Abstract
Background and goals During wine fermentation, grape berry components, including seeds, undergo extensive physical and chemical changes that result in the release of flavonoids, such as tannins, from seeds into wine. Understanding changes in seed morphology during fermentation is crucial for aiding the development of accurate prediction models for flavonoid extraction during winemaking, which enhances fermentation management and ensures consistency in wines from year to year.
Methods and key findings High-resolution x-ray microcomputed tomography (x-ray μCT) was used to investigate the effect of red wine fermentation on changes in grape seed morphology. Using a PyTorch-based implementation of a fully convolutional network with a Resnet-101 backbone for semantic segmentation of x-ray μCT images, we quantified extensive alteration to grape seed structure during fermentation. Image analyses revealed the development of a pore network breaking apart the seed endosperm by the end of fermentation, leading to an increase in surface area.
Conclusions and significance Fermentation significantly altered grape seed morphology. Such alterations could enable transport of seed flavonoids from inside the endosperm and integument to outside the seed. Further research on the physical processes occurring in seeds during wine fermentation is necessary to build better physiochemical models.
Introduction
Quantifying changes in winegrape (Vitis vinifera) berry components, including skins and seeds, during fermentation is an important research area in enology. The development of predictive models to elucidate flavonoid release by skins and seeds is an important tool for winemakers, as flavonoid compounds play a crucial role in determining a wine’s taste, color, and mouthfeel (Somers and Evans 1979, Skogerson et al. 2007, Lee et al. 2008, Zanoni et al. 2010, Miller et al. 2019, 2020a, 2020b, Le Mao et al. 2021). Flavonoids found in skins and seeds are primarily released by diffusion or mass transfer during fermentation and maceration (Somers and Evans 1979, Setford et al. 2017, Miller et al. 2019). This process is influenced by the temperature, time, and solvent composition of the fermentation, among other factors (Casassa et al. 2013a, 2013b, Bindon et al. 2014, Lerno et al. 2015, 2017, 2018, Beaver et al. 2020, Gutiérrez-Escobar et al. 2021).
Seed-derived flavonoids, especially tannins, can make up a significant portion of total wine flavonoids, though the specific mechanisms behind their extraction into wine are relatively unknown. Research to date has focused on the physical and chemical processes involved in flavonoid extraction from the exterior of grape seeds, with little focus on internal alterations to seed structure during fermentation, which could enhance flavonoid extraction from the interior of seeds. Very few studies have identified sources of flavonoids in specific grape seed tissues. Flavonoids were localized primarily in the epidermis, outer integument, and inner cell layer of the inner integument of Cabernet franc grape seeds (Cadot et al. 2006). There were no flavonoids in the endosperm. In contrast, there were flavonoids in both the epidermis and the endosperm of Pinot noir and Cabernet Sauvignon grapes, though this experiment did not measure flavonoids in integument tissues (Thorngate and Singleton 1994). A freeze-thaw treatment increased flavonoid extraction efficiency from Cabernet Sauvignon, Cabernet franc, Merlot, and Pinot noir seeds (Gombau et al. 2020). This process disrupted the seed integument, which might have increased flavonoid extraction efficiency (VanderWeide et al. 2020). Identifying the sources of flavonoids within the seed can be challenging and requires specific instrumentation to obtain accurate results.
Most studies of grape berries and seeds use thin sectioning and light microscopy to visualize changes in morphology during maturation or fermentation. These techniques are destructive and only enable 2-D interpretation of morphological changes. Minimally destructive techniques for imaging 3-D volumes, like x-ray microcomputed tomography (x-ray μCT), are used to study grapevine physiological processes, and were only recently applied to studying grape berries or seeds. For example, air spaces in Shiraz berries were quantified using x-ray μCT with a pixel resolution of 8.35 μm (Xiao et al. 2018). Additionally, air spaces in Chardonnay grape flowers and berries were quantified using x-ray μCT with a pixel resolution of 8.35 μm (Xiao et al. 2021). Botrytis cinerea infection of Cabernet berries was examined using x-ray μCT, although no resolution was given (Weiller et al. 2021). No studies have applied this technique to quantify fermentation-induced morphological changes in grape seeds.
