diacetyl, with some yielding concentrations above 12 mg L−1 and only one isolate producing measurable levels of the biogenic amine histamine (3.3 mg L−1). However, wine conditions may not have been optimal for biogenic amine production and this result will need to be tested in multiple wines to define any matrix-specific effect.

A number of isolates reduced colour in red wines (measured at 520nm) by over 10% while polymeric pigment content declined by almost 30% in wines inoculated with one strain of P. parvulus (Strickland et al. 2016). Such impacts may be undesirable depending on the wine style sought, i.e. rosé versus red wine. However, desirable sensory descriptors such as 'floral', 'overall fruit' and 'red fruit' were often higher in wines where Pediococcus sp. had grown compared to the control, indicating that growth of these bacteria may not always result in spoiled wine (Strickland et al. 2016). Based on the limited information about these LAB, it is clear that further characterisation of Pediococcus species and strains and their enzymatic potential will help in understanding the impact that these bacteria may have on wine (see also a recent review on Pedicoccus in wine; Wade et al. 2018).

Most LAB can degrade l-malic acid (l-malate) to l-lactic acid (l-lactate) by a direct decarboxylation (Caspritz and Radler 1983; Schümann et al. 2013) catalysed by the malolactic enzyme (MLE). This reaction is dependent on NAD and Mn2+ as cofactors, and the encoding gene is contained within an operon (Fig. 2) consisting of two genes encoding MLE (mleA in O. oeni; mleS in Lactobacillus sp.; EC 4.1.1.101), and a l-malate transporter (mleP). A LysR-type transcriptional regulator (mleR) is upstream of mleA and may control the transcription of both mleA and mleP (Labarre et al. 1996; Landete et al. 2013; Renault et al. 1989). However, regulation of the operon is yet to be fully elucidated and Galland et al. (2003) hypothesised that regulation of the malolactic operon in O. oeni may involve an alternative regulatory factor linked to metabolic energy. In this case, the connection between the malolactic system and H+-ATPase was not present in other LAB such as Lactococcus lactis and Leuconostoc mesenteroides and may be unique to O. oeni and/or other acidophilic bacteria, but this needs to be investigated further (Galland et al. 2003). A link between malic enzymes and contribution to cellular ATP has been reported in other bacterial malic enzyme systems (Meyer and Stülke 2013). Furthermore, when considering MLE activity, it is necessary to remember that structurally similar organic acids will act as competitive inhibitors for the active site of the malolactic enzyme and may affect l-malic acid metabolism. Succinic acid, citric acid and tartaric acid have all been reported to inhibit malolactic enzyme activity (Lonvaud-Funel and Strasser de Saad 1982; Naouri et al. 1990), but this inhibition effect was not tested in later papers (e.g. Schümann et al. 2013).Representation of genes and characterised pathways involved in metabolism of l-malic acid in LAB. (a, b) Schematic representation of the genetic organisation of (a) mleA/mleS and (b) mae orthologous group gene clusters present in LAB (NCBI (https://www.ncbi.nlm.nih.gov/); Landete et al. 2013; Miguel-Romero et al. 2017; Monedero et al. 2017). (c) Characterised pathways in LAB (Landete et al. 2010, 2013). Import of malic acid by O. oeni occurs through the malic acid transporter MleP, whereby l-malic acid is subsequently decarboxylated by maltate decarboxylase MleA producing l-lactic acid and CO2. MleA also requires NAD+ and manganese ions as co-factors. Lactobacillus also has malic acid transporters: MaeP and MleP. In contrast to O. oeni, Lactobacillus sp. has two pathways where malic acid may be utilised. Similar to O. oeni, Lactobacillus sp. has a malate decarboxylase MleS that is able to decarboxylate malic acid in the same way O. oeni does. The alternative pathway in Lactobacillus involves conversion of malic acid to pyruvate and CO2 by malate dehydrogenase MaeE; however, this can be repressed by glucose. When MaeE is not repressed, the pyruvate produced can be utilised for growth and other cellular processes, or as displayed here, converted to lactic acid by lactate dehydrogenase

PHOTO (COLOR)

