- Email: [email protected]
Dynamic changes in brewing yeast cells in culture revealed by statistical analyses of yeast morphological data
Journal of Bioscience and Bioengineering VOL. xx No. xx, 1e7, 2013 www.elsevier.com/locate/jbiosc
Dynamic changes in brewing yeast cells in culture revealed by statistical analyses of yeast morphological data Shinsuke Ohnuki,1 Kenichi Enomoto,2 Hiroyuki Yoshimoto,2 and Yoshikazu Ohya1, * Department of Integrated Bioscience, Graduate School of Frontier Sciences, University of Tokyo, Bldg. FSB-101, 5-1-5 Kashiwanoha, Kashiwa, Chiba 277-8562, Japan1 and Research Laboratories for Brewing, Kirin Brewery Company, Limited, 17-1 Namamugi 1-chome, Tsurumi-ku, Yokohama, Kanagawa 230-8628, Japan2 Received 15 June 2013; accepted 13 August 2013 Available online xxx
The vitality of brewing yeasts has been used to monitor their physiological state during fermentation. To investigate the fermentation process, we used the image processing software, CalMorph, which generates morphological data on yeast mother cells and bud shape, nuclear shape and location, and actin distribution. We found that 248 parameters changed significantly during fermentation. Successive use of principal component analysis (PCA) revealed several important features of yeast, providing insight into the dynamic changes in the yeast population. First, PCA indicated that much of the observed variability in the experiment was summarized in just two components: a change with a peak and a change over time. Second, PCA indicated the independent and important morphological features responsible for dynamic changes: budding ratio, nucleus position, neck position, and actin organization. Thus, the large amount of data provided by imaging analysis can be used to monitor the fermentation processes involved in beer and bioethanol production. Ó 2013, The Society for Biotechnology, Japan. All rights reserved. [Key words: Saccharomyces pastorianus; Bottom-fermenting yeast; CalMorph; Cell morphology; Fermentation; Principal component analysis; Budding profile]
Controlling propagation and fermentation in cellars is an important element in beer fermentation. After raw materials, including malted barley, hops, cereals, adjunct, and water, are converted into wort, brewing yeasts grow and produce ethanol. During fermentation, the production of a well-balanced aroma and the flavor of the final product are at least as important as efficient fermentation and high yield. Quality assurance management, especially of brewing yeasts, is important to maintain their good physiological condition (1). Several features of brewing yeasts have been used to characterize their physiological states during fermentation. Viability is the traditional method (2). Commonly used viability tests are based on the bright-field stains methylene blue (3) and methylene violet (4) and the fluorescent dye 1-anilino-8-naphthalenesulphonic acid (5). The ability to exclude the dye is dependent on cell viability, so any dead cells are stained. Viability provides information regarding the live population in culture but not about individual living yeast cells. In contrast, vitality reflects the physiological condition of individual cells during yeast proliferation. Several methods have developed to measure yeast vitality, such as detection of intracellular pH (ICP) (6), measurement of specific oxygen uptake rate (7,8),
* Corresponding author. Tel.: þ81 4 7136 3650; fax: þ81 4 7136 3651. E-mail address: [email protected] (Y. Ohya). Abbreviations: DAPI, 40 ,6-diamidino-2-phenylindole; DNA, deoxyribonucleic acid; FDR, false discovery rate; FITC-Con A, fluorescein isothiocyanate-Con A; KW test, KruskaleWallis test; PCA, principal component analysis; PCs, principal components; Rh-ph, rhodamine-phalloidin.
the acidification power test (9,10), carbon dioxide production (11), vicinal diketone reduction (12), glycogen and trehalose staining (13), and the budding ratio (14). Of the features analyzed, budding has become of great interest, because the budding ratio correlates with metabolism of polysaccharides and production of bioethanol and other metabolites, such as aroma and flavor compounds (15). Flow cytometry allows fine measurements, including parameters related to the cell cycle (16). More recently, methods combining these three features, viability, cell concentration, and budding ratio, have been developed to simultaneously monitor viability and vitality (17). To assess quantitatively the morphological features of yeast, we developed an automatic image processing system, CalMorph (18,19). CalMorph directly processes fluorescence micrographs and generates 501-dimensional quantitative data regarding mother cells and bud shape, nuclear shape and location, and actin distribution. Using CalMorph, we can easily, rapidly, and reproducibly generate various quantitative data (20). The program generates reproducible data consistent with those obtained manually. CalMorph can also monitor yeast morphological changes that accompany the budding cycle (20), such as the bud index, and specific morphological features in G1, S/G2 and M cells. The purpose of this study was to investigate the dynamics of brewing yeast during fermentation. Because the budding profile is a good indicator of cell vitality, we tried to use CalMorph to monitor the physiological state of brewing yeasts during fermentation. However, the hundreds of parameters generated by CalMorph distracted from the objective. Specific parameters could be selected
1389-1723/$ e see front matter Ó 2013, The Society for Biotechnology, Japan. All rights reserved. http://dx.doi.org/10.1016/j.jbiosc.2013.08.005
Please cite this article in press as: Ohnuki, S., et al., Dynamic changes in brewing yeast cells in culture revealed by statistical analyses of yeast morphological data, J. Biosci. Bioeng., (2013), http://dx.doi.org/10.1016/j.jbiosc.2013.08.005
2
OHNUKI ET AL.
J. BIOSCI. BIOENG.,
in advance based on biological meaning, but this would be arbitrary. The selection of pre-existing parameters was also problematic, because not all informative features were always used. As an alternative approach, we performed successive principal component analyses to extract important time-dependent morphological features. Our results suggest that statistical analyses of morphological data can facilitate yeast management during fermentation. MATERIALS AND METHODS Strains, culture conditions, and sample preparation The yeast strain used was the bottom-fermenting yeast Saccharomyces pastorianus KBY011 (21). KBY011 was pre-cultured in YPD10 medium containing 1% yeast extract (Becton, Dickinson and Co., USA), 2% peptone (Becton, Dickinson and Co.), and 10% glucose (Nacalai Tesquie, Japan) with shaking at 20 C for 3 days. Harvested cells were diluted to an optical density of OD600 ¼ 0.5, and then cultured at 20 C for 96 h in 500 mL of YPD10 medium with gentle stirring with a magnetic stir-bar under anaerobic conditions. Samples were collected at 0, 24, 48, 72, and 96 h, and cells harvested by centrifugation. Apparent extracts were analyzed according to a previous reference (22). Image acquisition and CalMorph analysis Fixation and staining of yeast cells, image acquisition, and CalMorph analysis were performed according to the CalMorph manual (http://scmd.gi.k.u-tokyo.ac.jp/datamine/calmorph/CalMorphmanual.pdf). Briefly, cells (8 106/mL) were fixed in 0.1 M potassium phosphate buffer (pH 6.5) containing 3.7% formaldehyde (Wako Pure Chemical Industries, Japan). To obtain fluorescence images of the cell-surface mannoprotein, actin cytoskeleton, and nuclear DNA, cells were stained with fluorescein isothiocyanateCon A (FITC-Con A, Sigma, USA), rhodamine-phalloidin (Rh-ph, Invitrogen Corp, USA) and 40 ,6-diamidino-2-phenylindole (DAPI, Sigma), respectively. CalMorph automatically characterizes each yeast cell by calculating 501 morphological parameters based on data from more than 200 cells. In total, five independent cultures grown under the same condition were analyzed. Successive principal component analysis To evaluate time-dependent morphological changes, we performed successive principal component analyses (PCA; Fig. 1). First, the 501 morphological parameters were screened to yield 248 parameters that change considerably during fermentation (Fig. 1A). Second, these 248 traits were subjected to PCA to extract the principal components (PCs) explaining the changes (Fig. 1B). Finally, PCA was again applied to the parameters correlating highly with each PC to extract independent parameters (Fig. 1C). All
501 parameters A) Kruskal-Wallis test 248 parameters B) First PCA 57 parameters PC1: 50 parameters
PC2: 7 parameters
C) Second PCA PC1a: 9 parameters
PC2a: 5 parameters
PC1b: 6 parameters
PC2b: 2 parameters
PC1c: 3 parameters
FIG. 1. Schematic flow chart for the successive principal component analysis. Down arrows indicate each step in application of the statistical test or PCA (AeC). The 501 parameters were systematically filtered into semantic fractions by each step. (A) KruskaleWallis test was used to select parameters that changed considerably during fermentation and storage. (B) PCA was applied to identify the patterns of the morphological variation. (C) PCA was applied to extract independent morphological features.
statistical analyses were performed using the ‘R’ software (http://www.r-project. org/). In Fig. 1A, the KruskaleWallis (KW) test (23) was applied to each parameter to identify large changes. For each parameter, the data consisted by five replications of five time points (25 samples). The false discovery rate (FDR) corresponding to each P value of the KW test was estimated by an empirical permutation test of 1000 iterations (24). In Fig. 1B, PCA was applied to the time-course data of the significantly changed 248 parameters (KW test, FDR ¼ 0.05), after five sets of five replicated sample values were combined across time points, ranked among the combined samples, and summed into one rank-sum value for each time point, as described previously (25). To standardize the rank-sum values among the parameters, the rank-sum values at 0 h were subtracted from the rank-sum values of the other time points. Then, eigenvalues (the variance) and eigenvectors (the rotation) of the rank-sum values were calculated from the covariance matrix for the 248 parameters, the contribution ratio was calculated as the ratio of variance, and the PC score was computed by matrix multiplication between the rank-sum values and the rotation. The PC loadings were computed by multiplication of the rotation (eigenvectors) by the square root of the eigenvalue and dividing by the square root of the variance of the ranksum. P-values for the loadings were computed using a t distribution, where t process was the same as that used to transform a Pearson’s productemoment correlation coefficient into a t value (26). In Fig. 1C, to determine morphological features accompanying PC1 of the first PCA, we selected 50 parameters showing significantly high absolute loadings in PC1 (FDR ¼ 0.15), and a second PCA was performed for the parameters selected using morphological data from 122 replicated wild-types as a null distribution. The 50 parameter values of the 122 replicated wild-type morphological data sets were transformed to a normal distribution using the BoxeCox power transformation, as described previously (26). The eigenvalues (the variance) and the eigenvectors (the rotation) of the 122 transformed wild-type data sets were calculated using the covariance matrix of the 50 parameters. The contribution ratio, PC scores, and loadings were calculated as for the first PCA. The PCs of the second PCA were named in alphabetical order (e.g., PC1, PC2, and PC3 were named PC1a, PC1b, and PC1c, respectively). For PC2, we selected seven parameters; the second PCA was performed in a similar manner.
