Each accession was characterized based on 66 varietal descriptors (Table 2): From these descriptors, 34 and 32 were quantitative and qualitative descriptors respectively, for plant, leaf, inflorescence (bract), flower, fruit, and seed (Table 2).

We evaluated 66 descriptors, qualitative characters were taken based on visual parameters, except for color that was taken as reference the Pantone® color chart; For the registration of the quantitative characters was used rule and electronic vernier mark Mitutoyo, Model No. CD-6 CS. For each character, 20 repetitions per accession were evaluated, considering a repetition to a leaf, a fruit or a seed according to the case.

The principal components analysis (PC) was applied to the data obtained using the PRINCOMP procedure of SAS Ver.6.12 [16] using the correlation matrix; (Eigenvalues), eigenvectors and pearson correlation coefficient between the original variables and the principal components [17] . The PCs were plotted on a Cartesian plane, to observe the distribution of the characterized accessions.

Hierarchical clusters were also analyzed by the PROC CLUSTER procedure of SAS Ver. 6.12 [16] and the algorithm was performed by hierarchical clustering, which generated a dendrogram that allowed to distinguish the groups conformed by the characterized accessions [17] [18] .

For the initial analysis, 25 accessions as well as 55 descriptors were used without including fruit and seed variables (Table 2). The PCA indicated that the first six components explained 69.4% of the total variability, with 26.1%, 14.0%, 9.0%, 7.8%, 6.5% and 5.9% respectively (Table 3). These results are very similar to those reported by Sosof et al. [19] with 43 cultivars of Heliconia but using only

PC = Principal Component.

three components (69.94).

The variables which contributed significantly were: for PC1: limb length (LLE, p < 0.001), limb width (LW, p < 0.001), limb petiole length (LPLE, p < 0.001), pseudostem petiole width (PPW, p < 0.001), pseudostem thickness at 60 cm (PPT, p < 0.001), pseudostem width at 60 cm (PSW, p < 0.001), rachis width (RW, p < 0.001), rachis thickness (RT, p < 0.001), second bract width (SNBW, p < 0.001), plant length (PL, p < 0.05), inflorescence length (ILE, p < 0.05), second bract thickness (SNBT, p < 0.05), pseudostem petiole thickness (PPT, p < 0.05), plant growth (PG, p < 0.05), pseudostem and leaf length ratio (PLRLEL, p < 0.05), bracts arrangement (BA, p < 0.05), and leaves growth (LG, p < 0.05). For PC2: inflorescence growth (IG, p < 0.001), type of inflorescence (TOI, p < 0.05), peduncle length (PLE, p < 0.05), rachis color (RCO, p < 0.05), wax in limb (WL, p < 0.05), pseudostem shape (PS, p < 0.05), and bracts color (BCO, p < 0.05). PC3: presence of pubescence on inflorescence (PPB, p < 0.001), open bracts (BNo, p < 0.05), inflorescence length-to-width ratio (ILEWR, p < 0.05), limb color (LCO, p < 0.05), rachis shape (RS, p < 0.05), leaves present (LNo, p < 0.05), and type of rhizome (TR, p < 0.05). All of the above suggested that 33 of 55 analyzed variables contributed significantly to explain the total variation within the 25 heliconia characterized accessions (Table 4).

According to PC1 and PC2, the distribution of the accessions within Cartesian axis permitted to observe the great variation among the Heliconia species (Figure 1). Likewise, Figure 2 shows that the clustering pattern of the accessions is consistent with their corresponding species, but inflorescence is now related to it.

With 55 variables (plant, leaf, inflorescence, and flower), a hierarchical cluster analysis (HCA) was carried out. According to a 0.05 semipartial correlation coefficient, six groups were determined and divided into different sub-groups, for example: group I was divided into IA and IB, group III into IIIA and IIIB, group V into VA and VB, and group VI into VI and VIB (Figure 3).

PC1: Principal Component 1, PC2: Principal Component 2; PC3: Principal Component 3. *. **= Significant with alpha = 0.05 and alpha = 0.01 respectively.

