Rice Regeneration in a Genebank: 21 Years of Data

Abstract

Genebanks, other than their pivotal role as diversity conservation repositories, regenerate part of their collection every year to maintain their material in optimal conditions. During regeneration cycles, morpho-physiological data are collected, contributing to the creation of large datasets that offer a valuable resource of information. In Italy, rice cultivation has been documented since the second half of the 15th century, and nowadays, Italy contributes more than 50% of the total European rice production. The ex situ collection of rice (mainly Oryza sativa L. subgroup japonica) held at the Research Center for Cereal and Industrial Crops (CREA-CI) of Vercelli is quite unique in Italy and its establishment dates back to the beginning of the 20th century. The collection is hereby presented through the analysis of 21 years of historic data, from 2001 to 2022, in 17 different locations in Northern Italy, for a total of 6592 entries, 677 genotypes analyzed and 9 phenotypic traits under investigation. An R script has been developed to analyze the dataset. The BLUEs calculation, heritability, PCA and correlation with weather data provided a comprehensive overview of the germplasm stored in the genebank. The great variability and phenotypic diversity were assessed, key aspects from the perspective of breeding programs. This work starts a re-evaluation of historic data, historic cultivars, and represents the first step toward the shift of the genebank to a bio-digital resource center.

1. Introduction

Rice (Oryza sativa L.) is one of the most important staple crops for more than half of the world’s population [1]. The details concerning the history of domestication of rice are still debated, but most researchers agree on three independent domestication events that took place about 8000 years ago. In southern China and northern India, the japonica and indica subspecies were domesticated from O. rufipogon and O. nivara, with later crossing and introgression of the two subgroups [2,3,4]. Around 3000 years ago, African farmers domesticated O. glaberrima starting from O. barthii [5].

O. sativa has been extensively studied because of its importance in agriculture, but also as a model species for cereals, due to its small diploid genome (389 Mb), whose sequence was firstly obtained for the cultivar Nipponbare and is constantly updated [6,7].

Italy is the first producer of rice in Europe [8], with 1,236,960 tons produced on 218,420 hectares in 2022 [9]; in the same year, it was the 14th exporter in the world, with 12,193 tons exported [9]. Rice cultivation has been documented in Italy since the second half of the fifteenth century, particularly in the swampy areas along the Po river [10]. Over time, it became well established in the northern region between Lombardy and Piedmont, where it became fundamental to the local economy and contributed to strongly modifying the landscape, mainly due to the extensive system of canals created to bring water from the nearby Alps, to ensure a steady supply of water for irrigation.

The first information on the germplasm used refers to a variety called Nostrale (literally “local”), probably a mixture of different ecotypes, as the most cultivated in the area in the 19th century. At the beginning of the 20th century, in order to face rice blast (Pyricularia oryzae), which caused dramatic epidemics, new varieties were introduced from Asia and subjected to mass selection, as confirmed by studies on the genetic closeness between Italian rice varieties and Northern Chinese cultivars [11]. Then, starting from 1925, crossing and selection drove the creation of new varieties, focusing on shortening the growth cycle and reducing plant height and blast susceptibility [12,13]. Rice breeding has also been influenced by the market, which has changed over time: from mainly the round rice type, breeding moved to “Long A” type with a high amylose content, for “risotto”, and in recent years, to “Long A” for parboiling and “Long B” varieties appreciated for Asian cuisine [12,13].

The Research Centre in Vercelli was founded in 1908 as “Stazione sperimentale di risicoltura” (“Rice Research Station”) and quickly became a center of excellence for rice research and breeding, being a pioneer in Italy in several fields, e.g., performing the first hybridization between two rice varieties (1925), and the first experiments of mutation through white light (1934) [10,13].

At the beginning of the 20th century, most of the research carried out in the Centre of Vercelli was based on importing rice varieties from abroad, assessing their adaptability to local conditions and eventually using them for breeding purposes. An example is represented by the varieties “Vary Lava A” and “Vary Lava 51A”, acquired in 1925 from Madagascar and extensively used for breeding in the following years. From the 1930s onwards, many successful cultivars have been developed in our Centre. Among them, Roma and Vialone Nano, still appreciated for their culinary characteristics.

