Comparing the hierarchy of inter- and intra-species interactions with population dynamics of wine yeast cocultures

Abstract In winemaking, the development of new fermentation strategies, such as the use of mixed starter cultures with Saccharomyces cerevisiae (Sc) yeast and non-Saccharomyces (NS) species, requires a better understanding of how yeasts interact, especially at the beginning of fermentation. Despite the growing knowledge on interactions between Sc and NS, few data are available on the interactions between different species of NS. It is furthermore still unclear whether interactions are primarily driven by generic differences between yeast species or whether individual strains are the evolutionarily relevant unit for biotic interactions. This study aimed at acquiring knowledge of the relevance of species and strain in the population dynamics of cocultures between five yeast species: Hanseniaspora uvarum, Lachancea thermotolerans, Starmerella bacillaris, Torulaspora delbrueckii and Sc. We performed cocultures between 15 strains in synthetic grape must and monitored growth in microplates. Both positive and negative interactions were identified. Based on an interaction index, our results showed that the population dynamics seemed mainly driven by the two species involved. Strain level was more relevant in modulating the strength of the interactions. This study provides fundamental insights into the microbial dynamics in early fermentation and contribute to the understanding of more complex consortia encompassing multiple yeasts trains.


Introduction
Wine is the result of the fermentation of grape must by a variety of micr oor ganisms.Wine yeast species show a great diversity, especially at the beginning of fermentation, including species belonging to the genera Hanseniaspora , Metschnikowia, Aureobasidium (yeast-like), Pichia , Starmerella , Torulaspora , Zygosaccharomyces , Rhodotorula , and others (Fleet 2003, 2007, Drumonde-Ne v es et al. 2021 ).Sacc harom yces cerevisiae , if detected, is only present at relativ el y low cell densities in the initial must, but increases during fermentation and is the main species responsible for the completion of the fermentation.Indeed, non-Sacc harom yces (NS) yeasts, sometimes after an initial r a pid gr owth, tend to decr ease during the latter stages of fermentation.This decrease has been attributed to se v er al abiotic factors such as decrease in oxygen, increase in toxic metabolites, including ethanol.Mor e r ecentl y biotic factors related to competitive interaction with other species have been highlighted as significant causes for these changes (Fleet 2003 ).During winemaking, must is often inoculated with S. cerevisiae to ensure completion of the fermentation process, howe v er, this is associated with standardization of the final product (Ciani et al. 2010 ).T hus , there is an increasing interest in introducing non-Sacc harom yces (NS) to impr ov e the quality of the pr oduct and meet consumers' expectations related to mor e natur al pr oducts (Galati et al. 2019 ).Indeed, NS secrete a broader spectrum of enzymes that might release aroma precursors from grapes and hav e div erse metabolic pathw ays that allo w for v ariable pr oduc-tion of secondary metabolites (Jolly et al. 2006, Polizzotto et al. 2016, Varela and Borneman 2017 ).Ho w ever, most of the time, NS alone are not able to ferment to dryness, which has led to the de v elopment of mixed starters including S. cerevisiae with NS such as Torulaspora delbrueckii , Metschnikowia fructicola , Lachancea thermotolerans (Binati et al. 2020 ).Se v er al NS starters ar e alr eady available on the market, which are advertised to increase wine ar oma complexity, r educe ethanol, or hav e biopr otection pr operties to name a few (Roudil et al. 2020 ).For instance, Renault et al. ( 2015 ) sho w ed that co-inoculation of S. cerevisiae with T. delbruec kii incr eases the acetate ester content.Torulaspora delbruec kii and Metschnikowia pulcherrima are also used for bioprotection (Simonin et al. 2018, Sipiczki 2020 ).Lachancea thermotolerans is mainly investigated for its potential for lactic acid production (Morata et al. 2018 ) and Starmerella bacillaris is related to increased glycer ol pr oduction, whic h impr ov es mouthfeel (Englezos et al. 2017, Binati et al. 2020 ).Cocultures of S. cerevisiae with Hanseniaspora uv arum wer e r elated to ov eryielding of gl ycer ol indicating positiv e interactions for this functional trait (Harlé et al. 2020 ).
Ho w e v er, ther e is limited understanding of how desirable properties emerge from application of multispecies starters, especially regarding the contribution of yeast-yeast interactions .T herefore , to manage fermentations using mixed starters, we need to better understand yeast interactions that might influence the process and final pr oduct.Inter actions betw een y easts can be positiv e, neutr al or negative.Besides the deleterious effect of toxic compounds such as ethanol produced during fermentation, sever al mec hanisms might explain yeast interactions in must (Ciani et al. 2016, Rossouw et al. 2018, Conacher et al. 2019, Bordet et al. 2020 ).For instance, some S. cerevisiae strains, as well as various other yeast species, can produce killer toxins, inhibiting cells either from other species or from the same species (Boynton 2019 ). S. cerevisiae also seems to induce cell death of L. thermotolerans and S. bacillaris through contact dependent interactions (Englezos et al. 2017, Petitgonnet et al. 2019, Luyt et al. 2021 ).Rossouw et al. ( 2015Rossouw et al. ( , 2018 ) ) also sho w ed that changes in adhesion properties of S. cerevisiae significantly affected the survival of other species.Presence of other yeast species have also been found to cause changes in gene expression, for instance S. cerevisiae tends to promote genes r elated to gl ycol ysis and aer obic r espir ation when in presence of T. delbrueckii or M. pulcherrima (Tronchoni et al. 2017, Mencher et al. 2021 ), which might increase its nutrient uptake .T hen, the different yeast species could also have ov erla pping nutritional r equir ements leading to competition for nutrients such as amino-acids or vitamins (Rollero et al. 2018, Evers et al. 2021 ).The evidence suggests that there are indeed species-specific y east-y east interactions.Ho w e v er, the str ain c hoice could also be an important parameter.Indeed, besides interspecific species diversity, yeasts also show a great intraspecific genetic diversity driven by geographic origin [e.g. S. bacillaris (Masneuf-Pomarede et al. 2015 )] or the technolog ical orig in [e.g S. cerevisiae or L. thermotolerans (Legras et al. 2007, Hranilo vic et al. 2017 )].T his genetic diversity is also associated with phenotypic variability between strains isolated from differ ent envir onments.For instance, S. cerevisiae strains from different environments sho w ed different fermentation performances (Camarasa et al. 2011 ) and competitive abilities (Pérez-Torrado et al. 2018 ).Even isolates from winery environments display variability in their phenotypic pr operties suc h as ethanol resistance, β-glucosidase activity, hydrogen sulphide production, and lactate pr oduction (Hr anilovic et al. 2017, Mor ata et al. 2018, Silv a-Sousa et al. 2022 ).This intr aspecific div ersity may cause variability in chemical composition of fermentation both in monoculture (Bordet et al. 2021 ) and in coculture (Wang et al. 2016 ) and therefore impact y east-y east inter actions and ultimatel y, the final wine product.
Ho w e v er, it is still unclear which taxonomic le v el most influences the nature of microbial interactions, or, in other w or ds, is it important to study str ain-str ain inter actions, or, ar e inter action mec hanisms gener alized at the species-species le v el?This ga p in understanding can lar gel y be attributed to challenges in differ entiating differ ent str ains of micr oor ganisms in natur e, since most known methodologies used in survey studies cannot distinguish differ ent str ains .T hus , the aim of this study was to investigate in simplified systems the determining level of interaction in five yeast species: H. uvarum , L. thermotolerans , S. bacillaris , and T. delbrueckii , and S. cerevisiae .To ac hie v e this, a flow cytometric methodology was de v eloped to distinguish differ ent str ains of yeast within mixed cultures, and a high-throughput methodology was used to quantify the population dynamics of all possible pairwise cocultures between 15 strains including 3 strains for the 5 species .T his study has applied a versatile methodology for interstr ain inter actions and has contributed to the understanding of taxonomic influence on y east-y east interactions, using wine yeast as a model.