RESULTS Time-dependent morphological changes in fermentation Bottom-fermenting yeast cells were cultured at 20 C with gentle stirring under anaerobic conditions. The changes in cell numbers and apparent extract are shown in Fig. S1. We quantified fermenting yeast cell morphology using 501 morphological parameters. We used image-processing software, CalMorph, after obtaining cell wall, actin, and nuclear DNA images (19). Fermenting yeast cells were sampled at 0, 24, 48, 72, and 96 h, fixed, stained with the fluorescent dyes, FITC-Con A, DAPI and Rh-ph, photographed (at least 200 cells), and quantified using CalMorph. The experiments were replicated five times independently for each time point. As shown in the heat map (Fig. 2), time-dependent changes in the morphological parameters differed in pattern. Of the 501 parameters, 248, 127, and 95 of the 501 differed significantly among the five time points with false discovery rates (FDRs) of 0.05, 0.01, and 0.005, respectively (Fig. 1A). Key variables in a high-dimensional morphological data set during fermentation PCA is an exploratory multivariate statistical technique for simplifying complex data sets (27). It has been used for analysis of time-dependent changes in gene expression data (28) and dose-dependent changes in morphology data (26). To summarize the morphological dynamics during fermentation, we applied PCA to the morphology data using the morphological measurements as the variables and the different time points as the observations. The data set contained values for 248 significantly different parameters (FDR ¼ 0.05) collected with five replicates at 0, 24, 48, 72, and 96 h during fermentation. Thus, the matrix to be analyzed had 25 rows of conditions and 248 columns of parameters. Five sets of five replicate sample values were combined across time points, ranked among the combined samples, summed at each time point, and used for the first PCA. Our analysis indicated that we could summarize the data using two variables. The cumulative contri
Please cite this article in press as: Ohnuki, S., et al., Dynamic changes in brewing yeast cells in culture revealed by statistical analyses of yeast morphological data, J. Biosci. Bioeng., (2013), http://dx.doi.org/10.1016/j.jbiosc.2013.08.005
VOL. xx, 2013
MORPHOLOGICAL ANALYSIS OF BREWING YEAST CELLS
3
The first component represented transient changes in morphology at 24 h. In Fig. 3A, the PCA score shows a peak at 24 h. This component was positive for a parameter that increased early and decreased later. In contrast, absolute loading values of the first component for a parameter whose relative values changed over time would be low. The second component represented morphological change over time. In Fig. 3B, the PCA score increased with time, from zero to positive values. This component was positive for parameters whose relative values increased with time, and negative for those whose relative values decreased with time. The third component measured concavity: note the parabolic nature of the scores in Fig. 3C. Morphological parameters significantly correlated with PC1 and PC2 To further understand the implications of the dynamic changes, we analyzed the parameters that correlated with PC1 (transient change with a peak) and PC2 (change over time). Of the 248 parameters, significant PC loadings in PC1 (FDR ¼ 0.20, 0.15, and 0.10) were observed for 75, 50, and 5 parameters, respectively. Accordingly, we focused on 50 parameters (FDR ¼ 0.15) with significant PC loadings for PC1 (Fig. 1B). These parameters are predicted to show transient changes, with a peak at 24 h. Seven parameters (FDR ¼ 0.15) with significant PC loadings for PC2 (Fig. 1B) likely represent changes over time. Extraction of independent features with a successive PCA Some morphological parameters are semantically related. To extract the independent morphological features of the PC1 (Fig. 1C), we applied a second PCA to a null-distributed data set of the 50 parameters, as described by Ohnuki et al. (26). The uniformly scattered data were a normalized data set obtained from a 122-replicate wild-type morphological data set used for extracting independent morphological features (26). By PCA for the 50 parameters, the first three PCs (PC1a, PC1b, and PC1c) explained more than 40% of the variance among PC scores of a null-distributed data set (Fig. S3). Regarding the PC scores projected from the rank-sum values of the time-course data, they explained w60% (Fig. S3). Examination of six parameters with significant PC loadings for PC1b (Fig. S4) showed that the PC1b represented a ratio of budded cells (Fig. 4). The PC1a and PC1c (Fig. S4) indicated the distance between two nuclei and the neck angle, respectively (Fig. 4). Based on this second PCA analysis, we identified three independent features: budding ratio, nuclear position, and bud angle. Independent morphological features of PC2 (PC2a and PC2b) were also extracted by a second PCA on a null-distributed data set of the seven parameters with significant PC loadings on PC2
FIG. 2. Heat-map of the time-course changes in cell morphology in the stored yeast cells. Numbers on the heat-map indicate the time-course (h) from the start of fermentation. The dendrogram at the right of the heat-map was generated by the average linkage based on the similarity calculated by the correlation coefficient between standardized rank-sum values of the parameters (see Materials and methods). Indices at the right of the heat-map indicate the detected parameters at FDR ¼ 0.05 (using the KruskaleWallis test). KW on the indices indicates the KruskaleWallis test.
bution ratio of PCA in the time-course data (Fig. S2) indicated that two principal components accounted for over 85% of the total variability; inclusion of the third component accounted for over 90%. The meanings of these components were distilled from their respective PCA scores.
200 400 600 800
B
C
PC 1
PC 2
PC 3
−200
0
PC score
A
0
20
40
60
80
100 0
20
40
60
80
100 0
20
40
60
80
100
Time (h) FIG. 3. Time-course changes of the principal component scores. Data from five replicate experiments were standardized by a rank-sum method, and PC scores were calculated as described in Materials and methods.
Please cite this article in press as: Ohnuki, S., et al., Dynamic changes in brewing yeast cells in culture revealed by statistical analyses of yeast morphological data, J. Biosci. Bioeng., (2013), http://dx.doi.org/10.1016/j.jbiosc.2013.08.005
OHNUKI ET AL.
J. BIOSCI. BIOENG.,
PC2b
PC2
Increase in ratio of unbudded cells with polarized actin patches
(Fig. 1C). The first two PCs (PC2a and PC2b) explained more than 90% of the variance among the PC scores, regardless of the data set used (Fig. S5). The PC2a and PC2b (Fig. S4) represented the number of actin patches and a ratio of unbudded cells with polarized actin patches, respectively (Fig. 4).
[D108_C] Distance between nuclear gravity center in mother and middle point of neck (µm)
C
D
1.0
[A122_A] Number of bright actin pathces
Identification and validation of representative parameters To extract representative parameters, parameters with significant PC loadings were plotted against the first two principal components (Fig. 5). The D108_C parameter (distance between nuclear gravity center in mother and middle point of neck) with significant PC1a loading (Fig. S4), in fact, had positive PC1 loadings and nearly zero PC2 loadings. As might be expected,
B
[D210] Nuclear A ratio to nuclear AA1BC cells (cells/cells)
FIG. 4. Independent features of morphological changes during fermentation. Representative shapes of the yeast cells were illustrated. Large circles in the cells of PC1a, PC2a and PC2b indicate nuclei. Small circles in the cells of PC2a and PC2b indicate the actin patches. Lines in the cell at PC1c indicate the neck angle to the long axis. Broad arrows at left side and right side of PCs indicate decreases and increases in the scores of the PCs. Text on the right side of the figure is a description of the independent features when the scores increased.
D108_C
0.3
[A106] Actin b ratio (cells/cells)
0.0
E
−0.5
PC2 (loading)
0.5
A106
C105_C D210
1.6
Decrease in number of actin patches
1.4
PC2
1.2
PC2a
A
1.0
Decrease in neck angle
0.8
PC1
0.6
PC1c
45
Increase of budded cells
35
PC1
25
PC1b
the time-dependent change in D108_C (Fig. 6A) showed a transient change with a peak at 24 h, similar to the change in PC1. In contrast, the changes in the D210 parameter (unbudded cell ratio, correlated with PC1b) and C105_C (neck position, anticorrelated with PC1c) with negative PC1 loadings and nearly zero
0.16
Increase in distance between two nuclei
0.08
PC1
0.5
PC1a
[C105_C] Neck position (degree)
4
−1.0
A122_A
0
24
48
72
96
Time (h) −1.0
−0.5
0.0 PC1 (loading)
0.5
1.0
FIG. 5. Distribution of principal component loadings in PC1 and PC2. Two-hundred and forty-eight parameters identified by the KruskaleWallis test at FDR ¼ 0.05 are plotted. Circles indicate 57 parameters with significant loadings in at least one of the two PCs at FDR ¼ 0.15 (t-test). Filled circles with the parameter names indicate the representative parameters for each PC.