The rice collection held at the CREA-CI Genebank (ITA383) consists of 701 O. sativa genotypes, mainly ssp. japonica, both Italian and international, 2 accessions of the ancestor O. rufipogon Griff., and 5 accessions of O. glaberrima Steud. Most of the accessions of O. sativa in the collection are Italian, while the most represented international varieties are American, Spanish, Portuguese, Chinese, and Egyptian. The oldest variety in the collection is Bertone: according to Giornale di Risicoltura (a journal dating back to 1909 with all the latest news and discoveries on rice obtained in the Vercelli Research Centre), it first appeared in 1829, and its cultivation was widespread in the first part of the twentieth century. Originario (also known as “Chinese Originario”) was imported from China or Japan in the early 1900s and had great success in the Vercelli area: it was the most cultivated variety until the 1960s, as stated in many technical reports, and it has been used in breeding programs for many years. Ostiglia, Lencino, and Ranghino, although less widespread than Bertone, were cultivated in Italy before 1900, representing the oldest cultivars in the collection.

O. rufipogon was proven to carry alternative alleles that modify the kernel shape, plant architecture, and yield [14] making it an interesting starting point for breeding programs. Introgression lines (ILs) of O. rufipogon in the genetic background of O. sativa cv Vialone Nano have recently been developed in our Centre and could represent valuable elements in the search for potentially useful genes and QTLs.

O. glaberrima is a good source of valuable genes that can be used to improve the resistance of O. sativa against pathogens, such as nematodes, insects, and viruses, as well as factors like drought stress [15]. This species also represents an interesting source of new alleles for salt and dehydration tolerance [16], panicle architecture [17,18], and photosynthetic activity under heat stress [19].

The maintenance and characterization of the collection are the main focus of the RGV-FAO Program, funded by the Italian Ministry of Agriculture, Food Sovereignty and Forestry (Masaf) in order to implement the International Treaty on Plant Genetic Resources for Food and Agriculture (ITPGRFA) [20]. During the regeneration cycles, the materials are characterized at the morpho-physiological level with international descriptors according to the “Standard Evaluation System for Rice” (www.irri.org) and the national guidelines for in situ, on-farm, and ex situ conservation of plant, animal, and microbial biodiversity [21], allowing the collection of a huge amount of historical data. In addition, many of the accessions have been characterized at the genomic level (genotyped), thus allowing the deep investigation of the loci responsible for important agronomic traits, e.g., resistance to biotic and abiotic stress, and for quality traits [12,22,23].

This work, through an overview of the historical data collected during years of regeneration and experiments, aims to describe the rice collection stored at CREA-CI Vercelli to infer its variability and diversity, fundamental characteristics for breeding programs that aim to overcome the climate change challenges we are doomed to face, a strategy suggested for the first time less than 10 years ago [24].

2. Materials and Methods

2.1. Materials

The collection stores a huge variability in terms of both the origin of the accessions, covering all the continents except Antarctica (

Table 1. Country of origin of the rice accessions held in the collection.

Country of Origin	Number of Accessions	Percentage
Italy	311	43.99
United States	60	8.49
Spain	53	7.50
Portugal	40	5.66
China	29	4.10
Egypt	24	3.39
Argentina	21	2.97
France	18	2.55
Japan	18	2.55
Philippines	14	1.98
Bulgaria	12	1.70
India	13	1.84
Brazil	10	1.41
Greece	10	1.41
Australia	9	1.27
Colombia	7	0.99
Hungary	6	0.85
South Korea	4	0.57
Other	28	3.96
Unknown	20	2.83

), and the historical period of their release (

Table 2. Release period of the rice accessions held in the collection.

Release Period	Number of Accessions	Percentage
Before 1900	4	0.56
1900–1927	33	4.66
1928–1950	28	3.96
1951–1970	81	11.45
1971–1990	119	16.83
1991–2005	198	28.00
2005–2022	121	17.11
Unknown	123	17.39

).

For the present work, the following parameters were analyzed: flowering time, maturity time, days between flowering and maturity (flo-mat), plant height, Thousand Kernel Weight (TKW), yield, and infection severity/susceptibility (in-field conditions) to the pathogens Pyricularia oryzae (rice blast), Magnaporthe grisea (neck rice blast), and Bipolaris oryzae (brown spot).