Group I (GI) was formed by six accessions, which can be easily distinguished by the oblique position of their leaves and short petiole, and the medium plant height. Accessions from group IA are separated from IB primarily because of the lack of wax in the limb. Although L3 is a tall-size plant, it is considered to be a hybrid between H. uxpanapensis and H latisphata, since it presents some morphological characteristics closer to H. latisphata [14] , for example: an erected growth of inflorescence and an absence of pubescence on inflorescence (glabrous).

Group II (GII) was made up by only one accession from H. vaginalis Benth. Subsp. Besides being Mathiasiae a-1 85.31 cm height short-size plant and having a-3 13 stem/leaf ratio, erected inflorescence, undulate rachis, bracts in distichous position, and an orange color; it is the only species that presents a growing zingiberidae-like shape.

Group III (GIII) was integrated by five accessions, which are easily distinguishable for presenting wax in the limb [13] , leaf growth in an oblique position, pubescence on inflorescence, undulated rachis, and a helicoidal arrangement in the bracts. Accession C1 was associated to this group due to the short size of the plant, which is probably in response to where the characterization was carried out [20] . However, as this accession comes from an altitude higher than 1200 m, it presents a pubescence just like the H. spisa species, which belong to GIII.

Group IV (GIV) was integrated by three accessions C2, C3, and C4 belonging to H. collinsiana Griggs var. Collinsiana. These samples were collected in Chiapas (C2, C3) and in Veracruz (C4) at an altitude of 596, 402, and 250 m respectively and present a greatest inflorescence length [21] .

Group V (GV) included five accessions, which are easily distinguishable for being tall-size plants (>400 cm), their growth in plant is of a musoide type, have an oval-shape stem, presence of wax in the limb, an erected growth of the inflorescence, absence of pubescence on the inflorescence, the bracts are in distichous position, and an erected growth. In the sub-group VA, two accessions of H. uxpanapensis are clustered, a third one was associated to H. champaneana because of its similarity in shape, bracts of short length and orange color.

Group VI (GVI) clustered five accessions that are characterized for being plants with a growing musoide habit, have an oval-shape stem, absence of wax in the pseudostem, number of leaves from 2 to 3, absence of wax in the limb, an erected growth of the inflorescence, pubescent, and bracts arrangement in one level. Accessions belonging to H. bourgaena and H. bihai clustered within sub- group VIA abide by the similarity in their inflorescence, while in the sub-group VIB two other accessions were clustered based on the size of the plant.

Guimarães et al. [13] registered the presence of wax in pseudostem on H. bihai whilst Costa et al. [21] reported a bigger length in the inflorescence. This variation can be highly influenced by the environment since temperature conditions and rainfalls vary from 26˚C to 1, 968 mm [20] . According to Robles [22] , limitations in water affect plants morphology, physiology, and metabolism. Among the physiological and metabolic changes that occur within these plants, there is an increase of wax in leaves’ surface.

Accessions clustering of H. bourgaena and H. bihai within sub-group VIA abide by the similarity in their inflorescence, while in sub-group VIB; two other accessions are clustering by the plant’s size.

In general, the groups reflect an association among the individuals that belong to the same species; this tendency has been observed by Londoño [23] who reported the association of samples from different in vitro cultures through AFLP markers into H. caribaea and H. orthotricha especies. Pereira et al. [24] using 16 morphological markers in heliconias found variation between the studied species, but these did not allow a clear differentiation between species.

It was found that in the most grown flower width (MDFW), in the most flourished flower thickness (MFFT), in limb’s apice shape (LAS), limb undulation (LUN), leaf nervation type (LNT), bract shape (BS), and sepals colouring (SCO) contribute significantly to groups formation (Data no show).

Ovary coloring (OCO) and width (OW) were the characters that presented the greatest variability regarding the fruit size. For seed, the most relevant characters were: seed thickness (SET), seed color (SECO), and seed width (SEW) (Data no show)

When including fruit and seed variables, a better clustering could be observed. The groups did not change on the first analysis, therefore; it indicates that fruit and seed are important variables to include when making a distinction among species accessions (Figure 4).

The great phenotypic diversity found on the 25 studied accessions, which are maintained in the Rosario Izapa genebank, will be a very useful strategy in programs of genetic improvement just as it is carried out in other countries of Central [19] and South America [3].