A total of 32 accessions were selected from the tomato collection of La Platina Germplasm Bank, Instituto de Investigaciones Agropecuarias (INIA) – Chile. These accessions represented a wide diversity in fruit morphology, year of collection, and collection territory inside Chile ( Table 1 ; Supplementary Figure 1 ). The 11 accessions assigned with the local name Limachino in the tomato collection were among the selected. The accession SLY74, collected in 1960 with the vernacular name Limachino, is taken as the Limachino reference sample because it was the oldest accession with the name assigned. Three commercial hybrids (Patrón, Carmelo, and Hybrid DRW 7742), widely cultivated commercial varieties in the Marga valley, were also included in the study ( Table 1 ).

Twenty-four tomato accessions were used in the morphological characterization ( Table 1 ). An open field essay was established during the season 2016–2017 at La Platina Regional Research Center of the Instituto de Investigaciones Agropecuarias, Santiago, Region Metropolitana, Chile (33°34’20.20”S, 70°37’31.22”O, 620 m.a.s.l.). The environmental conditions and the soil characteristics of the evaluation site are shown in Supplementary Table 1 . The experimental design was completely randomized with three blocks and a subsampling of six plants per accession per block. The seeds were sown in August, and seedlings were transplanted in the field in October 2016. The planting distance was 0.5 m within and 0.7 m between rows. The plants were individually tutored and conducted on one axis. Lateral shoots were removed regularly. Black polyethylene mulch was used for weed control. The essay was weekly fertirrigated following commercial guidelines, 0.75 kg of N and 1.3 kg of K₂O were applied after transplant, continuing later with 0.52 kg of N and 1.3 kg of K₂O at the flowering stage. On December 22, one preventive application of Karate Zeon_200 cc ha⁻¹ and Neres 50% SP_1.0 kg ha⁻¹ was made against aphids (Aulacorthum solani, Aphis craccivora, Aphis gossypii, and Myzus persicae) and tomato moth (Tuta absoluta), respectively.

Fruit harvest was made continuously throughout the season, until the end of February, 144 days after transplant, with a time between harvests of the same plant at most 10 days, reaching 16 harvest events during the season. Each cluster per plant was identified independently at harvest, measuring the number of fruits (cluster⁻¹, plant⁻¹, and harvest event⁻¹) and fresh weight of the fruits (g cluster⁻¹, g plant⁻¹, and g harvest event⁻¹). For the phenotypical analysis, only the total sum of these traits was considered.

At the end of the season, six traits related to yield and phenology ( Supplementary Table 2 ) were evaluated: fresh fruit yield (Y, g/m²), the total number of fruits harvested (NFM, fruits/m²), the average fresh fruit weight (FFW, g/fruit), the number of harvested clusters per plant expressed in clusters (CLU, clusters/plant/m²), and the average number of fruits per cluster (FRU, fruit/cluster). Growing degree days were estimated with a base temperature of 10°C (Scholberg et al., 2000), being registered as the date of the harvesting of each cluster’s first fruit. The growing degree days to the first tomato harvested on the last cluster of each plant were divided by the total number of clusters to estimate the thermic requirement for a cluster to develop (CGDD).

Sixty-five morphological traits were evaluated ( Supplementary Table 2 ). The vegetative traits were taken on leaves above the third cluster with the red fruit stage, following the guidelines of IPGRI (1996). In the case of flower and inflorescence descriptors, both were taken on the third or in the fourth cluster of each plant. One fruit from the third or fourth cluster was taken from each plant for high-resolution scanning of one-half. The images were analyzed using the software Tomato Analyzer (Rodríguez et al., 2010) for the quantitative characterization of the fruit. The fruit color was measured using a CR-400 Croma Meter colorimeter (Konica Minolta Inc, Tokyo, Japan), utilizing the L, a, b color space. Descriptions and codes of the morphological traits evaluated are in Supplementary Table 2 .