In total, the curated dataset comprises 6592 entries, 674 genotypes in a time span of 21 years, from 2002 to 2023. The phenotypic traits did not have the same number of data, with the flowering and maturity times being the most measured, having more than 6000 entries, while the TKW had the lowest number of datapoints (just over 2000). Almost half of the collection is represented by Italian varieties, with a good representation of cultivars originating from the USA, Spain, and China. The complete list of the rice cultivars used in this study can be found in the Supplementary Materials, Table S1.

The trial locations are 17, all in the region between Lombardy and Piedmont in Northern Italy, which is the main rice-growing area in the country (Figure 1).

2.2. Agronomic Information and Phenotypic Data Collection

The materials were sown on dry fields at a seedling rate of 450 seeds per square meter, in single plots of eight rows, 20 cm wide and 2.5 m long, with two to four replications, between April and May, and harvested between September and October, depending on the specific environmental conditions of each year. The agronomic practices followed the standard techniques of the area: permanent flooding condition, chemical control of weeds, and chemical control of pests and diseases when necessary. Fertilization was fractionated into three treatments: before sowing (24/0/29 NPK, 260 kg/ha), at tillering (24/0/29 NPK, 260 kg/ha), and at stem elongation (urea, 40 kg/ha).

The phenotypic traits analyzed were measured according to the Plant Trait Ontology [25,26]: total plant height (TO:0000207), excluding the panicle, recorded in the field at complete maturity; flowering time (TO:0002616), defined as the number of days since sowing when the panicles are fully flowering; maturity time (TO:0000933), defined as the number of days since sowing when the panicles are fully mature; Pyricularia oryzae, Bipolaris oryzae, and Magnaporthe grisea infections (all measured as 0 to 9, where 0 is no infection and 9 is severe infection); yield (in tons/hectare), measured on paddy at 14% humidity; and TKW (TO:0000533), recorded as the average weight of two 100-paddy rice samples, measured in grams and multiplied by 10.

2.3. Statistical Analyses

The analyses were performed using the R programming language, version 4.3.1 [27]. Since the data were collected over the span of 21 years, and not all the accessions had the same number of replications during the experimental trials, and the analysis was carried out under the assumption it was a longitudinal study. As previous works [28,29] suggest, a Linear Mixed Model was used with the package nlme, which employs the Laird–Ware form [30,31,32] of the mixed model shown in Equation (1):

$$Y_{ij}=X{\mathsf{\beta }}_i+\mathrm{Z}{u}_j+e_{ij}$$

(1)

Y_ij is the observed phenotype at the ith year and at the jth accession; β_i is the vector of the fixed effect coefficient for the year ith; u_j is the effect of the jth random variable (the accession, in this work also called the “genotype” or “cultivar”); e_ij is the error for the observation at the ith year and at the jth accession; and X and Z are matrixes for the effects. The equation has been performed individually for each trait under investigation.

Following the code available in Philipp et al. [28], the outliers have been pinpointed and eliminated from the dataset. The code has been slightly changed since, in this work, the R package nlme was used instead of asreml. The workflow was divided in two steps: (1) year effect and year-specific error variance were obtained from Equation (1) and used to calculate the coefficient of variation, and (2) then Equation (1) was adjusted based on the data just obtained, by assuming accession as the fixed effect and year as the random effect. The Bonferroni–Holm test was then applied to scan for outliers. The normality of the data was tested with the package nortest [33] and the Anderson–Darling test for normality. Skewness and kurtosis were calculated with the package psych [34,35].

The Best Linear Unbiased Estimates (BLUEs) of the phenotypic data, after outlier elimination, were calculated exploiting the R package polyqtlR [36].

The R package sommer [37] was used for the calculation of heritability, following Cullis’ heritability model shown in Equation (2):

$${\mathrm{H}}_{\mathrm{Cullis}}^2=1-\frac{{\mathrm{v}}_{\Delta }^{\mathrm{BLUP}}}{2{\ast \mathsf{\sigma }}_{\mathrm{g}}^2}$$

(2)

where ${\mathrm{v}}_{\Delta }^{\mathrm{BLUP}}$ is the mean variance of difference between genotypic BLUPs, while ${\mathsf{\sigma }}_{\mathrm{g}}^2$ is the genetic variance of the accession. Principal Component Analysis (PCA) was performed using the R packages factoextra and FactoMineR [38].

