Ethanol Production and Maximum Cell Growth Are Highly Correlated with Membrane Lipid Composition during Fermentation as Determined by Lipidomic Analysis of 22 Saccharomyces cerevisiae Strains

Abstract

Optimizing ethanol yield during fermentation is important for efficient production of fuel alcohol, as well as wine and other alcoholic beverages. However, increasing ethanol concentrations during fermentation can create problems that result in arrested or sluggish sugar-to-ethanol conversion. The fundamental cellular basis for these problem fermentations, however, is not well understood. Small-scale fermentations were performed in a synthetic grape must using 22 industrial Saccharomyces cerevisiae strains (primarily wine strains) with various degrees of ethanol tolerance to assess the correlation between lipid composition and fermentation kinetic parameters. Lipids were extracted at several fermentation time points representing different growth phases of the yeast to quantitatively analyze phospholipids and ergosterol utilizing atmospheric pressure ionization-mass spectrometry methods. Lipid profiling of individual fermentations indicated that yeast lipid class profiles do not shift dramatically in composition over the course of fermentation. Multivariate statistical analysis of the data was performed using partial least-squares linear regression modeling to correlate lipid composition data with fermentation kinetic data. The results indicate a strong correlation (R² = 0.91) between the overall lipid composition and the final ethanol concentration (wt/wt), an indicator of strain ethanol tolerance. One potential component of ethanol tolerance, the maximum yeast cell concentration, was also found to be a strong function of lipid composition (R² = 0.97). Specifically, strains unable to complete fermentation were associated with high phosphatidylinositol levels early in fermentation. Yeast strains that achieved the highest cell densities and ethanol concentrations were positively correlated with phosphatidylcholine species similar to those known to decrease the perturbing effects of ethanol in model membrane systems.

INTRODUCTION

Nearly 60% of all transportation fuel is consumed by passenger vehicles and ethanol derived from cellulosic biomass is considered to be the most viable short- to middle-term biofuel alternative to petroleum-based fuels (1, 2). Although the development of fuel alternatives has driven a growing interest in alcoholic fermentations, the process of producing ethanol through fermentation for the production of wine and other alcoholic beverages, studied for well over a hundred years, shares many of the same issues. While many yeast and other microorganisms have an innate ability to convert sugar to ethanol at relatively high concentrations, alcoholic fermentations may still stop at some concentration of ethanol that depends on the strain, as well as factors such as nutrient availability and temperature (3–8). Modulating ethanol output in production strains will allow for desirable process intensification in biomass-to-energy conversion, as well as increased flexibility in the production of ethanol for other purposes. In addition, understanding ethanol tolerance will allow for the introduction of this trait into potential production strains with other desirable characteristics, such as the simultaneous use of five- and six-carbon sugars.

Over the course of alcoholic fermentation, yeast experience numerous environmental and biological stresses that can affect fermentation efficiency and outcome (3–5, 9). Yeast strains commonly used in industrial-scale alcoholic fermentations have acquired stress response mechanisms involving complex genomic, signal transduction, and regulatory pathways to adapt and survive in dynamic and inhospitable environments (3, 9–11). However, one of the most important and least understood adaptive responses of this organism is its ability to withstand high ethanol concentrations. The yeast plasma membrane appears to be a primary target of the perturbing effects of ethanol exhibited by impacts on membrane integrity, as well as membrane-associated processes (3, 10, 12). There is significant evidence that the lipid composition of the strain contributes to its tolerance to increasing quantities of self-produced ethanol (5, 12, 13).

The cellular membranes of yeast are composed primarily of three types of biomolecules: phospholipids, sterols, and membrane associated proteins. The principal sterol in Saccharomyces cerevisiae is ergosterol, and the principal phospholipids have been shown to be phosphatidic acid, phosphatidylethanolamine, phosphatidylinositol, phosphatidylserine, and phosphatidylcholine with fatty acid chains that are predominantly oleic acid (C_18:1) and palmitoleic acid (C_16:1), with smaller amounts of palmitic acid (C_16:0) and stearic acid (C_18:0) (14, 15). Variations in the fatty acid moieties esterified to the glycerol backbone of phospholipids yield hundreds of different molecules that yeast cells utilize to maintain cellular function and adapt to their environment (16). Owing to the complex composition of these membranes, little is known about the physical responses of these lipid bilayers to increasing concentrations of ethanol. However, model membrane systems composed of phosphatidylcholines and sterols have demonstrated that lipid composition and structure can have a protective effect on membrane bilayers in the presence of ethanol by mitigating the membrane-thinning effect of ethanol (17–20). Ethanol-induced changes in the membrane thickness of fermenting yeast cells could potentially interfere with membrane-associated protein function (12, 21), e.g., proteins involved with sugar and nitrogen transport, as well as signal transduction (5, 21, 22).

Given the effect of ethanol on model membrane systems and modulation of this effect by lipid composition, it might be expected that yeast membrane composition would be correlated with ethanol tolerance per cell for various strains of S. cerevisiae. Previous studies have examined yeast strains from various sources under differing fermentation conditions, making it difficult to draw definitive conclusions regarding how yeast cell lipid composition contributes to ethanol tolerance (9). Furthermore, these studies lacked analytical methodologies that are amenable to rapid, quantitative analysis of lipid compositional changes that occur over the course of fermentation in a large set of yeast strains. Utilization of mass spectrometry (MS) methods to analyze the lipid composition, or lipidome, of an organism has gained widespread popularity due to higher resolution, increased sensitivity, and rapid data acquisition over previous methods (23). Recently, we reported an analytical MS method for rapid, quantitative analysis of phospholipids and ergosterol from yeast cell extracts, which facilitates the examination of a large set of yeast strains over multiple fermentation time points in a feasible time period (24). While the lipid composition is considered to be an important potential contributor to a yeast strain's ethanol tolerance, other mechanisms that contribute to fermentation capabilities of this organism have been proposed, such as conditions that lead to larger yeast cell populations (6–8). Differentiating between the various factors that contribute to a successful fermentation would facilitate directed construction of strains capable of producing higher levels of ethanol.