To expand the genotype panel, eight additional accessions (*) were added, increasing the molecular characterization to 32 tomato accessions ( Table 1 ). The total genomic DNA was extracted from the young leaves of each plant. Four to 24 individuals per accession and commercial hybrids were analyzed ( Table 1 ). The DNA was extracted following a modified CTAB protocol as described by Barra et al. (2012). DNA integrity was seen by electrophoresis in an agarose 1% gel. The purity and concentration of the DNA were calculated using a Nano-Drop (ACT Gene ASP-3700) spectrophotometer, for a posterior dilution for each sample to a final concentration of 10 ng μl. A set of 15 simple sequence repeat (SSR) markers widely used in tomato (Suliman-Pollatschek et al., 2002; He et al., 2003; Benor et al., 2008; Caramante et al., 2009; Albrecht et al., 2010) was selected according to the following criteria: (i) one marker per chromosome; (ii) amplification conditions, and (iii) a high level of polymorphism. Amplifications were performed using two methods: direct labeling and M13(-21) tailed protocol (Schuelke, 2000). In the first method, forward primers were labeled with any of the following fluorochromes: 6-FAM (Macrogen Inc., Seoul, Korea), NED, PET, or VIC (Applied Biosystems™, Foster City, USA). Polymerase chain reaction conditions were 20 ng of the DNA template, 200 μM of each dNTP, 0.1 μM of each primer, 1X Taq polymerase buffer, 1.5 to 3.5 mM MgCl₂, and 0.25 U of DNA Taq polymerase (5 U μl⁻¹, GoTaq^® G2 Flexi, Promega, Madison, USA). In the second method, the forward primer was modified with an M13 (-21) tail at its 5´ end and a third fluorescent labeled M13 primer was used in the PCR reaction. PCR mix contained the following: 30 ng of the DNA template, 150 μM of each dNTP, 0.025 μM of M13-forward primer, 0.1 μM of reverse primer, 0.1 μM of fluorescent-labeled M13 (-21) primer, 1X Taq polymerase buffer, 2.0 mM MgCl ₂ , and 0.25 U of DNA Taq polymerase (5 U μl ⁻¹ , GoTaq^® G2 Flexi, Promega, Madison, USA). The reaction was carried out in a Veriti Applied Biosystems Thermocycler (Thermofisher Scientific^®). The SSR characteristics and PCR amplification conditions are shown in Table 2 .

Visualization was done by capillary electrophoresis on an ABI 3130xl Prism Genetic Analyzer with NimaPOP TM-7 polymer (Nimagen, Nijmegen, The Netherlands), as follows: 2 μl of a pool plex of five PCR products was mixed with 10 μl of HiDi formamide (Applied Biosystems™, Foster City, USA) and 0.2 μl of GeneScan 500LIZ size standard (Applied Biosystems™, Foster City, USA). Allele calling was performed using ABI Prism GeneMapper software v4.0 and reviewed manually. Allele binning was done automatically with the TANDEM program, which transforms allele sizes before rounding to compensate for allelic drift and compression at large fragment sizes (Matschiner and Salzburger, 2009). To detect allele dropout errors, biological and/or technical replicates were performed at DNA extraction, PCR product amplification, and allele calling levels.

A phenotypic dendrogram made with the totality of the morphological descriptors is presented in Figure 1 , with a high cophenetic correlation (0.91). Three main clusters were observed, with most of the accessions grouped in cluster I. This cluster presents four subdivisions: (SLY49), (a) clustering with 4 accessions, (b) clustering with 14 accessions, and (SLY152). Accession SLY49, collected in 1938, is the most diverse among the phenotyped accessions. It has high variability in the number of leaflets ( Figures 2F, G ) and fruit morphotypes, all of them morphologically characterized by a high number of clusters (CLU), high number of fruits (NFM), and low fresh fruit weight (FFW). Subcluster a has well-sustained nodes, grouping materials grown in 1938 from the far north of Chile (Antofagasta, Chile and Tacna, Peru) and the south of Chile (Temuco, Chile), although the accession collected in Temuco (SLY47) was reported at the time of collection as grown in Talca. Inside the b subdivision, two well-sustained nodes give rise to further subdivisions, 1 and 2. Subcluster b1 clustered the accessions grown to 1938 in Limache (SLY30) and Arica (SLY39). SLY50 appears closer to this subcluster, collected in 1938 along with SLY30 from a garden in Limache. A distinguishable leaf morphotype of a reduced number of leaflets and a large terminal leaflet is a common phenotype for all the plants in subcluster b1. On the other subdivision of b, subcluster b2 is composed of the two Marmande tomatoes (SLY129 and SLY83), plus the totality of the denominated Limachino tomatoes collected between 1960 and 2015. The morphotypes of Limachino and Marmande appear close but distinguishable among them by flower morphology. The Marmande type has the most inserted stigmas and significantly differs from the Limachino type SLY74 ( Figure 2C ).