LAB can also utilise the malic enzyme (ME, EC:1.1.1.38) for l-malic acid degradation. ME converts l-malic acid into pyruvate (Fig. 2) enabling growth with l-malic acid as a carbon source (Landete et al. 2010). Although both pathways utilise the same substrate, the transcription of the corresponding genes is reported to be independently regulated (Landete et al. 2013). The genes involved in the ME pathway are arranged in two operons, maePE and maeKR (Fig. 2). They encode a putative l-malic acid transporter (MaeP), an ME (MaeE), and a two-component signal transduction system (TCS, MaeK and MaeR), which has been studied in Lb. casei and Lb. rhamnosus (Landete et al. 2013; Miguel-Romero et al. 2017; Monedero et al. 2017). In the study of Landete et al. (2013), the ME pathway in Lb. casei was related to higher growth rate by energy generation, while the function of the MleA pathway was reported to be solely deacidification. O. oeni contains putative ME (OEOE_RS02010, OEOE_RS002015) and TCS operons (OEOE_RS00545, OEOE_RS08540), but the function of these in O. oeni is not yet known (see also Kegg pathway https://www.genome.jp/kegg-bin/show%5fpathway?ooe02020+OEOE%5f0418)_I_._i_ It is possible that the presence and expression of duplicate l-malic acid degradation pathways in LAB may affect their ability to efficiently conduct MLF. Sternes et al. (2017) observed that OEOE_RS02010 (maeE) was up-regulated in one (AWRIB419) out of three O. oeni strains tested (AWRB419, AWRIB551 and AWRIB552). AWRIB419 took 16 days to complete MLF whereas AWRIB551 and AWRIB552 took 4 and 6 days, respectively (Sternes et al. 2017). The efficiency of these pathways is yet to be fully compared and it may be that the MLE pathway has a higher rate of l-malic acid degradation. While MLE has been studied in moderate detail, ME has so far gained less attention. More information is required to delineate the different pathways of MLF, including the functions and influential factors of the ME pathway as well as a comparison between ME and MLE. Further study of these pathways would improve basic understanding, methods for inducing successful and controllable MLF and would provide clear targets for selection/development of new starter cultures such as those with higher MLE activity.

The improvement of LAB strains for more reliable fermentation was described in some detail in a previous review (Sumby et al. 2014) and as such only a short summary (Table 2) and new research will be described here.

There are many methods to improve wine bacterial strains for more reliable MLF. Amongst these is the preferred non-GMO method of directed evolution (DE). In the first study reporting the use of DE to improve O. oeni, Betteridge et al. (2018) conducted continuous culture of commercial strain SB3 in the presence of increasing ethanol concentration to yield a more ethanol-tolerant isolate, A90. Although this strain had superior fermentation performance in laboratory MRSAJ (de Man, Rogosa and Sharpe medium + apple juice) medium, the ethanol-resistant phenotype was not evident when inoculated into the complex, multi-stressor environment of Red Fermented Chemically Defined Grape Juice Medium (RFCDGJM; Jiang et al. 2018). This was true at both low inoculation rates and those analogous to recommendations for high ethanol wines, suggesting that A90 had evolved to a particular niche and may have limited abilities outside of this.

In a follow-up study (Jiang et al. 2018), continuous culture was used to evolve strain A90 in a multi-stress environment resembling wine. Accordingly, the strain was grown for ∼ 350 generations in RFCDGJM before increasing the proportion of red wine mixed with RFCDGJM thereby raising the levels of stressors (e.g. low pH, ethanol and SO2). Three strains were selected through the course of the DE experiment based on their ability to consume more l-malic acid than the parent strain (which became stuck) when evaluated in a RFCDGJM/wine blend with 15.1% (v/v) ethanol, 26 mg L−1 SO2 at pH 3.35. Further assessment of these selected strains in four different red wines (pH values of 3.37-3.55; ethanol 13.9-16.7% (v/v)) revealed they also fermented faster and/or achieved a greater population than the parent. In this way, the effectiveness of using DE to improve O. oeni performance and reliability under winemaking conditions was clearly demonstrated.