2.4. Acquisition and Preparation of Bioclimatic Data

Bioclimatic data were collected from ClimatologyLab (https://www.climatologylab.org, accessed on 1 February 2024). From the available datasets, TerraClimate was selected [39]. For this study, only three climatic parameters were selected: minimum temperature (Tmin, in °C), maximum temperature (Tmax, in °C), and precipitation (Ppt, in mm). The climatic data were downloaded as raster files in .TIF format.

To produce the digital map and extract the required data, the software QGIS, version 3.36 “Maidenhead” (https://qgis.org, accessed on 1 February 2024), was utilized. For this work, a project was created in which all the rasters for the selected bioclimatic variables were collected and divided into groups based on the type of variable. The ED50/UTM zone 32N (authority ID: EPSG: 23032) Coordinate Reference System (CRS) was selected for this work.

For each bioclimatic variable, one layer for each year spanning from 2002 to 2023 was loaded into the project. A total of 66 raster layers for the bioclimatic data were added to the map. Experimental sites were included in the QGIS project using a dataset in .csv format. For each site, the latitude and longitude coordinates were provided. Due to the availability of the experimental sites, but not the precise location coordinates in degrees, the latitude and longitude were set at the nearest town or village, which was used as a reference location for the projects considered in this study. After loading the dataset for the locations, a new points vector layer was added to the digital map. These points were used to collect Tmin, Tmax, and Ppt for each considered location and for each year from 2002 to 2023. Before extracting the desired data, all the loaded layers were reprojected to the project CRS, using the “Warp” option (for rasters) and “Reproject layer” (for vectors) in QGIS. The ED50/UTM zone 32 CRS was selected. The Point sampling tool plugin was used to extract the climate data from the obtained digital map. With this plugin, a GeoPackage (.gpkg) file was obtained for each bioclimatic dataset, represented as a new point vector layer on the map—one layer for each Tmax, Tmin and Ppt. These new data points contain both location data and information relative to the given parameter. The final step to obtain datasets about climate data consisted of extracting the attribute table for each Tmax, Tmin and Ppt GeoPackage file in a .csv file. For this purpose, the MMQGIS plugin was used to extract the desired data from the layers. To reduce the amount of raw data, only those relative to the growing season (from April to October) were selected for each year.

3. Results

The complete dataset, which includes more than 11,000 entries, was manually curated to eliminate obvious mistakes and select, out of the 209 different experiments, only those that could be compared to each other and were performed under standard agronomic management conditions, as described in the Materials and Methods. The final curated dataset comprises 6592 entries, with 9 phenotypic traits analyzed over a span of 21 years, from 2002 to 2023, in 17 different locations in Lombardy and Piedmont, both located in Northern Italy.

The analyses showed great variability among the different cultivars and the different traits. As can be seen in

Table 3. Descriptive statistics of the original and outlier-corrected historical data. ^a before outlier removal; ^b after outlier removal. Pyricularia oryzae is the pathogen responsible for rice blast, Magnaporthe grisea for neck rice blast and Bipolaris oryzae for brown spot. Norm.test = Anderson–Darling normality test; TKW = Thousand Kernel Weight.

Trait	N. of Years a,b	N. Total Entries a	Heritability a	N. Total Entries b	Heritability b	Norm.test (p-Value) b	Skewness b	Kurtosis b
Flowering time	18	6317	0.75	6293	0.86	2.2 × 10−16	0.64	0.92
Maturity time	19	6371	0.54	6358	0.71	2.2 × 10−16	0.09	1.43
Flowering-Maturity days	18	6194	0.46	6191	0.61	1.2 × 10−13	−0.03	0.57
Plant height	19	5531	0.85	5511	0.94	2.2 × 10−16	0.65	0.7
Magnaporthe infection	15	3902	0.23	3860	0.39	2.2 × 10−16	1.59	2.65
Bipolaris infection	15	2982	0.08	2966	0.41	2.2 × 10−16	0.7	0.43
Pyricularia infection	17	4224	0.23	4202	0.39	2.2 × 10−16	1.31	2.23
Yield	16	2393	0.77	2393	0.77	2.2 × 10−16	−0.39	−0.37
TKW	5	2283	0.89	2216	0.96	2.2 × 10−16	0.41	−0.12

