Recent descriptions of hybrid animal species have spurred interest in this phenomenon, but little genomic data exist to support it. Here, we use frequency variation for 657 amplified fragment length polymorphism (AFLP) markers and DNA sequence variation from 16 genes to determine whether the genome of Heliconius pachinus, a suspected hybrid butterfly species, is a mixture of the putative parental species, Heliconius cydno and Heliconius melpomene. Despite substantial shared genetic variation among all three species, we show that the genome of H. pachinus is not a mosaic; both AFLP and DNA sequence data overwhelmingly associate H. pachinus with just one of the potential parents, H. cydno. This pattern also applies to the gene wingless, which is tightly linked to the locus that determines forewing colour—one specific H. pachinus trait that has been hypothesized to have originated from H. melpomene. As a whole, the data support a traditional, bifurcating model of speciation in which H. pachinus split from a common ancestor with H. cydno without a genetic contribution from H. melpomene. However, comparison of our data to DNA sequence data for another putative hybrid Heliconius species, Heliconius heurippa, suggests that the H. heurippa genome may be a mosaic.
Understanding how new species originate is a major focus of research in evolutionary biology. In general, the formation of new species follows a branching pattern, in which an ancestor gives rise to two (occasionally more) descendent lineages. An alternative to this standard bifurcation model of speciation is hybrid speciation, in which interbreeding between two species gives rise to a third distinct lineage. In plants, hybrid speciation as a result of allopolyploidy is well known (Rieseberg 1997; Levin 2002; Coyne & Orr 2004) and there are examples of allopolyploid hybrid animal species as well (Dowling & Secor 1997). Much less common is homoploid or recombinational hybrid speciation, in which hybridization generates a new species without a change in chromosome number. In theory, this process should be rare because a new homoploid hybrid lineage faces a variety of obstacles that are likely to prevent it from becoming established. Most critically, a hybrid lineage must become reproductively isolated from both parental species quickly otherwise it risks fusing with one of them (Buerkle et al. 2000). It also must be ecologically distinct otherwise it risks competitive exclusion (Coyne & Orr 2004). Owing to these obstacles, and a commonly held view that hybridization, in general, is quite rare in animals, hybrid speciation in animals has received little attention historically (Dowling & Secor 1997). Recently, however, multiple descriptions of putative homoploid hybrid animal species have spurred interest in this phenomenon (Salzburger et al. 2002; Smith et al. 2003; Schwarz et al. 2005; Gompert et al. 2006; Mavarez et al. 2006; Meyer et al. 2006).
One animal group where hybridization is strongly suspected to have played an important role in driving phenotypic and species diversification is the neotropical butterfly genus Heliconius (Linares 1989; Gilbert 2003; Mavarez et al. 2006). Hybridization between two species in particular, Heliconius cydno and Heliconius melpomene, is hypothesized to have fuelled mimetic convergence within the genus (Gilbert 2003) and may be responsible for the origin of at least four species: Heliconius pachinus; Heliconius heurippa; Heliconius timareta; and Heliconius tristero (Gilbert 2003; Mavarez et al. 2006). Despite having diverged in a variety of ecological characteristics and mimetic wing patterns, H. cydno and H. melpomene continue to hybridize (Mallet et al. 1998, 2006) and exchange genes (Bull et al. 2006; Kronforst et al. 2006a) throughout Central America and Andean South America. The four putative hybrid species are all currently recognized as very close relatives of H. cydno based on morphology, behaviour, degree of interfertility and genetic data, but each displays colour pattern elements characteristic of a sympatric race of H. melpomene. For example, H. pachinus is ecologically and behaviourally identical to H. cydno, it is completely interfertile with H. cydno and DNA sequence data for mitochondrial and nuclear genes generally cluster H. pachinus haplotypes with those of H. cydno (Brower 1994; Brower & Egan 1997; Beltrán et al. 2002; Kronforst et al. 2006a). However, at six of seven Mendelian colour patterning loci, H. pachinus shares functionally equivalent alleles with H. melpomene. The H. pachinus colour pattern elements that may be derived from H. melpomene include the yellow colour of the forewing, the yellow colour of the hindwing, the position of the distal forewing shutter, the presence of the proximal forewing shutter, the absence of the hindwing shutter and the presence of red at the ventral base of the wings (Gilbert 2003). Similarly, a variety of data indicate that H. heurippa, H. timareta and H. tristero are closely related to H. cydno (Brower 1996a; Brower & Egan 1997; Salazar et al. 2005), yet each has colour pattern elements that appear to be derived from H. melpomene. For H. heurippa, these include the presence of red on the forewing and at the base of the ventral fore- and hindwing, the absence of brown on the ventral hindwing and the absence of blue iridescence. For H. timareta, these include the size and shape of the yellow forewing patch and the presence of red hindwing rays. Heliconius tristero displays a wing pattern virtually indistinguishable from the local race of H. melpomene but has mitochondrial haplotypes and genitalic morphology that place it in the H. cydno clade (Brower 1996a).