While EMS or UV mutagenesis was not used in the previous study, the DE approach could possibly be enhanced further by pre-stressing strains and applying mutagens before initiation of DE. Still further opportunities to improve on DE methods arise by applying what is known about the cross-stress behaviour and DNA repair in LAB. Accordingly, Machielsen et al. (2010) demonstrated that even though mutation frequency was unaffected in Lb. plantarum during high temperature, low pH, osmotic or starvation stress, it was increased by a factor of 100 after exposure to sub-lethal levels of H2O2. Interestingly, preadaptation at 42 °C, a non-mutagenic condition, reduced the mutagenic effect of oxidative stress (Machielsen et al. 2010). Evidence of cross-stress behaviour has been well documented throughout the microbial kingdom. Early work on glucose- and nitrogen-starved Escherichia coli showed increased survival rates after heat shock or hydrogen peroxide (H2O2)-mediated stress compared with non-stressed cells (Jenkins et al. 1988). Other studies have used DE of LAB primarily to study the biology of these organisms. Genetic analysis of changes arising in Lb. rhamnosus strain GG following 1000 generations of growth in a rich medium under four different conditions (stress free, salt stress, bile stress, shear stress) showed that mutation rates were low under all conditions (Douillard et al. 2016). Deletion events, however, mediated by activation of IS elements arose during bile and shear stress (Douillard et al. 2016).

It would also be of interest to investigate why O. oeni improves so rapidly during DE with few detrimental phenotypes developing in the DE process, i.e. is it more than just the absence of the DNA repair mechanism MutS/MutL? In a study investigating mutS mutants, Overbeck et al. (2017) improved the DE of Lb. casei to increase lactic acid resistance at low pH, through the deliberate use of cells with inactivated DNA mismatch repair gene mutS. A two-step gene replacement method was used to delete mutS before strains were subjected to a 100-day DE process to increase lactic acid resistance at low pH (Overbeck et al. 2017). Genome sequencing confirmed that inactivation of mutS decreased DNA replication fidelity during DE and thereby allowed mutants to arise that grew better and produced more lactic acid at low pH than with wild-type cells undergoing the same DE process. However, it is also possible that hypermutation of the mutS mutants could affect other genes associated with replication fidelity thereby preventing stable genotype restoration. This would need to be tested.

Regardless of the improvement strategy used, it is also necessary to improve methods of selecting for improved MLF strains. As mentioned, it is difficult to isolate environmental samples of O. oeni, thus selection strategies tend to target wines undergoing MLF. Both Betteridge et al. (2018) and Jiang et al. (2018) used 96-well microplate (300 μL) screening methods to highlight their improved isolates in the medium used for DE. This approach is very useful when dealing with hundreds of DE isolates. However, results do not always translate to the larger scale and methods that utilise larger screening volumes, such as the automated 96 × 100-mL flask fermentation platform described by Peter et al. (2018) appear to deliver results more representative of those at the larger, pilot scale.

A number of strategies have been used to improve production of LAB starter cultures to better adapt them to wine. Most manufacturers apply an acclimation treatment prior to freeze-drying O. oeni cells for long-term conservation. However, altering adaptation conditions prior to freeze-drying does not always lead to differences in viable cell number between acclimated and non-acclimated cells. For example, O. oeni cells that were adapted with trehalose as a cryoprotectant prior to freeze-drying showed no significant differences in viable cell number when inoculated into synthetic wine and only showed increased adaptation towards the end of MLF (Bravo-Ferrada et al. 2018). However, irrespective of freeze-dried method used, O. oeni consumption of l-malic acid was higher with freeze-dried cells showing a better performance than fresh cultures (Bravo-Ferrada et al. 2018), and this effect warrants further investigation in this area.