Strains and medium
In this study, five wine yeast species were used: Saccharomyces cer e visiae and four non-Sacc har omyces (NS): Hanseniaspora uv arum, Lac hancea thermotoler ans, Starmer ella bacillaris, and Torulaspor a delbruec kii.For eac h species, thr ee str ains wer e included, all isolated from wine-related en vironments .All the 15 str ains wer e fluor escentl y ta gged by integr ating a fluor escent pr otein gene into the genome to ensure a better stability of the signal.Origin of each strain can be found in Table 1 .Strains were k e pt at −80 • C in yeast peptone dextrose YPD (Peptone; 20 g/L, Glucose 20 g/L; Yeast extract 10 g/L-Sigma-Aldrich, Johannesburg, South Africa) supplemented with 20% of gl ycer ol befor e being str eaked on Wallerstein (WL) nutrient agar (Sigma Aldrich, Darmstadt, Germany).
The high-thr oughput micr oplate gr o wth assay w ere performed using synthetic gr a pe m ust (SGM425) pr epar ed according to Bel y et al. ( 1990 ), with 100 g/L of glucose, 100 g/L of fructose and 425 mg/L of yeast assimilable nitrogen (as a mix of ammonium chloride and amino acids).

Gener a tion of fluorescently tagged yeast strains
For the yeast strains transformed in this study, EGFP (enhanced gr een fluor escent pr otein) was integr ated into the genome in fusion to TDH3 gene (or its orthologue in non-Sacc harom yces species) using homologous recombination.In S. cerevisiae , TDH3 promoter is a strong promoter known to be expressed throughout fermentation.Cassettes containing the fluorescent protein and an antibiotic selection marker were amplified from different plasmids listed in Supplementary Table 1 .