These data suggest a scenario in which introgressive hybridization may have moved alleles at certain colour patterning loci from H. melpomene to a largely H. cydno genetic background to produce the recombinant forms H. pachinus, H. heurippa, H. timareta and H. tristero. This hypothesis predicts that the genomes of these species should be a mosaic of the H. cydno and H. melpomene genomes. Furthermore, the hybrid-origin hypothesis makes specific predictions about what proportion of the genome should be derived from each parental species. Since female F1 hybrids between H. cydno and H. melpomene are sterile, the most direct route from initial hybridization to a phenotype similar to any of these putative hybrid species involves one or two generations of backcrossing to pure H. cydno, followed by mating between these new fertile female hybrids and first-generation or backcross male hybrids. Indeed, a series of crosses like this have recently been shown to result in a wing colour pattern identical to that of the suspected hybrid species H. heurippa (Mavarez et al. 2006). Under this scenario, the hybrid-origin hypothesis predicts a mosaic genome with approximately 65–75% derived from H. cydno and 25–35% derived from H. melpomene.
Recently, we generated two comparative genomic datasets which allow us to test the hypothesis that the genome of one of these suspected hybrid species, H. pachinus, is a mosaic (Kronforst et al. 2006a). The first dataset consists of hundreds of amplified fragment length polymorphism (AFLP) markers from a total of 127 H. cydno, H. pachinus and H. melpomene individuals. The second dataset consists of comparative DNA sequence data for 15 genes from all three species and the closely related outgroup Heliconius hecale. Previously, we used these data to infer the extent of admixture and the direction of interspecific gene flow among H. cydno, H. pachinus and H. melpomene in Costa Rica (Kronforst et al. 2006a). The results of these analyses indicated that all three species exhibit detectable contemporary admixture and high levels of introgression. While these findings suggest that the basic conditions necessary for the origin of new species via hybridization exist in Heliconius, they do not tell us whether this has occurred.
Here, we supplement these data with DNA sequences for an additional gene, decapentaplegic, and use them to test whether the genome of H. pachinus is a mixture of the genomes of H. cydno and H. melpomene. Furthermore, we explicitly test the hypothesis that the yellow forewing colour of H. pachinus is derived from H. melpomene by analysing DNA sequence data for the gene wingless, which is very tightly linked to the Mendelian locus that controls this trait (Kronforst et al. 2006b). Finally, we combine our DNA sequence data with published sequences for a second potential hybrid species, H. heurippa, to see whether our conclusions regarding the origin of the H. pachinus genome are likely to apply to this species as well. The results suggest that while there is a large amount of shared genetic variation among all of these species, the genome of H. pachinus is not a mosaic. In contrast, the limited data available for H. heurippa are consistent with the genomic mosaicism expected as a result of hybrid speciation.
2. Material and methods
We collected 56 H. cydno, 44 H. pachinus and 27 H. melpomene specimens from locations throughout Costa Rica. All samples were collected in the field as adults between June and August 2000 or 2002. Tissue was preserved in 95% ethanol and total genomic DNA was extracted using a DNeasy Tissue Kit (Qiagen, Valencia, CA) or a standard phenol/chloroform extraction protocol.
(b) Molecular methods
We genotyped each of the 127 H. cydno, H. pachinus and H. melpomene specimens with AFLPs (Vos et al. 1995) using the ABI Plant Mapping Kit (PE Applied Biosystems, Foster City, CA) and 4 selective primer combinations, i.e. EcoRI-ACT/MseI-CAT, EcoRI-ACT/MseI-CTG, EcoRI-ACA/MseI-CAT and EcoRI-ACA/MseI-CTG. Fragments were separated with an ABI Prism 3100 genetic analyzer (PE Applied Biosystems). AFLP fragments between 50 and 500 bp were sized and scored using ABI Genemapper software v. 3.7 (PE Applied Biosystems). Fragments with a peak height below 100 reflectance units were scored as an absence.