Recently, research has focused on the use of bacterial exopolysaccharide (EPS) production due to its protective nature to bacterial cells. There is great interest in the variation in ability of LAB strains to produce EPS as a protective layer around the cell. The EPS of O. oeni has recently been studied for its role in protection during freeze-drying for malolactic starter production (Dimopoulou et al. 2016). Interestingly, 10 of the 14 tested strains were surrounded by cell-linked polysaccharide (encapsulated) and were generally (but not always) more tolerant to freeze-drying than the non-capsulated ones. Strains that were able to synthesise dextran in industrial growth media containing sucrose were more tolerant to freeze-drying (Dimopoulou et al. 2016). Therefore, the detection of cell encapsulation or the production of dextran may be of use in predicting the ability of improved or new strains to survive scale-up and commercial processing. Following on from this work, Dimopoulou et al. (2018) evaluated gene expression and the conditions in which eps genes and EPS synthesis were most stimulated to improve the production of freeze-dried malolactic starters, acclimation procedures and MLF efficiency. The authors reported that O. oeni strains are equipped with at least one EPS biosynthetic pathway and eps genes are expressed at significant levels whatever the growth conditions (Dimopoulou et al. 2018). For a review on EPS and protective mechanisms of sugars during freeze-drying, see Santivarangkna et al. (2008).

Other researchers focusing on improvement of freeze-dried starter process have trialled the use of monosodium glutamate (MSG) addition to O. oeni cells during freeze-drying (Yang et al. 2018). Evaluation of protein expression and cellular integrity during freeze-drying of O. oeni strain SD-2a revealed less cell damage and higher production of proteins involved in cyclopropane fatty acid composition in MSG-treated cells (Yang et al. 2018). Proteins involved in EPS production and quorum sensing were up-regulated when compared with freeze-drying alone. However, the use of MSG may prove a barrier to some markets where addition of this is not well accepted by consumers. It would be of interest to look closer at the mechanism of protection and to seek alternatives that might by more widely accepted by consumers.

Another MLF improvement strategy that has been revisited is the encapsulation of O. oeni cells. One recent improvement in this area includes encapsulation of O. oeni in silica-alginate (SiO2-ALG) biocapsules (Simó et al. 2019). The authors included siliceous materials into ALG hydrogel thereby improving LAB stability and reducing cell shrinking under winemaking conditions (Simó et al. 2019). Another recent study used pulsed electric field (PEF) technology to reduce the microbial community in wines after AF to improve implantation of O. oeni (González-Arenzana et al. 2018). This method slightly shortened MLF duration in some of the tested wines (González-Arenzana et al. 2018). However, of concern with this method is the isolation of Dekkera bruxellensis and Candida zeylanoides from one of the PEF-treated wines and the use of only one O. oeni strain (Uvaferm alpha). This method needs to be studied further to determine if this effect was due to increased yeast cell lysis and consequently a spike in nutrition for the inoculated MLF bacteria or due to a reduction in microbial competition.

Monitoring of MLF is also important for winemakers and improvements in this area include the use of an ultrasonic sensor to measure MLF progression (Çelik et al. 2018). One of the main advantages of these sensors is their direct installation in standard stainless-steel tanks of large wineries for monitoring fermentation progress. For a recent review on the use of ultrasound technology in fermentations, please refer to Ojha et al. (2017). A further development in monitoring MLF is an electrochemical bienzymatic biosensor for the determination of l-malate (Giménez-Gómez et al. 2017). The biosensor comprises a thin-film gold electrode as the transducer, malate dehydrogenase (MDH) and diaphorase (DP) enzymes combined with nicotinamide adenine dinucleotide (NAD+) cofactor as the selective receptors and a redox mediator. The core reaction can be measured by either spectrophotometry or amperometry (via detection of ions in a solution based on electric current or changes in electric current), with the latter being suggested as suited to monitor MLF (Giménez-Gómez et al. 2017). This technology is not new with authors such as Gamella et al. (2010) having already described it. However, the recent biosensor presented good stability and retained more than 90% of its sensitivity after 37 days of monitoring MLF in red wines (Giménez-Gómez et al. 2017). Looking at the reaction from another angle, Bravo et al. (2017) have reported the use of biosensors to measure increases in lactate in wine, beer and yoghurt. A disposable biosensor was designed to act as an electro-catalytic platform, which was coupled to lactate oxidase and the reaction measured based on the oxidation of hydrogen peroxide (H2O2). This enabled direct determination of lactate with good sensitivity and stability for more than 1 month. This method is probably more useful in the dairy industry, but demonstrates quite nicely how winemakers could use alternative biosensors to monitor fermentations.