FIG. 6. Time-course changes in representative parameter values. Solid lines on the boxplots indicate the line chart of the mean value in each time point. (A) Time-course changes in the distance between nuclear gravity center in the mother cell and middle point of the neck (D108_C), exemplifying PC1a. (B) Time-course change of the ratio of unbudded cells (D210), exemplifying PC1b. (C) Time-course change of the neck position (C105_C), exemplifying PC1c. (D) Time-course change of the number of bright actin patches (A122_A), exemplifying PC2a. (E) Time-course change of the ratio of unbudded cells with polarized actin patches (A106), exemplifying PC2b.
Please cite this article in press as: Ohnuki, S., et al., Dynamic changes in brewing yeast cells in culture revealed by statistical analyses of yeast morphological data, J. Biosci. Bioeng., (2013), http://dx.doi.org/10.1016/j.jbiosc.2013.08.005
VOL. xx, 2013
MORPHOLOGICAL ANALYSIS OF BREWING YEAST CELLS
PC2 loadings (Fig. 5) also showed a transient change with a peak at 24 h, but the reverse of the change in PC1 (Fig. 6B, C). The changes in the A122_A parameter (number of bright actin patches in the unbudded cells, correlated with PC2a) with negative PC2 loadings and nearly zero PC1 loadings (Fig. 5) decreased over time (Fig. 6D), whereas the change in A106 (ratio of unbudded cells with polarized actin, anti-correlated with PC2b) with positive PC2 loadings and nearly zero PC1 loadings (Fig. 5) increased over time (Fig. 6E). The results indicate that we can use these representative parameters to monitor dynamic aspects of yeast during fermentation. The unbudded yeast cells start budding, nuclear division and bud elongation occur within 24 h, and the vegetative growth stops after incubation for 48 h (Fig. 7). The ratio of unbudded cells with polarized actin increased and the number of actin patches decreased over time (Fig. 7). DISCUSSION Monitoring the vitality of yeast cells is of great importance for the brewing industry, because such information is necessary to determine the optimal fermentation conditions. Budding and cell cycle profiles are good indicators of cell viability and vitality: highly proliferating cells have a high budding ratio and mitotic index, and fermentation conditions often alter cell polarity. In this study, statistical analyses of images of yeast cells were used to provide information regarding budding profiles during fermentation. We used the image-processing software, CalMorph, which determines 501 morphological features after fluorescence staining of the cell wall, actin, and DNA. We performed PCA on morphological data to simplify the analysis and visualize multidimensional data sets. This analysis allowed us to obtain important information on a yeast culture during fermentation. Gene expression data and morphological data Because high-dimensional data on yeast cell morphology are similar in form to gene expression data composed of expression ratios for 6000 genes from Saccharomyces cerevisiae, an analogous statistical methodology can be used. Gene expression patterns and morphological similarities have been used previously in various inference tasks. For example, these data have been used to identify gene clusters based on the profiles (24,25,29,30), to define metrics that measure involvement of genes in particular cellular functions (19,29), and to predict intracellular drug targets (31e34). Eliminating noisy components without discarding important information is also a common problem. PCA (28) and functional PCA (35) have been applied to time series of expression data. Likewise, we showed here that PCA can determine a reduced set of morphological variables that are useful for understanding morphological changes. Multidimensional morphological data also have some different properties from gene expression data. Unlike expression data, morphological data contain hidden dependencies among observations. For this reason, we extracted the independent morphological
M S/G2
0h
5
features by means of a second PCA. Another difference is that morphological parameters can be considered in PCA as observations or variables. In this study, we treated morphological parameters and time points as variables and observations, respectively, in our first PCA, because our main purpose was to obtain as much information as possible about dynamic properties. It is possible to treat morphological parameters as observations, as in the application of PCA to time series of expression data, as described by Raychaudhuri et al. (28). However, in this case, variation of the endpoint rather than change at a peak is more weighted. Systematic extraction of important cellular landmarks from the high-dimensional data We effectively extracted important cellular features from 501 parameters by successive PCA, as described previously (26,34). The first key point of this method is to summarize two types of time-course change by applying PCA to the morphological data at different time points. The second key point is to extract independent parameters by applying PCA to a nulldistributed data set. Together with the selection of highly affected parameters in advance, this novel method enabled us to unravel the complex high-dimensional data into the independent and important cellular features (26). Although high-dimensional data are useful to describe various aspects of cellular activities (36,37), identifying primary features can be problematic because of the complexity of the data. One approach is hypothesis-driven (38e40), in which accumulated knowledge in the field and/or some creative insights are required. In successive PCA, the important features are extracted systematically in a data-driven manner (26). Because successive PCA is a general-purpose approach, it can be applied to any data set in a high-dimensional data analysis of cellular traits under several experimental conditions; e.g., metabolome database analyses under several conditions. Morphological features used for yeast management The vitality of brewing yeast cells, which has been monitored for yeast management, is defined as the capacity of yeasts to initiate metabolism rapidly after transfer from a nutrient-poor to a nutrient-rich environment (1,41) and the physiological state of the viable cell population (42). Accordingly, the bud index has been used as an indicator of vitality (14,16,17). It can be used to estimate the reproduction rate of yeast populations, which directly correlates with metabolism of polysaccharides and bioethanol production (17). One of the important morphological features that we found in this study was the bud index, indicating that we could extract a vital parameter by successive PCA. It should be noted that other morphological parameters behaved similarly to the bud index. Two representative parameters, nuclear position and bud angle, as well as other parameters with high absolute loading values on PC1 may also reflect the vital state of yeast cells. Thus these could be used for monitoring the vitality of brewing yeasts. Regarding the parameters of actin patches and actin regions with high absolute loading values to PC2, these are
G1
24 h
96 h
FIG. 7. Morphological changes during fermentation. Large and small circles in the yeast cell indicate nucleus and actin patch, respectively. Pie charts show the ratio of cells at the G1, S/G2, and M phases of the cell cycle. Arrows indicate time interval (24 h per arrow).
Please cite this article in press as: Ohnuki, S., et al., Dynamic changes in brewing yeast cells in culture revealed by statistical analyses of yeast morphological data, J. Biosci. Bioeng., (2013), http://dx.doi.org/10.1016/j.jbiosc.2013.08.005
6
OHNUKI ET AL.
not indicators of vitality, but rather are correlated with incubation time. The morphology of yeast cells changes in culture and is thus a powerful tool for assessing the physiological state of yeast cells and the fermentation conditions. There are several advantages to using a large number of morphological parameters. First, we are able to focus on several vital parameters simultaneously to improve the accuracy of the monitoring system. A similar approach was used previously, focusing on three features: viability, cell concentration and budding ratio (17). Second, we are able to analyze correlations among the changes in morphological parameters under different fermentation conditions to estimate physiological states. If all of the morphological features behave similarly, this would suggest that the physiological state changed similarly. Third, we are able to target efforts toward the vital morphological parameters to improve fermentation conditions. Every effort should be made to achieve a high budding ratio during fermentation or to understand the reason for its low value. Finally, by collecting data under different culture and storage conditions, this system could be expanded to create a yeast management database. Imaging and other high-dimensional information could be integrated to contribute to yeast management. Further study is necessary to develop tools and methods to analyze and estimate better conditions for beer fermentation based on such a database. Supplementary data related to this article can be found online at http://dx.doi.org/10.1016/j.jbiosc.2013.08.005. ACKNOWLEDGMENTS We thank Kyoko Horikoshi, Satoru Nogami, Tomohiko Ichii for their cooperation, and members of the Laboratory of Signal Transduction for fruitful discussions. This work was supported by Grants-in-Aid for Scientific Research from the Ministry of Education, Culture, Sports, Science and Technology, Japan (21310127 and 24370002 to Y.O.). S.O. was a Research Fellow of the Japan Society for the Promotion of Science. References 1. Lodolo, E. J., Kock, J. L., Axcell, B. C., and Brooks, M.: The yeast Saccharomyces cerevisiae e the main character in beer brewing, FEMS Yeast Res., 8, 1018e1036 (2008). 2. Bendiak, D.: Review of metabolic activity tests and their ability to predict fermentation performance, pp. 34e45, in: Smart, K. (Ed.), Brewing yeast fermentation performance. Blackwell Science, Oxford, UK (2000). 3. Chilver, M. J., Harrison, J., and Webb, T. J. B.: Use of immunofluorescence and viability stains in quality control, J. Am. Soc. Brew. Chem., 36, 13e18 (1978). 4. Smart, K. A., Chambers, K. M., Lambert, I., and Jenkins, C.: Use of methylene violet staining procedures to determine yeast viability and vitality, J. Am. Soc. Brew. Chem., 57, 18e23 (1999). 5. McCaig, R.: Evaluation of the fluorescent dye 1-anilino-8-naphthalene sulphonic acid for yeast viability determination, J. Am. Soc. Brew. Chem., 48, 22e25 (1990). 6. Imai, T. and Ohno, T.: The relationship between viability and intracellular pH in the yeast Saccharomyces cerevisiae, Appl. Environ. Microbiol., 61, 3604e3608 (1995). 7. Wheatcroft, R., Lim., Y. H., Hawthorne, D. B., Clarke, B. J., and Kavanagh, T. E.: An assessment of the use of specific oxygen uptake measurements to predict the fermentation performance of brewing yeast, Proc. Int. Conv. Inst. Brew., 20, 193e199 (1988). 8. Peddie, F. L., Simpson, W. J., Kara, B. V., Robertson, S. C., and Hammond, J. R. M.: Measurement of endogenous oxygen uptake rates of brewer’s yeast, J. Inst. Brew., 97, 21e25 (1991). 9. Kara, B. V., Simpson, W., and Hammond, J. R. M.: Prediction of the fermentation performance of brewing yeast with the acidification power test, J. Inst. Brew., 94, 153e158 (1988). 10. Fernandez, S., Gonzalez, G., and Sierra, A.: The acidification power test and the behavior of yeast in brewery fermentations, Tech. Q. Master Brew. Assoc. Am., 28, 89e95 (1991). 11. Dinsdale, M. G., Lloyd, D., McIntyre, P., and Jarvis, B.: Yeast vitality during cider fermentation: assessment by energy metabolism, Yeast, 15, 285e293 (1999).