The purpose of this study was to further investigate the changes in grape seed morphology during red wine fermentation, using x-ray μCT. Changes in grape seed components in the field, at crushing, and at the end of fermentation were quantified and compared to determine whether structural physiological changes, such as the introduction of new diffusion pathways, could be another mechanism that influences seed flavonoid mobility and release into wine during fermentation. Using a PyTorch implementation of a fully convolutional network (FCN) with a Resnet-101 backbone, seed morphology data was extracted from x-ray μCT images of grape seeds collected before and after a 2018 fermentation study on Cabernet Sauvignon grapes (Vitis vinifera L.) at the UC Davis Teaching and Research Winery (Miller et al. 2019). These are the highest-resolution images of grape seeds to date. Findings from this research aim to support the work of predictive modeling to provide winemakers with more efficient and accurate decision-making tools.
Materials and Methods
Data collection
High-resolution x-ray μCT images of Cabernet Sauvignon grape seeds prior to harvest and before and after fermentation were collected from a separate research study, but analyzed for this current work. X-ray μCT scans of the grape seeds were conducted at the Advanced Light Source in the Lawrence Berkeley National Laboratory, Berkeley, CA, using a 2× objective lens with a pixel resolution of 3.25 μm at 23 keV. Raw tomographic image data was reconstructed using TomoPy (Gürsoy et al. 2014). Details of the fermentation and sampling are as described (Miller et al. 2019) and listed in Supplemental Information-Methods. Phenolic data and fermentation measurements were not determined (Miller et al. 2019). Nine x-ray μCT scans were obtained, where three seeds were imaged for each sampling time point. The three scans were treated as replicates for each group and seeds were assumed to be representative of the population of seeds in the vineyard. For analyses in this study, groups were labeled as: “Field” (sampled prior to harvest), “Fermentation Start” (FS; sampled during crushing), and “Fermentation End” (FE; sampled at the end of fermentation).
Image annotation and modeling
An x-ray μCT scan for one seed section produced 416 images and was referred to as an image sequence. Within an image sequence, four to five individual images were selected for annotation that aimed to capture the heterogeneity within the seed along the Z-axis of the scan (Figure 1). Physiologically important grape seed components, as defined previously (Ristic and Iland 2005), were annotated manually from x-ray μCT images using the open-source Fiji (Fiji Is Just ImageJ) image processing and analysis software in Java (Schindelin et al. 2012, Momayyezi et al. 2022). Images and associated annotations from three or four images per grape seed were then used to train a Python-based PyTorch implementation of an FCN with a ResNet-101 backbone on an NVIDA A40 graphics processing unit with 48 GB of VRAM (Oliphant 2007, He et al. 2015, Long et al. 2015, Paszke et al. 2019, Rippner et al. 2022). Results for the identification of grape seed components came from the best-performing model. Model performance was evaluated based on the accuracy, precision, recall, and f1 score calculated after evaluating one image from each of the seeds (nine total) that were not included in the training or validation data sets (Rippner et al. 2022). In our work, accuracy, precision, recall, and f1 scores are defined as:
Where TP = true positive, FP = false positive, TN = true negative, and FN = false negative. Further details about the model parameters can be found in the Supplemental Information-Methods.
Selected x-ray microcomputed tomography images from along the z-axis of grape seeds from each sampling time point. “Field” seeds were collected prior to harvest, “Fermentation Start” seeds were collected during crushing, and “Fermentation End” seeds were collected at the end of fermentation (total soluble solids [Brix] ≤ 0). This image has been altered for distribution. The full, original file is available in the Supplemental Data.
Statistics
All statistical analyses were conducted in R (R Core Team 2021). Normality of residuals, homogeneity of variance, and the leverage of individual observations were inspected visually to meet the assumptions for analysis of variance using the plot() command. Multiple comparisons among treatments were made using Tukey’s honestly significant difference test with significance declared at p ≤ 0.05 using the agricolae package (de Mendiburu 2019). Additional information about statistical analysis can be found in the Supplemental Information. All figures were created using the R package ggplot2, Fiji, or Dragonfly (Object Research Systems) (Schindelin et al. 2012, Wickham 2016).
Results
X-ray μCT imaging of grape seeds
X-ray μCT imaging of grape seeds revealed spatial and temporal changes in seed morphology and internal porosity (Figure 1). The outer layer of the seeds was identified as the integument, the inner portion as the endosperm, and the tip and base as the ventral face (VF) and dorsal face (DF), as defined previously (Cadot et al. 2006) (Supplemental Figure 1). Internal porosity (pore space) was identified in both tissue types based on their lack of x-ray absorbance and comprised the most significant changes in seed morphology. A PyTorch implementation of an FCN with a Resnet-101 backbone was trained to identify different seed components. Model accuracy was ≥99% for each seed component (Table 1). Model precision in identifying the endosperm was greatest at 97%, compared to the integument, endosperm pore space, and seed pore space, which were 96, 89, and 88%, respectively. Recall and F1 scores were >90% for all components and greatest for the endosperm layer (98 and 99%, respectively). Based on outputs from the model generated for each scanned seed, a network of pore spaces were observed that varied spatially in size and distribution within the seed and temporally over the course of sampling (https://youtu.be/KkAzzw4aQQ0). The most notable impact of these pore spaces occurred in the endosperm layer of the seeds, resulting in the greatest changes in seed morphology.