Although acid reduction is the principal sensory effect of LAB during MLF (Volschenk et al. 2006), Davis et al. (1985, 1988) showed that LAB strains have a specific sensory effect on wine. It is now clear that during MLF, LAB can produce or release several compounds that have either a positive or negative effect on the sensorial properties of the wine, depending on the nature of these compounds, their concentration and the physical-chemical properties of the wine. MLF can be used to modulate sensory attributes such as buttery aroma (diacetyl), fruity aroma, mouthfeel and colour (Swiegers et al. 2005). The underlying mechanisms involve the production of enzymatic activities that can have both aroma and/or visual effects. For example, anthocyanin pigments may be impacted by glycosidases (Vivas et al. 1997; Grimaldi et al. 2005a; Burns and Osborne 2013, 2015). For a recent review on the variability of bacterial enzymes, including glycosidases, esterases, proteases and other enzymes that can generate a wide spectrum of sensorially significant compounds in wine, refer to Cappello et al. (2017). Aroma compounds that can increase or decrease during MLF include esters (Antalick et al. 2012; Pozo-Bayón et al. 2005; Sumby et al. 2010, 2013a), higher alcohols (Ugliano and Moio 2005), aldehydes (Osborne et al. 2000), alcohols (e.g. 2-phenylethanol), aglycones (including terpenols and C13-norisoprenoids) from their glycosylated precursor (Boido et al. 2002) and 2,3-butanedione (diacetyl) (Bartowsky and Henschke 2004; Nielsen and Richelieu 1999).

Beyond strain choice, it appears the sensory impact of MLF can also be influenced by the timing of MLF inoculation (i.e. sequential vs. co-inoculation), pre-MLF wine matrix conditions (e.g. pH) or nutrient supplementation (Abrahamse and Bartowsky 2012; Costello et al. 2012; Gámbaro et al. 2001; Gammacurta et al. 2017; Knoll et al. 2012; Maarman 2014) and potentially inoculation rate.

A major compound produced during MLF in wine is diacetyl, which is described as having a buttery or butterscotch aroma and flavour. Diacetyl is formed through the metabolism of citric acid with the first step catalysed by citrate lyase (citE), which cleaves citrate into acetate and oxaloacetate (Mills et al. 2005). Oxaloacetate is then decarboxylated to pyruvate, leading on to the formation of diacetyl (Hugenholtz 1993). The organoleptic impact of diacetyl in wine has been debated for many years with low concentrations considered to contribute positively to the wine's bouquet.

The role of strain is regarded as important for the initial yield of diacetyl and some LAB strains do not have the genes needed for formation (Mtshali et al. 2010, 2012), but once produced, diacetyl can be enzymatically reduced to 2,3-butanediol. Thus co-inoculated MLF would be expected to produce wines with ultimately less diacetyl since yeast can metabolise diacetyl to acetoin and 2,3-butandiol (Lasik-Kurdyś et al. 2018; Mink et al. 2013). In addition, a faster MLF tends to result in lower diacetyl concentrations (Sternes et al. 2017). Post-MLF processing can also affect the final diacetyl content and influence the sensory profile of the wine. Ageing on lees can encourage diacetyl degradation, and bâtonnage (resuspension of yeast lees) can reduce or even eliminate the buttery aroma (Bartowsky and Henschke 2004; Martineau and Henick-Kling 1995; Martineau et al. 1995).

Acetaldehyde can be consumed by LAB during MLF, thereby limiting its conjugation with anthocyanins in wine and, in turn, colour stabilisation (Liu and Pilone 2000; Osborne et al. 2000). As a result, a loss of colour after MLF is one potentiality (Burns and Osborne 2015). Conversely, some LAB have the ability to produce acetaldehyde (Liu and Pilone 2000), which might be expected to benefit wine colour, although this does not seem to have been specifically examined. More recently, Wang et al. (2018) reported that O. oeni strains tested in model wine increased the level of acetaldehyde at the beginning of MLF, but that a dramatic decrease was observed after 4 days of MLF. However, when the model wine medium contained Lb. plantarum, a continuous accumulation of acetaldehyde (two to three times higher) was observed. This effect and possibly the effect of co-inoculation (O. oeni and Lb. plantarum together) on acetaldehyde levels could be explored further.

