Thirty-eight cultivated and named E. ventricosum landraces which are used in the treatment of seven different human diseases or disorders were identified with the help of knowledgeable village elders from four locations (administrative zones/special district) consisting of nine districts or special districts of the Southern, Nations, Nationalities, and Peoples (SNNP) regional state of Ethiopia (Figure 1). For comparison, 13 enset landraces that have other non-medicinal use values (principally used for human food) were also sampled. To test the consistency of naming of landraces within each location, up to four duplicate samples (based on their availability) were collected from different sites. Since the landraces are not scientifically characterized, each individual was considered as a separate sample so that a total of 92 plant samples were collected (Supplementary Table 1). The samples were collected from individual farmers’ fields, located at 18 Kebele (the lowest tier of civil administration unit) from across the enset distribution. Since different landraces may have been given the same vernacular name at different locations (Olango et al., 2015), landraces having identical names, but originated from different locations were labeled by including the first letter of names of location after a vernacular name of the second landrace. Healthy young cigar leaf (a recently emerged leaf still rolled as a cylinder) tissue samples were collected from individual plants from November to March 2017 and they were stored in zip-locked plastic bags containing silica gel and preserved until the extraction of genomic DNA. The dried leaf samples were ground and genomic DNA was isolated from 150 mg of each pulverized leaf sample following the modified cetyltrimethylammonium bromide (CTAB) extraction protocol (Borsch et al., 2003).

Twenty-one enset SSRs primer-pairs (14 from Olango et al., 2015, and 7 from Biswas et al., 2020) were initially screened for good amplification, polymorphism, and specificity to their target loci using 15 samples. This led to the selection of 15 primer pairs to genotype the landraces (Table 1).

A PCR amplification was carried out in a 20 μl reaction volume containing 1.5 μl (100 mM) template DNA, 11.5 μl molecular reagent water, W 4502 (Sigma, St. Louis, MO, United States), 0.75 μl dNTPs (10 mM) (Bio line, London), 2.5 μl Taq buffer (10× Thermopol reaction buffer), 1.25 μl MgCl₂ (50 mM), 1 μl forward and reverse primers (10 mM), and 0.5 μl (5 U/μl) BioTaq DNA polymerase (Bioline, London) and amplified using a PCR thermal cycler (BiometraTOne, Terra Universal, Germany). The three-step amplification program consisted of initial (1) denaturation for 2 min at 95°C, (2) 35 cycles of denaturation at 95°C for 1 min, annealing at a temperature specific to each primer set (Table 2), for 1 min, extension at 72°C for 1min, and (3) final extension at 72°C for 10 min. The PCR products were stored at 4°C until electrophoresis.

The separation of the amplified product was accomplished in a 4% (w/v) agarose (Bioline, London) gel in 1% (w/v) Tris-acetate-EDTA (TAE) buffer containing ethidium bromide, and electrophoresed at 80 V for 3 h. A standard DNA ladder of 100 bp (Q step 2, Yorkshire Bioscience Ltd., United Kingdom) was loaded together with the samples to estimate molecular weight. The banding pattern was visualized using a gel documentation system (NuGenius, SYNGENE, Cambridge, United Kingdom) and the pictures were documented for scoring.

