Seed Morphological Changes Associated with the Domestication of Vitis Cultivars

1 Introduction

Vitis vinifera L. is one of the most culturally and economically significant crops globally, with over 7 million hectares under cultivation [1,2]. While thousands of cultivars exist, a small group of varieties—such as the Spanish Airén and Tempranillo, or the French Cabernet Sauvignon and Chardonnay—dominates international production [3,4]. A central challenge in viticultural research remains the accurate characterization of these cultivars and the reconstruction of their complex histories. While molecular techniques have revolutionized the field, morphometric approaches provide a vital phenotypic complement for understanding the “domestication syndrome” [5].

The genus Vitis presents a complex evolutionary landscape marked by a shift from dioecy in wild populations to hermaphroditism in domesticated vines. Centuries of vegetative propagation have preserved the genetic signatures of ancient cultivars while maintaining high levels of heterozygosity. Significant progress in resolving these lineages was made through microsatellite (SSR) and single nucleotide polymorphism (SNP) markers [6,7,8,9,10,11,12,13,14,15], which identified foundational ancestors such as Gouais Blanc, Muscat of Alexandria, Pinot Noir, and Hebén. Pangenome sequencing of 3525 accessions by Dong et al. [16] defined six primary genomic haplotype groups (CG) aligned with eco-geographic origins: the Muscat group (CG3), Balkan varieties (CG4), Western European wine grapes (CG5), and Iberian varieties (CG6). Recent pangenome studies further reveal that structural variants are key drivers of this diversification [17].

Recent studies have shown that seed shape reflects these genetic groupings. The domestication process typically involves a transition from the small, rounded seeds of wild V. vinifera subsp. sylvestris toward the elongated, piriform shapes found in cultivated varieties [18]. This shift is primarily captured by two metrics: aspect ratio (representing elongation) and solidity (representing the smoothness of the seed outline). For instance, cultivars related to the Iberian ‘Hebén’ (CG6) display an intermediate morphotype, while “advanced” cultivars in the Muscat group (CG3) exhibit the most extreme elongation and lowest solidity [18,19,20].

Despite these advances, two gaps remain: the need for a standardized method to quantify seed curvature and the difficulty in distinguishing true sylvestris from “feral” individuals (domesticated escapes) using molecular data alone. Elliptic Fourier Analysis (EFA) provides a mathematical framework to address this by reconstructing average outlines across populations [21,22,23]. In this study, we refine this framework by analyzing 121 Vitis populations across four morphotype groups: wild species, and three cultivated lineages corresponding to the groupings by Dong et al. [16] (CG3, CG5, and CG6). By integrating the J-Index [18,19,20] with curvature analysis [24,25], we aim to validate a multi-parametric method for varietal identification.

We hypothesize that curvature and solidity metrics allow consistent discrimination between wild, cultivated, and feral groups. We propose that these geometric markers provide a stable taxonomic signal, conserved across diverse environments and years, which complements current genomic classifications.

2 Materials and Methods

2.1 Plant Material and Seed Sampling

This study analysed 121 seed samples representing 48 distinct varieties or species within the genus Vitis. Detailed information regarding the origin of each sample and its taxonomic classification is provided in Table 1. Of these, seven accessions correspond to Vitis species other than V. vinifera. Twenty-one samples represent wild-type populations, categorized here as sylvestris. These samples were obtained from female vines surveyed between 2003 and 2009 in riparian forests across various locations in France and Spain and subsequently maintained at the IMIDRA germplasm collection. We acknowledge the potential for genomic overlap in these populations; thus, our “sylvestris” category includes both V. vinifera subsp. sylvestris and potential feral individuals (cultivars escaped from cultivation) that exhibit wild-type morphological traits.