A comprehensive review on the reported changes in ester concentration during wine fermentation, including changes during MLF, appeared in 2010 (Sumby et al. 2010). While the berry fruit aroma characteristics of red wine vary according to grape variety and winemaking, the fruit aroma of red wine is a complex interaction between fruity esters, norisoprenoids, dimethyl sulfide, ethanol and other components. Strains of O. oeni have been shown to vary in their ability to generate or release these compounds.

Recent research has focused on characterising the enzymes that are responsible for ester synthesis and hydrolysis during MLF (Costello et al. 2013; Esteban-Torres et al. 2014; Sumby et al. 2009, 2013a, 2013b). This group characterised intracellular esterases from O. oeni and Lactobacillus hilgardii under wine-like conditions (Sumby et al. 2009, 2013a, 2013b). Two heterologously expressed and purified esterases were stable and active under conditions that would be encountered in wine and therefore offer the potential to reduce short-chain ethyl esters such as ethyl acetate. Following on from this, O. oeni strain AWRI B551 was shown to produce significant levels of ethyl hexanoate and ethyl octanoate following growth in an ethanolic test medium (Costello et al. 2013).

Characterisation of esterase enzymes from Lb. plantarum strain Lp_1002 defined the first arylesterase from a wine LAB under wine-like conditions (Esteban-Torres et al. 2014). The enzyme was reported able to withstand ethanol, sodium metabisulfite and tartaric, lactic and citric acids, with only malic acid slightly inhibiting activity. While it is generally accepted that MLF has a significant influence on the ester composition of wines, there is as yet no consensus on the effect of individual bacterial strains. Gammacurta et al. (2018) found that after MLF using two different commercial O. oeni starters compared with un-inoculated MLF, the branched hydroxylated esters, ethyl 2-hydroxy-3-methylbutanoate and ethyl 2-hydroxy-4-methylpentanoate were the only compounds to be influenced by strain choice, irrespective of the vineyard or the vintage considered.

Clearly more research is necessary to determine if sensory differences, related to ester concentrations, reported between strains are due to true strain differences or if there are other causes such as a matrix effect. It is clear that oenological parameters such as pH, ethanol and SO2 (Knoll et al. 2011a) affect ester metabolism by LAB, but further research is called for.

The presence of volatile phenols has been related to the action of Brettanomyces yeast particularly in relation to winemaking practices involving wine ageing in oak barrels. Their presence is considered detrimental for wine aroma and flavour. Volatile phenols, including 4-ethyl guaiacol and 4-ethyl phenol, may also increase after MLF because some Lactobacillus species are able to produce low concentrations of these compounds relative to Brettanomyces yeast (Brizuela et al. 2018; Couto et al. 2006). A greater understanding of this metabolic pathway is needed and a screening of strains for their production propensity should be conducted.

Volatile sulphur compound (VSC) production by wine-related LAB is not well documented despite there being numerous reports of this from the dairy industry (e.g. Al-Attabi et al. 2008; Curioni and Bosset 2002). VSCs can contribute positively to wine aroma, but some compounds are considered detrimental to wine quality, depending on their concentration. As an example, the metabolism of the sulphur-containing amino acid, methionine, has an impact on wine aroma and has been studied for the production of VSCs (Pripis-Nicolau et al. 2004; Vallet et al. 2008). Both Lactobacillus sp. and O. oeni can metabolise methionine to form methanethiol and dimethyl disulfide (Antalick et al. 2012; Pripis-Nicolau et al. 2004). However, 3-(methylsulfanyl)propan-1-ol and 3-(methylsulfanyl) propionic acid were formed in more significant quantities by O. oeni than Lactobacillus (Pripis-Nicolau et al. 2004). The cloning and characterisation of an O. oeni enzyme able to degrade sulphur-containing amino acids has been reported (Knoll et al. 2011b). Amongst other activities, the ability to demethiolate methionine to methanethiol, an unpleasant VSC in terms of wine aroma, was observed.