Cluster II grouped two distant genetic accessions: (a) the American type Marglobe (SLY82) and (b) Limachino Italiano (SLY123). The morphology of the fruits supports the lack of node consistency: Marglobe is a uniform commercial type with red, round, smooth tomatoes, while Limachino Italiano is a mixed population of beefsteak pear-shaped tomatoes and medium caliber Marmande-type tomatoes ( Supplementary Figure 2 ). Cluster III included SLY157 and SLY159, two accessions of pink beefsteak tomatoes collected in south-central Chile (Talca and Chillan), characterized by fruits of more than 500 g but with a low number of fruits per plant ( Supplementary Figure 2 ).

The dendrogram nodes ( Figure 1 ) were validated using a linear discriminant analysis ( Supplementary Table 3 ), yielding a high classification score (92.7%) inside the defined clusters. The accession SLY123 had the most significant error (27.3%), which was expected as the accession seems to be composed of a seed mixture. Limachino (Ib2L) and Marmande (Ib2M) were evaluated for distinctiveness, resulting in a high accuracy rate when distinguishing between the two clusters ( Supplementary Table 3 ). The classification error was 3.8% for the Limachino (L) cluster and 10% for the Marmande (M) cluster. Individual accessions were also distinguishable among the accessions panel, with classification errors of 0% for SLY152 and Marglobe (IIb).

The PCA allowed identifying the descriptors that discriminate between the accessions ( Figure 3 ), with 45% of the observed variability explained by the two main components. The variables with higher correlation with principal component 1 ( Supplementary Table 4 ) were the fruit ellipse area relations (EAI), fruit ellipse relation to height (ECD), internal versus external fruit rectangle relation (RET), the relation between fruit dimensions (STR), fruit perimeter (PER), fruit width at mid-height (WMH), maximum height in curvature (MXW), and the length of the internode (LEN). In contrast, the phenology of the plant (CGDD) correlated negatively with principal component 1. Principal component 2 correlated positively to yield components WG and FFW, and negatively to NFM, FRU, NSF, ECC, and RFC. At the same time, the fruit eccentricity descriptors (EAI, ECD, and ECC) presented a relevant weight in both principal components. The accessions with a more distinctive phenotype were the pink beefsteak tomatoes collected in 2015 (SLY157 and SLY159) ( Figure 3 ). The accessions more closely related to the reference Limachino (SLY74) were the ones grown in Limache from different years: 2015 (SLY121), 1980 (SLY150), and 1930 (SLY30). Most of the seeds collected from Limache in 2015 (SLY122 and SLY123) are close to SLY124 and they are distant from the tomatoes collected in 1938.

Fifteen SSRs were used to screen a set of 397 individuals, representing 32 diverse tomato cultivars (29 landraces and three commercial hybrids). All markers were polymorphic, and 88 alleles were detected ( Table 2 ). Mean PIC for all the alleles was 0.24. The loci SSR70 (PIC = 0.57) and SSR248 (PIC = 0.61) presented a higher diversity, while LEcaa001 showed the lowest diversity in terms of allele number (2) with a PIC of 0.01. Markers with a locus on chromosome 9 had differential PICs, independently of the number of alleles observed (A). An association pattern was observed between He, Ho, FIS, and flower morphology (stigma-height dimorphism; Figure 4 ). Plants with more exerted stigmas or longistila ( Figure 2A ) have higher Ho (r = 0.61; p-value = 0.0030) and He ( r = 0.52; p-value = 0.0154), and a negative FIS ( r = −0.46; p-value = 0.0340), indicating excess heterozygosity. This association between exerted stigmas and heterozygosity is attributable to higher rates of cross-pollination. However, in accessions with slight or without stigma exposure or brevistila, the components of the endogamy coefficient of Wright showed a high value of FIS (0.76) and a low value of FST (0.39), thus being associated mainly with autogamous reproduction in tomatoes ( Figures 2B, C ), supported by an endogamy coefficient (FIT) of 0.86 ( Table 3 ).