We examine here a series of 22 industrial S. cerevisiae strains that exhibit a range of fermentation capabilities and final ethanol concentrations. We describe how the final ethanol concentrations achieved by each strain correlate with lipid composition over the course of alcoholic fermentation. We also examine how maximum cell density is correlated with lipid composition in fermenting yeast strains. Atmospheric pressure ionization (API)-MS was utilized to facilitate the analysis of the effects of lipid composition on yeast cell growth and ethanol tolerance for a large set of yeast strains over several fermentation time points. We describe the use of multivariate statistical techniques, commonly used with the analytical procedures utilized in the present study, to deconvolute the large and complex data sets that were generated to determine possible mechanisms of tolerance and identify candidate lipids that appear important to these mechanisms.

MATERIALS AND METHODS

Materials.

All chemicals were acquired from Sigma-Aldrich (St. Louis, MO), and all Nanopure water used was obtained from a Milli-Q Synthesis A-10 water purification system (Millipore, Billerica, MA) unless noted otherwise. Internal standards (ISTDs) were used during method development and to construct standard curves for quantitation. They were chosen based on their not being endogenous and not having the same molecular weight (isobaric) as lipids previously identified in the S. cerevisiae lipidome (25, 26) or reference (REF) lipids representing each lipid class being analyzed. Specifically, these were 1,2-dilauroyl-sn-glycero-3-phosphate (PA 12:0-12:0) (ISTD), 1,2-dilinoleoyl-sn-glycero-3-phosphoethanolamine (PE 18:2-18:2) (ISTD), l-α-phosphatidylinositol (liver, bovine) (PI 18:0-20:4) (ISTD), 1,2-dilinoleoyl-sn-glycero-3-phospho-l-serine (PS 18:2-18:2) (ISTD), 1,2-dipentadecanoyl-sn-glycero-3-phosphocholine (PC 18:2-18:2) (ISTD), cholesterol (plant derived) (ISTD), 1,2-diheptadecanoyl-sn-glycero-3-phosphate (PA 17:0-17:0) (REF), 1,2-diheptadecanoyl-sn-glycero-3-phosphoethanolamine (PE 17:0-17:0) (REF), l-α-phosphatidylinositol (soy) (PI 16:0-18:2) (REF), 1,2-diheptadecanoyl-sn-glycero-3-phospho-l-serine (PS 17:0-17:0) (REF), and 1,2-dilinoleoyl-sn-glycero-3-phosphocholine (PC 15:0-15:0) (REF), and they were purchased from Avanti-Polar Lipids (Alabaster, AL). Ergosterol (REF) was purchased from Sigma-Aldrich. Lipid extraction solvents were high-pressure liquid chromatography (HPLC)-grade chloroform with 0.5 to 1.0% (vol/vol) ethanol stabilizer, HPLC-grade methanol, and HPLC-grade water. Mobile-phase and gas chromatography (GC) solvents and standards were HPLC-grade hexanes, absolute ethanol, isopropanol, and water and LC-MS-grade formic acid and triethanolamine. Yeast extract-peptone-dextrose (YEPD) agar plates were prepared using 10 g of Bacto yeast extract/liter, 20 g of Bacto peptone/liter, 20 g of glucose (dextrose)/liter, and 20 g of Bacto agar (Becton Dickinson, Sparks, MD)/liter according to the method of Amberg et al. (27). Gases necessary for instrumentation operation were medical-grade nitrogen and helium and hydrogen of 99.99% grade (Praxair, Danbury, CT).

Low-sugar (220 g/liter; ∼22.0 °Brix), as a 1:1 mixture of glucose (110 g/liter) and fructose (110 g/liter), MMM synthetic grape juice medium was prepared according to the method of Giudici and Kunkee (28) as previously reported (24).

Lipid extraction solvent A was composed of 2.5:1:1 methanol-chloroform-water, extraction solvent B was composed of 1:1 methanol-chloroform, and extraction solvent C was composed of 5% (vol/vol) formic acid in water. The injection solvent and mobile phase A were composed of 7:3 hexanes-isopropanol with 0.5% (vol/vol) formic acid and 0.5% (vol/vol) triethylamine, and mobile phase B was composed of 92:8 isopropanol-water with 0.5% (vol/vol) formic acid and 0.5% (vol/vol) triethylamine.

Yeast strains and inoculation.

All S. cerevisiae strains were acquired from the UC Davis Enology Culture Collection and are listed in Table 1. The strains were stored and plated on YEPD plates for single colony isolation according to the method of Amberg et al. (29). Yeast strains were stored at 4°C on plates for no longer than 1 month. The optical density at an absorbance wavelength of 600 nm (OD₆₀₀) of the yeast culture was determined using a UV-1201 spectrophotometer (Shimadzu Scientific Instruments, Inc., Kyoto, Japan) as previously described (24).

Table 1.

Origins of S. cerevisiae strains used in this study

Strain	UCD accession no.	Source description	Subspecies
Montrachet	522	Isolated from Montrachet wine in 1958
Prise de Mousse	594	From the original Prise de Mousse culture in the Pasteur Institute collection in Paris, France; isolated from French champagne	bayanus
Sake A18	612	Isolated from sake, Kurashi
Bread yeast	668	Commercial bread yeast; Universal Foods, Lesaffre Co., Lille, France
EC1118	777	Commercial wine yeast; Institute Oenologique de Champagne	bayanus
M2	906	Commercial wine yeast; Stellenbosch, South Africa
Lalvin ICVD47	963	Commercial wine yeast; Suze-la-Rousse, Cotes du Rhone
FST 40-427	1427	Distillery yeast; Renmark Growers Distillery, 1948; Renmark, Australia
Cote de Blanc	2031	Commercial wine yeast; Red Star
Uvaferm 43	2032	Commercial wine yeast; Lallemand	bayanus
Cepage chardonnay	2061	Commercial wine yeast; Gist-Brocades, Delft, Netherlands
Zymaflore VL-1	2074	Commercial wine yeast; Laffort, Burgundy France
Premier Cuvee	2212	Commercial wine yeast; Red Star	bayanus
CY3079	2497	Commercial wine yeast; Bourgoblanc
DV10	2498	Commercial wine yeast; Champagne	bayanus
ICV D254	2499	Commercial wine yeast; Rhone region
Enoferm Simi White	2501	Commercial wine yeast; Lallemand
Enoferm T306	2502	Commercial wine yeast; Hunter Valley, New South Wales, Australia
ICVK1 (V-1116)	2537	Commercial wine yeast; Montpellier
Lalvin Rhone L2226	2545	Commercial wine yeast; Cotes du Rhone
C0490GP2-B11	V3	Strains generated by directed evolution at the University of Adelaide
FM16-7	V4	Strains generated by directed evolution at the University of Adelaide

To inoculate experimental cultures, a preculture was prepared by transferring a single colony from the YEPD agar plate to 15 ml of MMM medium and allowed to grow aerobically for 24 to 48 h at 25°C in an orbital incubator (New Brunswick Scientific, Edison, NJ). Next, an aliquot was taken from this preculture to inoculate 50 ml of MMM in a 250-ml Erlenmeyer flask to an OD₆₀₀ of ∼0.1. This inoculum was grown in an orbital incubator at 25°C until a corrected OD₆₀₀ of ∼3 was reached. Finally, a volume of the inoculum to give an initial OD₆₀₀ of ∼0.1 (∼15 ml) was added to 400 ml of MMM medium in a 500-ml Erlenmeyer flask fitted with a Senior Airlock (Buon Vino, Ontario, Canada).

