Upland cotton, G. hirsutum L., is a species with wild, feral, and semi-domesticated populations (Brubaker and Wendel, 1994). All cultivated forms, including the highly improved varieties or genetically engineered varieties, cannot be considered as fully domesticated, because they are able to survive even if human intervention stops.

Wild G. hirsutum flowers all year round. Flowers are white, hermaphrodite, cup shaped, with a single central style surrounded at the bottom by stamens (Meade, 1918; Smith and Cothren, 1999). Some plants exhibit flowers with a colored disk inside of the base of the cup that ranges from deep red to light yellow (Tan et al., 2013). Flowers remain open between 8 and 11 h; at the start of the day, they are all white and when they close the sepals start turning pink at the base (Smith and Cothren, 1999). Anthesis takes place in the morning, as soon as the flower completely opens, and the stamens start to release pollen soon afterward (Smith and Cothren, 1999). Flowers produce both pollen and nectar as a reward for visitors (Wäckers and Bonifay, 2004).

Sampling was performed in coastal dunes and dry forests of Mexico, in three of the eight wild cotton metapopulations defined genetically, geographically, and ecologically by Wegier et al. (2011), namely: Central Pacific Metapopulation (CPM), Yucatan Peninsula Metapopulation (YPM), and South Pacific Metapopulation (SPM). SPM is of particular interest due to evidence of recent introgressive hybridization with domesticated plants (Wegier et al., 2011). Given the distinctive extinction-colonization dynamic observed in cotton metapopulations (Wegier, 2013), the full extent of CPM, YPM, and SPM was surveyed to find G. hirsutum patches with enough flowers (Figure 1). Hand-pollination treatments (Tate and Simpson, 2004; Machado and Sazima, 2008; Hernández-Montero and Sosa, 2016) were carried out during the dry season, between November 2012 and May 2013: CPM in November 2012, YPM in December 2012, and SPM in February 2013. Sites were revisited for fruit collection after 86 days, on average. In addition, domesticated cotton plants, bought in local markets, were kept under greenhouse conditions in Mexico City to maintain a suitable temperature (Figure 1).

In order to execute the hand-pollination treatments to test for different mating systems, a search was conducted for flower buds before anthesis (Tate and Simpson, 2004; Machado and Sazima, 2008; Hernández-Montero and Sosa, 2016). In each metapopulation, 40 replicates of the five pollination treatments were set up (Table 1), anticipating the risk of collecting too few fruits afterward: assisted self-pollination, automatic self-pollination, assisted cross-pollination (Kearns and Inouye, 1993), emasculated control (to avoid automatic self-pollination), and control (open-pollination). Multiple treatments were placed on the same plant where possible to control for individual variation; however, due to the variability of the number of flowers, not all of the plants held the same type or number of treatments. Moreover, when flowers were scarce, treatments were placed daily within each study site, in up to four patches per metapopulation, until the 40 replicates per treatment were completed. Special care was taken to avoid changing or altering the environment (i.e., without introducing new genotypes or changing plant abundances or distributions). The same experimental design was applied for domesticated plants in a greenhouse that allowed the entry of local insects. Some treatments required mesh bags to exclude any pollinator access that could alter the results (Table 1). The treatments that did not include bagging before anthesis were bagged after flower closure to help control for mechanical damage from bagging.

Fruit-set was calculated as the percentage of recorded fruits produced by each treatment in each metapopulation (Dafni, 1992). In each study site, 20 flowers not involved in the pollination-treatments were collected and brought back to the laboratory in separate sealed containers with 70% alcohol. Each flower was dissected to count the number of ovules present. An average number of ovules was calculated for each wild metapopulation and for domesticated plants. Afterward, seed-set (Schoper et al., 1987; Burd, 1994) was calculated as the percentage of seeds obtained from each fruit for each pollination treatment in relation to the average number of ovules of the study population to which these fruits belonged. Additionally, all seeds were weighed individually to estimate the seed weight per treatment in each study site. Later, all the seeds were germinated individually. Each seed was washed with 2% Captan (PESTANAL^®, Merck) solution and covered with a damp cotton swab; tissue culture lids were used. Seeds were checked daily until all reached emergence of the radicle. While some studies on seed germination consider only a set of seeds (Schemske, 1983; Gil and López, 2015; Raphael et al., 2017; Farooq et al., 2018), we took into account all of the collected seeds for the analysis.