Plasmid construction
Plasmids containing the EGFP and different antibiotic resistance gene or specific homology regions were constructed by Gibson assembly (Gibson et al. 2009 ) using the NEB Builder HiFi DNA Assembly Master Mix (New England Biolabs) and transformed into Esc heric hia coli DH5 ɑ (New England Biolabs) following the manufacturer instructions.EGFP in the pFA6 backbone as well as antibiotic resistance genes were obtained from plasmids ordered from AddGene (#44 900, #44 645, Table S1 ) (Sheff andThorn 2004 , Lee et al. 2013 ).Wher e necessary, homologous sequences of a ppr oximately 1 kb were amplified from the target species ( T. delbruecki CLIB3069, H. uvarum CLIB3221, S. bacillaris CLIB3147).All plasmids wer e c hec ked by enzymatic digestion (Ne w England Biolabs).A list of the primers and templates used for the amplification of the different Gibson fragments can be found in Supplementary Table 2 .
Plasmid DN A w as extr acted fr om 3 mL of ov ernight E. coli LB culture with the NucleoSpin Plasmid extraction kit (Macherey Na gel, Dür en, German y) according to manufacturer instructions.Cassettes used for transformation were amplified with a high fidelity enzyme, either the KAPA HiFi kit (Cape Town, South Africa) or Phusion High-Fidelity DNA pol ymer ase (Thermo Fisher Scientific , Vilnius , Lithuania), using primers specific to each species, as listed in Supplementary Table 3 .

Lithium acetate transformation
Cells were transformed according to Güldener et al. ( 1996 ) with some modifications.Fifty milliliters of fresh culture grown in YPD to OD 600 = 2 were centrifuged at 4415 g for 5 min.Pellets were washed with 20 mL of Tris 10 mM, pH 7.5 and suspended in 25 mL of lithium acetate 0.1 M in Tris-HCl 10 mM, pH 7.5.Cells were incubated for 40 min at room temperature under gentle shaking.After incubation, cells were pelleted at 430 g for 5 min and suspended in 1,125 mL of lithium acetate 0.1 M in Tris-HCl 10 mM,

Electroporation
The protocol used for the transformation of cells by electr opor ation was adapted from Gordon et al. ( 2019 ).Briefly, cells were inoculated at OD 600 = 0.5 in 50 mL YPD and grown to OD 600 = 2. Cells w ere pelleted b y centrifugation at 4415 g for 5 min and rinsed in 25 mL of water.Incubation in 0.1 M LiOAc in 1X TE and DTT 1 M were done according to Gordon et al. 2019 .At the final pr epar ation step, cells were suspended in 1 mL of 1 M sorbitol instead of 250 μL.
Electr o-competent cells wer e stor ed at − 80 • C. Before electropor ation, cells wer e thawed at r oom temper atur e, then centrifuged for 5 min at 430 g.Supernatant was replaced by fresh sorbitol 1 M and cells were k e pt on ice.Eighty microliters of cells were electr opor ated at 1.5 kV, 600 , and 10 μF in 0.2 mm cuvettes using Eppor ator electr opotator (Eppendorf, Hambur g, German y).After electr opor ation, 1 mL of YPD/sorbitol mix (50:50) was immediately added to the cells.Cells were transferred to test tubes and incubated overnight at 28 • C without shaking.Cells were plated onto YPD supplemented with antibiotics and allo w ed to gro w at 28 • C for one week.

Clone selection and constructions control
For each yeast strain transformation, 8 clones were selected and str eaked on fr esh YPD supplemented with antibiotic.S. cerevisiae and H. uvarum transformants were selected on YPD supplemented with 200 μg/mL of G418 (Sigma-Aldrich), L. thermotolerans and T. delbruec kii tr ansformants wer e selected on YPD supplemented with 100 μg/mL of nourseothricin (Jena Bioscience, Jena, German y), S. bacillaris tr ansformants wer e selected on YPD supplemented with 800 μg/mL of hygromycin (Sigma).The correct integration of the cassette at the TDH3 locus was verified with two PCR using primers outside the integration site and in the cassette.A list of primers used for verification PCR can be found in Supplementary Table 4 .Fluorescence was observed by fluorescent microscopy and flow cytometry.Growth of selected clones was compared to WT in YPD and SGM425 to ensure genetic modification did not influence strain behaviour.

Micr oplate gr owth
A single colony picked from a WL agar plate was propagated in 5 mL YPD for 17 hours at 25 • C with shaking at 40 rpm.Then, 100 μL of this pr ecultur e was pr opa gated in 5 mL SGM425 for 24 hours at 25 • C with shaking at 40 rpm.One mL of culture was harvested, centrifuged, and washed in physiological saline (3000 g-5 min).Cell density and fluorescence of all cells in the culture wer e measur ed by cytometry and a ppr opriate volume of cultur e was used to inoculate SGM425 at 10 6 cells/mL.Microplates were pr epar ed by mixing 100 μL of SGM425 cell suspension of two str ains in eac h w ell, as sho wn on the schematic plan in Fig. 1 .
Monocultur es wer e inoculated with both the WT str ain and its fluor escentl y ta gged counter part, while cocultur es wer e inoculated