2.1.1 Cultivar Selection and Genetic Clustering

To facilitate comparative analysis, the 121 samples were categorized into five distinct groups based on their taxonomic status and established genetic lineages:

• Group 1: Wild Vitis species (Outgroup). This group consists of seven non-vinifera species: V. amurensis, V. berlandieri, V. californica, V. candicans, V. doaniana, V. riparia, and V. rupestris. These were included as a morphological outgroup to establish the range of seed variation across the genus Vitis and to provide a comparative baseline for the vinifera lineage.
• Group 2: Wild Vitis vinifera subsp. sylvestris. This group includes the 21 samples surveyed from riparian forests in France and Spain. As previously noted, this group represents the wild phenotypic spectrum, potentially including both ancestral sylvestris lineages and naturalized feral individuals.
• Group 3: Iberian Cultivars (CG6). Cultivars associated with chloroplast haplotype CG6 [16], representing the traditional varieties of the Iberian Peninsula.
• Group 4: Central European Cultivars (CG5). Cultivars associated with chloroplast haplotype CG5 [16], representing the core lineages of Central European viticulture.
• Group 5: Muscat Group (CG3). Cultivars associated with chloroplast haplotype CG3 [16], representing the distinct Muscat lineage characterized by unique aromatic profiles and specific seed morphotypes.

Assignment to these groups was determined either by the documented haplotype of the cultivar [16] or through established pedigree analysis showing direct descent from these foundational lineages [3,9,10,12,13,19].

2.1.2 Provenance and Harvest Standardization

To account for potential environmental and temporal variation in seed development, samples were harvested across multiple years and locations. Seeds were collected from the IMIDRA collection during the 2020 (label 1), 2024 (2), and 2025 (3) seasons. Additional samples were harvested in 2025 from three distinct Spanish viticultural sites to broaden the dataset’s environmental representation: Abadía Retuerta (Sardón de Duero, Valladolid), Jardín Ampelográfico de Valbusenda (Toro, Zamora), and Bodegas Ayuso (Villarrobledo, Albacete). The specific harvest site and year for each sample are codified in the legend of Table 1.

2.2 Photography

Images were captured with a Sony α5100 24-megapixel camera (Sony Corp., Tokyo, Japan) equipped with an AF-S Micro NIKKOR 60 mm f/2.8G ED lens (Nikon Corp., Tokyo, Japan). To minimize geometric distortion and ensure maximum reproducibility, the focal distance of the objective was held constant at an object distance of 17 cm, with all specimens positioned strictly in the center of the frame. Each seed was oriented with its ventral side facing the objective. The longitudinal axis—defined by the line connecting the apex and the stalk base—was precisely aligned with the vertical Y-axis of the image field. This standardized orientation is a critical prerequisite to ensure consistency in subsequent elliptical Fourier descriptors and curvature landmark analyses. Camera settings were configured manually to optimize image quality and depth of field, utilizing a shutter speed of 1/5 s, an aperture of f/18, and an ISO sensitivity of 100. Lighting conditions were maintained between 800 and 1200 lumens with a color temperature of 6400 K. No optical or digital noise-reduction filters were applied during or after image acquisition, as the low ISO values (≤400) guaranteed negligible electronic noise. Furthermore, chromatic correction was unnecessary since the original images were computationally binarized into high-contrast black contours for shape analysis. The raw image dataset is publicly available repository via Zenodo (see Supplementary Materials).

2.3 General Morphological Measurements

2.3.1 Seed Morphometric Measurements

Quantitative data, including Area (A), Aspect Ratio (AR), and Solidity (S), were obtained for 20–30 seeds per cultivar using ImageJ v.1.54h [26].

Aspect Ratio (AR) is the ratio of the major axis (L) to the minor axis (W) of the image: A R = L W

Solidity (S) is the ratio of the seed area (A) to the area of its convex hull (Ahull). It measures the “smoothness” of the perimeter: S = A A h u l l

2.3.2 Spatial Calibration and Image Processing

To ensure accuracy in measurements, spatial calibration was performed for each image set:

• 1.Scaling: A physical ruler included in each high-resolution image served as the reference. Using the “Set Scale” function under the “Analyze” menu, pixel distances were converted into millimeters (mm). For example, a mean area of 42,320 pixels for the cultivar Hebén (2020) corresponds to a calibrated area of 21 mm2.
• 2.Binarization: Images were converted to 8-bit grayscale format. A global threshold was manually adjusted to optimize the contrast between the seed silhouette and the background, ensuring a clean extraction of the seed perimeter.
• 3.Automated Extraction: Morphological parameters were recorded using the “Analyze Particles” function, which automatically calculates geometric indices based on the binarized outlines.