Another sulphur compound, 3-sulfanylhexan-1-ol (3SH), needs to be considered due to its important contribution to fruity notes of wine. Although 3SH has been reported not to change significantly during MLF with O. oeni (Antalick et al. 2012) it may be produced by other LAB species (Takase et al. 2018).

Further genetic and enzymatic studies are needed to more fully characterise these potential contributions to wine VSC composition.

Through ongoing research, there is now a better understanding of how MLF could be used to influence wine style. The choice of LAB strain, as well as timing of bacterial inoculation, makes it possible to modulate MLF sensory influence in wine. Future work should likely focus on diversity profiling and the sensory differences between uninoculated and inoculated MLF at different inoculation rates, by way of metabolomics (Bokulich et al. 2016; Lee et al. 2009) and genomics (Bartowsky and Borneman 2011; Sternes et al. 2017).

Future work may also include evaluation of the effect of strain origin. For example, Campbell-Sills et al. (2017b) sequenced 14 isolates from red and white wine and determined that they share a common ancestor that probably colonised two different substrates. MLF was undertaken and the volatile composition varied between the strains and was dependent on their group of origin. El Khoury et al. (2017) reported that strains could be grouped according to the beverage they were isolated from, and there was no correlation with geographical origin. This has interesting implications for the concept of terroir.

Although we have come a long way in recent years towards our understanding of MLF and how best to undertake it when desirable, much remains unknown. It is clear that the influx of 'omics data over the next few years will be massive as it becomes more economically and technically available. The importance of microbial diversity and grape must ecology is becoming more widely appreciated, and it is expected that the study of interactions between microbes in both inoculated and uninoculated ferments will uncover a range of new information on how best to control both AF and MLF. We now know how to conduct MLF within a range of conditions. But what can we do when we have problematic ferments, and how can we predict MLF outcomes? It would appear that it is first necessary to obtain a better understanding of the response of O. oeni cells to ethanol and other inhibitory compounds. As well as to continue investigations of the presumably synergistic effect of wine stressors and why some cells are more affected than others.

This review has highlighted that there are still considerable potential future research directions in this area. The potential for the isolation of improved strains suited to a region's unique climate, white versus red wine, or even differences amongst the grape varieties has not been investigated thoroughly. There is the potential that there a 'super strain' suited all conditions/varieties that simply has not been isolated. The use of indigenous bacterial strains is still common practice, while strategies and conditions to best encourage indigenous strains have not been examined in detail.

Alternate strategies to improve LAB other than Oenococcus is an area for further investigation. While the improvement of O. oeni strains using DE has been successful, presumably due to their ineffective DNA repair systems, it is likely that other strategies, e.g. mutagenesis, will be need to be employed for other LAB (e.g. UV or EMS mutagenesis followed by screening in a competitive environment). Further development of transformation systems for O. oeni is necessary. It is not yet possible to routinely modify gene presence and expression in O. oeni. The ability to do so will help to rapidly delineate metabolic pathways and stress-response systems, and assist in investigation of methods to improve the expression of mleA. To directly assist winemakers, it would be of benefit to improve the sensing technologies used to assess LAB growth and measure MLF progress so they are robust, economical and industrially viable. With further possibilities to eliminate additives or processing steps during winemaking through the use of bacteriocin-producing strains to increase their competitiveness over indigenous, potentially unwanted LAB, to reduce SO2 usage by addition of prophage to selectively inhibit growth of unwanted microorganisms during fermentation (Bondy-Denomy et al. 2016) or even the addition of LAB cultures, or their cell free supernatants, after AF that display antifungal activity and therefore assist in process control.

This review was supported by The Australian Research Council Training Centre for Innovative Wine Production (http://www.arcwinecentre.org.au; project number IC170100008) funded by the Australian Government with additional support from Wine Australia and industry partners. The University of Adelaide is a member of the Wine Innovation Cluster (http://www.thewaite.org/waite-partners/wine-innovation-cluster/).

This article does not contain any studies with human participants or animals performed by any of the authors.

The authors declare that they have no competing interests.

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.