Fermentation sampling.

Fermentations were performed in triplicate for each strain. Initial cell concentration determinations (OD₆₀₀) and Brix measurements were carried out at the beginning of fermentation and subsequently every 24 h until the °Brix fell below 1 or remained constant for two consecutive measurements, at which point the fermentations were stopped. Brix measurements were performed on clarified supernatants (900 × g, 5 min) using a DMA 35N densitometer (Anton Paar USA Inc., Ashland, VA). Yeast cells were harvested from a volume of fermenting medium that yielded an ∼150-mg cell pellet (15 to 50 ml [wet weight]). The harvested cells were washed three times in Nanopure water, and the cell pellet was stored at −80°C. Clarified supernatants were sterile filtered with 0.45-μm-pore-size syringe filters (Millipore) and stored at −20°C.

Ethanol analysis.

The final volume percent ethanol concentration of the fermentations was determined utilizing GC coupled to a flame ionization detector (GC-FID) according to the method of Zoecklein et al. (30) using isopropanol as the quantitative standard. The separation of ethanol and isopropanol was achieved using a Hewlett-Packard 5890 GC equipped with a 7673A autosampler on a Restek SilcoSmooth (2 m by 2 mm [inner diameter]) and 5% Carbowax (20 m) on a 80/120 mesh CarboBlack B support packed column (Restek Corp., Bellefonte, PA). The carrier gas was nitrogen at a column head pressure of 35 lb/in², with a corresponding flow rate of ∼30 ml/min. The air and hydrogen gas flows for the FID were 300 and 30 ml/min, respectively. The injector and detector temperatures were held at 150°C. The initial oven temperature was 85°C, with no hold. Oven temperature was ramped at 65°C per min to 150°C with an 18-s hold for a total run-time of 5 min. Clarified and filtered supernatants were diluted 1:99 in 0.20% isopropanol prior to analysis and the ethanol/isopropanol peak area ratios were correlated to a standard curve constructed using quantitatively prepared standards containing between 0 and 20 volume percent absolute ethanol. The weight percent ethanol was calculated based upon the ratio of the density of ethanol (0.789 g/cm³) to the density of pure water (31).

Lipid extraction and sample preparation.

The lipid extraction procedure was a modified Bligh-Dyer method adopted from Weckwerth et al. (32) with the addition of a 5% formic acid extraction step to improve recovery of acidic phospholipids. The lipid extracts stored under N₂ (gas) were removed from −20°C and allowed to warm to room temperature prior to analysis. The samples were then resuspended in the appropriate injection solvent and loaded into the LC auto-sampler (Agilent Technologies, Santa Clara, CA) for analysis.

HPLC separation.

Phospholipid separations were achieved using normal-phase chromatography as previously described (24) on an Agilent 1100 Series HPLC equipped with a temperature programmable column compartment and temperature controlled autosampler (Agilent Technologies) equipped with a YMC 2.0-by-150-mm column packed with PVA-Sil (5-μm particle size) and a guard column of the same material (Waters Corp., Milford, MA) at a temperature of 40°C. The samples were kept at a temperature of 10°C prior to injection (5 μl). A binary gradient of mobile-phase A and B sequentially eluted the polar lipids according to the following schedule: (i) from 0 to 10 min, mobile phase B increased from 12.5 to 15% at a flow rate of 400 μl/min; (ii) from 10 to 15 min, mobile phase B was increased to 100% at the same flow rate; (iii) from 15 to 15.5 min, mobile phase B remained at 100%, and the flow rate was decreased to 350 μl/min; (iv) from 15.5 to 25 min, mobile phase B remained at 100%; (v) from 25 to 26 min, mobile phase B returned to 12.5%; (vi) from 26 to 31 min, mobile phase B and the flow rate remained unchanged; and (vii) from 31 to 36 min, mobile phase B was kept at 12.5%, and the flow rate was returned to 400 μl/min. The total chromatographic cycle between injections was 36 min. The column eluent was directed to a micro-splitter valve (IDEX Health and Science, Oak Harbor, WA) to reduce solvent flow to the electrospray ionization (ESI) source to 50 μl/min.

Flow injection analysis.

The analysis of ergosterol from yeast extracts was adapted from the method of Toh et al. (33) as previously described (24). Sample injections of 10 μl were carried using the HPLC binary pump system (Agilent Technologies) described above at a flow rate of 400 μl/min, without a chromatographic column. The total time between injections was 6 min. The samples were kept at a temperature of 10°C. The sample/carrier solvent was directed to the atmospheric pressure chemical ionization (APCI) source of the ion-trap MS instrument.

MS analyses.

MS analysis was performed on an Agilent 6330 series quadrupole ion-trap mass spectrometer (Agilent Technologies). For phospholipid analysis, the instrument was equipped with an ESI source operating in negative-ion detection mode. For ergosterol analysis, the ionization source on the instrument was changed to an APCI source operating in positive-ion detection mode. The instrument was tuned and calibrated using the ESI and APCI calibration standards from Agilent Technologies. For phospholipid analysis, the MS settings were as follows: the capillary voltage was +3.5 kV, the nebulizer pressure was 25.0 lb/in², and the dry gas flow rate and temperature were 5 liters/min and 350°C, respectively. For ergosterol analysis, the MS settings were as follows: corona voltage, +4,000 nA; capillary voltage, −3.5 kV; nebulizer pressure, 60.0 lb/in²; dry gas flow rate, 5 liters/min; and dry and vaporizer temperatures, 350 and 400°C, respectively. For phospholipid analysis, the scan range was m/z of 100 to 900, and for ergosterol analysis, it was was m/z 100 to 500. The UltraScan scan mode was used in both analyses (21,000 m/z/s). The nebulizing and drying gas was nitrogen, and the collision-induced dissociation (CID) and ion-cooling gas was helium. The instrument control and HPLC-MS and HPLC-multistage tandem mass spectrometry (MSⁿ) data were collected using the ChemStation software package that accompanied the instrument (Agilent Technologies).

