The study was conducted at the Research Institute of Tropical Roots and Tuber Crops (INIVIT) in Santo Domingo, Villa Clara, Cuba (22°35´00´´N, 80°14’18´´W; 50 m above sea level), during 2021 and 2024. The site has a calcareous brown soil (Hernández et al., 2015). Meteorological data were recorded at the automatic meteorological station belonging to the institute (national meteorological network code: 78326, data: http://www.insmet.cu).

A total of 731 sweetpotato accessions were characterized, of which 387 were native, 146 foreign, and 198 were improved ( Table 1 ).

Planting was conducted on October 10, 2021, establishing for each sweetpotato genotype to be characterized, a plot of two rows two meters long ( Figure 1 ), for an effective area of 3.6 m², using a planting distance of 0.9 x 0.30 m, with 14 total plants per plot, and a distance between plots of two meters (on both sides). The type of irrigation used was sprinkler, with a weekly frequency, at a net partial norm of 250 m³ ha^-1. No chemical or organic fertilization, nor chemical or biological pesticides were applied. The agronomic management used was that recommended in the Technical Instructions for this crop, proposed by INIVIT (2012).

A morphological characterization was performed based on 8 qualitative (nominal and ordinal) varietal descriptors ( Table 2 ). Additionally, sweetpotato genotypes were characterized using morphometric traits with 16 continuous quantitative variables ( Table 3 ), along with an agronomic evaluation of resistance/susceptibility to C. formicarius.

All morphological traits were described at 90 days after planting (DAP). Quantitative descriptors were recorded from the means obtained from 10 plants per accession, while qualitative descriptors were derived from the average expression of the trait observed in the section located at the center of the main stem.

Of the 16 morphometric variables used, 10 were focused on the shape, dimensions, and color of storage roots, and six on the shape and dimensions of leaves. Samples were individually photographed using a Canon EOS 600D camera (Fukushima Canon Inc., Fukushima, Japan). For root color quantification: washed roots were gently patted dry for skin color measurements to ensure surface consistency, while flesh color was assessed immediately after making fresh transverse cuts to prevent oxidation effects. Three measurement points were recorded for both skin and flesh on each of three representative roots per accession.

Measurements were conducted under controlled environmental conditions: temperature (20 ± 2°C), relative humidity (75 ± 5%), and illumination (500 lx). For shape quantification of roots and leaves, we used the professional digital image analysis software ImageJ version 1.46 (NIH, USA), programmed in Java. The color space employed was the CIE 1976 L*a*b* system from the International Commission on Illumination, where: L* = lightness, a* = red/green coordinates (+a indicates red, -a indicates green), and b* = yellow/blue coordinates (+b indicates yellow, -b indicates blue).

In August 2022 (10 months after planting), we conducted an agronomic evaluation of resistance/susceptibility to C. formicarius. All roots were harvested by accession and collected individually. The infestation percentage was determined per cultivar:

Scale:

Following the identification of putative resistant accessions from the germplasm collection, experimental trials were conducted with these selections. Plantings were established in September 2022 (dry season) across two spatially separated fields (1,500 m apart) within INIVIT’s experimental station. The trials employed Talekar’s (1982) methodology, featuring. Two border rows (0.90 m wide, spaced 2.70 m apart) planted two months prior to the test genotypes using the susceptible cultivar ‘CEMSA 78-354’ to establish uniform pest pressure. Two central rows maintained as buffer zones. Following rotary tillage of central rows, test genotypes were planted in two row plots (0.90 m × 5 m x 2 = 9 m²) with three replicates per genotype. This design positioned each genotype equidistant from infested borders, ensuring consistent weevil exposure. No chemical or biological insecticides were applied throughout the trial period.

Harvest occurred at 150 days after planting (DAP), with all storage roots collected and evaluated individually by genotype. Infestation percentages were calculated using standardized protocols consistent with previous experiments. Dual trait selection was performed with independent thresholds: minimum yield of 3 t ha⁻¹ and maximum infestation of 10%.

The most promising resistant genotypes identified in the 2022 trials were reevaluated in March 2023 using identical methodology. Final harvest occurred at 150 DAP in August 2023, maintaining consistent evaluation parameters across both trial periods.