Statistical analysis
Statistical analyses were performed using R software, version 4.2.2 (R Core Team 2022 ).To obtain the growth rate and area under the curv e (AUC) gr owth par ameters, gr owth curv es fr om the plate r eader wer e anal ysed with the gr owthcurv er pac ka ge based on a logistic model (Sprouffske and Wagner 2016 ).To assess the species and strain effect on growth in monocultures, mixed model analysis were applied to the growth parameters values using the following model: For the growth rate, AUC and maximum population in monocultur es, since we observ ed a str ong species effect, differ ences between strains were tested for each species by ANOVA using the a gricolae pac ka ge (Mendiburu and Yaseen 2020 ).For the latency, no species effect was observed thus all 15 strains were compared by ANOVA.For cocultures, an interaction index was calculated using the values of both monocultures using the following formula applied to the AUC example: AUC coculture is the value of the AUC for the coculture of strain S1 and S2, AUC S1 and AUC S2 are the value of the AUC for the monocultures of strain of S1 and strain S2 respectively.Clustering analyses on the interaction index were performed using the hclust function from the stats (R Core Team 2022 ) pac ka ge and clusters wer e v erified by bootstr a pping done with clusterboot pac ka ge (using 'subset' and 'complete' method, on 1000 bootstr a p iter ation) function from the fpc package (Hennig 2010 ).Heatmaps were built following the clustering and interaction index with the Complex-Heatma p pac ka ge (Gu 2022 ).

Results
In this w ork, w e studied the population dynamics in cocultures between 15 different strains from five yeast species: Saccharom yces cerevisiae , Lac hancea thermotoler ans, Torulaspor a delbrueckii, Starmerella bacillaris, and Hanseniaspora uvarum .For this purpose, we first generated all strains with a fluorescent tag to measure the population r elativ e abundance of eac h str ain in the cocultur es.
We then tested all monocultures (coculture consisting of the wildtype strain and its genetically modified counterpart, which carries a fluorescent protein) and pairwise cocultures in synthetic must, following the OD 600 in microplates (Fig. 1 ) to enable testing numerous combinations.

Construction of fluorescently tagged strains
We successfull y mana ged to integr ate the cassette containing the EGFP gene at the locus in the five species we studied.Howe v er, protocol has been adapted for each species, which should inform futur e a pplicability of this methodology for other target nonconventional yeast.For example, LiAc transformation method did not result in transformants for L. thermotolerans .T hus , only S. cerevisiae was transformed with the LiAc method, whereas electroporation method was applied for all the non-Saccharomyces (NS) strains since it is reported to be more effective (Lin-Cereghino et al. 2005, Gordon et al. 2019 ).Concerning the homologous recombination, in S. cerevisiae and L. thermotolerans, short homology arms (60 bp) flanking the cassette were sufficient to obtain transformants with the cassette integrated at the locus .T her efor e, for S. cerevisiae and L. thermotolerans , cassettes were amplified from nonspecific plasmids using primers containing overhang with homologous sequence to the target integration site (stop codon of the TDH3 gene) of 60 bp ( Table S3 ).Ho w e v er, for T. delbruec kii as well as H. uvarum , short homologous sequences were not sufficient to obtain targeted integration, which may be explained by a predominance of the non-homologous-end-joining (NHEJ) DNA repair mechanism in some species (Cai et al. 2019 , Nav arr ete and L. Martínez 2020 ).Consequently, we used homologous sequences of 1 kb upstream and downstream the target integration site to pr omote homologous r ecombination for T. delbruec kii , H. uv arum and S. bacillaris (Nambu-Nishida et al. 2017, Badura et al. 2021 ).This r equir ed the construction of specific plasmid containing the homologous sequences flanking the cassette with the EGFP and antibiotic resistance genes ( Table S1 ).Every coculture, as well as monoculture, was then constituted of one tagged strain (or clone for monoculture) with one untagged strain (or clone) to enable their discrimination.

Inter-and intra-specific variability in monocultures
Gr owth of cocultur es and corr esponding monocultur es wer e assessed through four kinetics par ameters, namel y latency time ( Latency = time in hours for the OD 600 to exceed 0.25), intrinsic gr owth r ate ( r ), maxim um observ ed OD 600 ( maxOD ) and the ar ea under the curve ( AUC ).Growth rate ( r ), and area under the curve ( AUC ) w ere obtained b y fitting gro wth data with a logistic model.The AUC is a convenient metric to study microbial growth since it includes all pr e vious metrics (Spr ouffske and Wa gner 2016 , Piccardi et al. 2019 ).MaxOD and Latency were directly measured.Gr owth par ameters wer e assessed after 24 hours of gr owth since population then r eac hed signal satur ation.Mor eov er, pr eliminary test did not show significant difference in maximum population after 30 hours compared to 24 hours of growth (data not shown).
While there is no significant effect of the species for Latency, there is a significant strain effect (p.value Latency/strains << 0.001, Fig. 3 ).Here the separation is not structured by species but by strains .For example , L. thermotolerans , S. cerevisiae , T. delbrueckii and S. bacillaris all presented one strain whose latency value was different from the others (Fig. 3 ).For the 15 strains, the latency ranged from 3.48 ± 1.21 h for Td 3337 to 7.66 ± 0.59 h for Td LO544 with an av er a ge of 5.98 ± 1.51 h.