Data analysis methods.

Fermentation kinetic data was entered into SigmaPlot version 12.0 (Systat Software, Inc., San Jose, CA) for statistical analysis and used to plot Brix and the cell culture OD₆₀₀ at the following time points (in h): 48, 120, 192, and end (i.e., the time at which the °Brix had fallen below 1 or decreased by <1 °Brix over two consecutive measurements).

Chromatography, MSⁿ, and m/z profile data were exported in mzXML format and imported into MZmine version 2.1 (34) for MS feature extraction and processing. The processed MS data were exported to Excel for quantitative analysis. Phospholipid identification was performed using a chromatographic and MSⁿ database that was constructed using elution times, m/z values, and expected fragmentation patterns and/or characteristic losses following MS² and MS³ determined from ISTD and REF standards, as well as values that had been previously reported (35). Sterol identification was performed based upon the m/z values and MS² fragmentation profiles ascertained using cholesterol and ergosterol standards. These lipid m/z values were compiled into an MSⁿ database that was used to identify phospholipids and ergosterol in yeast lipid extracts. Lipid composition data was entered into Excel for statistical analysis and SigmaPlot version 12.0 (Systat Software) to generate plots of the lipid data.

Lipid quantitation.

Standard curves for quantification of lipids using MS data were generated as follows. Multiple samples were made varying the final concentration of REF lipids from 1 to 600 μM and were extracted, prepared for analysis, and analyzed in the same manner as for the yeast lipid extracts. Lipid standard data were plotted as the ratio of the concentrations (μM) of the REF standard to that of the ISTD (i.e., [REF]/[ISTD]) along the ordinate and the ratio of molecular ion intensity peak areas of the REF standard to the ISTD (i.e., MIA_REF/MIA_ISTD) along the abscissa and the least-squares linear models, and R² values of these data were generated in Excel, as previously described (24).

Multivariate statistical analysis.

The fermentation kinetic data (Y-Block) and quantitative lipid data (X-Block) were imported into MATLAB (version 7.11.0; MathWorks, Natick, MA) for partial-least-squares (PLS) regression analysis and interval-partial-least-squares (iPLS) variable selection using the PLS Toolbox (version 4.0; Eigenvector Research, Inc., Wenatchee, WA). Parameters for iPLS were chosen according to the method of Wise et al. (36) and Anderson and Bro (37). Reverse-analysis-mode iPLS selection was performed with an interval size of 1 variable with a maximum of 20 latent variables (LVs). The step size (distance between interval centers) and number of variables selected was automatically determined such that the algorithm was terminated when there was no improvement in the “root mean squared error in cross-validation” (RMSECV).

Upon completion of iPLS, the selected variables were loaded into PLS Toolbox for structured equation modeling. PLS regression analysis was performed using the SIMPLS algorithm (38) at a confidence limit of 0.95. Preprocessing of the X- and Y-blocks consisted of autoscaling, followed by mean centering. Cross-validation of the PLS model was performed using Venetian Blinds with 20 data splits and a maximum of 20 LVs. The correlation coefficient (R²), cross-validated correlation coefficient (Q²), “root mean squared error of correlation” (RSMEC), and RMSECV were determined to assess the quality and predictability of the PLS model. Scores, loadings, and regression vector plots of the selected lipid variables were generated to determine their influence on the PLS model and to determine potential lipid candidates that contributed to the final ethanol production and yeast cell growth during fermentation.

RESULTS

Fermentations.

Fermentations were performed for each of the 22 Saccharomyces strains in triplicate to assess the sugar utilization and ethanol production abilities of the strains relative to one another under identical growth conditions using the same defined medium, MMM. Because the fermentations were started at the same sugar concentrations, the final °Brix was well correlated with the final ethanol concentration. Cell biomass (estimated from the OD₆₀₀) and °Brix levels were assessed at regular intervals. Examples of the cell density and sugar utilization curves are shown in Fig. 1. Key data from all fermentations are compiled in Table 2.

Fig 1.

Representative experimental culture optical density at 600 nm (A) and °Brix curves (B) for three industrial yeast strains: DV10 (●), ICV D254 (○), and ICV K1 (V-1116) (▼). Each data point represents the average from fermentations carried out in triplicate.

Table 2.

Fermentation characteristics of S. cerevisiae strains used in this study

Straina	UCD accession no.	Mean ± SD			End time point (h)
		Max OD600	Final °Brix	% (wt/wt) ethanol
Enoferm T306	2502	5.89 ± 0.16	0.60 ± 0.36	9.93 ± 0.13	312
ICVK1 (V-1116)	2537	5.84 ± 0.42	0.67 ± 0.61	9.83 ± 0.59	264
Sake A18	612	5.76 ± 0.49	0.10 ± 0.17	10.36 ± 0.90	312
Cepage chardonnay	2061	5.64 ± 0.48	0.43 ± 0.45	10.18 ± 0.50	312
FM16-7	V4	5.61 ± 0.14	0.47 ± 0.32	10.67 ± 0.17	264
Enoferm Simi White	2501	5.51 ± 0.32	0.90 ± 0.79	10.67 ± 0.23	312
EC 1118	777	5.49 ± 0.40	0.33 ± 0.58	10.62 ± 0.37	264
Lalvin Rhone L2226	2545	5.20 ± 0.21	0.30 ± 0.30	10.44 ± 0.29	312
ICV D254	2499	5.03 ± 0.15	0.50 ± 0.36	10.34 ± 0.11	312
C0490GP2-B11	V3	4.99 ± 0.32	1.20 ± 0.53	10.57 ± 0.25	312
M2	906	4.97 ± 0.12	0.36 ± 0.02	9.50 ± 0.21	552
Uvaferm 43	2032	4.78 ± 0.51	1.17 ± 1.11	9.08 ± 0.56	312
Lalvin ICVD47	963	4.78 ± 0.14	0.73 ± 0.87	10.10 ± 0.51	312
Cote de Blanc	2031	4.37 ± 0.18	0.30 ± 0.26	10.55 ± 0.16	360
Bread yeast	668	4.20 ± 0.21	0.27 ± 0.46	10.13 ± 0.38	312
Zymaflore VL-1	2074	4.01 ± 0.20	0.27 ± 0.46	9.69 ± 0.28	384
Premier Cuvee	2212	3.21 ± 0.07	1.17 ± 1.00	8.88 ± 0.31	576
FST 40-27*	1427	3.14 ± 0.17	4.40 ± 0.90	7.96 ± 0.44	408
Montrachet*	522	2.98 ± 0.08	3.63 ± 0.32	7.96 ± 0.14	576
Prise de Mousse*	594	2.93 ± 0.06	2.53 ± 0.86	8.57 ± 0.35	432
CY3079*	2497	2.77 ± 0.08	2.60 ± 0.30	8.46 ± 0.20	576
DV10*	2498	2.68 ± 0.07	2.80 ± 0.26	8.33 ± 0.07	576