The sizes of the clearly amplified fragments were estimated across all the sampled landraces. The number of different alleles (N_a), the effective number of alleles (N_e), Shannon’s information index (I), observed heterozygosity (H_o), expected heterozygosity (H_e), un-biased expected heterozygosity (uH_e), and Fixation index for each locus were computed using GENALEX version 6.503 (Peakall and Smouse, 2012). The Polymorphism Information Content (PIC) for each locus was computed using PowerMarker version 3.25 (Liu and Muse, 2005). The Genetic differentiation (F_ST) between the two groups of landraces was estimated using GENALEX. An analysis of molecular variation (AMOVA) was performed to evaluate the relative level of genetic variations among groups, and among individuals within a group using GENALEX. The neighbor-joining (NJ) tree was constructed using the software DARwin (Perrier et al., 2003) based on Nei’s genetic distance (Nei, 1972) to reveal the genetic relationships among the groups and individual landraces. The resulting trees were displayed using Fig Tree var.1.4.3 (Andrew, 2016). Discriminant Analysis of Principal Components (DAPC) was implemented using R, version 4.4.1 in ‘adegenet’ package (Jombart, 2008). Detection of admixture was inferred using a Bayesian model-based clustering algorithm implemented in STRUCTURE version 2.3.4 (Pritchard et al., 2000), To determine the most likely number of populations (K), the simulation method of Evanno et al. (2005) was implemented using the web-based STRUCTURE HARVESTER ver. 0.6.92 (Earl and von Holdt, 2012). Each of the probable K was run 10 times with K = 1–10, and the length of burning period was set at 50,000 and 500,000 Markov chain Monte Carlo (MCMC) iterations.

Principal coordinates analysis was carried out using R version 3.6.3 (R Core Team, 2017) to further evaluate the genetic similarity between the landraces.

The polymorphic nature of some of the SSR markers was as shown in Supplementary Figure 1. A total of 38 alleles were detected across 15 SSR loci in 92 genotypes (Table 2). The number of alleles generated per locus ranged from 2 to 3, with an average of 2.53 alleles. The PIC values for the markers varied from 0.16 (primer EnBedSSR020585) to 0.52 (primer Evg2) with an average of 0.41. The observed heterozygosity (H_o) and expected heterozygosity (H_e) ranged from 0 to 0.64 and 0.18 to 0.63, respectively, and Shannon’s information index (I) ranged from 0.31 to 1.04.

The AMOVA showed that 97.6% of the total variation was assigned to individuals within a group; while only 2.4% variation was contributed by variation among the groups (Table 3). The overall genetic divergences among the two groups of enset landraces (“medicinal” and “non-medicinal”), measured in coefficients of genetic differentiation (F_ST) was 0.024 (Table 3).

The unweighted neighbor-joining tree cluster analysis performed using Nei’s genetic distance showed that landraces used in the treatment of a specific disease traditionally were not grouped into the same cluster or sub-cluster; instead, they were mixed with those landraces having other use values (Figure 2). Similarly, the landraces originating from each location were scattered into all 4 clusters (data not presented). In the neighbor-joining tree, it was also observed that some of the landraces with the same vernacular name (replicate samples) were found to be identical, while the others show a difference. Whereas, except two (bishaeset and mekelwesa), all landraces with the different vernacular names were distinct. The Bayesian clustering result showed the presence of four subpopulations, with some shared admixture memberships (Figure 3A), which is in agreement with the results of the neighbor-joining tree. The DAPC grouped the studied enset landraces into four clusters irrespective of their use value (Figure 3B and Supplementary Table 2). Although the medicinal and non-medicinal landraces were not separately clustered, the majority of the later were grouped in to cluster 3 and 4.

We assessed the genetic diversity of 38 cultivated enset landraces used in traditional medicine and 13 landraces that have non-medicinal values. According to our results, a moderate level of genetic diversity (H_e = 0.47) was detected. A relatively higher H_e values (0.55 and 0.59) of enset were reported (Getachew et al., 2014) and (Olango et al., 2015; Gerura et al., 2019), respectively. The value of I (0.74) in the current study was also lower as compared with the earlier report (1.08) (Gerura et al., 2019). The variation of the result is probably due to the fact that our study was focused on a selected group of landraces, those used in traditional medicine. Lower genetic diversity estimates were reported earlier using ISSR (Tobiaw and Bekele, 2011) and RAPD (Birmeta et al., 2002) markers. However, comparisons of detailed diversity estimates from marker systems that have different properties and origins of variation do not allow useful conclusions (Powell et al., 1996; Hamza et al., 2013).