(columns three and five), all the traits, except for the yield, showed some outliers, ranging from 3 datapoints excluded for flowering-maturity days up to 42 datapoints excluded for Magnaporthe grisea infection. For each trait, the heritability and number of experimental years were calculated before and after outlier removal. The Anderson–Darling normality test, as well as the calculation of skewness and kurtosis, was conducted after outlier removal.

Removing the outliers is a critical step when a huge amount of data collected over a large timeline is analyzed. Heritability was calculated for all the traits, both before and after the removal of the outliers (Table 3). In all cases (except yield), the parameter improved after filtration, especially for brown spot (B. oryzae) infection (from 0.08 to 0.41). Among all the traits analyzed, pathogen infections showed the lowest heritability, confirming how this trait is heavily influenced by both the environment, such as the weather conditions and location (some regions are more heavily affected by the diseases), and the field treatments. This is also confirmed by the results reported in Table S2 (Supplementary Materials): Bipolaris infection is shown to be influenced by some of the experimental sites but not by the experimental year.

After elimination of the outliers, the Anderson–Darling normality test carried out on the entire dataset did not suggest the normal distribution of the data (Table 3 and Figure 2, lower graphs), as confirmed by the skewness and kurtosis, with multiple peaks. The flowering and maturity time showed the same tendency to have a lower peak in the middle.

The flowering time, maturity time, and flowering-maturity days (Figure 2a–c, upper graphs) visibly fluctuated in different years, while other parameters such as the plant height (Figure 2d) were more stable during the years, as expected. This last result is confirmed by the data shown in Table S2 (Supplementary Materials) where, as for the previously described Bipolaris oryzae infection, this parameter is not affected by the years of experiment as it is by the different environments. The yield (Figure 2e) seemed to be more affected by the location (Figure S1 and Table S2) than by the year of the experiment.

Principal Component Analysis (PCA) was used to visualize the relationship among the traits and their contribution to the overall variability (Figure 3). The results were calculated using the complete dataset, after removing outliers. The reproductive stages of the plants (flowering, maturity, and flowering-maturity times) generated most of the variation in the dataset in the first two dimensions. In the first dimension, the diseases under investigation accounted for between 15% and 18% of the total variability. The PCA also includes weather information extracted for each year in each location. Since all the locations are quite close to each other (Figure 1), there are no big differences between the environment’s weather conditions, but it is possible to appreciate the great variation in the total rain, minimum and maximum temperatures during the years (Figure S2): from 2002 to 2023, a slight decrease in precipitation associated with an increase in temperature (both minimum and maximum) can be observed. There have been many peaks of high temperature; for example, in 2022 (average maximum temperature in August of 33.08 °C), 2006 (32.41 °C), and 2010 (31.09 °C). The minimums of the precipitation along the rice-growing season (April to October) were recorded in 2003 (315 mm), 2004 (333 mm), 2017 (364 mm), and 2022 (376 mm).

The correlation matrix was calculated using the weather data and from the complete dataset after removing outliers (Figure 4). The matrix did not show any statistically significant relationships, but some trends can be observed. For instance, there is a negative relationship between maturity and total rainfall of the season, as well as between TKW and plant height. Additionally, the yield and total rainfall are negatively correlated (−0.41).

After the BLUEs calculation, a statistical description of the varieties under investigation was performed (

Table 4. Statistical description of the rice accessions held in the collection over the 21 years of experiments (BLUEs data). St. Dev. = standard deviation; σ² = variance.