The decomposition of the genetic variance ( Table 3 ) indicates a high genetic differentiation between accessions (40%) and within individuals inside the populations (46%), which is high for a population of endogamous species. This is consistent with a mainly endogamic reproduction with variable rates of cross-pollination and further evidenced by the low percentage of molecular variance within individuals (14%).

The private alleles in some accession ( Table 4 ) occur in almost all the genetic loci evaluated, existing alleles that are underscored because of their high frequency. The high-frequency private alleles distinguished SLY147, Hybrid DRW 7742 (H), SLY159, SLY157, SLY129, SLY49, and SLY74. The microsatellite SSR70 presented two private alleles for the reference Limachino (SLY74), allowing fingerprinting accession. The accession SLY49 had the highest number of private alleles and the highest frequencies for those alleles.

The FST by pairs of populations showed groups with a significant degree of differentiation ( Figure 5 ; p-value ≤ 0.05). The accessions SLY147, SLY148, SLY149, SLY150, and SLY151, collected in Limache in 1980, had no differentiation between them, but also with SLY122 and Marmande accessions (SLY83 and SLY129). All these accessions shared a genetic pattern ( Supplementary Figure 3 ). The San Marzano landraces showed diverse degrees of genetic differentiation between them, with the accession collected in Argentina (SLY103) being closer to the San Marzano collected in Italy (SLY104) than the San Marzano collected in Spain (SLY102). Pink tomato accessions from the south of Chile (SLY157 and SLY159) had a low FST between them, but both have high FST, with the third pink tomato from the central valley (SLY158) indicating differences between them. Among the Limachino accessions collected in 2015, two (SLY121 and SLY122) showed little differentiation from the Limachino type collected in 1960 (SLY74). Supplementary Figure 3 shows that the reference Limachino is a population with individuals from two predominate genotypes; when comparing the genetic patterns of each population, SLY121 is observed as similar to Marmande, but some individuals share the Limachino (SLY74) genetic pattern. SLY121 is the closest accession to the reference Limachino ( Figure 5 ).

The UPGMA clustering using Nei’s genetic distance ( Figure 6 ) reflects the genetic structure observed in the FST by pairs of accessions. The highest genetic distance was between SLY65 and the commercial variety Patrón (P), with a value of 0.643. The dendrogram allows us to infer the hypothesis posed in this work, as three clusters are generated: cluster I, composed of European materials Marmande (French) and San Marzano (Italian), the USA type Marglobe, commercial materials such as Patrón (Syngenta), Hybrid DRW 7742 (Seminis), Carmelo (French), Limachinos collected in 1980, the reference Limachino (SLY74), and accessions SLY122 (Marmande-type) and SLY121 (Limachino-type), both collected in 2015; cluster II, which clusters the Chilean landraces grown from Tacna to Talca to 1938 (except SLY37), with two tomatoes grown in Limache to 2015 (SLY123 and SLY124), plus two pink tomatoes grown to 2015 in San Clemente, Talca (SLY157) and San Carlos, Chillán (SLY159); and cluster III, which is represented by one accession (SLY65) grown in Antofagasta to 1938. Clusters II and III clustered accessions collected in Chile in 1938, indicating that, in Chile, tomatoes different from commercial types have been cultivated from at least 1938 to the present, differentiated on phenotype and genotype.

The classification made by structure for the optimal K was analyzed for the first peak of ΔK with K = 2, showing a division between Chilean accessions (Blue) and the accessions from Europe (Yellow) ( Figure 7 ). Marglobe (SLY82) showed Q values of 0.471 (blue) and 0.529 (yellow). Similarly, SLY123 showed Q values of 0.372 (blue) and 0.628 (yellow). Both accessions were collected in the territory of Limache in 2015, representing an admixtured group of individuals.

The second peak of ΔK, with K = 4, was also evaluated ( Figure 7 ). Under the classification of four clusters, the Marglobe admixture condition is lost but is increased in the accessions collected in Chile in 1938, where the individuals of the same accession showed high variability. When subdivided into four clusters the within groups explained variability increased by 5% respect to the subdivision into two clusters ( Table 5 ). In both classifications, the main variance element of the observed molecular variance is due to individuals within populations, consistent with the results in Table 3 .