^a*, fermentation was stopped due to slow progress.

For the 22 strains that were evaluated, the mean weight percent ethanol ranged from 7.96 to 10.67 (the mean final °Brix values ranged from 0.10 to 4.40), and the mean maximum cell densities ranged from 2.68 to 5.89 OD units (Table 2). A number of strains failed to achieve final ethanol concentrations greater than 9.0%, indicating a lower ethanol tolerance or other problems during fermentation. Five of these yeast strains—FST 40-27, Montrachet, Prise de Mousse, CY3079, and DV10—experienced stuck fermentations with final °Brix values between 2.5 to 4.4 (the final weight percent ethanol concentrations ranged from 7.96 to 8.57%) and maximum OD₆₀₀ values no higher than 3.2 upon cessation of fermentation. Therefore, these strains, along with the other 16 strains that completed fermentation, represent a range of final ethanol concentrations and the ability to convert the same nutrients to biomass.

Lipid profiles during fermentation.

Quantitative MS analysis of yeast lipid extracts was performed using normal-phase HPLC-MS and APCI-MS for phospholipids and ergosterol, respectively. Calibration curves were generated and were approximately linear for all phospholipids in the sample concentration range from 3 to 600 μM for each lipid class being analyzed, with typical R² values of 0.989 to 0.998 (data not shown). Linear calibration curves were generated for ergosterol, spanning a concentration range from 7 to 600 μM with a typical R² value of 0.997 (data not shown).

Dramatic shifts in the relative concentrations of phospholipids within each lipid class (i.e., lipid head group) were not observed for most yeast strains during fermentation, while significant variations in phospholipid concentrations between lipid classes were seen in these strains (e.g., phosphatidylinositol versus phosphatidylcholine). As an example, Fig. 2A through D represent the molecular profiles of major phospholipid and ergosterol concentrations over four fermentation time points for the S. cerevisiae strain, ICV K1 (V-1116). Figure 2A represents the lipid profile of this strain at 48 h or the exponential growth phase. The most abundant phospholipid species were the phosphatidylcholines, in particular, PC 16:0-18:1. The next most abundant class was the phosphatidylethanolamines, followed by the phosphatidylinositols, in particular PI 34:1 and PI 18:1-18:0. By 120 h, or pre-stationary-phase growth, phosphatidylethanolamine levels increased and the concentrations of PC 16:0-18:1 and PI 34:1 were nearly equivalent (Fig. 2B). At 192 h, or stationary-phase growth (Fig. 2C), PI 34:1 was the single most abundant lipid detected in this strain and PI 18:1-18:0 had increased significantly. By the end of fermentation at 264 h (Fig. 2D), PI 34:1 remained the most abundant individual phospholipid detected and the concentration of PC 16:0-18:1 decreased, while remaining the most abundant phosphatidylcholine. Ergosterol levels remained relatively unchanged throughout fermentation in this strain, never exceeding 5 mol% (Fig. 2). Some of the same trends seen in ICV K1 (V-1116) were exhibited in the remaining 21 yeast strains. However, all strains had unique lipid profiles and changes in their lipid composition (see Fig. S1 to S21 in the supplemental material).

Fig 2.

(A to D) Major PL species and ergosterol profiles for yeast strain, ICV K1 (V-1116), at the following time points: exponential phase (48 h) (A), prestationary phase (120 h) (B), stationary phase (192 h) (C), and end stationary phase (264 h) (D). Missing bars represent lipids below the threshold of detection. The bars represent measurements from three separate fermentations (n = 3).

Correlation of lipid composition with fermentation ability.

The original hypothesis for this work was that lipid composition contributes to ethanol tolerance in S. cerevisiae in two ways. (i) Lipid components contribute to a stable cellular membrane that enhances yeast cell growth during fermentation, resulting in a higher maximum cell concentration. (ii) Specific lipid species minimize the perturbing effect of ethanol on yeast cell membranes, allowing increased maximum ethanol concentrations. To test this hypothesis, the lipid composition for all 22 yeast strains was quantitatively determined and correlated with the final ethanol concentration and maximum cell density (maximum OD₆₀₀).

Due to the dimensionality of the data, the multivariate statistical technique PLS linear regression was utilized to correlate the contributions of individual lipid species to the fermentation abilities of the yeast strains. Statistical selection of the lipid variables that contributed to the greatest amount of variation in the fermentation kinetic data was performed using iPLS regression analysis using the lipid composition of all 22 Saccharomyces strains over three fermentation time points, covering exponential growth through stationary-phase metabolism. The final ethanol concentration and maximum OD₆₀₀ were the response variables that composed the Y-block. The lipid composition data from three fermentation time points were the predictor variables used to construct the X-block. Next, PLS analysis indicated which variables in both the Y-block and the X-block contributed to the greatest amount of variation in the data and how the variables were correlated and then grouped the variables into a new latent variable (LV).

The first analysis that was performed was to determine how lipid composition at three time points contributed to the final ethanol concentration of the fermentations. The results from this analysis are listed in Table 3 and indicate that this model yielded six LVs capturing 71.68% of the variation in the lipid composition data and 90.78% of the variation in the final ethanol concentrations. A correlation coefficient of R² = 0.91 indicated a strong linear relationship between the measured final ethanol concentration versus that predicted according to the lipid composition of these yeast strains (Fig. 3A). Cross-validation (CV) was performed to assess how the model would perform on a new data set and CV correlation coefficients (Q²) were generated to indicate the predictive strength of the model to ensure that it has not been over-fit to the data (39). The RMSEC for the final ethanol concentration as a function of lipid composition was 0.29 weight percent ethanol, the RMSECV was 0.39 weight percent ethanol, and the Q² value for this model was 0.83, which indicated that lipid composition was a very good predictor of the final ethanol concentration achieved by these strains (Fig. 3A). Given that the final ethanol concentration (a likely indicator of ethanol tolerance) is likely to have two components, ethanol tolerance per cell and the maximum cell density formed, these results indicate that lipid composition is correlated with one or both of these components.