Ev alua ting species-species and str ain-str ain interaction in cocultures
To determine how the species and strain effect affected growth in cocultures, we tested whether there was a significant effect on the four growth parameters in all cocultures.To this end, we distinguished three types of culture: S 1 strain 1 monoculture, S 2 strain 2 monoculture and Co the corr esponding cocultur e comprising strain 1 and strain 2. Theoretically, four typical outcomes can be distinguished (Fig. 4 ): when the T-test did not show significant difference between the parameter value of the coculture and the aver a ge of the parameter value of the monocultures, it was called F igure 2. Gro wth of the 15 monocultures (composed of 50% of fluorescently tagged cells and 50% of WT cells of the same strain) for the 5 species tested in this study: H. uvarum (Hu), L. thermotolerans (Lt), S. bacillaris (Sb), S. cer e visiae (Sc) and T. delbrueckii (Td).All growth curves are represented in gr ey, gr owth curv es of all thr ee str ains of a species ar e r epr esented in color ed lines and species ar e separ ated in facets.Monocultur es wer e done in biological triplicates.Curves were ploted using the Loess smoothing method from R tidyverse package.case A [for example: AUC Co ≈ (AUC S1 + AUC S2 )/2], and there is no perceiv ed c hange in dynamics.If the cocultur e gr owth par ameter is statistically superior to the best monoculture or inferior to the worst monoculture, it was r espectiv el y called cases B and C, wher e ther e is a clear positive or negative interaction.In all other cases (case D) the parameter value of the coculture is between monocultur e par ameters v alues but it is not possible to determine if there is an interaction or only the effect of r espectiv e population density (Fig. 4 D).In addition to classifying results according to these categories, we also calculated an interaction index (Id) by comparing for each parameter (growth rate, maximum population, latency and AUC) the coculture value to both monoculture values.We used this index to perform a clustering analysis of all cultures.
When focusing on the AUC, a great majority of cocultures were in the A (77%) and D (33%) cases (Fig. 4 , Table 2 ) wher e ther e is little to no perceived interactions.For the other parameters, most cocultur es wer e also in case A or D (Table 2 ).A few underyielding cases were identified for the maximum population (5%) such as in the coculture of S. cerevisiae 59A with L. thermotolerans V7-21 (Fig. 5 ).Only one overyielding case (B) was found for the growth rate in the cocultur es of H. uv arum 3221 with T. delbruec kii 3337 r espectiv el y (Fig. 5 ).Ov er all, these r esults would suggest that the measured gr owth par ameters did not a ppear to identify str ong inter actions in cocultures.
As we did not observe many extreme changes in population dynamics such as over-and under-yielding, we thus used an interaction index for all four growth parameters to assess the strength of the interaction in addition to the quality of the interaction.Noteworthy, for positive interactions, it was not possible to determine whether the interaction was positive for both species or only one.Regarding the AUC-based heatmap (Fig. 6 ), negative interaction were seen in cocultures of all strains of S. cerevisiae with T. delbrueckii CLIB3337 and L. thermotolerans V7-21.The strain S. cerevisiae 59A also sho w ed significant negativ e inter actions with all T. delbruec kii str ains and all L. thermotolerans str ains.On the contr ary, cocultur es of S. bacillaris with S. cerevisiae, H. uvarum or T. delbrueckii tended to have positive index but only cocultures of S. bacillaris CLIB3147 with S. cerevisiae VIN13 or S. bacillaris with either T. delbrueckii CLIB3337 or T. delbrueckii CLIB3069 sho w ed statistically significant positive interactions.
The other growth parameters also corresponded to the interactions quantified in terms of AUC.Cocultures of S. bacillaris with other species tended to be positive.Ho w ever, contrary to the other par ameters, negativ e inter actions wer e r e v ealed by the latency parameter in cocultures of S. bacillaris with L. thermotolerans or T. delbruec kii with e v en underyielding for S. bacillaris 3147 and L. thermotolerans Y1240 ( Fig. S2 ).For the latency, a positiv e inter action index indicates a longer latency time, hence a delay ed gro wth so a negative interaction.
Negativ e inter actions between S. cerevisiae and T. delbruec kii or L. thermotolerans were also revealed by the interaction index of gr owth r ate, maxim um population, and latency.Underyielding (Fig. 3 C) was observed for growth rate in cocultures of S. cerevisiae VIN13 with L. thermotolerans CLIB3053 or L. thermotolerans V7-21 as well as for the maximum population in cocultures of S. cerevisiae 59A with T .delbrueckii 3069, T .delbrueckii LO544 and L. thermotolerans V7-21 ( Fig. S3 ).The latency also r e v ealed some negative interactions between L. thermotolerans and T. delbrueckii .
In addition to interspecific inter actions, the maxim um population par ameters r e v ealed intr a-specific negativ e inter actions for the cocultures of T. delbrueckii 3069 with T. delbrueckii LO544, and L. thermotolerans CLIB3053 with L. thermotolerans Y1240 with negative inter action str ong enough to induce underyielding ( Fig. S2 ).
F igure 3. Gro wth parameters of the 15 monocultures: observed maximum population (maxOD), growth rate hours −1 (r), area under the curve (AUC) and latency in hours (lat, time to r eac h OD600 = 0.25).r and AUC were computed using modelling from the GrowthCurver R package.Monocultures were done in biological triplicates.Letters a,b,c indicate the statistical group of the species.Species effect and strain effect were evaluated using a nested ANOVA: P -values of both fixed (species effect) and random effect (strain effect) are indicated for each growth parameter.H. uvarum (Hu), L. thermotolerans (Lt), S. bacillaris (Sb), S. cer e visiae (Sc) and T. delbruec kii (Td).
Table 2. Count and percentage of cocultures types for all four growth parameters.A: no statistical difference between the coculture par ameter v alue and the av er a ge of both monocultur es (e v aluated by T-test).B: ov eryielding-the cocultur e par ameter v alue is higher than the maximum of both monocultures (T-test with maximum).C: underyielding-the coculture parameter value is lo w er than the minimum of both monocultures (T-test with minimum).D: the coculture parameter value is between both monocultures and statistically differ ent fr om the av er a ge of both monocultur es.