The clusters in the dendrogram based on Nei’s genetic distance ( Figure 6 ) are consistent with the clusters observed in the phenotypic distance dendrogram ( Figure 1 ) and the classification of individuals made by Structure ( Figure 7 ). In the Structure classification with K = 4, clusters A (light blue), B (blue), C (orange), and D (yellow) were observed, having consistent admixture levels inside the clusters identified along the phenotypic dendrogram. Phenotypic clusters are supported by genotypic classification in the following: (i) cluster I.a, which corresponds to the genotypic dendrogram cluster II.a.6, composed of the accessions with a reduced number of leaflets ( Figure 2E ), a larger terminal leaflet (SLY47, SLY70, and SLY66) ( Figure 2D ), and the Structure classification related to A and B; associated with this cluster in terms of phenotypic leaf morphology and Structure classification is SLY65; (ii) the phenotypic cluster I.b.1 that groups the landraces SLY39, SLY50, and SLY30, having complete correspondence with the genotypic node (II.a.6) and classified by structure as an admixture for K = 4; and (iii) the phenotypic cluster III characterized by pink tomatoes that can weigh over 500 g, corresponding to cluster II.a.7 from the genotypic dendrogram, which was classified as light blue (A) by Structure (K = 4).

One of the cultivated tomatoes collected in Limache in 1938 (SLY49) is an exception that does not cluster on a phenotypic and genotypic basis with the rest of the accessions; the Structure classification supports this as an admixture. This is confirmed by the high diversity of phenotypes observed among plants of the SLY49 accession ( Figures 2F, G ). Marglobe accession, a US commercial round, red, and smooth-type tomato, was classified as C grouping distant in genotype and phenotype to the rest of the accessions. A relatively close accession to Marglobe was one of the pink tomatoes (SLY158), which, despite being described as a close phenotype to accessions SLY157 and SLY159, was observed as a commercial variety due to its molecular and phenotypic uniformity. This is in accordance with the information given by the seed donor. Limachino accessions SLY152 and SLY124 collected in 1980 and 2015, respectively, are also observed as populations close to Marglobe.

The European materials were classified as D by Structure, including the San Marzano accessions with different levels of mixture with the C group. Most of the Limachinos collected in 1980 and the other two accessions collected in 2015 appear as populations closest to European materials, with different levels of introgression. The reference Limachino SLY74 is classified as a mixture of individuals close to European or US accessions ( Figure 7 ).

The morphological variability was evaluated in 24 tomato accessions, with 46% corresponding to Limachino type, 42% to other Chilean landraces, and 12% being commercial varieties grown in Chile in the 1980s and in 2015. The potential of the local tomato genetic resources is seen in the broad diversity of phenotypes and existing significant differences for multiple traits of importance for agriculture ( Supplementary Table 5 ), mainly traits associated with phenology, fruit morphology, and yield components. Identified morphological groups correspond to the scenario of four genetic groups proposed via Bayesian modeling. The yield components fresh fruit weight (FFW) and the number of fruits (NFM) showed higher CV in the genetic classification groups of structure associated with Chilean materials (A and B); thus, the utilization of this germplasm could present a potential impact on breeding programs ( Supplementary Table 6 ).

The distinctiveness of the landraces is associated with the discrimination among tomato accessions. Figàs et al. (2015), using Tomato Analyzer, defined the shape indexes (SI and SII) as the descriptors with a higher discrimination capacity, unlike our work in which the eccentricity indexes (EAI, ECD, and ECC) had the highest discrimination capacity, associated with consistent distinctive fruit shapes. Figàs et al. (2015) also found no strong correlations ( ± 0.2) between descriptors and principal components, explaining 34% of the variability observed in the first two components, while morphological descriptors in the first two principal components of the present work, mainly associated with fruit, yield, and phenology, explained 45% of the total observed variability ( Figure 3 , Supplementary Table 4 ). In both cases, fruit descriptors presented the highest discrimination capacity among accessions. Bota et al. (2014) had similar results utilizing Tomato Analyzer; the first two principal components explained 47% of the variability observed. The shape indexes showed the highest correlation to the first component, and the fruit size descriptors (ARE, PER, and MXW), in conjunction with postharvest traits and fruit number, correlated to the second component. Principal component analysis has had a similar result in the present work, with PER, MXW, and NFM correlating with the second principal component. The low correlation of shape indexes with the principal components is probably associated with a lack of landraces with distinctive shapes such as cylindric, heart, or piriform.