2.3.3 Geometric Indices

• Aspect Ratio (AR): Calculated as the ratio of the major axis to the minor axis of the fitted ellipse, representing the degree of elongation.
• Solidity (S): Solidity is a measure of the geometric “density” or convexity of a plane figure. It is defined as the ratio of the actual area of the seed (A) to the area of its convex hull (the smallest convex polygon that entirely encloses the object) [27]. A perfectly convex shape has a Solidity of 1. In this study, Solidity values are expressed as ×1000 to facilitate statistical comparison of fine-scale marginal variations.

2.4 Fourier Equations and Extraction of Fourier Coefficients from the Images with Momocs

Seed outlines are treated here as closed-plane curves, and, as such, may be approximated using sums of trigonometric functions as truncated Fourier expansions [28,29]. When the parametric components are piecewise linear functions x t = ∑ i = 1 K ∆ x i , y t = ∑ i = 1 K ∆ y i , these expansions are: XNt=A0+∑n=1Nancos2nπtT+bnsin2nπtT,(1) YNt=C0+∑n=1Ncncos2nπtT+dnsin2nπtT,(2) where A 0 and C 0 represent the centroid coordinates, and each addend corresponds to an ellipse (the n-th harmonic) determined by the coefficients a n , b n , c n , d n , which are calculated as:

The coefficients a n , b n , c n , d n of EFT equations can be extracted automatically with the software Momocs [21]. The number N of harmonics determines the degree of precision in the adjustment of the closed curve to the image outline and it may vary depending on the objectives of the study as well as on the characteristics of the image under analysis. While the low-order coefficients ( a i , b i , c i , d i , i = 1 , 2 ) define the basic structure of the curves, the higher-order coefficients ( a i , b i , c i , d i , i > 4 ) are responsible for surface details. In addition, b and c coefficients determine the symmetry of the curve. Curves with bilateral symmetry may have b i , = d i = 0 , i = 1,2 , …

Images (JPG) containing 20–30 seeds were converted into .TPS files, the coordinates for seed outline extracted, and the corresponding coefficients were obtained following the protocol outlined in Momocs v1.4.0 [21]. The coefficients were applied to Mathematica equations to determine the average outlines (Aos) relative to each of the seed populations. In the parametric equations representing the curves, coefficients belonging to N = 6 harmonics were considered for curvature analysis. The values a 1 = 1 , b 1 = c 1 = 0 were fixed, thus leaving a total of 6 × 4 − 3 = 21, varying coefficients. Equations with 21 varying coefficients defined the Aos that served to calculate curvatures. Although it was estimated that N = 7 accounts for 95% of the total of shape variation in Vitis [22,23], and values over it resolve details of the outlines in particular cultivars, for curvature analysis N = 6 results in 10 points well conserved in most cultivars and varieties simplifying the comparisons. In the models N is variable between 7 and 9, giving 4N − 3 coefficients.

2.5 Curvature Analysis

Seed outlines were reconstructed using Elliptic Fourier Descriptors (EFD). Curvature analysis was performed on average outlines derived from Fourier coefficients calculated for six harmonics. This level of approximation was selected to achieve an optimal equilibrium between morphological fidelity and computational simplicity. While increasing the number of harmonics improves the resolution of fine-scale irregularities, it also generates a higher density of curvature points that complicates direct inter-varietal comparisons. At six harmonics, the fundamental geometric structure of the Vitis seed—characterized by ten primary curvature extrema—is consistently captured across all accessions, facilitating a standardized comparative framework (see Fig. 1 and Supplementary Materials).