Analysis of the interaction matrices
Another inter esting r esult obtained fr om the heatma p is the structure of the interaction matrix.The clustering of rows and columns was based only on similarity of the interaction index between strains (Euclidean distance with complete linkage), and the r ele v ance of clusters was c hec ked by bootstr a pping.We observed that the resulting clusters based on AUC interaction index values fit species le v el for H. uv arum , S. cerevisiae and S. bacillaris (Fig. 6 ) .Cluster 2 includes both L. thermotolerans strains and T. delbruec kii str ains (CLIB 3069 and CLIB 3337).Cluster 5 includes T. delbruec kii LO544 onl y, ho w e v er bootstr a pping anal ysis of the clusters sho w ed the significance of cluster 5 w as lo w (J accar d index = 0.64).Altogether, this would suggest that species rather than strain is the main le v el determining interactions in cocultures with two species, even though strain can impact the strength of the interaction since values of the interaction index varied between strains of the same species.For instance, T. delbrueckii 3069 and 3337 sho w ed great positive interactions with S. bacillaris V8-1 whereas this interaction was more neutral with S. bacillaris 3147.
Figure 6.Heatmap of the AUC index: positive index means an overall higher growth of the coculture.* denotes cocultures whose AUC is significantly differ ent fr om the av er a ge AUC of both monocultur es .Colors of cluster and species , as w ell as addition of the * w er e edited manuall y fr om the PDF file .T he original figure from R is available in the data repository.
Clustering on the gr owth r ate inter action index also fitted mostly to the species level; with the only discrepancy being S. cerevisiae 59A that clustered with T. delbrueckii strains ( Fig. S1 ).There were some examples of strain-level interactions on growth rate, for example, T. delbrueckii 3069 and 3337 exhibited opposing negative and positive interactions respectively with H. uvarum .For the maximum population and latency time, there were differing clustering patterns ( Figs S2 and S3 ).For the interaction index data from the latency time, strains of S. bacillaris and strains of T. delbrueckii did not group together, while strains of other species did gr oup into r espectiv e clusters ( Fig. S2 ).For the maximum population metric, S. cerevisiae, L. thermotolerans and two strains of T. delbruec kii str ains gr ouped together in a single cluster, while strains of other species did group into respective clusters ( Fig. S3 ).
The fact that structures of the interaction matrix didn't follow the species le v el for the maxim um population may be related to limitations of measures by OD that tend to be quic kl y satur ated (Ste v enson et al. 2016 ).This might also result from the intrinsic growth phenotypes of each strain since monocultures that were alr eady gr ouped together for the maxim um population, namel y S. cerevisiae , T. delbrueckii , and L. thermotolerans , wer e cluster ed together.The structure found for the phenotype of monocultures might also explain the fact that in cocultures, as for monocultur es, no pr edominant species effect was observ ed for the latency.