The outcrossing rate (Te) was calculated for each study site following Barrett et al. (1996):

where S is the selfing rate, estimated with the fruit-set results from our selfing (Ws) and outcrossing (Wx) treatments, i.e., automatic self-pollination and emasculated control, respectively. For CPM, Wx was obtained with the fruit-set from the assisted cross-pollination treatment, because none of the emasculated control results were found when revisiting the metapopulation for fruit collection.

To test if there were significant differences in seed-set and seed weight among treatments, a Generalized Linear Mixed Model GLMM (Zuur et al., 2009) was used considering the plant as a random factor, because the pollination treatments were not equally represented in each plant (as explained in section 2.3). For GLMM analyses, a Quasi-Poisson distribution was considered for seed-set and a Gaussian distribution for seed weight (Cayuela, 2009). Afterward, a Tukey post hoc test was performed to evaluate the significance of the results. To compare germination frequencies and percentage of fruit-set, a chi square test was used with the post hoc standardized residue test for each one. Outliers were identified using the method described by Viechtbauer and Cheung (2010); to summarize, a multivariate detection method (Cook distance) was used to calculate the distance among all data points, and those that were not included in the general model were identified as “influential data points” or outlier values. Germination was calculated as the number of germinated seeds in relation to the total number of seeds (Gómez, 2004; Gil and López, 2015). All tests were carried out with the lme4, multcomp, stats, and ggplot2 packages of R version 3.4.3 (R Core Team, 2017). The scripts utilized for the analyses are available online at https://github.com/conservationgenetics/BiologiaReproductiva.git.

All treatments produced fruits regardless of the metapopulation (Table 2). The CPM open-pollinated control showed the highest value of all treatments among all groups; YPM showed the highest fruit-set produced by outcrossing, and the lowest by automatic self-pollination and control treatment. On the other hand, the highest fruit-set was observed for all the treatments in SPM, with exception of the open-pollination control.

Seeds were produced both by outcrossing and selfing treatments in all metapopulations (Figure 2). The average number of seeds per fruit was 15.9 in SPM, 15.4 in domesticated, 12.3 in CPM, and 10.0 in YPM, while the average number of ovules was 16.3 in SPM, 28.6 in domesticated, 15.6 in CPM, and 13.4 in YPM. Regarding seed-set, the control treatment of domesticated cotton was lower than that of wild and introgressed plants [P(χ²) = 4.13 × 10^-4, df = 3] (Figure 3). When evaluating the results of each metapopulation individually, CPM presented seed-set differences between the control and the rest of the treatments (P(χ²) = 0.1 × 10^-4, df = 3), in SPM the differences were found between cross-pollination and all treatments, except emasculated control [P(χ²) = 0.001, df = 4], while in YPM and the domesticated there were no significant differences among treatments (Figure 2). On the other hand, seed weight only presented differences in SPM, between cross-pollination and both assisted and automatic self-pollination [P(χ²) = 0.026, df = 4] (Figure 2). In YPM, CPM, and the domesticated cotton, no significant differences among treatments were observed (Figure 2).

Less than 20% of the seeds from wild metapopulations CPM and YPM germinated. Regarding the domesticated group, 40–63% of the seeds germinated, except for the seeds produced by cross-pollination that only reached 28%. The seeds from the five treatments assessed at SPM showed germination percentages above 83%, with cross-pollination reaching the highest value of 96% (Table 3).

With regard to the germination rate of the seeds that germinated (Figure 4), the slope of the curve suggests that wild upland cotton presents some kind of inhibition to the completion of germination, whereas domesticated populations do not display this behavior. As shown in Figure 4A, domesticated seeds germinated faster, within the first 6 days, whereas seeds presenting evidence of introgressive hybridization (SPM) reached 95% of germination within the first 7 days and continued germinating for 48 days. Unlike domesticated and SPM seeds, the seeds of wild plants germinated over the course of 73 days (Figure 4A). Concerning the pollination treatments from all study sites, 50% of the seeds of all treatments germinated within the first 5 days; however, after the 5th day the difference in germination rate is evident between autogamy and the rest of the treatments (Figure 4B).

All study groups presented outcrossing rates different from 0 and 1 (i.e., 1 > Te > 0), which is indicative of a mixed mating system. Wild metapopulations (YPM and CPM) recorded a higher outcrossing rate (0.72 and 0.71, respectively) than SPM (0.40) and domesticated (0.65) (Supplementary Material 1).