To simplify the geometric model and account for the bilateral symmetry of the seed, the ten detected curvature points were reduced to six representative values:

• 1.Unique Points: Points 1 (Apex) and 6 (Base) are unique landmarks. Point 1 consistently represents the maximum positive curvature value across all samples.
• 2.Symmetric Pairs: The remaining eight points were treated as four bilateral pairs. Average values were calculated for points 2/2′, 3/3′, 4/4′, and 5/5′, and assigned as values 2, 3, 4, and 5, respectively.
• 3.Minimum Curvature: Value 2 (the mean of the 2/2′ pair) typically yields a negative curvature, representing the minimum curvature (inflection zones) of each outline.

The final dataset, comprising the six curvature values, average curvature and curvature ratio (Max to average) and the three morphological indices (A, AR, S), is detailed in Table S1.

Figure 1: Left: Seed outline resulting from the representation of a Fourier equation with 6 harmonics. Right: 10 notable points of curvature conserved in the cultivars.

2.6 Models and J-Index Measurements

The J-Index was employed to quantify the percentage of geometric similarity between a seed’s digital outline and a reference model. This index provides a robust measure of how closely an individual accession adheres to a generalized cultivar or population “archetype.”

2.6.1 Model Definition and Standardization

Reference models were created by calculating the arithmetic mean of the Fourier coefficients (an, bn, cn, dn) from all populations within a group (e.g., the “Hebén model”). Reference models were developed at two hierarchical levels using Elliptic Fourier Descriptors (EFDs):

• Population Models: For each sample, a specific geometric model was generated using the corresponding Elliptic Fourier equations, utilizing between 6 and 9 harmonics to ensure an accurate representation of the mean population shape.
• Cultivar Models: For cultivars represented by multiple accessions, models were constructed by calculating the arithmetic mean of the Fourier coefficients (an, bn, cn, dn) from all constituent populations. This ensured that the resulting “consensus” outline represented the centroid of the cultivar’s morphospace.

To ensure consistency during image processing, seed orientation was standardized by aligning the longitudinal axis (the line connecting the apex and the base) with the vertical Y-axis of the image field.

2.6.2 Image Processing and Calculation Protocol

The comparison between seed outlines and reference models followed a standardized digital protocol to ensure maximum alignment (Fig. 2):

• •Superimposition: Using Corel PhotoPaint v.24.5 (Corel Corporation, Ottawa, ON, Canada), reference models were superimposed onto the individual seed outlines. Unlike standard rigid-body transformations, this protocol permitted adjustments in the aspect ratio to optimize the fit between the model and the sample, focusing the analysis on the underlying geometric curvature rather than absolute dimensions.
• •Area Measurement: Two distinct files were generated for each superimposition: one with the model perimeter in black and one in white. Using ImageJ v.1.53 [26], two area values were recorded:
  • ○Total Area (T): The total surface occupied by the union of the seed and the model (logical OR operation).
  • ○Shared Area (S): The surface area where the seed and the model overlap (logical AND operation).
• •The J-Index is a measure of percentage similarity between a seed’s outline and a reference model. It is not a black-box metric but a calculation of area overlap. The J-Index is defined as:J-Index=ST×100,where a score of 100 indicates a perfect geometric match.

Figure 2: Schematic representation of the protocol to measure J-Index (the percent of similarity between a seed and a given model). From left to right: The model, a seed, region of the seed (gray) with the superimposed model showing total area (T), and finally the image in red representing the area shared between the image and the model (S).

2.6.3 Methodological Reliability and Reproducibility

To address inter-operator reproducibility, a subset of 100 seeds was processed independently by two different researchers. The resulting J-Index values showed a high correlation (R2 > 0.96), indicating that the manual alignment protocol in Corel PhotoPaint is robust and highly reproducible. Intra-sample repeatability was assessed by measuring a single population three times on different days, resulting in a coefficient of variation (CV) of less than 2%.

All raw images and superimposition files are archived and accessible via Zenodo (see Supplementary Materials).