The genetic differentiation between the landraces used in traditional medicine and those having other use-values was very low (0.024). Genetic differentiation values (0.037) among locations were reported on enset (Gerura et al., 2019), although the direct comparison of different populations is difficult. The AMOVA also showed that the proportion of genetic variation among the two groups of enset landrace was very much limited (2.4%), while the majority was contributed by variation among individuals.

From the landraces that have the same vernacular names (replicate samples), the majority were closely similar genetically and placed together in the neighbor-joining tree. This indicates that farmers have rich indigenous knowledge in identifying and naming enset landraces based on phenotypic traits, and the knowledge is shared across the growing region. However, few other replicates of landraces were placed in different clusters, indicating that genetically different landraces were given the same vernacular name. Perfect identification of genotypes using morphological traits is difficult, and the existence of homonyms has been reported previously (Olango et al., 2015).

Except for two, all landraces with distinct vernacular names were found to be different, showing that vernacular names are good indicators of genetic distinctiveness in these specific groups of landraces. Whereas, the existence of 37 and 8 duplicates of landrace in diversity analysis of enset using four AFLP (Negash et al., 2002) and 12 RAPD (Birmeta et al., 2002) markers, respectively, was reported. Gerura et al. (2019), who studied 83 enset genotypes using 12 SSR markers, also reported 10 duplicates of landraces. Although full identity among the landraces can only be determined when the entire genomes are compared, it is expected that the SSR markers used in the current study could sufficiently discriminate the landraces than the studies that reported a higher number of duplicates. The variation of the results, therefore, could be due to the sample collection method followed in the current study, which involved focusing mainly on specific landraces used in traditional medicine.

The use of some of the enset landraces in traditional treatment of various human ailments in the major enset growing region of Ethiopia, SNNPR, was reported by several authors (Tsehaye and Kebebew, 2006; Olango et al., 2014; Ayenew et al., 2016; Daba and Shigeta, 2016). However, landraces that are used in the treatment of the same types of diseases did not show distinct grouping; instead, landraces used to treat different diseases were mixed with each other and even with those having other use values in the neighbor-joining tree, indicating that “medicinal” properties do not appear to be monophyletic. Furthermore, the DAPC also showed that the two groups of landraces neither formed a separate cluster nor did one group show greater spread or genetic diversity. From these results, it can be argued that landraces that are used in traditional medicine are not genetically distinct from other landraces.

There are several possible explanations for these observations. First, all enset landraces may have a degree of medicinal value, but specific genotypes are preferred for phenotypic or cultural reasons. Second, the medicinal value may arise through genotype-environment interactions or management practices specific to those landraces i.e., they may have non-differentiated genotypes, but in situ, they generate unique biochemistry with medicinal properties. Thirdly, a number of important medicinal landraces may have been omitted, or medicinal value incorrectly assigned to non-medicinal landraces. This could serve to hinder our analysis and make it more difficult to detect real genetic differentiation. This would also be an indication of a decline in the quality of indigenous knowledge. We also note that it is unlikely that the strong trust of society upon these landraces could not be developed after a very long period of use, and we have observed remarkable similar enset medicinal claims across a wide variety of distinct ethnic groups in multiple languages. Moreover, anti-bacterial and anti-fungal activities of a compound extracted from the unspecified E. ventricosum landrace (Hölscher and Schneider, 1998), and a report (Sreekutty and Mini, 2016) on the medicinal property of a related species, Ensete superbum, suggests that at least some of the enset landraces have real medicinal property. The higher mineral concentration (that has a relation with bone health) landraces used in traditional medicine was reported (Nuraga et al., 2019b). Finally, a biochemical survey of enset landraces (Tamrat et al., 2020a) detected high levels of arginine, compared to other amino acids, in three medicinal landraces (Koshkowashiye, Astara, and Lochingiya). Arginine is involved in collagen formation, tissue repair, and wound healing via proline, indicating a possible biochemical basis for the medicinal properties of some of the landraces.