Richards (1997) defined autogamy as within-flower or self-pollination, and allogamy as the pollination between pollen and ovules of different flowers; moreover, he further divided allogamy into geitonogamy (i.e., pollination between different flowers on the same genet) and xenogamy (i.e., pollination between pollen and ovules of different genets). Our results show that wild and domesticated cotton produce offspring in all pollination treatments (Figures 2, 4B and Tables 2, 3); thus, the analyzed plants have the capacity to produce progeny by both autogamy and xenogamy. To discard geitonogamy, it is necessary to perform a molecular genetic analysis of paternity. However, since autogamy is common in our system, there is no need to discard this type of allogamy. Furthermore, previous studies (see Supplementary Material 1), together with our own, indicate that the G. hirsutum wild-domesticated complex has a mixed mating system. This result is particularly relevant in upland cotton’s center of origin, because of its significance on strategies for long-term conservation of genetic diversity in the event of gene flow between wild and domesticated relatives (Ellstrand, 1992).

Barrett and Eckert (1990) and Barrett et al. (1996) described the outcrossing rate (Te), which indicates that when the value is 0.5 the mating system is equally balanced between self and cross-pollination. Any value different from 0 (completely self-pollinated) or 1 (completely cross-pollinated) implies a mixed mating system; when Te > 0.5, the system is predominantly allogamous-xenogamous, whereas when Te < 0.5, the system is predominantly autogamous. Our observed rates vary from Te > 0.5, e.g., 0.71 (CPM), 0.72 (YPM) and 0.65 (domesticated), to Te < 0.5, e.g., 0.40 (SPM). Domesticated plants, and wild CPM and YPM, have a greater contribution of seeds from cross-pollination in the next generation, although the contribution of self-pollination is high and important, and it contributed to maintenance of genetic structure. The high contribution of self-pollinated seeds in SPM is striking, far from being similar or intermediate between wild and domesticated; local factors may be affecting the result and should be addressed in a future study.

To further explore the mixed mating system of the species, we compared the germination rate of seeds produced by different pollination treatments. We found that within the first 5 days the seeds for all treatments reach 50% of germination. After the 10th day, a notorious difference on germination rate (<15%) among autogamy and the other treatments is observed. Such discrepancy is due to the difference in number of seeds produced in each treatment (Figure 4B). As suggested theoretically, when germination does not differ among treatments, self-pollination is not the cause of inbreeding depression (Charlesworth and Willis, 2009). The mating system described in our study coincides with Baker’s law of reproductive assurance (Pannell and Spencer, 1998), where species that migrate long distances colonize or recolonize patches initially by self-fertilization; then, because of its perennial nature, generations overlap in the same area and plants are pollinated by close relatives or by themselves in the absence of pollinators (Kalisz et al., 2004). The information described here, agrees with the ecological and genetic evidence that describes the metapopulation dynamics of G. hirsutum, along with the ability to migrate long distances, historically and currently (Wegier et al., 2011).

In addition, Wegier et al. (2011) reported high values of gene flow among metapopulations in the same study area, which could homogenize genetic variation, but their data exhibit population structure (k = 8) and high F_ST. Self-pollination and cross-pollination seem to maintain the genetic diversity of the species in the wild, although crossings with domesticated members of the complex (Wendel et al., 1992; Wegier et al., 2011) or even domesticated plants of Gossypium barbadense (Brubaker et al., 1993; Brubaker and Wendel, 1994; Ellstrand et al., 1999; Ellstrand, 2014) might be contributing to these results. In addition, gene flow with feral cotton can also take place (Rache Cardenal et al., 2013; de Menezes et al., 2015).

Finally, one of the fitness components measured in plants is seed weight (Primack and Kang, 1989), due to the fact that larger seeds perform better because of the higher amount of resources they possess (Armstrong and Westoby, 1993; Westoby et al., 1996). In our research, seed weight showed no significant difference between treatments within metapopulations (Figure 2).

Our analyses show differences in characters linked to some of the reproductive structures of upland cotton, which can be associated with the domestication syndrome and will be discussed below.