2.7 Methodological Pipeline

The workflow followed a four-step sequence:

• 1.Image Capture: Standardized orientation (ventral view, vertical axis).
• 2.Pre-processing: Binarization and EFA coefficient extraction via Momocs.
• 3.Shape Quantification: Calculation of curvature points (P1–P6) and the J-Index via area overlap.
• 4.Statistical Integration: Multivariate analysis (PCA) to distinguish between groups

2.8 Statistics

The dataset for each seed population comprised eleven quantitative variables (see Table S1). Eight variables were derived from the reconstructed average outlines: curvature values at six specific points (P1–P6), the average curvature, and the curvature ratio (maximum to average curvature). The remaining three variables—Area (A), Aspect Ratio (AR), and Solidity (S)—represent the mean values calculated from the individual measurements of 20–30 seeds per population.

All statistical analyses were performed using IBM SPSS Statistics v.28 (SPSS Inc., Chicago, IL, USA) and R v.4.2 [30]. The normality of the data was assessed using the Shapiro-Wilk test; as several variables did not follow a Gaussian distribution, non-parametric methods were employed for all group comparisons.

For analyses involving three or more groups (e.g., comparing the five classification groups), the Kruskal–Wallis H test was used. Significant results were further investigated using the Campbell and Skillings stepwise comparison procedure to identify specific inter-group differences while controlling for Type I error. For comparisons between two independent populations, the Mann–Whitney U test was applied. A Principal Component Analysis (PCA) was conducted in R to visualize the morphometric space and identify the primary variables driving the differentiation between wild and cultivated accessions. Effect sizes for Kruskal–Wallis tests were estimated using epsilon-squared (η2) calculated as: η2=H−k+1n−k where H is the Chi-Square value, K is the number of groups and n the total number of samples in the comparison.

3 Results

3.1 Variations in Curvature in the Seed Outlines

Vitis seed images typically present a prominent apical pole (stalk) and a broader basal pole, resulting in a modified elliptical outline (Fig. 3).

Figure 3: Left: Image of a Vitis seed of cultivar Airén (Bodegas Ayuso); Right: Curve from Fourier equation with 8 harmonics. P1: Apex with maximum curvature. P2: Point of negative curvature. P3, P4 and P5 curvature points at the lower lateral surface. P6: Curvature point at the basis.

Reconstructing these outlines using Elliptic Fourier equations with six harmonics captured ten geometric landmarks, reduced to six primary curvature points (P1–P6) for analysis.

The mean values and measures of dispersion across 121 samples are summarized in Table 2. The highest conservation was observed in average curvature (CV = 6.6%) and P5 (CV = 12.0%). In contrast, P6 (basis) and P4 showed the greatest inter-varietal variability (CV > 40%). Cultivars representing the extreme ends of the morphological spectrum are highlighted in Fig. 4. Notably, P2 (negative curvature) reached its minimum absolute value in V. amurensis (−0.02) and its maximum in ‘Albillo Mayor’ (−2.45), representing a critical landmark for domestication tracking.

Figure 4: Outlines of the samples having the highest (above) and lowest (below) values of (from left to right) curvature at points P1 to P6, average curvature and ratio maximum to average curvature.

3.2 Factor Reduction by PCA

Principal Component Analysis (PCA) was performed on the eleven morphometric variables (seed size, aspect ratio, solidity, and curvature metrics) across all 121 samples. The analysis identified three principal components accounting for 79% of the total variance.

• PC1 (36.6%): Positively correlated with P2 (negative curvature), solidity, and P4; inversely related to average curvature and P3.
• PC2 (28.4%): Driven by P1 (apex curvature), curvature ratio, area, and P5; inversely related to P6.
• PC3 (14.0%): Associated primarily with aspect ratio and P6 (Table 3).

PERMANOVA results confirmed significant differences between the centroids of the pre-defined groups (F = 14.17, p = 0.001). However, low silhouette values (ranging from −0.015 for Muscats to 0.124 for wild species) indicate a phenotypic continuum with substantial morphological overlap between groups.