In January, February, and March 2024 (an experiment conducted three consecutive times), the sweetpotato genotypes with the highest reported resistance to the weevil were evaluated, along with a susceptible control (CEMSA 78-354). The experiments were conducted under controlled laboratory conditions to determine the level of infestation and the subsequent emergence of adult weevils.

The trials were carried out in a square wooden box (50 × 50 × 30 cm) covered with anti-aphid mesh to prevent external contamination while allowing adequate ventilation. Inside the box, sweetpotato roots of each genotype were randomly arranged around a central inoculum source composed of heavily infested sweetpotatoes (previously confirmed to have the presence of C. formicarius larvae and adults) ( Figure 2 ). This arrangement ensured homogeneous exposure of the genotypes to insect attack.

After one month, the percentage of flesh damage in each sweetpotato was quantified by image analysis using ImageJ version 1.46 software (NIH, USA). To do this, cross sections of the roots were made, and each slice was photographed under standardized conditions. The total flesh area and the area damaged by C. formicarius were measured, and the percentage of damage was calculated for each genotype.

After 15 days of initial colonization in the infestation chambers, roots were individually transferred to ventilated Magenta containers (GA-7 vessels, 77×77×97 mm) to monitor adult emergence. This protocol prevented cross genotype contamination. Weevil counts were conducted daily from day 10 to day 14 post-transfer, with a final recording on day 24, enabling quantification of total emerged adults per genotype, temporal emergence patterns, and developmental time from infestation to adult emergence.

To characterize the accessions while simultaneously considering multiple traits and their interrelationships, we applied interdependence multivariate methods. The analytical procedure involved two main steps: an ordination approach (principal component analysis, PCA) followed by a classification method (cluster analysis).

For continuous quantitative variables, we performed PCA to identify associations between descriptors and determine whether they contributed similarly or oppositely to the observed variation. The eigenvector matrix was interpreted to assess variable contributions. A bar plot was constructed to display the absolute and cumulative proportion of variance (Y-axis) explained by each principal component (X-axis). Correlations between the original variables and selected principal components were calculated using Pla’s (1986) formula. Additionally, variable correlations were projected onto the first two principal axes (PC1 and PC2) to visualize their relationships.

To classify accessions into relatively homogeneous groups based on shared characteristics, we conducted a cluster analysis combined with a heatmap. For multi-state (nominal and ordinal) morphological data, Gower’s (1971) metric distance (dissimilarity) was applied, while Euclidean distance (ED) was used for continuous morphometric variables. An agglomerative hierarchical clustering method (multi-level grouping) was employed, with results presented as a dendrogram.

Analyses were performed in R version 4.0.0 using various packages for statistical processing and data visualization. Graphical representation included bar charts and scatter plots using the ggplot2 package, correlation analysis using networks implemented in the corrr package, complemented by correlation network visualizations using qgraph, chord diagrams to visualize associations between variables using the circlize package, and combined representations of dendrograms and heatmaps generated with ComplexHeatmap. Data were standardized using the scale() function when necessary. All figures were exported in vector format (600 dpi).

Out of the 731 accessions, 66 (9%) did not produce tuberous roots. The skin color histogram reveals a clear predominance of red (376 accessions) and yellow (190 accessions) values. The low frequency of white (8 accessions) and purple (55 accessions) skins suggests that these traits are minority features in the collection ( Figure 3A ). The scatter plot generated from the CIELab color space coordinates a* (green-red) and b* (blue-yellow) for the skin of the analyzed accessions shows a heterogeneous distribution, indicating broad variability in the chromatic characteristics of the evaluated samples. The points in the graph exhibit wide dispersion, with a* values ranging approximately from –5 to 35 and b* values from 0 to 40. This variability suggests that the accessions display a diverse range of skin tones, from whitish hues (negative a* values) to reddish tones (positive a* values), as well as from purplish (negative b* values) to yellowish (high b* values) ( Figure 3B ).

The flesh color distribution shows a notable frequency of yellow (260 accessions), cream (181 accessions), and white (172 accessions), followed by orange (33 accessions) and purple (20 accessions) in smaller proportions ( Figure 3C ). The CIELab color space scatter plot for flesh color reveals a concentration of accessions within the a* range of –5 to 5 and the b* range of 0 to 40, suggesting that most samples exhibit intermediate tones with a slight tendency toward whitish and yellowish colors ( Figure 3D ).