J. BIOSCI. BIOENG., 12. Boulton, C., Box, W., Quain, D., and Molzahn, S.: Vicinal diketone reduction as a measure of yeast vitality, Tech. Q. Master Brew. Assoc. Am., 2, 89e93 (2001). 13. Hutter, K.: Flow cytometric determinations of glycogen content of yeast during fermentation, J. Inst. Brew., 108, 52e53 (2002). 14. Pretorius, I. S., Toit, M. D., and Rensburg, P. V.: Designer yeasts for the fermentation industry of the 21st century, Food Technol. Biotechnol., 41, 3e10 (2003). 15. Hu, X. H., Wang, M. H., Tan, T., Li, J. R., Yang, H., Leach, L., Zhang, R. M., and Luo, Z. W.: Genetic dissection of ethanol tolerance in the budding yeast Saccharomyces cerevisiae, Genetics, 175, 1479e1487 (2007). 16. Novak, J., Basarova, G., Teixeira, J. A., and Vicente, A. A.: Monitoring of brewing yeast propagation under aerobic and anaerobic conditions employing flow cytometry, J. Inst. Brew., 113, 249e255 (2007). 17. Laverty, D. J., Kury, A. L., Kuksin, D., Pirani, A., Flanagan, K., and Chan, L. L.: Automated quantification of budding Saccharomyces cerevisiae using a novel image cytometry method, J. Ind. Microbiol. Biotechnol., 40, 581e588 (2013). 18. Ohtani, M., Saka, A., Sano, F., Ohya, Y., and Morishita, S.: Development of image processing program for yeast cell morphology, J. Bioinform. Comput. Biol., 1, 695e709 (2004). 19. Ohya, Y., Sese, J., Yukawa, M., Sano, F., Nakatani, Y., Saito, T. L., Saka, A., Fukuda, T., Ishihara, S., Oka, S., and other 15 authors: High-dimensional and large-scale phenotyping of yeast mutants, Proc. Natl. Acad. Sci. USA, 102, 19015e19020 (2005). 20. Suzuki, G., Sawai, H., Ohtani, M., Nogami, S., Sano-Kumagai, F., Saka, A., Yukawa, M., Saito, T. L., Sese, J., Hirata, D., Morishita, S., and Ohya, Y.: Evaluation of image processing programs for accurate measurement of budding and fission yeast morphology, Curr. Genet., 49, 237e247 (2006). 21. Yoshida, S., Imoto, J., Minato, T., Oouchi, R., Sugihara, M., Imai, T., Ishiguro, T., Mizutani, S., Tomita, M., Soga, T., and Yoshimoto, H.: Development of bottom-fermenting Saccharomyces strains that produce high SO2 levels, using integrated metabolome and transcriptome analysis, Appl. Environ. Microbiol., 74, 2787e2796 (2008). 22. European Brewery Convention: Method 9.4, original, real and apparent extract of beer, pp. 1e3, in: Analytica-EBC. Verlag Hans Carl Getranke-Fachverlag, Nürnberg (1998). 23. Kruskal, W. H. and Wallis, W. A.: Use of ranks in one-criterion variance analysis, J. Am. Stat. Assoc., 47, 583e621 (1952). 24. Ohnuki, S., Nogami, S., Kanai, H., Hirata, D., Nakatani, Y., Morishita, S., and Ohya, Y.: Diversity of Ca2þ-induced morphology revealed by morphological phenotyping of Ca2þ-sensitive mutants of Saccharomyces cerevisiae, Eukaryot. Cell, 6, 817e830 (2007). 25. Okada, H., Abe, M., Asakawa-Minemura, M., Hirata, A., Qadota, H., Morishita, K., Ohnuki, S., Nogami, S., and Ohya, Y.: Multiple functional domains of the yeast 1,3-b-glucan synthase subunit Fks1p revealed by quantitative phenotypic analysis of temperature-sensitive mutants, Genetics, 184, 1013e1024 (2010). 26. Ohnuki, S., Kobayashi, T., Ogawa, H., Kozone, I., Ueda, J. Y., Takagi, M., ShinYa, K., Hirata, D., Nogami, S., and Ohya, Y.: Analysis of the biological activity of a novel 24-membered macrolide JBIR-19 in Saccharomyces cerevisiae by the morphological imaging program CalMorph, FEMS Yeast Res., 12, 293e304 (2012). 27. Everitt, B. S. and Dunn, G.: Applied multivariate data analysis. Oxford University Press, New York (1992). 28. Raychaudhuri, S., Stuart, J. M., and Altman, R. B.: Principal components analysis to summarize microarray experiments: application to sporulation time series, Pac. Symp. Biocomput., 455e466 (2000). 29. Eisen, M. B., Spellman, P. T., Brown, P. O., and Botstein, D.: Cluster analysis and display of genome-wide expression patterns, Proc. Natl. Acad. Sci. USA, 95, 14863e14868 (1998). 30. Michaels, G. S., Carr, D. B., Askenazi, M., Fuhrman, S., Wen, X., and Somogyi, R.: Cluster analysis and data visualization of large-scale gene expression data, Pac. Symp. Biocomput., 3, 42e53 (1998). 31. Spellman, P. T., Sherlock, G., Zhang, M. Q., Iyer, V. R., Anders, K., Eisen, M. B., Brown, P. O., Botstein, D., and Fucher, B.: Comprehensive identification of cell cycle-regulated genes of the yeast Saccharomyces cerevisiae by microarray hybridization, Mol. Biol. Cell, 9, 3273e3297 (1998). 32. Bharucha, N. and Kumar, A.: Yeast genomics and drug target identification, Comb. Chem. High Throughput Screen., 10, 618e634 (2007). 33. Ohnuki, S., Oka, S., Nogami, S., and Ohya, Y.: High-content, image-based screening for drug targets in yeast, PLoS ONE, 5, e10177 (2010). 34. Iwaki, A., Ohnuki, S., Suga, Y., Izawa, S., and Ohya, Y.: Vanillin inhibits translation and induces messenger ribonucleoprotein (mrnp) granule formation in Saccharomyces cerevisiae: application and validation of high-content, image-based profiling, PLoS ONE, 8, e61748 (2013). 35. Barra, V.: Analysis of gene expression data using functional principal components, Comput. Methods Programs Biomed., 75, 1e9 (2004). 36. Nogami, S., Ohya, Y., and Yvert, G.: Genetic complexity and QTL mapping of yeast morphological traits, PLoS Genet., 3, e31 (2007). 37. Yamagishi, H., Ohnuki, S., Nogami, S., Ogata, T., and Ohya, Y.: Role of bottomfermenting brewer’s yeast KEX2 in high temperature resistance and poor proliferation at low temperatures, J. Gen. Appl. Microbiol., 56, 297e312 (2010).
Please cite this article in press as: Ohnuki, S., et al., Dynamic changes in brewing yeast cells in culture revealed by statistical analyses of yeast morphological data, J. Biosci. Bioeng., (2013), http://dx.doi.org/10.1016/j.jbiosc.2013.08.005
VOL. xx, 2013 38. Kikuchi, Y., Mizuuchi, E., Nogami, S., Morishita, S., and Ohya, Y.: Involvement of small GTPase Rho1p in cell size control in yeast, FEMS Yeast Res., 7, 569e578 (2007). 39. Kono, K., Nogami, S., Abe, M., Nishizawa, M., Morishita, S., Pellman, D., and Ohya, Y.: G1/S cyclin-dependent kinase regulates small GTPase Rho1p through phosphorylation of RhoGEF Tus1p in Saccharomyces cerevisiae, Mol. Biol. Cell, 19, 1763e1771 (2008).
MORPHOLOGICAL ANALYSIS OF BREWING YEAST CELLS
7
40. Watanabe, M., Watanabe, D., Nogami, S., Morishita, S., and Ohya, Y.: Comprehensive and quantitative analysis of yeast deletion mutants defective in apical and isotropic bud growth, Curr. Genet., 55, 365e380 (2009). 41. Kaprelyants, A. S. and Kell, D. B.: Rapid assessment of bacterial viability and vitality by rhodamine 123 and flow cytometry, J. Appl. Bacteriol., 72, 410e422 (1992). 42. Smart, K. A.: Nutritional requirements and performance of yeast, Proc. Eur. Brew. Conv., 24, 146e157 (1996).