The PCA biplot (Fig. 5) visualizes the distribution of these groups across four quadrants (Q1–Q4). Q4 is largely occupied by wild species and sylvestris accessions, characterized by higher solidity and P2 values. Domesticated cultivars are distributed across the remaining quadrants: the Hebén group (CG6) predominates in Q1, Muscat types (CG3) in Q2, and the Traminer lineage (CG5) in Q3. Sylvestris samples exhibit an intermediate distribution, overlapping with the Hebén and Traminer groups but notably absent from the Muscat cluster.

While the significant differences in centroids confirm distinct morphological ‘signatures’ for each group, low silhouette values confirm a high degree of overlap—consistent with a common genomic background, and a large variation in the groups (phenotypic continuum). The silhouette values oscillated between −0.015 (Muscat group) and 0.124 (Vitis no vinifera group), thus indicating low separation between the groups. The ellipses of groups sylvestris, Traminer and Hebén are elongated indicating heterogeneity among the samples in these groups. Neverthelelss, there were notable differences among the cultivars offering the possibility to define morphotypes. The samples of Hebén 2020 and 2024 are very close to each other and, in their proximity, there is a good representation of cultivars in the progeny of Hebén (Miguel de Arco, Merseguera, Montúa, Plant Fina, Quigat, Señá, Trepat, Verdejo de Salamanca and Xarello).

Figure 5: Biplot with the Results of PCA. Dark blue: group of Vitis species other than V. vinifera; light blue: sylvestris; red: Hebén group; purple: muscat. Green: Traminer group. PC1 (Dim1) is related to P2, solidity and P4, and inversely related to average curvature and P3. PC2 (Dim2) is related to P1, curvature ratio, area and P5, and inversely related to P6.

The progeny of Savagnin (Traminer) expands to the left in the lower part of the diagram in a region occupied by Chenin Blanc, Gewuertztraminer, Godello, Alfrocheiro, Riesling, Mencía and Pinot Noir, having all these cultivars a related pedigree. Verdejo Blanco, in the progeny of Savagnin and Castellana Blanca, expands to the upper left part of the diagram. In the region between Hebén and Traminer-related cultivars there is a concentration of Muscat-type cultivars, sharing parent-offspring relationships between them, such as Alphonse Lavallee, Cardinal, Imperial, Muscat Hamburg, Muscat a Petits Grains. Afus Ali and Aledo are on top of the diagram with high values of P1 and average curvature.

Those results justify the use of ANOVA to identify specific canalized traits—such as aspect ratio related to seed elongation, solidity or particular variations in curvature points.

3.3 Comparison between Groups

Significant differences were identified across the groups using Kruskal–Wallis tests (Table 4).

• Group of non-vinifera species (Group 1): Characterized by the highest solidity (968.0), lowest aspect ratio (1.36), and lowest average curvature.
• Iberian/Hebén group—CG6 (Group 3): Displayed intermediate solidity (954.6) and P2 values.
• Groups of Traminer (CG5; Group 4): and Muscats (CG3; Group 5): Exhibited the lowest solidity (946.5 and 945.3, respectively) and the highest absolute values for negative curvature (P2).
• Group of Muscats (Group 5): Showed significantly larger seed area (20.1 mm2) compared to other groups (p < 0.05).

Effect sizes were large for all magnitudes except area (moderate).

3.4 Analysis by J-Index

The J-Index quantified the morphological fit between seed samples and reference models for Cabernet Sauvignon and Hebén (Fig. 6).

Figure 6: Models for Hebén and Cabernet Sauvignon were made with the average outlines of three and four populations respectively for each cultivar.

• Cabernet Sauvignon Model: Testing against multiple cultivars revealed a significantly higher J-Index for Cabernet Sauvignon (92.4%) compared to Garnacha Tinta (89.9%) and Gewuerztraminer (89.8%) (Table 5).

• Hebén Model: This model successfully identified Hebén (92.1%) and its direct progeny, Mollar Cano (92.1%), showing no significant difference between the two. However, it significantly discriminated against the more morphologically distant Muscat Hamburg (88.6%) and Gewuerztraminer (90.9%) (Table 6).