Trait	Mean	Median	Min	Max	Mode	St. Dev.	σ2
Flowering time	95.52	94.57	65.90	139.55	90.53	8.27	68.5
Maturity time	152.54	152.84	126.92	183.89	154.87	7.81	61.1
Flowering-Maturity days	57.68	57.85	16.77	94.77	60.77	7.21	52.1
Plant height	90.73	88.41	58.79	137.56	88.76	14.2	203.65
Magnaporthe infection	1.71	1.56	0.01	6.50	0.50	1.10	1.22
Bipolaris infection	1.91	2.11	0.10	5.00	2.11	0.86	0.74
Pyricularia infection	2.20	2.10	0.10	6.10	2.24	0.74	0.58
Yield	6.51	6.67	1.76	10.74	1.76	1.35	1.84
TKW	31.46	31.46	19.06	47.39	31.06	5.56	30.94

). The variance is particularly high for the plant height, reproductive stages, yield, and yield-related traits (TKW), while the mode showed results more comparable to modern commercially available cultivars.

4. Discussion

Historical data are gaining increasing interest in the scientific community for their potential [40,41,42]. These studies have already proven fundamental for understanding the importance of genebanks [28], their data, and the information extracted from them, especially if associated with genomic data [43].

From this perspective, this work aims to present the collection held in the CREA genebank in Vercelli, exploiting historical data collected during over 20 years of field regeneration. To provide a proper description and inference of the general characteristics of each cultivar in the collection, outliers were calculated and eliminated. Despite this step, the Anderson–Darling normality test did not suggest the normal distribution of the data. The diseases’ distribution and the density of the data are very oddly shaped, but these results are expected since they are non-continuous data. Regarding flowering and maturity days, it is possible to see a double high peak with a lower central peak: indeed, the collection is mainly represented by early or late varieties, with few varieties having a medium growing cycle.

Even though the distributions of the data were not normal, the heritability after outlier filtering was quite high for the flowering time (0.86), plant height (0.94), and TKW (0.96). These results are expected since these traits are under strong genetic control [44]. The flowering time, connected with the circadian clock and photoperiod sensitivity of the plant, is one of the first aspects domestication put pressure on, since it was fundamental for adapting the plant to different environments [3,45,46]. Oryza sativa is a short-day plant, but it was induced to mutate into a long-day plant to complete its life cycle in European conditions and be cultivated in the region.

The heritability was medium–high for the yield (0.77). This parameter is ruled by many genes and loci, but it is heavily influenced by environmental conditions, as inferred from similar work [41] and confirmed in this study: there is not a detectable influence of the agronomic year, rather there is a quite remarkable influence by the different environments, where the genotype influence, or rather the differences in performance of the different cultivars are quite strong (Table S2).

Regarding the plant height, great variability was observed among the rice cultivars, with the tendency, driven by the breeding selection, to shorten it over time to avoid lodging. Ostiglia and Lencino (both cultivars developed in the 1800s) are about 110 cm in height; among the oldest varieties, Bombon (reaching more than 130 cm), Gigante Vercelli (literally “the giant of Vercelli”, 128 cm), and Fortuna (126 cm), bred between 1900 and 1930, were the tallest. Generally, the shortest varieties have been bred in the last 50 years, even if the decrease in stature is not as dramatic in rice (in the first two dimensions of the PCA plant height count for 1.60% of the total variability) as in other cereals such as wheat, for example [47]. An interesting exception is represented by the cultivar Nano, developed in 1912, which strikingly showed an average height of 55 cm during the years of the experiments.

The pathogens infection had a peak in the early 2000s: the years 2002 and 2007 had the highest amount of rain and lowest mean temperature during the rice-growing season. Rice cultivation in Europe is traditionally performed under constantly flooded conditions, as in these experiments; therefore, rain should not directly affect yield. However, rain events are often connected to higher humidity and lower temperature, as can be inferred from the results presented in Figure 3, which are highly connected with the incidence of pathogen infections that can consequently impact yield. Fungal susceptibility showed the lowest heritability in the dataset, 0.39 for Magnaporthe grisea and Pyricularia oryzae, and 0.41 for Bipolaris oryzae. Many genes are well-known to be involved in building resistance against illnesses [48,49,50], but much of the plant infection rate seems to be dependent on the agronomic season, the weather, and the field treatments. The results shown in Table S2 make it possible to infer the same conclusion: rice blast and rice neck blast (P. oryzae and M. grisea, respectively) are influenced by the experimental year. In general, all three infectious diseases addressed in this work show how some locations are more prone to the illnesses: Borgolavezzaro, Ronsecco, Rovasenda, Valle Lomellina and Vercelli. A major part of the data were collected in Vercelli, so this could explain why this location has shown to be correlated with most traits. Bipolaris infection is not affected by the year but mostly by the different environments: this illness is more virulent in certain sites, also depending on the field treatment (i.e., pesticides) applied and, in this scenario, is influenced by the genotypes that have different resistance to the pathogen.