The definition of tomato landraces based on cultivars with distinctive fruit shapes grown in the same or different territory has shown no correlation between the genotype and phenotype of the landraces (Cebolla-Cornejo et al., 2013). In contrast, in this study, the genetic groups of cultivars shared specific morphological traits. The observed diversity in morphological traits associated with yield (FFW, NFM, CGDD) is highly important for agriculture for sustainable intensification of agriculture (Khoury et al., 2016). Cortés-Olmos et al. (2015) found similar groups to the ones presented in this work by using SNPs, identifying at a molecular and phenotypical level: Rosa accession, characterized for being large pink tomatoes; and the Centenares accession, a cherry-type tomato. In our study, the pink beefsteak tomatoes (SLY157 and SLY159) and the cherry-like accession SLY49, which can be classified as a high-weight cherry tomato, were clearly differentiated, although SLY49 is highly diverse at a molecular level and at a morphological level, presenting a wide range of leaves, fruits, and plant architecture ( Figures 2F, G ). Cortés-Olmos et al. (2015) studied SNP and SSR variability, establishing that it is possible to define an association between molecular diversity and agro-morphological diversity in contraposition to evaluations of landraces from Spain (Cebolla-Cornejo et al., 2013), Greece (Terzopoulos and Bebeli, 2008), and Italy (Mazzucato et al., 2010). In contradistinction to García-Martínez et al. (2013), who found no differences between landraces from Spain and Italy, the genetic structure of the Chilean germplasm can be associated with its provenance. Further phylogeographic studies of the materials are needed to explore the history of the genetic structure.

From 1999 to date, only the Italian tomato landrace San Marzano (Rao et al., 2006; Mazzucato et al., 2010), the Vesuvian Pomodorino del Piennolo (Manzo et al., 2018), and the Greek tomato Santorini (Koutsika-Sotiriou et al., 2016) have received a Protected Denomination of Origin, becoming the standard on which the distinctiveness of tomato landraces has been proved. The molecular fingerprint of San Marzano has been developed utilizing the oligonucleotide (GATA)4 and microsatellites (Rao et al., 2006; Caramante et al., 2009). In the case of the utilization of SSR, Caramante et al. (2009) found all alleles to be monomorphic for the evaluated accessions of San Marzano, which is not the case for the reference Limachino, supporting its classification as a landrace. The Limachino-type tomatoes ( Supplementary Figure 4 ) are observed as close to Marmande and distant to old Chilean landraces, probably associated with an origin in the hybridization of Marmande-type tomatoes with local tomatoes as proposed by Rick (1958). However, a total of nine private alleles ( Table 4 ) in the reference Limachino (SLY74) allows us to distinguish it at a molecular level from the Chilean accessions evaluated and the foreign accessions, such as San Marzano, Marmande, and commercial cultivars. The high number of low-frequency alleles and, in this case, private ones might be associated with a foundation and posterior population expansion scenario. This hypothesis needs to be tested but is consistent with the historical records of tomato cultivated surface and the 163 smallholder tomato growers registered in Limache to 1955 (Donoso et al., 2022). The selection of growers has generated genetic differences within the Limachino type, as landraces are dynamic populations, having still a Limachino-type phenotype according to Figure 1 , and have been grown in the territory of Limache from at least 1960 (SLY74) to 2015 (SLY121). The same pattern was observed in San Marzano accessions (SLY102, SLY103, and SLY104) collected in different countries ( Figure 6 ).

Considering our results, there is the background to support the hypothesis of this work, existing tomato landraces that have been cultivated (SLY39, SLY30, SLY50, SLY65, SLY47, SLY70, SLY66, and SLY49) and that are being cultivated (SLY159, SLY157, SLY123, and SLY124) in Chile continuously before 1938 until 2015. These landraces are different from the European and US varieties, demonstrating the existence of an in situ conservation of tomato landraces with novel alleles.

High molecular and morphological diversity was observed, with an association of phenotypes to differentiated genotypes. Accessions of Chilean tomato landraces could have the potential to generate yield increments in breeding programs by having a higher variability in yield components (Donoso and Salazar, 2023). Our results are consistent with those obtained by other researchers that had proposed Mesoamerica and the Andes Region as a center of diversity for tomatoes.