We sequenced multiple haplotypes for 1 mitochondrial and 15 nuclear loci from H. cydno, H. pachinus, H. melpomene and H. hecale (table 1 in electronic supplementary material). The loci are apterous (ap), cubitus interruptus (ci), cinnabar (cn), Cytochrome oxidase (mtDNA—CO), decapentaplegic (dpp), Distal-less (Dll), Elongation factor 1α (Ef1α), engrailed (en), invected (inv), Mannose phosphate isomerase (Mpi), patched (ptc), scalloped (sd), scarlet (st), Triose phosphate isomerase (Tpi), white (w) and wingless (wg). All 15 nuclear loci have been placed on a H. cydno genetic map (figure 1 in electronic supplementary material). Sequences have been deposited in GenBank under accession numbers AY744577–AY744672, AY745254–AY745278, AY745315–AY745335, AY745356–AY745490, DQ448305–DQ448516 and EF041105–EF041122. Additional details regarding sampling and molecular methods can be found in Kronforst (2005) and Kronforst et al. (2006a).
(c) Amplified fragment length polymorphism data analyses
To determine whether H. pachinus originated with a genetic contribution from both H. cydno and H. melpomene, we analysed the AFLP data in three ways. First, to assess whether H. pachinus is genetically intermediate to H. cydno and H. melpomene, we plotted the relative genetic distance among all H. cydno, H. pachinus and H. melpomene individuals using multidimensional scaling (MDS). For this analysis, we computed pairwise Euclidean square distances among individuals using Arlequin v. 2.0 (Schneider et al. 2000) and these distances were then used in a metric MDS analysis which was performed with NCSS v. 2000 (Hintz 2001). To visualize relationships among species, we plotted individuals based on the two dimensions which encompassed the most inter-individual variation. This analysis was performed with three partitions of the AFLP data: the full dataset; the subset of loci that consisted of ancestry informative markers (AIMs)—loci that had significant allele frequency differences between H. cydno and H. melpomene (p<0.05); and the subset of AIMs that retained a significant frequency difference between H. cydno and H. melpomene after Bonferroni correction for multiple comparisons. AIMs were identified by performing a locus-by-locus analysis of molecular variance (AMOVA) with the H. cydno and H. melpomene AFLP data using Arlequin. The p-values for the locus-specific FST estimates that resulted from this analysis were based on a distribution of values obtained by permuting individuals between species 100 000 times.
For our second analysis, we used the model-based clustering approach implemented in the program Structure (Pritchard et al. 2000; Falush et al. 2003) to assess whether the H. pachinus genome is a mixture of the H. cydno and H. melpomene genomes. If H. pachinus originated due to hybridization between H. cydno and H. melpomene, we expect Structure-based clustering to partially assign each H. pachinus individual to each of the parental clusters when clustering is based on the assumption of two populations. Without using prior information regarding the identity of individuals, we inferred genetic clusters and individual genomic admixture for all 127 individuals, assuming both K=2 and 3 populations using the admixture model and 106 iterations of data collection following a 10 000 iteration burn-in. This analysis was performed separately with each of the three partitions of the AFLP dataset discussed previously.
Finally, we further investigated the relationship between H. pachinus and the putative parental species at each of the AIMs. The hybrid-origin hypothesis predicts that the allele frequencies should be similar between H. pachinus and H. cydno for some AIMs, while at others the allele frequencies between H. pachinus and H. melpomene should be similar. To test this hypothesis, we performed pairwise locus-by-locus AMOVAs between H. pachinus and each of the putative parental species for those loci that distinguished the parents. Based on these results, we assigned each locus to one of the four categories: those that associated H. pachinus with H. cydno (loci for which the allele frequency in H. pachinus was not significantly different from the frequency in H. cydno but was significantly different from the frequency in H. melpomene); those that associated H. pachinus with H. melpomene; those that were distinct in H. pachinus (the frequency in H. pachinus was different from both H. cydno and H. melpomene); and those for which H. pachinus was not significantly different from either putative parent. As controls for this analysis, we treated each of the parental species as a potential hybrid between the other two species and repeated the steps: identifying AIMs that distinguished the ‘parental’ species; comparing the ‘hybrid’ allele frequency to each of the ‘parent’ allele frequencies at each AIM; and assigning each locus to one of the four possible categories. We performed these analyses with AIM datasets obtained before and after correcting for multiple comparisons.