Please cite this article in press as: Ohnuki, S., et al., Dynamic changes in brewing yeast cells in culture revealed by statistical analyses of yeast morphological data, J. Biosci. Bioeng., (2013), http://dx.doi.org/10.1016/j.jbiosc.2013.08.005
Dynamic changes in brewing yeast cells in culture revealed by statistical analyses of yeast morphological data Shinsuke Ohnuki,1 Kenichi Enomoto,2 Hiroyuki Yoshimoto,2 and Yoshikazu Ohya1, * Department of Integrated Bioscience, Graduate School of Frontier Sciences, University of Tokyo, Bldg. FSB-101, 5-1-5 Kashiwanoha, Kashiwa, Chiba 277-8562, Japan1 and Research Laboratories for Brewing, Kirin Brewery Company, Limited, 17-1 Namamugi 1-chome, Tsurumi-ku, Yokohama, Kanagawa 230-8628, Japan2 Received 15 June 2013; accepted 13 August 2013 Available online xxx
The vitality of brewing yeasts has been used to monitor their physiological state during fermentation. To investigate the fermentation process, we used the image processing software, CalMorph, which generates morphological data on yeast mother cells and bud shape, nuclear shape and location, and actin distribution. We found that 248 parameters changed significantly during fermentation. Successive use of principal component analysis (PCA) revealed several important features of yeast, providing insight into the dynamic changes in the yeast population. First, PCA indicated that much of the observed variability in the experiment was summarized in just two components: a change with a peak and a change over time. Second, PCA indicated the independent and important morphological features responsible for dynamic changes: budding ratio, nucleus position, neck position, and actin organization. Thus, the large amount of data provided by imaging analysis can be used to monitor the fermentation processes involved in beer and bioethanol production. Ó 2013, The Society for Biotechnology, Japan. All rights reserved. [Key words: Saccharomyces pastorianus; Bottom-fermenting yeast; CalMorph; Cell morphology; Fermentation; Principal component analysis; Budding profile]
Controlling propagation and fermentation in cellars is an important element in beer fermentation. After raw materials, including malted barley, hops, cereals, adjunct, and water, are converted into wort, brewing yeasts grow and produce ethanol. During fermentation, the production of a well-balanced aroma and the flavor of the final product are at least as important as efficient fermentation and high yield. Quality assurance management, especially of brewing yeasts, is important to maintain their good physiological condition (1). Several features of brewing yeasts have been used to characterize their physiological states during fermentation. Viability is the traditional method (2). Commonly used viability tests are based on the bright-field stains methylene blue (3) and methylene violet (4) and the fluorescent dye 1-anilino-8-naphthalenesulphonic acid (5). The ability to exclude the dye is dependent on cell viability, so any dead cells are stained. Viability provides information regarding the live population in culture but not about individual living yeast cells. In contrast, vitality reflects the physiological condition of individual cells during yeast proliferation. Several methods have developed to measure yeast vitality, such as detection of intracellular pH (ICP) (6), measurement of specific oxygen uptake rate (7,8),
* Corresponding author. Tel.: þ81 4 7136 3650; fax: þ81 4 7136 3651. E-mail address: [email protected] (Y. Ohya). Abbreviations: DAPI, 40 ,6-diamidino-2-phenylindole; DNA, deoxyribonucleic acid; FDR, false discovery rate; FITC-Con A, fluorescein isothiocyanate-Con A; KW test, KruskaleWallis test; PCA, principal component analysis; PCs, principal components; Rh-ph, rhodamine-phalloidin.
the acidification power test (9,10), carbon dioxide production (11), vicinal diketone reduction (12), glycogen and trehalose staining (13), and the budding ratio (14). Of the features analyzed, budding has become of great interest, because the budding ratio correlates with metabolism of polysaccharides and production of bioethanol and other metabolites, such as aroma and flavor compounds (15). Flow cytometry allows fine measurements, including parameters related to the cell cycle (16). More recently, methods combining these three features, viability, cell concentration, and budding ratio, have been developed to simultaneously monitor viability and vitality (17). To assess quantitatively the morphological features of yeast, we developed an automatic image processing system, CalMorph (18,19). CalMorph directly processes fluorescence micrographs and generates 501-dimensional quantitative data regarding mother cells and bud shape, nuclear shape and location, and actin distribution. Using CalMorph, we can easily, rapidly, and reproducibly generate various quantitative data (20). The program generates reproducible data consistent with those obtained manually. CalMorph can also monitor yeast morphological changes that accompany the budding cycle (20), such as the bud index, and specific morphological features in G1, S/G2 and M cells. The purpose of this study was to investigate the dynamics of brewing yeast during fermentation. Because the budding profile is a good indicator of cell vitality, we tried to use CalMorph to monitor the physiological state of brewing yeasts during fermentation. However, the hundreds of parameters generated by CalMorph distracted from the objective. Specific parameters could be selected
1389-1723/$ e see front matter Ó 2013, The Society for Biotechnology, Japan. All rights reserved. http://dx.doi.org/10.1016/j.jbiosc.2013.08.005
Please cite this article in press as: Ohnuki, S., et al., Dynamic changes in brewing yeast cells in culture revealed by statistical analyses of yeast morphological data, J. Biosci. Bioeng., (2013), http://dx.doi.org/10.1016/j.jbiosc.2013.08.005
2
OHNUKI ET AL.
J. BIOSCI. BIOENG.,
in advance based on biological meaning, but this would be arbitrary. The selection of pre-existing parameters was also problematic, because not all informative features were always used. As an alternative approach, we performed successive principal component analyses to extract important time-dependent morphological features. Our results suggest that statistical analyses of morphological data can facilitate yeast management during fermentation. MATERIALS AND METHODS Strains, culture conditions, and sample preparation The yeast strain used was the bottom-fermenting yeast Saccharomyces pastorianus KBY011 (21). KBY011 was pre-cultured in YPD10 medium containing 1% yeast extract (Becton, Dickinson and Co., USA), 2% peptone (Becton, Dickinson and Co.), and 10% glucose (Nacalai Tesquie, Japan) with shaking at 20 C for 3 days. Harvested cells were diluted to an optical density of OD600 ¼ 0.5, and then cultured at 20 C for 96 h in 500 mL of YPD10 medium with gentle stirring with a magnetic stir-bar under anaerobic conditions. Samples were collected at 0, 24, 48, 72, and 96 h, and cells harvested by centrifugation. Apparent extracts were analyzed according to a previous reference (22). Image acquisition and CalMorph analysis Fixation and staining of yeast cells, image acquisition, and CalMorph analysis were performed according to the CalMorph manual (http://scmd.gi.k.u-tokyo.ac.jp/datamine/calmorph/CalMorphmanual.pdf). Briefly, cells (8 106/mL) were fixed in 0.1 M potassium phosphate buffer (pH 6.5) containing 3.7% formaldehyde (Wako Pure Chemical Industries, Japan). To obtain fluorescence images of the cell-surface mannoprotein, actin cytoskeleton, and nuclear DNA, cells were stained with fluorescein isothiocyanateCon A (FITC-Con A, Sigma, USA), rhodamine-phalloidin (Rh-ph, Invitrogen Corp, USA) and 40 ,6-diamidino-2-phenylindole (DAPI, Sigma), respectively. CalMorph automatically characterizes each yeast cell by calculating 501 morphological parameters based on data from more than 200 cells. In total, five independent cultures grown under the same condition were analyzed. Successive principal component analysis To evaluate time-dependent morphological changes, we performed successive principal component analyses (PCA; Fig. 1). First, the 501 morphological parameters were screened to yield 248 parameters that change considerably during fermentation (Fig. 1A). Second, these 248 traits were subjected to PCA to extract the principal components (PCs) explaining the changes (Fig. 1B). Finally, PCA was again applied to the parameters correlating highly with each PC to extract independent parameters (Fig. 1C). All
501 parameters A) Kruskal-Wallis test 248 parameters B) First PCA 57 parameters PC1: 50 parameters
PC2: 7 parameters
C) Second PCA PC1a: 9 parameters
PC2a: 5 parameters
PC1b: 6 parameters
PC2b: 2 parameters
PC1c: 3 parameters
FIG. 1. Schematic flow chart for the successive principal component analysis. Down arrows indicate each step in application of the statistical test or PCA (AeC). The 501 parameters were systematically filtered into semantic fractions by each step. (A) KruskaleWallis test was used to select parameters that changed considerably during fermentation and storage. (B) PCA was applied to identify the patterns of the morphological variation. (C) PCA was applied to extract independent morphological features.
statistical analyses were performed using the ‘R’ software (http://www.r-project. org/). In Fig. 1A, the KruskaleWallis (KW) test (23) was applied to each parameter to identify large changes. For each parameter, the data consisted by five replications of five time points (25 samples). The false discovery rate (FDR) corresponding to each P value of the KW test was estimated by an empirical permutation test of 1000 iterations (24). In Fig. 1B, PCA was applied to the time-course data of the significantly changed 248 parameters (KW test, FDR ¼ 0.05), after five sets of five replicated sample values were combined across time points, ranked among the combined samples, and summed into one rank-sum value for each time point, as described previously (25). To standardize the rank-sum values among the parameters, the rank-sum values at 0 h were subtracted from the rank-sum values of the other time points. Then, eigenvalues (the variance) and eigenvectors (the rotation) of the rank-sum values were calculated from the covariance matrix for the 248 parameters, the contribution ratio was calculated as the ratio of variance, and the PC score was computed by matrix multiplication between the rank-sum values and the rotation. The PC loadings were computed by multiplication of the rotation (eigenvectors) by the square root of the eigenvalue and dividing by the square root of the variance of the ranksum. P-values for the loadings were computed using a t distribution, where t process was the same as that used to transform a Pearson’s productemoment correlation coefficient into a t value (26). In Fig. 1C, to determine morphological features accompanying PC1 of the first PCA, we selected 50 parameters showing significantly high absolute loadings in PC1 (FDR ¼ 0.15), and a second PCA was performed for the parameters selected using morphological data from 122 replicated wild-types as a null distribution. The 50 parameter values of the 122 replicated wild-type morphological data sets were transformed to a normal distribution using the BoxeCox power transformation, as described previously (26). The eigenvalues (the variance) and the eigenvectors (the rotation) of the 122 transformed wild-type data sets were calculated using the covariance matrix of the 50 parameters. The contribution ratio, PC scores, and loadings were calculated as for the first PCA. The PCs of the second PCA were named in alphabetical order (e.g., PC1, PC2, and PC3 were named PC1a, PC1b, and PC1c, respectively). For PC2, we selected seven parameters; the second PCA was performed in a similar manner.