Precision, recall, accuracy, and F1 scores for model inference performance on the test data set.
Integument
Seeds sampled at each timepoint had some degree of pore space connectivity within the integument throughout the entire seed (Figure 2). Prior to harvest, seeds contained very small pore spaces along the perimeter of the outer integument, which decreased at the start of fermentation and eventually disappeared by the end. A network of small pores attached to the outer integument on the VF of the grape seeds surrounded two voids that penetrated the center of the seed, one on each side (Figure 2). These voids remained inside the integument prior to harvest and the start of fermentation; however, at the end of fermentation, the outer integument perimeter separating the inside of these voids from the grape flesh had dissipated and there were no longer pores within the integument (Figure 1). Overall, the integument had the largest average surface area of any seed structure for all three sampling timepoints (Supplemental Table 1 and Supplemental Figure 2), which was likely due to this network of small pores and the morphology of the two larger, central pores. Surface area for the integument decreased by 4.7% from in the field to the start of fermentation and another 1.6% from the start to end of fermentation. However, volume of the integument increased by 21% from in the field to the start of fermentation, then decreased by 18.4% at the end of fermentation. Although changes occurred in the integument before and after fermentation, the differences were not statistically significant (Figure 3).
Annotated grape seeds generated by a Python-based PyTorch implementation of a fully convolutional network (FCN) model. Two versions of annotated grape seed images are displayed for each sampling time point: A) Replicate 1 of “Field” seeds, B) Replicate 1 of “Fermentation Start” seeds, and C) Replicate 1 of “Fermentation End” seeds. The top image represents basic model annotations of each grape seed component. Bottom images show a transparent endosperm layer to reveal the intricate pore networks obtained through use of a Python-based PyTorch implementation of an FCN model. Brown is the seed integument, dark gray is integument pore space, white is the endosperm, and the multicolored portions are pores in and around the endosperm, where unique pore colors represent unconnected pores.
Surface area and volume of grape seed components from in the field, and immediately before and after fermentation. Box plots containing distribution of replicate seed surface area and volume measurements for grape seed components across sampling time points: A) integument volume, B) integument surface area, C) endosperm volume, D) endosperm surface area, E) endosperm pore space volume, F) endosperm pore space surface area, G) total pore space volume, and H) total pore space surface area. Results for Replicate 3 of “Fermentation Start” seeds were removed for statistical analyses due to a hollow endosperm layer. Letters represent significant differences based on analysis of variance with post-hoc Tukey’s honestly significant difference test where α = 0.05.
Unlike the other sampling time points, the amount and distribution of pore space present within the integument varied across FE seed replicates (Supplemental Figure 2). Replicates 2 and 3 had delamination around the perimeter of the VF of the seeds that was not present in Replicate 1. Furthermore, a thin layer of perimeter delamination in the integument of Replicate 3 occurred around the entire seed and expanded to create a larger pore along the DF, which then increased in area and split into several smaller interconnected pores. Overall, there was no significant difference in total seed pore space across sampling points (p < 0.05).
Endosperm
Endosperm morphology and pore space changed the most compared to the other grape seed components. Seeds sampled prior to harvest and at the end of fermentation contained three long pores between the inner perimeter of the integument and the outer perimeter of the endosperm, one on each side of the seed and the third along the VF, which remained present throughout the inside of the seed (Figure 2). These pores were minimal or absent at the start of fermentation (Supplemental Table 1). This change in porosity led to a significant increase in endosperm surface area between the beginning and end of fermentation.
Endosperm pore distribution and size varied across replicates and sampling time points (Supplemental Figure 2). Replicate 1 of the Field seeds showed greater pore distribution across the endosperm and inside the seed compared to the other replicates. Replicate 1 also had a deformity toward the seed edge, where part of the integument seeped into the endosperm and created a new object within the endosperm composed of integument pieces and additional pore space. Replicate 3 of FS seeds was excluded from statistical comparisons due to the complete absence of an endosperm, which left a hollow seed. This phenomenon has been observed previously, where the cells of the endosperm layer became deformed at veraison and then disappeared at maturity (Cadot et al. 2006). The remaining two replicates of FS seeds showed little endosperm pore space and, when present, the pores were disconnected and mostly present within the middle of the seed. Additionally, only Replicate 1 contained pore space connecting the outer perimeter of the endosperm and inner perimeter of the integument.