Table 3.

Results from partial least-squares regression analysis

Response variable (predictor variable)	No. of latent variables	RMSEC	RMSECV	R2	Q2	% variance captured
						X-block	Y-block
Final ethanol (maximum OD600)	1	0.55	0.57	0.66	0.63	100.00	65.75
Final ethanol (lipid composition)	6	0.29	0.39	0.91	0.83	71.68	90.78
Maximum OD600 (lipid composition)	6	0.18	0.27	0.97	0.94	67.51	97.41

Fig 3.

Results from partial least-squares regression modeling for all strains correlating lipid variables to the fermentation response variables final ethanol concentration (wt/wt) (A) and maximum OD₆₀₀ (B). Final ethanol concentration <9.0%, ▼; final ethanol concentration >9.0%, ●. The partial least-squares linear regression fit represents the measured versus predicted Y-block parameter. The linear regression equation for the model for panel A is Y = 0.91X + 0.84, and that for panel B is Y = 0.92X + 0.35, such that Y = Y _predicted and X = Y _measured.

Because we were able to measure maximum cell density directly, a PLS model was first generated to examine the correlation of the maximum OD₆₀₀ of the fermentations with the lipid composition of the yeast strains over the three time points from exponential growth through the stationary phase. The results listed in Table 3 indicate that this model also yielded six LVs capturing 67.51% of the variation in the lipid composition data and 97.41% of the variation in the yeast cell concentration data. The RMSEC for the maximum OD₆₀₀ as a function of lipid composition was 0.18 OD units, and the RMSECV was 0.27 OD units (Table 3). The R² value for this model was 0.97, and the Q² value was 0.94, indicating the yeast cell lipid composition was highly correlated with the maximum cell density achieved during alcoholic fermentation (Fig. 3B).

The next analysis performed was to determine how the final ethanol concentration correlated with maximum OD₆₀₀ (Table 3), i.e., what fraction of the ability to reach high ethanol concentrations was attributable to maximum cell concentration versus another component, such as ethanol tolerance per cell. Partial least-squares regression modeling of these two variables gave a coefficient of determination (R²) of 0.66 (R² value for simple linear regression analysis was identical). The results from a cross-validation of this model shown in Table 3 gave a Q² of 0.63 and RMSEC and RMSECV values of 0.55 and 0.57, respectively. Taken together, these data indicate a modest correlation between the final ethanol concentration and maximum cell growth during fermentation. Therefore, it is likely that the remaining level of correlation between final ethanol concentration and lipid composition is due to another mechanism such as ethanol tolerance per cell (e.g., specific lipids contributing to the membrane bilayers ability to mitigate the perturbing effects of ethanol), although a direct measurement of ethanol tolerance per cell is not yet available, and other explanations might be possible as discussed below.

Specific lipids associated with final ethanol concentration and increased biomass.

PLS and iPLS provide information about how specific lipid variables in the model correlate with the response variables, providing potential lipid candidates associated with ethanol tolerance and yeast cell density. Low-abundance phospholipids (e.g., those with 14- and 12-carbon fatty acid chains) were measured but excluded from the PLS analysis, since the amount of missing data artificially skewed the multivariate statistical analysis toward phospholipids containing these fatty acids. The sample scores plot from both PLS models for maximum OD₆₀₀ (Fig. 4A) and final ethanol concentration (Fig. 4B) indicated that two distinct subpopulations formed; strains that achieved a final ethanol concentration of <9.0% (wt/wt) (final °Brix > 2) and strains that achieved a final ethanol concentration greater than ∼9.0% (wt/wt) (final °Brix < 2). All of the yeast strains that had a maximum OD₆₀₀ of <4 and a final weight percent ethanol below 9.0% produced a subpopulation in the two left quadrants of Fig. 4A and B. The remaining strains were spread out over the two remaining quadrants and were not distributed according to either maximum OD₆₀₀ or final ethanol concentrations. The loadings plots shown in Fig. 5 indicated that the yeast strains that failed to produce ethanol levels above 9.0% were highly correlated with phosphatidylinositol, whereas strains with higher cell densities and final ethanol concentrations were highly correlated with phosphatidylcholine and phosphatidic acid. In the second LV, there did not appear to be any significant correlation of phosphatidylethanolamine and phosphatidylserine with any of the strains that completed fermentation. However, ergosterol appeared to be more correlated with the non-wine-yeast strains in the second LV (Fig. 4A and 5A).

Fig 4.

Sample scores for maximum OD₆₀₀ (A) and final ethanol concentration (wt/wt) (B) for all strains, indicating the relationship between yeast strains based upon similar correlations with specific lipid variables in the first two latent variables. Final ethanol concentration <9.0%, ▼; final ethanol concentration >9.0%, ●.

Fig 5.

Variable loadings for maximum OD₆₀₀ (A) and final ethanol concentration (wt/wt) (B) for all strains, indicating the relationship between yeast strains and lipid variables in the first two latent variables. Phosphatidic acid, ▼; phosphatidylethanolamine, ; phosphatidylinositol, ■; phosphatidylserine, ●; phosphatidylcholine, ◆; ergosterol, ▲.

Regression vector plots were generated (Fig. 6) to determine the statistical weights of the lipid predictor variables on the PLS models, i.e., the degree to which lipid variables were positively or negatively correlated with the response variable. Of the original 96 lipid variables, iPLS variable selection yielded 45 lipids in the maximum OD₆₀₀ model and 34 lipids in the final ethanol concentration model. Both models had a number of lipid variables in common, whereas a few lipids were determined to be unique to each model.

Fig 6.

Regression vector plot for maximum OD₆₀₀ (A) and final ethanol concentration (wt/wt) (B) for all strains, indicating each lipid's contribution (i.e., weight) to the PLS regression model through the first three fermentation time points. Positive values for the variable indicate a positive correlation between the predictor and response variables, whereas negative values indicate a negative correlation between the predictor and response variables.