Among the 665 accessions that produced tuberous roots, 484 (72.78%) exhibited skin surface defects (longitudinal fissures and horizontal constrictions), while 181 (27.22%) were defect-free. The Circularity (C) vs. Aspect Ratio (AR) distribution plot reveals significant morphological patterns in the analyzed sweetpotato germplasm. The density distribution shows a predominant concentration in high circularity values (0.5–0.7) and aspect ratio values (2.5–3.5). Low-density areas in regions of low circularity (< 0.4) and high aspect ratio (> 4) indicate a lower frequency of roots with pronounced elongated shapes. Defect-free roots were scattered without a defined pattern ( Figures 4A, B ).

The 2D density plot uses a color gradient to highlight that the highest-density zones (dark tones) correspond to intermediate to high values of both traits, while peripheral areas (yellow tones) represent less common morphologies ( Figure 4C ).

The analysis of the scatter plot between Leaf Circularity and Aspect Ratio in the sweetpotato germplasm revealed no positive correlation between these variables. However, the observed data dispersion indicates significant variability in leaf morphology among accessions, suggesting differences between these parameters and leaf contour complexity. The density distribution shows a predominant concentration in high circularity values (0.6–0.8) and aspect ratio values (1–1.3), implying that most accessions have relatively compact leaves with well-defined contours. Several outlier points were identified, corresponding to accessions with: highly lobed leaves (low circularity:< 0.2) or exceptionally compact leaves (high circularity: > 0.7) ( Figure 5 ).

The network graph obtained reveals interesting patterns of interrelationship among phenotypic traits in the sweetpotato germplasm. The network structure displays clear modularity, with traits grouping according to their nature, showing four main modules: tuberous root skin color traits, tuberous root flesh color traits, leaf morphometric traits, and root morphometric traits. The strongest connections (thicker edges, r ≥ 0.5 and r ≤ -0.5) are observed between shape and color parameters within the same tissue, particularly among the L*, a*, and b* coordinates, suggesting highly coordinated pigmentation patterns. Notably, root morphometric traits (circularity, aspect ratio, roundness, and solidity) exhibit high interconnectivity, possibly indicating a shared genetic basis or strong ontogenetic integration during storage organ development. However, the variable of susceptibility to the weevil did not show significant correlations with any of the other morphological or morphometric characters analyzed. The network also highlights a significant bridge between tuberous root defects and root solidity ( Figure 6 ).

The results demonstrate a clear association between skin color and flesh color in sweetpotato tuberous roots. The most frequent combination was red skin with yellow flesh (160 cases), followed by red skin with cream flesh (105 cases), suggesting that genotypes with red skin tend to exhibit yellow or cream flesh. In contrast, purple skin showed exclusivity with purple flesh (20 cases) ( Figure 7A ).

A higher incidence of defects was observed in roots with red skin (268 defectives vs. 108 defect-free) and yellow skin (146 defectives vs. 44 defect-free). Conversely, roots with white skin showed the lowest defect frequency (8 cases) ( Figure 7B ). Roots with yellow flesh had the highest defect count (186 cases), while those with purple flesh showed the lowest incidence (7 cases). Notably, orange flesh though less frequent exhibited a substantial proportion of defects (28 cases), indicating a non-linear relationship between color and defect susceptibility ( Figure 7C ).

The variance associated with each principal component differed and decreased sequentially. The first component explained 17.5% of total variance and the second accounted for 16%. Together, the first four components explained 57.3% of the cumulative variance ( Figure 8A ). Leaf morphometric variables (Circularity and Solidity) showed strong positive correlations with each other and were the primary contributors to Component 1 variance. In contrast, tuberous root morphometric variables (Circularity and Roundness) were highly correlated and represented the main variance contributors for Component 2. The weevil susceptibility variable, located near the origin, indicated little to no relationship with either component ( Figure 8B ).