Yield and plant height show a pattern similar to Bipolaris: there is no influence by the agronomic year, but mainly by the environment. Some locations are more fertile naturally and in this context it is possible to better appreciate the influence of the different genotypes, as also confirmed by the high heritability scores for these two parameters.

Dividing the cultivars according to their release period (Figure S3, Supplementary Materials), a good overview of the collection and its diversity can be appreciated. Ostiglia (purple dot), selected from Asian material imported in the 19th century, is related to a long flowering time; Ranghino (pink dot), from 1887, is quite susceptible to rice blast, neck blast, and brown spot. The first news regarding rice blast (P. oryzae) infections in Italy was reported in the beginning of 1900, and since then, Italian breeders have worked to produce new resistant varieties [48,49,51]. The first variety reported to be resistant to rice blast is Precoce Rossi, developed in 1953. Analyzing the correlation between the continent of origin and the phenotypic variability for each trait, it is possible to see how the European group seems the most affected by the illnesses (Figure S4): this group consists mostly of Italian varieties, many of which are quite old, not yet selected for disease resistance. Furthermore, this pathogen is known for its rapid evolution, which jeopardizes efforts to build plant resistance [52,53]. Many accessions from South America, North Africa (Egypt), and Oceania (Australia) are primarily breeding lines of modern origin and generally show lower infection rates, as illustrated in the PCA in Figure S4 (Supplementary Materials).

The statistical description of the varieties based on the BLUEs values (Table 4) provides a good overview of the collection. The variance is high, while the mode and the average are more in line with the values recorded for currently commercially grown rice cultivars. This confirms the great variability of the collection, mainly due to the performance of the old varieties. Further variability could be ascribed to the influence of environmental conditions (Figure S1 and Table S2 in Supplementary Materials), especially for the reproductive stages, plant height, and yield. Regarding the pathogen infections, the collection showed a medium–low average value (Table 4). A possible explanation could be that almost half of the collection is composed by varieties developed in the last 40 years, when breeding was heavily focused on finding solutions against pathogens.

The PCA highlights how infectious diseases, maturation time, flowering time, and temperature are the inputs that provide most of the variability in our dataset. As expected, yield is negatively correlated with late flowering and maturity and infectious disease severity. Although the TKW does not significantly contribute to the overall variability, it can be related to the yield, a fundamental trait of agronomic interest. Yield is proven to be severely affected by infectious diseases that can harm the plants in the field. It is estimated that up to 30% of the global annual rice production is lost due to plant pathogens [19].

The Pearson correlation matrix does not provide any statistically significant correlation, but some trends are visible: between plant height and TKW, and between maturity time and total rain in the agronomic season. The latter trend could be due to the decrease in temperature caused by heavy rains, which increases the time needed for completing the life cycle.

5. Conclusions

The aim of this work was to shed light on the potential of our genebank, its diversity, and the breeding treasure it can represent, especially when historic and easy-to-access data are curated. The collection was found to exhibit great variability for the traits analyzed, the period of development (ranging from 1829 to 2020), the genetic diversity, and the phenotypic performance in field. Some experimental sites have proven to be more productive and less prone to some illnesses, while the influence of the genetic background was always quite strong: the most resistant varieties are the most recent ones (within the last 40 years). The most productive varieties (more than 8 tons/ha) are Baldo, the best among the Italian varieties, and Handao 297 (a Chinese cultivar), the best performing within all the collection.

This study is a valuable source for future multi-environment analysis and, strengthened by the multitude of traits recorded, for multi-trait and selection models analysis. Furthermore, this approach could form the basis for developing a common protocol for genebank data analysis, making it easier and faster for other research centers, even with other plants under investigation, to perform similar analyses.