FE seeds had the greatest total endosperm porosity and endosperm surface area, yet the pore space distribution was also variable across replicates. Replicate 1 had three primary pores around the endosperm that remained relatively consistent throughout the seed. Replicates 2 and 3 had delamination occurring around most of the endosperm perimeter, creating pore space that further separated the endosperm from the inner perimeter of the integument. All three replicates contained a network of pores within the endosperm.
Overall, median individual pore space surface area and volume decreased significantly by >4x and >10x, respectively, between the Field and FS (p < 0.001), then increased significantly by >3x and >3.5x, respectively, between FS and FE (p < 0.001; Figure 2 and Supplemental Figure 3). Despite numerically smaller median individual pore surface areas at FE compared to Field, total mean endosperm surface area increased between Field and FE, caused by formation of many small pores and a few very large pores during fermentation (p < 0.01; Figure 2 and Supplemental Figure 3).
Discussion
Accurate model performance for segmenting x-ray μCT scans is crucial to accurately quantify changes in grape seed morphology during fermentation. The FCN model used here had a high degree of accuracy, precision, and recall. The results of this model were as good or better than was achieved previously when identifying and segmenting tissues in other plant organs (Théroux-Rancourt et al. 2020, Momayyezi et al. 2022, Rippner et al. 2022). Such accuracy is crucial for elucidating changes in grape seed morphology during fermentation. This can enable a mechanistic understanding of seed flavonoid release during winemaking, which is necessary to improve predictive flavonoid extraction models.
Seed flavonoids are located mostly in the epidermis, outer integument, and inner integument (Cadot et al. 2006). Flavonoids are also present in the seed endosperm (Thorngate and Singleton 1994). While seed flavonoid extraction during winemaking is complicated and impacted by flavonoid composition and molecular weight, grape variety, selective adsorption by skin or flesh water-insoluble materials, and polyphenol reactions/interactions leading to precipitation, structural changes in grape seeds during fermentation affect the rate of flavonoid diffusion (Abi-Habib et al. 2021, 2022, 2023). Epifluorescence microscopy was used to image changes in a grape seed’s outer layer over the course of the same fermentation that provided the grape seeds used here (Miller et al. 2019). The epiflourescence showed a decrease in average endocarp (outer integument) thickness, which was a possible explanation for the zero-order rate of seed flavonoid release observed in their mechanistic model of phenolic extraction. It was suggested that shrinking of the outer layer helped reduce mass transfer resistance to the release of flavonoids from deeper in the seed. Our x-ray μCT analyses and FCN modeling on seeds from the same fermentation revealed no significant change in the seeds’ integument (outer and inner integument combined) thickness. Differences in the results could be attributed to slight differences in the tissues being measured and the greater number of cross sectional measurements (416 per seed in this case) that can be attained using x-ray μCT compared to traditional microscopy techniques. Measurements of grape seed integuments before and after fermentation varied across the seeds sampled and showed no clear evidence that, on average, the integument thickness was changing.
The x-ray μCT results did suggest the possibility of an alternative mechanism that could allow for the release of flavonoids located deeper in the seed. Specifically, the formation of pores inside the seed during fermentation could aid in the reduction of mass transfer resistance for the release of flavonoids from deeper in the seed without a change in integument thickness. Increased endosperm porosity also increased endosperm surface area, which increased flavonoid flux from the endosperm and could help explain the previously measured zero order rate release of seed flavonoids (Lerno et al. 2017, Miller et al. 2019). Our results, in combination with previous research, imply that multiple physiochemical reactions and diffusion pathways, especially those caused by changes in seed morphology during fermentation, should be considered when applying quantitative mechanistic models to flavonoid extraction in industry settings.
Conclusion
This work provides unique insight into seed morphological changes that could influence flavonoid release during fermentation. From x-ray μCT images of grape seeds, a deep learning model was developed to identify and quantify changes in grape seed components at harvest and over the course of fermentation. Results revealed pore spaces formed throughout the grape seed and endosperm pore spaces increased significantly in surface area and volume by the end of fermentation. This study, combined with insight from previous research, suggests that there are several physiochemical factors contributing to flavonoid extraction in wine that should be considered when developing quantitative predictive models, including changes in seed morphology. Further research is needed to determine the extent to which endosperm porosity impacts flavonoid release and the impact of extended maceration on pore development. Overall, creating predictive models that represent industry fermentation conditions can be a valuable tool. These models, combined with a winemaker’s knowledge of flavonoid extraction, will greatly improve grape quality assessments, development of wine style and fermentation management, and ensure consistency in wines from year to year.