(d) DNA sequence analyses
The datasets for the genes Mpi, Tpi, CO and Ef1α were supplemented with sequences available on GenBank (accession numbers Mpi: AF413731, AF413734, AF413739–AF413744, AF516220, AY332417–AY332422, AY332461–AY332464; Tpi: AF413778, AF413782–AF413790, AY329804, AY329805, AY329839–AY329843; CO: U08482, U08483, U08500, U08518, U08520, U08523, U08524, U08544, AF413672–AF413674, AF413679, AF413683, AF413707; Ef1α: AY090168). All of these sequences have previously been reported by Brower (1994, 1996b), Beltrán et al. (2002), Wahlberg et al. (2003) or Flanagan et al. (2004). To examine the evolutionary history of each locus, we estimated a gene genealogy using the neighbour-joining method implemented in Mega v. 2.1 (Kumar et al. 2001) based on uncorrected pairwise proportional differences with gaps excluded in pairwise comparisons. The strength of support for each node was assessed by bootstrapping (1000 replicates). We quantified the overall genetic similarity among H. cydno, H. pachinus, H. melpomene and H. hecale at each locus by computing pairwise FST values using Arlequin. The p-values for these FST estimates were based on a distribution of values obtained by permuting haplotypes between species 1000 times. For the gene wingless, we also tested the support for alternative evolutionary histories using a Bayesian phylogenetic approach. Using MrBayes v. 3.0 (Huelsenbeck & Ronquist 2001), we constrained H. pachinus wg haplotypes to form a clade with those of either H. cydno or H. melpomene, and we compared the marginal likelihoods for these two hypotheses using Bayes factor (Kass & Raftery 1995; Nylander et al. 2004). For each analysis, four Metropolis-Coupled Markov chains were run for 750 000 generations following 250 000 burn-in generations, sampling every 100 generations, starting from a random tree. Parameters were estimated based on the GTR+I+Γ model.
Finally, we supplemented our DNA sequence datasets for two genes, Dll and inv, with published sequences from another putative hybrid species, H. heurippa, to determine whether the genomic patterns uncovered for H. pachinus are likely to apply to other potential hybrid Heliconius species. These additional data, which are from Mavarez et al. (2006), include sequences from H. heurippa as well as the local races of the putative parental species, Heliconius cydno cordula and Heliconius melpomene melpomene (accession numbers Dll: DQ445384–DQ445415; inv: DQ445416–DQ445457). Using the combined data, we estimated neighbour-joining gene trees and computed pairwise FST values among H. pachinus, H. heurippa, H. cydno, H. melpomene and H. hecale.
(a) Amplified fragment length polymorphism data associate H. pachinus with only one potential parental species, H. cydno
We genotyped each H. cydno, H. pachinus and H. melpomene specimen for the presence or absence of a fragment at 664 AFLP loci. Of the markers, 7 were monomorphic (fixed for the presence of a fragment in all individuals) and 88 were singletons (fragment present in only one individual). All three analyses of the AFLP data supported a close genetic relationship between H. pachinus and H. cydno and none suggested a genetic contribution from H. melpomene in the origin of H. pachinus. MDS and Structure-based clustering both correctly identified the three genetic clusters corresponding to the three species. Neither analysis suggested that H. pachinus was genetically intermediate to H. cydno and H. melpomene; the H. pachinus cluster was not located between the H. cydno and H. melpomene clusters in the MDS plot and Structure clustering assuming K=2 populations did not partially assign H. pachinus individuals to each of the H. cydno and H. melpomene clusters (figure 1). These conclusions apply to analyses based on the full AFLP dataset (figure 2 in electronic supplementary material), the subset of 173 loci with a significant (p<0.05) frequency difference between H. cydno and H. melpomene (figure 1) and the subset of 62 loci with a significant (p<0.00008) frequency difference between H. cydno and H. melpomene after correcting for multiple comparisons (figure 2 in electronic supplementary material). Of the 173 loci that distinguished H. cydno and H. melpomene (only one of which was a fixed difference), 116 associated H. pachinus with one parent or the other; 83 had similar allele frequencies between H. pachinus and H. cydno while 33 had similar allele frequencies between H. pachinus and H. melpomene (table 1). The remaining 57 loci did not associate H. pachinus with a particular parent because the allele frequencies in H. pachinus were distinct from (35 loci) or not different from (22 loci) both parents. The control analyses, in which each of the parental species was treated as a potential hybrid, resulted in distributions very similar to the experimental analysis of H. pachinus (table 1). Correcting for multiple comparisons in all steps reduced the number of AIMs in each analysis but resulted in qualitatively similar distributions.