RESULTS Time-dependent morphological changes in fermentation Bottom-fermenting yeast cells were cultured at 20 C with gentle stirring under anaerobic conditions. The changes in cell numbers and apparent extract are shown in Fig. S1. We quantified fermenting yeast cell morphology using 501 morphological parameters. We used image-processing software, CalMorph, after obtaining cell wall, actin, and nuclear DNA images (19). Fermenting yeast cells were sampled at 0, 24, 48, 72, and 96 h, fixed, stained with the fluorescent dyes, FITC-Con A, DAPI and Rh-ph, photographed (at least 200 cells), and quantified using CalMorph. The experiments were replicated five times independently for each time point. As shown in the heat map (Fig. 2), time-dependent changes in the morphological parameters differed in pattern. Of the 501 parameters, 248, 127, and 95 of the 501 differed significantly among the five time points with false discovery rates (FDRs) of 0.05, 0.01, and 0.005, respectively (Fig. 1A). Key variables in a high-dimensional morphological data set during fermentation PCA is an exploratory multivariate statistical technique for simplifying complex data sets (27). It has been used for analysis of time-dependent changes in gene expression data (28) and dose-dependent changes in morphology data (26). To summarize the morphological dynamics during fermentation, we applied PCA to the morphology data using the morphological measurements as the variables and the different time points as the observations. The data set contained values for 248 significantly different parameters (FDR ¼ 0.05) collected with five replicates at 0, 24, 48, 72, and 96 h during fermentation. Thus, the matrix to be analyzed had 25 rows of conditions and 248 columns of parameters. Five sets of five replicate sample values were combined across time points, ranked among the combined samples, summed at each time point, and used for the first PCA. Our analysis indicated that we could summarize the data using two variables. The cumulative contri
Please cite this article in press as: Ohnuki, S., et al., Dynamic changes in brewing yeast cells in culture revealed by statistical analyses of yeast morphological data, J. Biosci. Bioeng., (2013), http://dx.doi.org/10.1016/j.jbiosc.2013.08.005
VOL. xx, 2013
MORPHOLOGICAL ANALYSIS OF BREWING YEAST CELLS
3
The first component represented transient changes in morphology at 24 h. In Fig. 3A, the PCA score shows a peak at 24 h. This component was positive for a parameter that increased early and decreased later. In contrast, absolute loading values of the first component for a parameter whose relative values changed over time would be low. The second component represented morphological change over time. In Fig. 3B, the PCA score increased with time, from zero to positive values. This component was positive for parameters whose relative values increased with time, and negative for those whose relative values decreased with time. The third component measured concavity: note the parabolic nature of the scores in Fig. 3C. Morphological parameters significantly correlated with PC1 and PC2 To further understand the implications of the dynamic changes, we analyzed the parameters that correlated with PC1 (transient change with a peak) and PC2 (change over time). Of the 248 parameters, significant PC loadings in PC1 (FDR ¼ 0.20, 0.15, and 0.10) were observed for 75, 50, and 5 parameters, respectively. Accordingly, we focused on 50 parameters (FDR ¼ 0.15) with significant PC loadings for PC1 (Fig. 1B). These parameters are predicted to show transient changes, with a peak at 24 h. Seven parameters (FDR ¼ 0.15) with significant PC loadings for PC2 (Fig. 1B) likely represent changes over time. Extraction of independent features with a successive PCA Some morphological parameters are semantically related. To extract the independent morphological features of the PC1 (Fig. 1C), we applied a second PCA to a null-distributed data set of the 50 parameters, as described by Ohnuki et al. (26). The uniformly scattered data were a normalized data set obtained from a 122-replicate wild-type morphological data set used for extracting independent morphological features (26). By PCA for the 50 parameters, the first three PCs (PC1a, PC1b, and PC1c) explained more than 40% of the variance among PC scores of a null-distributed data set (Fig. S3). Regarding the PC scores projected from the rank-sum values of the time-course data, they explained w60% (Fig. S3). Examination of six parameters with significant PC loadings for PC1b (Fig. S4) showed that the PC1b represented a ratio of budded cells (Fig. 4). The PC1a and PC1c (Fig. S4) indicated the distance between two nuclei and the neck angle, respectively (Fig. 4). Based on this second PCA analysis, we identified three independent features: budding ratio, nuclear position, and bud angle. Independent morphological features of PC2 (PC2a and PC2b) were also extracted by a second PCA on a null-distributed data set of the seven parameters with significant PC loadings on PC2
FIG. 2. Heat-map of the time-course changes in cell morphology in the stored yeast cells. Numbers on the heat-map indicate the time-course (h) from the start of fermentation. The dendrogram at the right of the heat-map was generated by the average linkage based on the similarity calculated by the correlation coefficient between standardized rank-sum values of the parameters (see Materials and methods). Indices at the right of the heat-map indicate the detected parameters at FDR ¼ 0.05 (using the KruskaleWallis test). KW on the indices indicates the KruskaleWallis test.
bution ratio of PCA in the time-course data (Fig. S2) indicated that two principal components accounted for over 85% of the total variability; inclusion of the third component accounted for over 90%. The meanings of these components were distilled from their respective PCA scores.
200 400 600 800
B
C
PC 1
PC 2
PC 3
−200
0
PC score
A
0
20
40
60
80
100 0
20
40
60
80
100 0
20
40
60
80
100
Time (h) FIG. 3. Time-course changes of the principal component scores. Data from five replicate experiments were standardized by a rank-sum method, and PC scores were calculated as described in Materials and methods.
Please cite this article in press as: Ohnuki, S., et al., Dynamic changes in brewing yeast cells in culture revealed by statistical analyses of yeast morphological data, J. Biosci. Bioeng., (2013), http://dx.doi.org/10.1016/j.jbiosc.2013.08.005
OHNUKI ET AL.
J. BIOSCI. BIOENG.,
PC2b
PC2
Increase in ratio of unbudded cells with polarized actin patches
(Fig. 1C). The first two PCs (PC2a and PC2b) explained more than 90% of the variance among the PC scores, regardless of the data set used (Fig. S5). The PC2a and PC2b (Fig. S4) represented the number of actin patches and a ratio of unbudded cells with polarized actin patches, respectively (Fig. 4).
[D108_C] Distance between nuclear gravity center in mother and middle point of neck (µm)
C
D
1.0
[A122_A] Number of bright actin pathces
Identification and validation of representative parameters To extract representative parameters, parameters with significant PC loadings were plotted against the first two principal components (Fig. 5). The D108_C parameter (distance between nuclear gravity center in mother and middle point of neck) with significant PC1a loading (Fig. S4), in fact, had positive PC1 loadings and nearly zero PC2 loadings. As might be expected,
B
[D210] Nuclear A ratio to nuclear AA1BC cells (cells/cells)
FIG. 4. Independent features of morphological changes during fermentation. Representative shapes of the yeast cells were illustrated. Large circles in the cells of PC1a, PC2a and PC2b indicate nuclei. Small circles in the cells of PC2a and PC2b indicate the actin patches. Lines in the cell at PC1c indicate the neck angle to the long axis. Broad arrows at left side and right side of PCs indicate decreases and increases in the scores of the PCs. Text on the right side of the figure is a description of the independent features when the scores increased.
D108_C
0.3
[A106] Actin b ratio (cells/cells)
0.0
E
−0.5
PC2 (loading)
0.5
A106
C105_C D210
1.6
Decrease in number of actin patches
1.4
PC2
1.2
PC2a
A
1.0
Decrease in neck angle
0.8
PC1
0.6
PC1c
45
Increase of budded cells
35
PC1
25
PC1b
the time-dependent change in D108_C (Fig. 6A) showed a transient change with a peak at 24 h, similar to the change in PC1. In contrast, the changes in the D210 parameter (unbudded cell ratio, correlated with PC1b) and C105_C (neck position, anticorrelated with PC1c) with negative PC1 loadings and nearly zero
0.16
Increase in distance between two nuclei
0.08
PC1
0.5
PC1a
[C105_C] Neck position (degree)
4
−1.0
A122_A
0
24
48
72
96
Time (h) −1.0
−0.5
0.0 PC1 (loading)
0.5
1.0
FIG. 5. Distribution of principal component loadings in PC1 and PC2. Two-hundred and forty-eight parameters identified by the KruskaleWallis test at FDR ¼ 0.05 are plotted. Circles indicate 57 parameters with significant loadings in at least one of the two PCs at FDR ¼ 0.15 (t-test). Filled circles with the parameter names indicate the representative parameters for each PC.