Many nations want to defend the rights of the next generations to enjoy and decide about biodiversity and its services, aware that decisions made today will have an impact on the natural resources available in the future (Bennett et al., 2015; Steffen et al., 2015; Morales et al., 2017). Upland cotton is a remarkably important plant for humanity, not only due to the versatile uses of its fiber, but for many other applications (Wegier et al., 2016). It follows that cotton’s wild-to-domesticated complex and its environment should be a conservation priority. Mesoamerican dry forests and coastal dunes contain the ecosystems and evolutionary processes that originated, mold, and maintain wild cotton diversity and its interactions. These evolutionary services (Faith et al., 2010; Bailey, 2011; Rudman et al., 2017) are essential for species conservation, because preserving this genetic diversity allows the capacity to adapt to environmental changes (Ellstrand, 1992; Hartl, 2000). However, the factors that mold each part of the wild-to-domesticated complex are different; for example, the conservation of native traditional varieties depends to a great extent on the communities that cultivate them, their management techniques, and interests (Zhang et al., 2007). Hence, the parts of the complex that could be used for crop improvement will depend on the objectives of the new processes of domestication and breeding (Ellstrand, 2018; Mastretta-Yanes et al., 2018).

Gene flow between crops and wild relatives should be examined on a case-by-case basis (Stewart et al., 2003), especially when genetically modified organisms (GMO) are involved, because the consequences depend on the nature of the transferred genes and their regulatory mechanisms (Ellstrand, 2003). For instance, a recent study has demonstrated that genetic modifications can affect fitness traits in the long-term (Hernández-Terán et al., 2017). An important issue to keep in mind is that for gene flow to occur, a crop must be within pollination distance of a compatible population (Ellstrand and Hoffman, 1990), but in the case of domesticated plants the distances can be shortened by human activities (Dyer et al., 2009; Wegier et al., 2011). Several studies have documented hybridization events between crops and their wild relatives; for instance, in the United Kingdom, one-third of the 36 species analyzed by Raybould and Gray (1993) hybridize with at least one element of the local flora; in the Netherlands, a quarter of 42 species does (de Vries et al., 1992); and all but one of the 13 crops reviewed by Ellstrand et al. (1999) hybridize naturally with their wild relatives in some part of their agricultural distribution (including G. hirsutum and other species of subgenus Karpas). These hybridization events could lead to a decline in wild genetic diversity, as opposed to native semi-domesticated varieties in traditional Mesoamerican systems where there is evidence that domesticated genomes have formed not only by selection under domestication, but also by gene flow with other closely related populations and species (Rendón-Anaya et al., 2017). For this reason, the wild-to-domesticated dynamics in terms of genetic diversity, reproductive biology, and gene flow should be well understood in the natural distribution of the species of interest, because extrapolating conclusions based on external or incomplete information about species complexes is inconsistent with the objectives of conservation and biosafety (Beebe et al., 1997; Acevedo et al., 2016).

In this study we found that the reproductive capacity of introgressed cotton is greater than that of wild and domesticated plants. This reveals a scenario that de Wet (1968), de Wet and Harlan (1975) and Keeler et al. (1996) had already described, where wild relatives of some introgressed crops can become weeds that are difficult to control. The wide genetic diversity of G. hirsutum, along with factors modified by traditional genetic improvement and modern genetic engineering, will be problematic for agroecosystems (Altieri, 2000) and ecosystem conservation if they increase cotton’s weediness or invasiveness (Schafer et al., 2011). Cotton has already been reported to persist in a few tropical regions, such as the north of Australia, Vietnam, México, the continental United States and Hawaii (Hawkins et al., 2005; Andersson and de Vicente, 2010; USDA, 2018), so it will be necessary to monitor these changes in wild populations given the species great capacity for long distance migration by natural and anthropogenic means.

Finally, local conditions can influence the results of reproductive biology studies (Ellstrand and Foster, 1983; Hucl, 1996; Murray et al., 2002); hence, it was essential to assess the mating system of G. hirsutum within its natural distribution. Some of the factors that have an effect on the results can be associated with the environment (pollen viability, nectar production, and pollinator activity due to environmental conditions; Ahrent and Caviness, 1994; Ibarra-Pérez et al., 1997; Chaves-Barrantes et al., 2014), ecological interactions (foraging rate, floral consistency, efficiency of pollen deposition, interactions with arthropodofauna, and composition of pollinator species; Rudgers, 2004; Kessler et al., 2012; Johnson et al., 2015), as well as the landscape (species abundance and surrounding species distributions; Murray et al., 2002). In this study, the results of automatic self-pollination and the emasculated control provide evidence that autogamy and allogamy occur naturally in upland cotton’s natural distribution. The occurrence of the latter highlights the importance of native pollinators on the reproductive biology of G. hirsutum and, consequently, conservation strategies should take this key interaction into consideration.