At the 48 h time point, or exponential growth phase, the model correlating maximum OD₆₀₀ with lipid composition (Fig. 6A) indicated that strains exhibiting the highest concentrations of the phosphatidylinositol species PI 18:1-18:1, PI 16:1-18:1, and PI 32:1 during this time point were correlated with lower maximum cell densities. Figure 7 confirms that the phosphatidylinositol concentrations during this growth phase were highest among the strains that had the lowest maximum OD₆₀₀ values for yeast strains exhibiting the three highest and three lowest maximum cell densities in the present study. Yeast strains with higher concentrations of PC 18:0-18:1, PC 16:1-16:1, and ergosterol during this growth phase were negatively correlated with maximum OD₆₀₀ (Fig. 6A). By 192 h, or stationary-phase growth, a number of phosphatidylcholine species with long-chain unsaturated fatty acids were positively correlated with yeast cell growth, namely, PC 18:0-18:1, PC 18:1-18:1, PC 16:1-18:1, and PC 16:1-16:1. Ergosterol had a positive correlation with maximum yeast cell density during this growth phase, though much less so than other lipid species in the model.

Fig 7.

Levels of the three phosphatidylinositol species correlated with the maximum OD₆₀₀ and the final ethanol concentration during exponential growth for the six yeast strains representing the three highest (Enoferm T306, ICV K1 [V-1116], and Sake A18) and the three lowest (Prise de Mousse, CY3079, and DV10) maximum OD₆₀₀ values and the ethanol concentrations of all 22 fermentations. Note that the iPLS variable selection did not select for PI 32:1 in the final ethanol concentration model and that PI 34:1 was not selected for by iPLS in either model but is included as a reference. Each bar represents measurements from three separate fermentations (n = 3).

The model correlating the final weight percent ethanol concentration with lipid composition also indicated that during exponential growth higher concentrations of the phosphatidylinositol species PI 18:1-18:1 and PI 16:1-18:1 were negatively correlated with final ethanol concentrations; however, their effect was not as significant as other lipids in the model at this time point (Fig. 6B). Higher concentrations of the phosphatidylcholine species, PC 16:0-18:1, during exponential- and stationary-phase growth were positively correlated with higher ethanol production (Fig. 6B). In addition, the doubly unsaturated phosphatidylcholine, PC 18:1-18:1 was correlated with higher ethanol concentrations during stationary-phase growth.

In both models, increased concentrations of phosphatidylinositols during exponential growth (48 h) were negatively correlated with cell growth and ethanol production. Furthermore, higher levels of phosphatidylcholines with longer fatty acid chains and more degrees of unsaturation during stationary phase were positively correlated with increased biomass and higher ethanol concentrations. In general, both PC 18:1-18:1 and PC 16:0-18:1 were the most abundant phosphatidylcholine species in S. cerevisiae, whereas the remaining phosphatidylcholine species in the models varied depending on the strain, growth phase, and fermentation performance (see Fig. S1 to S21 in the supplemental material). Ergosterol was modestly correlated with yeast strains exhibiting high cell densities during fermentation, although less so than other lipids, and it was not correlated at all with ethanol production (Fig. 6). Interestingly, the phosphatidylserine species, PS 16:1-18:1, had a very strong negative correlation with maximum cell biomass and ethanol production during stationary-phase growth (Fig. 6).

DISCUSSION

Significant efforts have been made to understand the mechanisms by which S. cerevisiae can thrive in a high-ethanol environment (4, 10, 12, 13, 40). The lipid membrane bilayer has been shown to be a vulnerable target to the toxic effects of ethanol at both the fundamental level (17–20, 41) and the organismal level (42–47). In yeast, a number of lipid membrane components have been implicated in ethanol tolerance, including unsaturated fatty acids, ergosterol, and phospholipid content (42, 43, 47). However, due to differences in both the experimental design and the yeast strains used in these studies, it is difficult to ascertain which membrane components were most critical to a successful fermentation. To this end, fermentations were carried out with 22 commercial S. cerevisiae strains with diverse ethanol tolerance and growth characteristics under the same experimental conditions in order to analyze the major lipid species and yeast strains that correlated with a successful fermentation.

The results from the analysis of the major phospholipids and ergosterol in yeast lipid extracts indicated that the lipid compositional profiles of S. cerevisiae were somewhat independent of time over the course of fermentation. In the example of ICV K1 (V-1116), a number of phospholipids increased or decreased in concentration over the course of fermentation; however, the relative lipid concentrations within each lipid class did not experience dramatic shifts over the course of fermentation. The most significant change in the lipid composition of this strain was the transition from the single most abundant phospholipid being PC 16:0-18:1 to PI 34:1 between the exponential and stationary phases. The transition in the maximum relative abundance of phosphatidylcholine to phosphatidylinositol lipid classes has previously been observed in fermenting yeast (43) and is likely due to nitrogen or other nutrient depletion since ethanol concentrations began to rise and the cells transition to stationary-phase metabolism (48–50). The ergosterol content of ICV K1 (V-1116) never exceeded 5 mol% and was relatively consistent throughout fermentation. It has previously been reported that elevated ergosterol levels are associated with ethanol tolerance in yeast (42, 43, 51); however, studies involving wine yeast have reported that elevated levels of ergosterol were not critical to the ability to tolerate increasing levels of self-produced ethanol (45, 46, 52), an observation that was corroborated with this strain and others studied here.

The results from PLS linear regression modeling confirm that the final weight percent ethanol was a strong function of lipid composition. Unexpectedly, this was likely more because of the strong correlation between ethanol tolerance and maximum yeast cell concentration during fermentation than because of a strong correlation of ethanol tolerance with ethanol tolerance per cell. Previous studies have demonstrated that nitrogen and other nutrient levels early in fermentation are extremely important factors in determining yeast cell growth and a successful fermentation outcome (6–8). The availability of specific nutrients (e.g., nitrogen, zinc, myo-inositol, etc.) during exponential growth alters the transcriptional regulation of genes involved with phospholipid biosynthesis in yeast (9, 48, 49, 53). Indeed, upon entering stationary-phase metabolism, S. cerevisiae experiences a decrease in total phospholipid biosynthesis (50). Therefore, the strong correlation of cell growth with lipid composition indicates that lipid biosynthesis in yeast may be intrinsically related to factors affecting yeast cell proliferation during fermentation by genetic mechanisms that have evolved to optimize adaptation to extreme environmental conditions.