FIG. 6. Time-course changes in representative parameter values. Solid lines on the boxplots indicate the line chart of the mean value in each time point. (A) Time-course changes in the distance between nuclear gravity center in the mother cell and middle point of the neck (D108_C), exemplifying PC1a. (B) Time-course change of the ratio of unbudded cells (D210), exemplifying PC1b. (C) Time-course change of the neck position (C105_C), exemplifying PC1c. (D) Time-course change of the number of bright actin patches (A122_A), exemplifying PC2a. (E) Time-course change of the ratio of unbudded cells with polarized actin patches (A106), exemplifying PC2b.
Please cite this article in press as: Ohnuki, S., et al., Dynamic changes in brewing yeast cells in culture revealed by statistical analyses of yeast morphological data, J. Biosci. Bioeng., (2013), http://dx.doi.org/10.1016/j.jbiosc.2013.08.005
VOL. xx, 2013
MORPHOLOGICAL ANALYSIS OF BREWING YEAST CELLS
PC2 loadings (Fig. 5) also showed a transient change with a peak at 24 h, but the reverse of the change in PC1 (Fig. 6B, C). The changes in the A122_A parameter (number of bright actin patches in the unbudded cells, correlated with PC2a) with negative PC2 loadings and nearly zero PC1 loadings (Fig. 5) decreased over time (Fig. 6D), whereas the change in A106 (ratio of unbudded cells with polarized actin, anti-correlated with PC2b) with positive PC2 loadings and nearly zero PC1 loadings (Fig. 5) increased over time (Fig. 6E). The results indicate that we can use these representative parameters to monitor dynamic aspects of yeast during fermentation. The unbudded yeast cells start budding, nuclear division and bud elongation occur within 24 h, and the vegetative growth stops after incubation for 48 h (Fig. 7). The ratio of unbudded cells with polarized actin increased and the number of actin patches decreased over time (Fig. 7). DISCUSSION Monitoring the vitality of yeast cells is of great importance for the brewing industry, because such information is necessary to determine the optimal fermentation conditions. Budding and cell cycle profiles are good indicators of cell viability and vitality: highly proliferating cells have a high budding ratio and mitotic index, and fermentation conditions often alter cell polarity. In this study, statistical analyses of images of yeast cells were used to provide information regarding budding profiles during fermentation. We used the image-processing software, CalMorph, which determines 501 morphological features after fluorescence staining of the cell wall, actin, and DNA. We performed PCA on morphological data to simplify the analysis and visualize multidimensional data sets. This analysis allowed us to obtain important information on a yeast culture during fermentation. Gene expression data and morphological data Because high-dimensional data on yeast cell morphology are similar in form to gene expression data composed of expression ratios for 6000 genes from Saccharomyces cerevisiae, an analogous statistical methodology can be used. Gene expression patterns and morphological similarities have been used previously in various inference tasks. For example, these data have been used to identify gene clusters based on the profiles (24,25,29,30), to define metrics that measure involvement of genes in particular cellular functions (19,29), and to predict intracellular drug targets (31e34). Eliminating noisy components without discarding important information is also a common problem. PCA (28) and functional PCA (35) have been applied to time series of expression data. Likewise, we showed here that PCA can determine a reduced set of morphological variables that are useful for understanding morphological changes. Multidimensional morphological data also have some different properties from gene expression data. Unlike expression data, morphological data contain hidden dependencies among observations. For this reason, we extracted the independent morphological
M S/G2
0h
5
features by means of a second PCA. Another difference is that morphological parameters can be considered in PCA as observations or variables. In this study, we treated morphological parameters and time points as variables and observations, respectively, in our first PCA, because our main purpose was to obtain as much information as possible about dynamic properties. It is possible to treat morphological parameters as observations, as in the application of PCA to time series of expression data, as described by Raychaudhuri et al. (28). However, in this case, variation of the endpoint rather than change at a peak is more weighted. Systematic extraction of important cellular landmarks from the high-dimensional data We effectively extracted important cellular features from 501 parameters by successive PCA, as described previously (26,34). The first key point of this method is to summarize two types of time-course change by applying PCA to the morphological data at different time points. The second key point is to extract independent parameters by applying PCA to a nulldistributed data set. Together with the selection of highly affected parameters in advance, this novel method enabled us to unravel the complex high-dimensional data into the independent and important cellular features (26). Although high-dimensional data are useful to describe various aspects of cellular activities (36,37), identifying primary features can be problematic because of the complexity of the data. One approach is hypothesis-driven (38e40), in which accumulated knowledge in the field and/or some creative insights are required. In successive PCA, the important features are extracted systematically in a data-driven manner (26). Because successive PCA is a general-purpose approach, it can be applied to any data set in a high-dimensional data analysis of cellular traits under several experimental conditions; e.g., metabolome database analyses under several conditions. Morphological features used for yeast management The vitality of brewing yeast cells, which has been monitored for yeast management, is defined as the capacity of yeasts to initiate metabolism rapidly after transfer from a nutrient-poor to a nutrient-rich environment (1,41) and the physiological state of the viable cell population (42). Accordingly, the bud index has been used as an indicator of vitality (14,16,17). It can be used to estimate the reproduction rate of yeast populations, which directly correlates with metabolism of polysaccharides and bioethanol production (17). One of the important morphological features that we found in this study was the bud index, indicating that we could extract a vital parameter by successive PCA. It should be noted that other morphological parameters behaved similarly to the bud index. Two representative parameters, nuclear position and bud angle, as well as other parameters with high absolute loading values on PC1 may also reflect the vital state of yeast cells. Thus these could be used for monitoring the vitality of brewing yeasts. Regarding the parameters of actin patches and actin regions with high absolute loading values to PC2, these are
G1
24 h
96 h
FIG. 7. Morphological changes during fermentation. Large and small circles in the yeast cell indicate nucleus and actin patch, respectively. Pie charts show the ratio of cells at the G1, S/G2, and M phases of the cell cycle. Arrows indicate time interval (24 h per arrow).
Please cite this article in press as: Ohnuki, S., et al., Dynamic changes in brewing yeast cells in culture revealed by statistical analyses of yeast morphological data, J. Biosci. Bioeng., (2013), http://dx.doi.org/10.1016/j.jbiosc.2013.08.005
6
OHNUKI ET AL.
not indicators of vitality, but rather are correlated with incubation time. The morphology of yeast cells changes in culture and is thus a powerful tool for assessing the physiological state of yeast cells and the fermentation conditions. There are several advantages to using a large number of morphological parameters. First, we are able to focus on several vital parameters simultaneously to improve the accuracy of the monitoring system. A similar approach was used previously, focusing on three features: viability, cell concentration and budding ratio (17). Second, we are able to analyze correlations among the changes in morphological parameters under different fermentation conditions to estimate physiological states. If all of the morphological features behave similarly, this would suggest that the physiological state changed similarly. Third, we are able to target efforts toward the vital morphological parameters to improve fermentation conditions. Every effort should be made to achieve a high budding ratio during fermentation or to understand the reason for its low value. Finally, by collecting data under different culture and storage conditions, this system could be expanded to create a yeast management database. Imaging and other high-dimensional information could be integrated to contribute to yeast management. Further study is necessary to develop tools and methods to analyze and estimate better conditions for beer fermentation based on such a database. Supplementary data related to this article can be found online at http://dx.doi.org/10.1016/j.jbiosc.2013.08.005. ACKNOWLEDGMENTS We thank Kyoko Horikoshi, Satoru Nogami, Tomohiko Ichii for their cooperation, and members of the Laboratory of Signal Transduction for fruitful discussions. This work was supported by Grants-in-Aid for Scientific Research from the Ministry of Education, Culture, Sports, Science and Technology, Japan (21310127 and 24370002 to Y.O.). S.O. was a Research Fellow of the Japan Society for the Promotion of Science. References 1. Lodolo, E. J., Kock, J. L., Axcell, B. C., and Brooks, M.: The yeast Saccharomyces cerevisiae e the main character in beer brewing, FEMS Yeast Res., 8, 1018e1036 (2008). 2. Bendiak, D.: Review of metabolic activity tests and their ability to predict fermentation performance, pp. 34e45, in: Smart, K. (Ed.), Brewing yeast fermentation performance. Blackwell Science, Oxford, UK (2000). 3. Chilver, M. J., Harrison, J., and Webb, T. J. B.: Use of immunofluorescence and viability stains in quality control, J. Am. Soc. Brew. Chem., 36, 13e18 (1978). 4. Smart, K. A., Chambers, K. M., Lambert, I., and Jenkins, C.: Use of methylene violet staining procedures to determine yeast viability and vitality, J. Am. Soc. Brew. Chem., 57, 18e23 (1999). 5. McCaig, R.: Evaluation of the fluorescent dye 1-anilino-8-naphthalene sulphonic acid for yeast viability determination, J. Am. Soc. Brew. Chem., 48, 22e25 (1990). 6. Imai, T. and Ohno, T.: The relationship between viability and intracellular pH in the yeast Saccharomyces cerevisiae, Appl. Environ. Microbiol., 61, 3604e3608 (1995). 7. Wheatcroft, R., Lim., Y. H., Hawthorne, D. B., Clarke, B. J., and Kavanagh, T. E.: An assessment of the use of specific oxygen uptake measurements to predict the fermentation performance of brewing yeast, Proc. Int. Conv. Inst. Brew., 20, 193e199 (1988). 8. Peddie, F. L., Simpson, W. J., Kara, B. V., Robertson, S. C., and Hammond, J. R. M.: Measurement of endogenous oxygen uptake rates of brewer’s yeast, J. Inst. Brew., 97, 21e25 (1991). 9. Kara, B. V., Simpson, W., and Hammond, J. R. M.: Prediction of the fermentation performance of brewing yeast with the acidification power test, J. Inst. Brew., 94, 153e158 (1988). 10. Fernandez, S., Gonzalez, G., and Sierra, A.: The acidification power test and the behavior of yeast in brewery fermentations, Tech. Q. Master Brew. Assoc. Am., 28, 89e95 (1991). 11. Dinsdale, M. G., Lloyd, D., McIntyre, P., and Jarvis, B.: Yeast vitality during cider fermentation: assessment by energy metabolism, Yeast, 15, 285e293 (1999).