Supplemental Data
The following supplemental materials are available for this article in the Supplemental tab above:
Supplemental Information
Supplemental Table 1 Calculated surface area (SA) and volume (V) measurements from annotated x-ray microcomputed tomography images. SA and V were calculated using the Porespy package in Python for each seed replicate at three sampling time points: Field, Fermentation Start (FS), and Fermentation End (FE). Thickness (T) of the integument layer was calculated by dividing V over SA.
Supplemental Figure 1 Annotation of an x-ray microcomputed tomography (x-ray μ-CT) grape seed image using a fully convolutional network (FCN) model. Comparison of original x-ray μ-CT grape seed images to human and computer-generated annotations: A) individual image slice from an x-ray μ-CT scan of a grape seed from Fermentation End samples; B) human annotation of the seed components; C) FCN model prediction of grape seed components; and D) FCN model prediction of the individual seed components with endosperms, connected in 3-D space, uniquely colored.
Supplemental Figure 2 Representative slices from seed scans. Representative slices from each of the individual seed scans for the Field, Fermentation Start, and Fermentation End treatments.
Supplemental Figure 3 Endosperm porosity density plots. Surface area (SA) and volume (V) of individual pores located within the endosperm layer of each seed was used to assess pore size distribution across sampling time points using density plots created in RStudio software. Results for Replicate 3 of Fermentation Start seeds were removed for statistical analyses due to a hollow endosperm layer. Levene’s test for homogeneity of variance and the Kruskal-Wallis test of analysis of variance were used to assess the normality of the data. Both tests confirmed the data was non-parametric (p < 0.001). A Dunn test was then used to compare the differences between sampling time points for SA and V of the endosperm pores. Significant differences are represented in letters within each density plot, along with a dotted line indicating the median SA and/or V for Field, Fermentation Start, and Fermentation End samples.
Footnotes
The authors would like to thank Brandon Peterson, Dr. Katherine East, Dr. Garet Heineck, and Dr. Collins Wakholi (USDA-ARS HCPGIRU) for technical assistance. The authors would also like to thank Dilworth Parkinson at the Advanced Light Source in Lawrence Berkeley National Laboratory (Berkeley, CA) for their support. The work presented in this article was funded by the United States Department of Agriculture - Agricultural Research Service - CRIS Projects No. 2072-21000-057-000-D and 2032-21220-007-000-D. Additional support was provided by the American Vineyard Foundation. X-ray microcomputed tomography beamtime was provided by the Advanced Light Source, which is supported by the Director, Office of Science, Office of Basic Energy Sciences, of the U.S. Department of Energy under Contract Number DE-AC02-05CH11231. The authors declare no conflicts of interest. Mention of trade names or commercial products in this publication is solely for the purpose of providing specific information and does not imply recommendation or endorsement by the U.S. Department of Agriculture. The U.S. Department of Agriculture prohibits discrimination in all its programs and activities on the basis of race, color, national origin, age, disability, and where applicable, sex, marital status, familial status, parental status, religion, sexual orientation, genetic information, political beliefs, reprisal, or because all or part of an individual’s income is derived from any public assistance program (not all prohibited bases apply to all programs). Persons with disabilities who require alternative means for communication of program information (Braille, large print, audiotape, etc.) should contact USDA’s TARGET Center at (202) 720-2600 (voice and TDD). To file a complaint of discrimination, write to USDA, Director, Office of Civil Rights, 1400 Independence Avenue, S.W., Washington, D.C. 20250-9410, or call (800) 795-3272 (voice) or (202) 720-6382 (TDD). USDA is an equal opportunity provider and employer. Code and data used for the production of this work are available at: https://github.com/daripp/grape-seed-morphology.
Gillispie EC, Miller KV, McElrone AJ, Block DE and Rippner DA. 2023. Red wine fermentation alters grape seed morphology and internal porosity. Am J Enol Vitic 74:0740030. DOI: 10.5344/ajev.2023.23025
By downloading and/or receiving this article, you agree to the Disclaimer of Warranties and Liability. If you do not agree to the Disclaimers, do not download and/or accept this article.
- Received April 2023.
- Accepted July 2023.
- Published online October 2023
This is an open access article distributed under the CC BY 4.0 license.
References
Vol 74 Issue 2
Jump to section
Related Articles
Cited By...
More from this TOC section
Similar Articles