Population dynamics highlighted by change in relati v e a bundance
To e v aluate the influence of cocultur e on the population composition, which is an important metric in determining competitive phenotypes, we calculated fold change with the r elativ e abundance of both strains in eac h cocultur es at start (T0) and after 24 hours of growth (T24).
The population abundances of most strains within coculture stayed consistent throughout the measured samples.In Figur e 7. P opulation fold change after 24-hour growth in SGM425 for each strain (panel) in coculture with the strain indicated in the column.Colors of the dot correspond to the species of the strain in the column.* denotes significant difference of the fold change to 1 as tested by t-test.particular, for monocultures, no c hange in r elativ e abundance of the wild-type and tagged strains were observed, indicating transformants did not exhibit differences in fitness as compared to wild-type strains (Fig. 7 ).On the contrary, all strains of S. bacillaris sho w ed significant decrease in r elativ e abundance after 24 h when in coculture with other species, with a 2-fold reduction on average for the 3 strains (fold change = 0.52 ± 0.02).This is logically associated with an increased abundance of the counterpart species .T he fold change in favour of better growing strains when cocultured with S. bacillaris could explain the positive interaction observed with maximum population and AUC for these cocultures (Fig. 6 , Fig. S3 ) since the maximum population observed is the result of their r espectiv e maxim um OD600.Fold c hange of cocultur es of other species were less sizable and fe w wer e significant.Sur prisingly, besides with S. bacillaris, S. cerevisiae abundance increased only in cocultures with T. delbrueckii.For instance, S. cerevisiae strain 1152 had a fold change of 1.55 ± 0.09 and 1.68 ± 0.2 when cocultured with T. delbrueckii 3069 and LO544 respectively.L. thermotolerans had significant increased abundance in some strainspecific cases such as L. thermotolerans 3053 with H. uvarum 3118 (1.52 ± 0.13) or L. thermotolerans Y1240 in coculture with T. delbrueckii LO544 (2.13 ± 0.26).
In terms of str ain-str ain differ ences, the data shows foldc hange v ariations in intr aspecific cocultur es of T. delbruec kii (Fig. 6 ).For instance, T. delbrueckii LO544 relative abundance declined when in presence of either T. delbrueckii 3069 or 3337 (fold change of 0.70 ± 0.06 and 0.66 ± 0.08, r espectiv el y).Two factors could explain this intraspecific effect.It could either be related to the longer latency phase of the strain LO544 (Fig. 3 ).It might also be the results of intraspecific negative interactions between strains of T. delbrueckii as observed with the significantly reduced maximum population in cocultures of T. delbrueckii LO544 with T. delbrueckii 3069 or 3337 compared to monocultures.On the contrary H. uvarum , S. bacillaris , and S. cerevisiae sho w ed no fold change when cocultured with a strain of the same species.