J. BIOSCI. BIOENG., 12. Boulton, C., Box, W., Quain, D., and Molzahn, S.: Vicinal diketone reduction as a measure of yeast vitality, Tech. Q. Master Brew. Assoc. Am., 2, 89e93 (2001). 13. Hutter, K.: Flow cytometric determinations of glycogen content of yeast during fermentation, J. Inst. Brew., 108, 52e53 (2002). 14. Pretorius, I. S., Toit, M. D., and Rensburg, P. V.: Designer yeasts for the fermentation industry of the 21st century, Food Technol. Biotechnol., 41, 3e10 (2003). 15. Hu, X. H., Wang, M. H., Tan, T., Li, J. R., Yang, H., Leach, L., Zhang, R. M., and Luo, Z. W.: Genetic dissection of ethanol tolerance in the budding yeast Saccharomyces cerevisiae, Genetics, 175, 1479e1487 (2007). 16. Novak, J., Basarova, G., Teixeira, J. A., and Vicente, A. A.: Monitoring of brewing yeast propagation under aerobic and anaerobic conditions employing flow cytometry, J. Inst. Brew., 113, 249e255 (2007). 17. Laverty, D. J., Kury, A. L., Kuksin, D., Pirani, A., Flanagan, K., and Chan, L. L.: Automated quantification of budding Saccharomyces cerevisiae using a novel image cytometry method, J. Ind. Microbiol. Biotechnol., 40, 581e588 (2013). 18. Ohtani, M., Saka, A., Sano, F., Ohya, Y., and Morishita, S.: Development of image processing program for yeast cell morphology, J. Bioinform. Comput. Biol., 1, 695e709 (2004). 19. Ohya, Y., Sese, J., Yukawa, M., Sano, F., Nakatani, Y., Saito, T. L., Saka, A., Fukuda, T., Ishihara, S., Oka, S., and other 15 authors: High-dimensional and large-scale phenotyping of yeast mutants, Proc. Natl. Acad. Sci. USA, 102, 19015e19020 (2005). 20. Suzuki, G., Sawai, H., Ohtani, M., Nogami, S., Sano-Kumagai, F., Saka, A., Yukawa, M., Saito, T. L., Sese, J., Hirata, D., Morishita, S., and Ohya, Y.: Evaluation of image processing programs for accurate measurement of budding and fission yeast morphology, Curr. Genet., 49, 237e247 (2006). 21. Yoshida, S., Imoto, J., Minato, T., Oouchi, R., Sugihara, M., Imai, T., Ishiguro, T., Mizutani, S., Tomita, M., Soga, T., and Yoshimoto, H.: Development of bottom-fermenting Saccharomyces strains that produce high SO2 levels, using integrated metabolome and transcriptome analysis, Appl. Environ. Microbiol., 74, 2787e2796 (2008). 22. European Brewery Convention: Method 9.4, original, real and apparent extract of beer, pp. 1e3, in: Analytica-EBC. Verlag Hans Carl Getranke-Fachverlag, Nürnberg (1998). 23. Kruskal, W. H. and Wallis, W. A.: Use of ranks in one-criterion variance analysis, J. Am. Stat. Assoc., 47, 583e621 (1952). 24. Ohnuki, S., Nogami, S., Kanai, H., Hirata, D., Nakatani, Y., Morishita, S., and Ohya, Y.: Diversity of Ca2þ-induced morphology revealed by morphological phenotyping of Ca2þ-sensitive mutants of Saccharomyces cerevisiae, Eukaryot. Cell, 6, 817e830 (2007). 25. Okada, H., Abe, M., Asakawa-Minemura, M., Hirata, A., Qadota, H., Morishita, K., Ohnuki, S., Nogami, S., and Ohya, Y.: Multiple functional domains of the yeast 1,3-b-glucan synthase subunit Fks1p revealed by quantitative phenotypic analysis of temperature-sensitive mutants, Genetics, 184, 1013e1024 (2010). 26. Ohnuki, S., Kobayashi, T., Ogawa, H., Kozone, I., Ueda, J. Y., Takagi, M., ShinYa, K., Hirata, D., Nogami, S., and Ohya, Y.: Analysis of the biological activity of a novel 24-membered macrolide JBIR-19 in Saccharomyces cerevisiae by the morphological imaging program CalMorph, FEMS Yeast Res., 12, 293e304 (2012). 27. Everitt, B. S. and Dunn, G.: Applied multivariate data analysis. Oxford University Press, New York (1992). 28. Raychaudhuri, S., Stuart, J. M., and Altman, R. B.: Principal components analysis to summarize microarray experiments: application to sporulation time series, Pac. Symp. Biocomput., 455e466 (2000). 29. Eisen, M. B., Spellman, P. T., Brown, P. O., and Botstein, D.: Cluster analysis and display of genome-wide expression patterns, Proc. Natl. Acad. Sci. USA, 95, 14863e14868 (1998). 30. Michaels, G. S., Carr, D. B., Askenazi, M., Fuhrman, S., Wen, X., and Somogyi, R.: Cluster analysis and data visualization of large-scale gene expression data, Pac. Symp. Biocomput., 3, 42e53 (1998). 31. Spellman, P. T., Sherlock, G., Zhang, M. Q., Iyer, V. R., Anders, K., Eisen, M. B., Brown, P. O., Botstein, D., and Fucher, B.: Comprehensive identification of cell cycle-regulated genes of the yeast Saccharomyces cerevisiae by microarray hybridization, Mol. Biol. Cell, 9, 3273e3297 (1998). 32. Bharucha, N. and Kumar, A.: Yeast genomics and drug target identification, Comb. Chem. High Throughput Screen., 10, 618e634 (2007). 33. Ohnuki, S., Oka, S., Nogami, S., and Ohya, Y.: High-content, image-based screening for drug targets in yeast, PLoS ONE, 5, e10177 (2010). 34. Iwaki, A., Ohnuki, S., Suga, Y., Izawa, S., and Ohya, Y.: Vanillin inhibits translation and induces messenger ribonucleoprotein (mrnp) granule formation in Saccharomyces cerevisiae: application and validation of high-content, image-based profiling, PLoS ONE, 8, e61748 (2013). 35. Barra, V.: Analysis of gene expression data using functional principal components, Comput. Methods Programs Biomed., 75, 1e9 (2004). 36. Nogami, S., Ohya, Y., and Yvert, G.: Genetic complexity and QTL mapping of yeast morphological traits, PLoS Genet., 3, e31 (2007). 37. Yamagishi, H., Ohnuki, S., Nogami, S., Ogata, T., and Ohya, Y.: Role of bottomfermenting brewer’s yeast KEX2 in high temperature resistance and poor proliferation at low temperatures, J. Gen. Appl. Microbiol., 56, 297e312 (2010).
Please cite this article in press as: Ohnuki, S., et al., Dynamic changes in brewing yeast cells in culture revealed by statistical analyses of yeast morphological data, J. Biosci. Bioeng., (2013), http://dx.doi.org/10.1016/j.jbiosc.2013.08.005
VOL. xx, 2013 38. Kikuchi, Y., Mizuuchi, E., Nogami, S., Morishita, S., and Ohya, Y.: Involvement of small GTPase Rho1p in cell size control in yeast, FEMS Yeast Res., 7, 569e578 (2007). 39. Kono, K., Nogami, S., Abe, M., Nishizawa, M., Morishita, S., Pellman, D., and Ohya, Y.: G1/S cyclin-dependent kinase regulates small GTPase Rho1p through phosphorylation of RhoGEF Tus1p in Saccharomyces cerevisiae, Mol. Biol. Cell, 19, 1763e1771 (2008).
MORPHOLOGICAL ANALYSIS OF BREWING YEAST CELLS
7
40. Watanabe, M., Watanabe, D., Nogami, S., Morishita, S., and Ohya, Y.: Comprehensive and quantitative analysis of yeast deletion mutants defective in apical and isotropic bud growth, Curr. Genet., 55, 365e380 (2009). 41. Kaprelyants, A. S. and Kell, D. B.: Rapid assessment of bacterial viability and vitality by rhodamine 123 and flow cytometry, J. Appl. Bacteriol., 72, 410e422 (1992). 42. Smart, K. A.: Nutritional requirements and performance of yeast, Proc. Eur. Brew. Conv., 24, 146e157 (1996).
Please cite this article in press as: Ohnuki, S., et al., Dynamic changes in brewing yeast cells in culture revealed by statistical analyses of yeast morphological data, J. Biosci. Bioeng., (2013), http://dx.doi.org/10.1016/j.jbiosc.2013.08.005