Sweetpotato accessions clustered into three distinct groups in the factorial space based on their response to C. formicarius: susceptible (532 accessions, 80%), tolerant (90 accessions, 13.5%), and resistant (43 accessions, 6.5%). The overwhelming predominance of susceptible genotypes reflects the high vulnerability of the evaluated germplasm. While scarce, the resistant accessions represent valuable genetic resources. Notably, these resistant accessions did not form defined clusters in the biplot but were instead dispersed throughout the principal components space. This pattern suggests that resistance is not associated with the evaluated morphological traits ( Figure 8C ).

Of the 43 accessions that showed resistance, 36 are Cuban, 1 from Japan (Ja-pan-1-2016), 2 from China (YAN SHU-1, YUI-BEI-BUINI), 2 from Nigeria (KOKEINO, IITA-TIS-8250), 1 from Mexico (Mexico-20) and 1 from the United States (Excel).

The combined heatmap and dendrogram analysis revealed interesting patterns regarding the susceptibility, tolerance, and resistance of sweetpotato accessions to weevils. Resistant accessions tended to exhibit lighter skin coloration (high L*skin values and low a*skin values) and less pigmented flesh (low b*flesh values). Many of these resistant accessions also showed greater root solidity. In contrast, tolerant accessions occupied an intermediate position in the dendrogram, sharing characteristics with both resistant and susceptible groups. These tolerant accessions typically dis-played moderate skin pigmentation (intermediate a*skin values) and mild surface defects. The susceptible accessions clustered in distinct branches of the dendrogram, often associated with darker skins (high a*skin values) and intensely pigmented flesh (high b*flesh values). Notably, these susceptible accessions frequently presented more pronounced surface defects, which likely facilitates infestation. A particularly relevant finding was the apparent correlation between stem color and weevil susceptibility. Accessions with purple stems tended to be more susceptible. Regarding the relationship between susceptibility to the weevil and the morphometric characteristics of the leaf, the analysis did not show significant associations. The parameters of circularity, aspect ratio and leaf solidity did not show clear patterns that would allow linking the morphology of the leaves with resistance or susceptibility to the insect. ( Figure 9A ).

The data revealed a weak but positive correlation (r=0.17) between skin color and resistance to C. formicarius. Red skinned accessions contained the highest number of resistant individuals (25), followed by yellow (11) and purple (6) skins. However, the high proportion of susceptible individuals across all skin colors (particularly in red: 306 susceptible versus 25 resistant) suggests that this morphological trait alone is not a reliable predictor of resistance. White skinned accessions showed the lowest number of resistant individuals (1) ( Figure 9B ).

A moderate negative correlation (r=-0.38) was observed between flesh color and resistance to C. formicarius. White and yellow flesh accessions concentrated the highest number of resistant individuals (17 each), while cream and orange flesh accessions showed greater susceptibility. Particularly striking was the case of purple flesh, which presented only 2 resistant accessions ( Figure 9C ).

A moderate positive correlation (r=0.31) was found between the presence of surface defects (fissures and constrictions) and susceptibility to C. formicarius. Among susceptible accessions, 79% exhibited defects, while resistant accessions showed a similar proportion with and without defects (49% versus 51%). This pattern suggests that surface defects may result from insect damage rather than representing an intrinsic trait, though it could also indicate that malformed roots are more vulnerable to attack. The presence of 21 resistant accessions with defects raises the hypothesis that some genotypes may develop tolerance mechanisms, while others exhibit antixenosis resistance (insect avoidance) ( Figure 9D ).

From the 43 accessions initially identified as resistant in the germplasm collection screening, only 18 genotypes met the dual selection criteria (yield ≥3 t ha⁻¹ and infestation ≤10%) in the first experiment (designed to assess resistance). While accessions with outstanding combinations of yield and resistance were identified such as INIVIT B-25 (60.13 t ha⁻¹ with 9.31% infestation) most genotypes exhibiting complete resistance (0% infestation) showed moderate yields (<20 t ha⁻¹), revealing a clear trade-off between these traits. However, two exceptional genotypes exceeded 50 t ha⁻¹ with 0% infestation ( Figure 10A ).