However, by comparing the correlations of final ethanol concentration with lipid composition (R² = 0.91) and final ethanol concentration with maximum OD₆₀₀ (R² = 0.66), it seems likely that lipid composition is having some effect on ethanol tolerance that is not related to maximum cell growth. One possibility is that specific yeast membrane lipids may minimize the perturbing effects of ethanol and a potential mechanism known as lipid interdigitation, where the fatty acyl chains of the phospholipids cross the bilayer mid-plane resulting in an up to 25% reduction in bilayer thickness (20, 41). Thinning of yeast cell membranes due to ethanol could lead to disruption of membrane-associated protein function, aggregation of membrane-bound proteins, and compromised membrane integrity (21, 22), which could result in yeast cellular inactivation during fermentation (6). Studies utilizing model membrane systems composed of phosphatidylcholine and ergosterol and subsequently exposed to ethanol concentrations commonly observed in wine fermentations have demonstrated that lipid composition provides a protective effect on the membrane by delaying the onset of the interdigitated phase (18, 20). Furthermore, it was recently reported by Vanegas et al. (19) that increased levels of long-chain unsaturated phosphatidylcholines in ternary lipid bilayer systems mitigated the membrane-thinning effects of ethanol. The results presented here indicate that S. cerevisiae strains with more robust fermentation characteristics (i.e., higher final ethanol concentrations and/or maximum OD₆₀₀) were highly correlated with these types of lipids. Since ethanol tolerance per cell cannot yet be directly measured, this potential mechanism can only be inferred from studies with purified lipid bilayers and other mechanisms, such as nutrient limitation/availability, substrate inhibition, osmotolerance, and/or pH, could be suggested.

The sample scores plot from both PLS models indicated that two subpopulations of yeast strains formed based on similarities in lipid composition could be differentiated based on final ethanol concentrations. According to the variable loadings, nearly all of the phosphatidylinositol species were highly correlated with the six yeast strains that failed to achieve a final ethanol concentration of >9.0% (wt/wt) (Fig. 5). Most of the phosphatidylcholine, phosphatidylserine, and phosphatidic acid species correlated with yeast strains that had completed fermentation. The remaining phospholipid classes were not observed to have as strong an association with any individual group as phosphatidylinositol, with the exception of ergosterol in the maximum OD₆₀₀ model, which was more correlated with the non-wine-yeast strains.

Chi and Arneborg (45) reported that low phosphatidylinositol levels in yeast were associated with growth media lacking myo-inositol, while inclusion of this cyclohexanehexol resulted in yeast exhibiting higher content of phosphatidylinositol during stationary-phase metabolism and increased ethanol tolerance. Furthermore, in S. cerevisiae, it has been shown that increased phosphatidylinositol levels are associated with yeast entering stationary-phase growth or the growth medium becoming depleted of nitrogen, zinc, or other nutrients (48, 49, 53). These results indicate that the five strains that exhibited lower cell densities and ethanol production may have experienced a problem with nutrient utilization early on in fermentation. Alternatively, phosphatidylinositol is the precursor to the phosphoinositides, which are involved in a multitude of yeast cell signaling and regulatory networks (54, 55). The high levels of phosphatidylinositol associated with these five yeast strains may indicate a problem with phosphoinositide turnover or synthesis that subsequently led to a breakdown in cellular metabolism, resulting in limited cell growth and slow sugar utilization. It could also be that these strains were not well adapted to growth conditions used here, which led to slower metabolic rates compared to other strains. This seems unlikely, though, since the conditions used in the present study mimicked wine fermentations and five of the six strains are commonly used to ferment grape must. Regardless of the reasons for the stuck fermentations experienced by these strains, it is clear that these problem fermentations were strongly correlated with higher relative phosphatidylinositol concentrations early in fermentation.

The regression vector plots from both final ethanol concentration and maximum OD₆₀₀ indicated a number of lipids in common between the models and the degree to which the lipids species had influenced the models. Both models demonstrated that higher relative concentrations of phosphatidylinositol species during exponential growth were negatively correlated with increased yeast cell density and final ethanol levels. Furthermore, both models showed that strains with higher levels of longer fatty acyl chains and unsaturated fatty acids, particularly phosphatidylcholines, late in fermentation were correlated with increased biomass and final ethanol concentrations. This correlation is likely due to increasing levels of ethanol during stationary-phase metabolism (6), where elevated levels of unsaturated fatty acids and increased fatty acid chain length in phospholipids reduce the fluidizing effect of ethanol (47, 56, 57).

In model lipid bilayers composed of phosphatidylcholine and sterols, ergosterol has been shown to have a protective effect against the membrane perturbing effects of ethanol (18, 20). In yeast, a number of studies investigating how lipid composition contributes to alcohol tolerance have concluded that elevated levels of ergosterol are important in mitigating the toxic effects of ethanol (42, 43, 51). However, other studies investigating wine fermentation yeast strains have found that ergosterol levels do not significantly contribute to ethanol tolerance (45, 46, 52). A possible explanation for this disparity could be that ergosterol levels are already significantly higher in S. cerevisiae compared to other yeast species (42). The results presented here indicate that ergosterol has a modest positive correlation with cell density and no correlation with the ethanol concentration.

The 22 yeast strains examined here were chosen based upon their ability to tolerate ethanol and their favorable fermentation characteristics to compare and contrast differences in their lipid composition over the course of anaerobic fermentation. Utilizing API-MS, it was possible to rapidly and quantitatively analyze lipid extracts from this diverse set of yeast strains over multiple fermentation time points, which provided unique insight into the molecular lipid profiles of these strains and how they varied over time and between strains. Partial least-squares regression analysis revealed the correlation of the lipid composition with the fermentation kinetic parameters of cell density and ethanol production in S. cerevisiae, as well as the lipid molecules that potentially contribute to this organism's ability to remain viable in a high ethanol milieu. Although there are a number of possible explanations for why yeast experience fermentation arrest, the results presented here indicate that among the lipids that were highly correlated with yeast cell growth and ethanol production, increased phosphatidylinositol concentrations early in fermentation were associated with strains that failed to complete fermentation. In addition, lipid species known to protect model membrane bilayers from the perturbing effect of ethanol were positively correlated with higher final ethanol concentrations and biomass during fermentation. The correlations observed here between lipid composition and ethanol production should help to direct further studies in elucidating the causative nature of this relationship.