Discussion
In the present study, we conducted an experiment aiming to explore ecological questions regarding the relative importance of species and strain for determining the nature and intensity of y east-y east interactions.For this purpose, we studied populationspecific and ov er-all gr owth kinetics of all pairwise cocultures of 15 total strains comprising 5 different yeast with a simplified fr ame work of the wine environment.
The a ppr oac h used in this study involv ed ta gging all str ains with fluorescent proteins to enable species detection with cytometry at the end of growth.With the current lack of information on the genetics of wine related NS, molecular tools available are still scarce and transformation of NS remains highly challenging (Masneuf-Pomarede et al. 2016 ).To our knowledge, only one article reported transformation with homologous recombination for H. uv arum (Badur a et al. 2021 ), and one article reported transformation of Starmerella bombicola (Gonçalves et al. 2018 ) but none specificall y r eported tr ansformation of S. bacillaris.The str ains we constructed will ther efor e be valuable tools in the future for studies on y east.Ho w e v er, mor e r esearc h should focus on tr ansformation of NS since ther e ar e still species that wer e r eported to be unable to integrate cassette at the target locus, such as M. pulcherrima (Gordon et al. 2019, Moreno-Beltrán et al. 2021 ).The tagging strategy coupled with cytometry, which is a po w erful analytical tool for population analysis of fermentation (Longin et al. 2017 ), enabled us to discriminate two strains in cocultures .T his a ppr oac h can also be applied to the detection of more species (Conacher et al. 2020 ).A limitation of the study is that our data only focus on the growth patterns of species, which limit our understanding on the r espectiv e gr owth of eac h str ain, as well as the impact of str ains on the resulting wine composition.Moreover, oxygen availability in microplates do not reflect real fermentation conditions.Howe v er, this method makes it possible to monitor growth of multiple pair-combination, and is very useful for high-throughput protocols.
In this study, comparison of growth kinetics of cocultures and monocultures based on an interaction index suggested a predominant effect of the species le v el ov er the strain on the interactions structure .T he cluster analysis resulting from the interaction matrix for the AUC and the growth rate displayed an ov erla p of clusters with the species le v el, despite the initial intra-specific diversity observed in monocultures.Altogether, our results would indicate that population dynamics between two species are mainly driven by the species type, while the strain would mostly affect the strength of the interaction.This is an important consideration in the design of synthetic communities.
It is interesting to note that most cocultures displayed little to no perceived interactions (cases A Table 2 , Fig. 3 ).These results would reflect those observed in some bacterial cocultures, either fr om str ains of the same species or spanning se v er al families and gener a, wher e inhibition inter actions constituted less than 15% of pairwise interactions (Russel et al. 2017, Ramia et al. 2020 ).It could also result from the microtiter plate method that do not allow a detailed analysis of minor changes, especially in the respective growth of each strain.Nevertheless, our study provides a broader insight in yeast inter actions, especiall y NS/NS inter actions that ar e still poorl y documented (Zilelidou and Nisiotou 2021 ).Most interactions that have been studied to date are negative interactions found between S. cerevisiae and NS, but some positive interactions thr ough cr ossfeeding wer e also identified between L. thermotolerans and Zygosacc harom yces spp.for example (Csoma et al. 2020 ).Further r esearc h is needed to confirm our findings, especially with more species to include more genera as well as species from the same genera similarly to a recent study that investigated cocultures of 60 strains of wine yeast in coculture with S. cerevisiae (Ruiz et al. 2023 ) .In addition, our findings are limited to only one synthetic media, whereas interactions are known to be modulated by environments (Piccardi et al. 2019, Gao et al. 2021 ).T hus , it would be r ele v ant to test these combinations in environments closer to actual wine fermentation, for instance using different natur al gr a pe m usts .Indeed, the wine en vir onment includes v arious stressors that have been shown to influence population dynamics, e v en at the strain level as shown by Schmidt for S. cerevisiae (Schmidt et al. 2020 ).The importance of strains variability might then lie in the adaptability of one species to different envir onments.Mor eov er, the str ains e v aluated her e hav e all been isolated from wine en vironments .Evidence clearly supports that this anthr opic envir onment has e volutionaril y sha ped the associated yeast community (Conacher et al. 2019, De Guidi et al. 2023 ).The interactions between the species and strains evaluated here might ther efor e be the result of wine-specific evolutionary adaptations linked to direct interspecies biotic selection pressures.It would be interesting to add strains isolated from other environments to our analysis.
Phylogenetic or metabolic distance might be part of the explanation of the r ele v ance of species in pairwise interactions.For instance, Russel et al. sho w ed that bacterial species phylogenetically closer tended to display higher competition, the assumption being that phylogenetically closer species have closer niches (Russel et al. 2017 ).Peay et al. ( 2012) obtained similar results for yeasts in a floral nectar flo w er community assembly.Our data would be in accordance with these findings since we observed significant negativ e inter actions between species suc h as S. cerevisiae , T. delbruec kii or L. thermotolerans , e v en though Ruiz et al. ( 2023 ) identified positiv e inter actions between S. cerevisiae and L. thermotolerans or T. delbrueckii .In our study, these three species sho w ed similar growth patterns in monocultures, except for a lo w er gro wth rate for T. delbrueckii strains (Fig. 3 ), and are known to be phylogenetically closer together than S. bacillaris and H. uvarum (Kurtzman 2011, Lemos Junior et al. 2018 ).Mor eov er, they also seem to have similar amino-acid consumption and are reported to be intermediate or good fermentativ e species, whic h might r esult in higher competition (Prior et al. 2019, Roca-Mesa et al. 2020 ).Howe v er, if it was only a question of phylogenetic distance, then there would be very high intraspecific competition which we observed onl y for T. delbruec kii and L. thermotolerans (fold change and maximum population; Fig. 7 ; Fig. S3 ) while interactions in intraspecific cocultures for the other 3 species were mostly neutral.This might indicate other interaction mechanisms are also involved, such as contact-dependent interactions for example.For instance, S. cerevisiae seems to induce contact-dependent cell-death of other species such as L. thermotolerans (Petitgonnet et al. 2019, Luyt et al. 2021 ).Although, for T. delbrueckii , Taillandier et al. ( 2014 ) excluded contact-mediated interactions between T. delbrueckii and S. cerevisiae but instead hypothesized that T. delbrueckii was sensitive to a killer toxin produced by S. cerevisiae .On the other hand, negativ e inter actions mediated by cell-contact hav e been r eported between S. bacillaris and S. cerevisiae , whereas we mostly identified positive interactions between S. bacillaris strains and the other

Figur e 1 .
Figur e 1. Experimental la yout of the mono-and coculture microplate cultures.

Figure 4 .
Figure 4. Classification of the possible outcomes of co-cultures compared to monocultures.A: No interaction-co-culture relates to the average of both monocultures, B: Overyielding-coculture is greater than both monocultures, C: Underyielding-co-culture is worse than both monocultures, D: Little inter action-cocultur e significantl y differs fr om the av er a ge of both monocultur es but r emains in the r ange of both.S1 = str ain 1, S2 = S2, in dashed line = the calculated av er a ge of both monocultures.

Figure 5 .
Figure 5. Examples of case B (left panel) and C (right panel).The onl y ov eryielding observ ed was for the gr owth r ate of the cocultur e of H. uv arum 3221 and T. delbrueckii 3337.Underyielding was observed with the maximum population of the coculture of L. thermotolerans V7-21 and S. cerevisiae 59A.Four biological replicates were run for each cocultures.

Table 1 .
List of yeast strains used in this study.
, Y i : gr owth par ameter, βSpecies i fixed term r elated to species, b i Strain i term related to the strain effect, since species and strains constitute hierarchical variables .T he model was tested with the lmer function (lmerTest pac ka ge , Kuznetso va et al. 2017 ).