In a second evaluation cycle, only 6 of the 18 tested genotypes met the dual selection criteria, simultaneously achieving yields above 10 t ha⁻¹ and infestation below 10%. These were classified into three groups based on performance. High yield and resistance: Highlighted by INIVIT B-25 (58.8 t ha⁻¹ with only 4.76% infestation), which emerged as the most promising genotype due to its exceptional productivity and phenotypic stability. Good yield and good resistance: Represented by INIVIT B-2022 (39.2 t ha⁻¹, 8.93%), IB Morado-16 (34.3 t ha⁻¹, 4.08%), and INIVIT B-90 (31.5 t ha⁻¹, 7.11%). Moderate yield but good resistance: comprising Catalina (23.1 t ha⁻¹, 9.09%) and V-525 (16.1 t ha⁻¹, 7.39%) ( Figure 10B ). All six resistant genotypes are of Cuban origin, five being improved varieties and one traditional (Catalina).

Highly significant differences (p< 0.05) were found in the tuberization depth among the evaluated sweetpotato genotypes (six resistant and one susceptible control), suggesting that this morphological trait may be associated with resistance to C. formicarius attack. The susceptible control (CEMSA 78-354) exhibited the shallowest tuberization depth (3.41 cm). In contrast, the genotypes V-525 (13.32 cm) and IB Morado-16 (12.50 cm) showed the highest values, which were statistically superior to the control and other intermediate genotypes. However, not all resistant genotypes followed this trend. INIVIT B-25 (4.53 cm), despite being considered resistant, did not differ significantly from the susceptible control in tuberization depth, indicating that resistance in this case may be mediated by other mechanisms ( Figure 11 ).

Sweetpotato genotypes varied significantly in susceptibility to C. formicarius (p< 0.05). The V-525 genotype, despite being considered resistant in previous findings, showed the highest infestation percentage (92.67% ± 6.13), even surpassing the susceptible control CEMSA 78-354 (89.33% ± 8.99). This contradictory finding suggests that V-525’s resistance may be associated with mechanisms other than infestation reduction, such as tuberization depth. On the other hand, INIVIT B-2022 exhibited the lowest infestation percentage (39.67% ± 13.72), standing out as the most promising genotype for direct insect resistance. The genotypes IB Morado-16 (46.67% ± 17.52) and INIVIT B-25 (53.33% ± 9.46) showed intermediate infestation levels. Catalina (61.33% ± 14.73) and INIVIT B 90 (77% ± 9.63) occupied intermediate-high positions ( Figure 12A ).

While INIVIT B-2022 demonstrated exceptional resistance with near-total suppression of adult emergence (2.0 ± 1.5 adults across replicates), the other resistant genotypes exhibited distinct temporal patterns in weevil development. The genotypes INIVIT B-25 (10.3 ± 5.2 adults) and IB Morado-16 (15.1 ± 9.4 adults) also displayed low emergence but with different dynamics: INIVIT B-25 showed a significant delay (first adults on day 12.5 ± 7.7), while IB Morado-16 had more gradual emergence (linear increase of 1.2 ± 0.3 adults/day). This suggests their resistance mechanisms may act at different stages of the insect’s life cycle (e.g., larval toxicity vs. pupation inhibition). V-525, despite its high infestation, showed intermediate emergence (28.4 ± 12.8 adults), indicating that although the insect colonizes the tubers, its development is partially affected (p< 0.05). Catalina and INIVIT B-90 showed moderate emergence (43.2 ± 22.1 and 17.3 ± 9.3 adults, respectively), while INIVIT B-90 peaked earlier (10.2 ± 4.8 adults by day 14) ( Figure 12B ).

Visual documentation of root phenotypes ( Figure 13 ) from the six most resistant genotypes reveals wide variation in flesh pigmentation (1 white, 2 yellow, 1 light orange, and 2 purple), while skin color was predominantly dark red/purple, with the exception of INIVIT B-2022 (yellow skin with orange flesh). This suggests that flesh color is not a consistent predictor of resistance. Instead, shared morphological traits such as anthocyanic skin color (5/6 genotypes), deep tuberization (5/6 genotypes), and elliptical root shape with smooth surfaces may contribute more significantly to resistance. The absence of pronounced skin defects in resistant genotypes could reduce soil surface cracking and consequently limit oviposition sites for C. formicarius. Notably, INIVIT B-2022 maintained its resistance (both in field and laboratory conditions), implying compensatory mechanisms.