(b) DNA sequence data suggest no detectable contribution of H. melpomene in the origin of H. pachinus
We analysed a total of 557 H. cydno, H. pachinus, H. melpomene and H. hecale haplotypes from 16 loci (table 1 in electronic supplementary material). Neighbour-joining gene genealogies revealed substantially shared genetic variation among species. For instance, for none of the 16 loci did haplotypes from all four species form monophyletic clades (figure 1 in electronic supplementary material). However, for most loci, there was a tendency for H. pachinus haplotypes to group with those from H. cydno. This association between H. pachinus and H. cydno was supported by locus-specific FST estimates; 15 out of 16 loci (all except white) exhibited significant differentiation among species and for all of these the level of differentiation between H. cydno and H. pachinus was less than that between H. melpomene and H. pachinus (table 2).
(c) Analyses of wingless do not support the hybrid-origin hypothesis for H. pachinus
We explicitly tested alternative evolutionary hypotheses for the gene wingless because it is tightly linked to the locus that controls the white versus yellow forewing colour. The ln(L) of the phylogeny in which H. pachinus haplotypes were constrained to form a clade with those of H. cydno (which was the best tree) was −2526.73, while that for the tree in which H. pachinus haplotypes were constrained to form a clade with those of H. melpomene was −2547.70 (figure 3 in electronic supplementary material). Bayes factor (2 ln B10) comparing these hypotheses was thus 41.94, which is indicative of very strong support for the tree grouping H. pachinus haplotypes with those from H. cydno versus the tree linking H. pachinus haplotypes with those from H. melpomene.
(d) The genome of H. heurippa may be a mosaic
By combining our Dll and inv sequence data for H. cydno, H. pachinus, H. melpomene and H. hecale from Costa Rica with 32 Dll and 42 inv sequences for H. heurippa, H. cydno and H. melpomene from Colombia, we found different distributions of genetic variation between the two putative hybrid species. Unlike H. pachinus, which shared only limited genetic variation with H. melpomene across 16 loci, H. heurippa exhibited a discordant pattern consistent with the hybrid-origin hypothesis. At Dll, the majority of H. heurippa haplotypes grouped with those from H. melpomene, while at inv, the majority grouped with H. cydno (figure 2). FST estimates supported this conclusion with less genetic differentiation between H. heurippa and H. melpomene at Dll and less differentiation between H. heurippa and H. cydno at inv (table 2 in electronic supplementary material).
Our analyses of two independent genomic datasets failed to reveal any robust evidence for genomic mosaicism in the putative hybrid butterfly species, H. pachinus. In fact, only the analysis in which we grouped AFLP loci according to whether they associated H. pachinus with H. cydno or H. melpomene provided results that were consistent with H. pachinus being a genetic mixture. That 72% of 116 AIMs associated H. pachinus with H. cydno while 28% associated H. pachinus with H. melpomene is remarkably consistent with the theoretical expectations under the hybrid-origin hypothesis (65–75% from H. cydno and 25–35% from H. melpomene). However, the control analyses showed that a similar distribution results regardless of which species is treated as a hybrid. This finding underscores the substantially shared genetic variation among all three of these species and it highlights the importance of proper controls in comparative genomic analyses of this type.
It is important to note that our finding of no genomic mosaicism in H. pachinus does not definitively answer the question of whether this species originated due to hybridization. This is primarily because hybridization and introgression following the origin of H. pachinus could have erased virtually all evidence of a hybrid origin. For instance, recent work has shown that H. cydno, H. melpomene and H. pachinus hybridize and exchange genes (Bull et al. 2006; Kronforst et al. 2006a), with interspecific gene flow between H. cydno and H. pachinus being particularly extensive. Coalescent simulations based on the isolation with migration model of speciation (Hey & Nielsen 2004) indicate that the population migration rate (2 Nm) from H. cydno to H. pachinus is approximately 0.5 while that from H. pachinus to H. cydno is very large at approximately 4.3 (Kronforst et al. 2006a). Such extensive ongoing hybridization has the potential to largely homogenize the genomes of H. cydno and H. pachinus over time and thereby erase the genome-wide signature of hybrid speciation sought in this study.