4 Discussion

4.1 Morphometric Indicators of the “Wild Syndrome” and Domestication Trajectories

Discriminating between wild and domesticated Vitis seeds has been a central objective of viticultural research for decades [31,32,33,34,35,36]. The transition from traditional measurements to automated image analysis now allows for a high-resolution characterization of both modern cultivars and fossil remains [23,37,38,39]. Our results reinforce the concept of the “wild syndrome,” characterized by rounded seeds with short stalks and high solidity [22,35,36]. A key finding of this study is the geometric relationship between solidity and negative curvature across Vitis species and cultivars. Because solidity represents the ratio of a shape’s area to its convex hull, the presence of negative curvature (inflection zones) at the stalk-chalaza intersection directly reduces solidity values. The observed morphometric gradient—extending from the high solidity and minimal negative curvature of Vitis species outgroups to the low-solidity, elongated forms of the Muscat and Savagnin groups—suggests a progressive shift in seed geometry. While accessions within the Hebén lineage (CG6) occupy an intermediate morphospace, these results should be interpreted as morphometric tendencies rather than definitive taxonomic classifications. These “transitional” patterns likely reflect cultivars that have retained ancestral morphological features while undergoing different selective pressures than more recently bred varieties.

4.2 Interpreting the Phenotypic Continuum: Wild vs. Feral Populations

A significant portion of our sylvestris samples were distributed across the PCA plot, overlapping with both non-vinifera species and domesticated cultivars. This high degree of overlap—evidenced by low silhouette values—confirms a phenotypic continuum rather than discrete, isolated clusters. While some populations may represent true relictual V. vinifera subsp. sylvestris, the clustering of certain “wild” samples within domesticated quadrants (Q1 and Q3) suggests the presence of “feral” plants (domesticated escapes) or wild-domestic hybrids [40,41,42]. Because true sylvestris is strictly dioecious, the morphological proximity of some wild samples to hermaphroditic cultivars supports the need for a re-evaluation of European wild Vitis populations [43,44]. However, these patterns are plausible morphometric inferences and require independent molecular validation. High-throughput sequencing is necessary to confirm whether these similarities stem from shared ancestry, recent introgression, or feralization.

4.3 Morphotypes in Pedigree and Haplotype Context

The morphometric groups defined here show a notable correspondence with the molecularly defined haplotypes (CG3, CG5, CG6) established by Dong et al. [16]. The relative consistency of the Hebén (CG6) and Savagnin (CG5) progeny clusters suggests that seed geometry can serve as a phenotypic complement to pedigree analysis. For instance, the proximity of ‘Mollar Cano’ to the ‘Hebén’ model aligns with its documented parentage [12,13]. Conversely, instances where progeny deviate from the maternal model, such as ‘Sabro’ or ‘Pedro Ximénez’, likely reflect the geometric influence of an unknown paternal parent or transgressive segregation.

Furthermore, the J-Index analysis of Hebén and Cabernet Sauvignon highlights the stability of these “morphotypes” across diverse geographic origins. This stability suggests that while environmental factors (such as hot summers or water stress) may alter seed size or aspect ratio [45,46,47], the underlying geometric “blueprint” defined by curvature and solidity remains relatively conserved during long periods of vegetative reproduction [48].

4.4 Limitations and Future Directions

Despite the clear trends identified, the substantial overlap in the multivariate space remains a primary limitation. This phenotypic continuum suggests that seed shape is subject to significant variation, likely driven by the complex maternal nature of seed tissues and environmental plasticity. Our study highlights morphometric affinities, but these should not be used as a standalone proxy for genetic identity.

Future research should integrate these geometric markers with independent molecular datasets to isolate the specific genetic factors responsible for seed curvature and solidity. By integrating J-Index modeling with curvature analysis, we provide a non-invasive tool that complements molecular data. This approach offers powerful insights into the phenotypic boundaries of Vitis, provided it is used within a multidisciplinary framework to decode the evolutionary history of the grapevine.