Similarly, our analysis of the evolutionary history of the gene wingless does not absolutely preclude the possibility that the yellow forewing colour of H. pachinus originated from H. melpomene. The gene wingless is tightly linked to the locus that controls the white versus yellow forewing colour in H. cydno and H. pachinus (Kronforst et al. 2006b) and it is an excellent candidate gene for this trait. On the developing forewings of another nymphalid butterfly, Junonia coenia, wingless is expressed in bands that eventually contain the ommochrome pigment ommatin D (Carroll et al. 1994; Nijhout 1997). The yellow pigment on Heliconius wings is also an ommochrome, while the white colour is structural (Gilbert et al. 1988). Thus, a gene with ommochrome-associated expression on the forewing of one nymphalid butterfly is also perfectly linked to an ommochrome on/off switch locus on the forewing of another nymphalid butterfly. Together, these observations suggest that wingless may be the Heliconius forewing colour locus (Kronforst et al. 2006b). If this is true, the hybrid-origin hypothesis predicts Heliconius pachinus wingless haplotypes to have originated from H. melpomene haplotypes rather than H. cydno haplotypes. In contrast, our analyses strongly favour an evolutionary scenario in which wingless haplotypes from H. pachinus group with those from H. cydno and thus support a non-hybrid origin for H. pachinus. However, it remains possible that wingless is the gene responsible for wing colour and the yellow colour of H. pachinus was in fact derived from H. melpomene, but recombination between H. pachinus and H. cydno obscured this phylogenetic signal along the entire gene except the actual sequence variation responsible for the colour difference. Future identification of the nucleotide variation actually responsible for the colour switch will truly resolve this question.
It is also important to note that just as introgression has the potential to erase evidence of a hybrid origin, it also has the potential to produce patterns of shared genetic variation that can appear consistent with hybrid speciation. This is especially true with limited genetic data. Hence, the preliminary data available for H. heurippa should not be over-interpreted. The observations that H. heurripa shares genetic variation with both H. cydno and H. melpomene and that the two loci show a discordant pattern in which H. heurippa is more similar to H. melpomene at Dll but more similar to H. cydno at inv are consistent with a recent hybrid origin for H. heurippa (Mavarez et al. 2006). Alternatively, it is possible that H. heurippa originated without a genetic contribution from H. melpomene but has since experienced variable rates of interspecific gene flow at the two loci. It is also possible that the observed patterns are simply a result of ancestral genetic variation that is sorting out differently in different descendent lineages. However, it is perhaps telling that with data for just two loci, H. heurippa exhibits the discordance among loci predicted to result from hybrid speciation. Heliconius pachinus, on the other hand, does not exhibit any obvious discordance among loci even with data from 16 genes. While the H. heurippa genetic data are limited, the contrast with the substantial data available for H. pachinus strengthens the argument for a hybrid origin for H. heurippa. As with H. pachinus, analyses of the evolutionary relationships among species at the actual genes that control the colour pattern elements believed to have originated from H. cydno and H. melpomene will be the real test of whether H. heurippa originated via hybridization.
We thank Ulrich Mueller, Joan Strassmann and Dave Queller for use of laboratory facilities, Laura Young and Lauren Blume for their assistance in the laboratory, Kenny Kronforst and Andres Vega for their assistance in the field, and Chris Jiggins, Andrew Brower and two anonymous reviewers for their comments on the manuscript. Butterflies were collected under permits provided by Costa Rica's Ministerio del Ambiente y Energia. This work was funded by Organization for Tropical Studies' Graduate Research Fellowship and National Science Foundation grants DEB 0206613 and DEB 0415718. M.L. and C.S. were funded by the Fondo Colombiano de Investigaciones Cientificas y Proyectos Especiales Francisco Jose de Caldas COLCIENCIAS grants 1204-05-11414 and 7155-CO, Banco de la Republica and private donations from Continautos S. A., Proficol El Carmen S. A., Didacol S. A. and F. Arango, Colombia.