Martín Tena Meza · Rafael Ma. Navarro-Cerrillo ·Diego Brizuela Torres

Abstract Wild plants represent relatively unexplored resource of high economic potential, especially as an alternative to developing new crops and, even more relevant, for improving existing crops and contributing to nutrition and health. The wild species Malpighia mexicana (manzanita)has a wide tradition of food, medicinal and ornamental use in Mexico. It is part of the American-origin group of tropical shrubs that produce edible red fruits, such as Acerola, which is considered the most important natural source of vitamin C in the world. Given the role played by M. mexicana in Mexico, and particularly in Barranca del Río Santiago(Santiago River Canyon), we modelled its potential distribution in both geographical areas. We used species’ records from databases, local herbaria and records collected by the authors as well as climatic variables representing long term average, variability and extreme conditions of temperature and precipitation. To f it the models we used the modelling algorithm Maxent and selected an adequate conf iguration by testing a range of model complexity settings. The results indicate a clear species preference for warm-dry tropical forest, most extensively in the Balsas river depression and the central valleys of Oaxaca. The probability of the species presence in the western region was also high, although the probability was also high for smaller surface areas, such as the region of Santiago river canyons, which are covered by warm-dry tropical forests.

Keywords Malpighia mexicana · Maxent · Warmdry tropical forest · Río Santiago · Genetic resources ·Ecological niche model · Manzanita

Introduction

Malpighiaceaeis a family of tropical and subtropical plants with 75 genera and 1300 species, of which 80% and 90%,respectively, are considered native to the Americas (Anderson 2013). Some species in theMalpighiaceaefamily are cultivated for their attractive f lowers and sometimes for their winged or edible fruits. The genusMalpighiaL. comprises 50 or more species that produce small drupes, which are dispersed by birds; about half of the species grow in Mexico and Central America. Of the 19 species in Mexico, 12 are endemic and the rest are distributed in Central America(Davis and Anderson 2010; Anderson 2013).

Diff erent species of theMalpighiagenus are known by the name acerola, particularlyM. glabraL., andM. emarginataDC. (and its synonym,M. punicifoliaL.), which began to be cultivated in diff erent parts of the world in the 1940s for commercial purposes. Not only is it the most important natural source of ascorbic acid for food, cosmetic and pharmaceutical industries (Asenjo 1959; Johnson 2003;Anderson et al. 2006), but its fruits are also rich in phenolic compounds, including benzoic acid derivatives, phenylpropanoids, f lavonoids, anthocyanins and carotenoids (Belwal et al. 2018). Extracts and bioactive compounds from the fruit are strong antioxicants and have high antifungal, antitumoral, antihyperglycemic, and radioprotective activity, and they are used to treat infections and bleach skin (Cáceres et al. 1993; de Rosso and Mercadante 2005; Düsman et al.2014; Belwal et al. 2018).

Other species such asMalpighia mexicanaA. Juss(manzanita) are grown in backyards for their edible fruits,although the details of this use are not fully documented. Is distributed in Mexico in the states of Durango to Oaxaca and Morelos (Stanley 1920). Although it is well represented in the western part of the country, it had been more widely distributed and consumed in the region of Barranca del Río Santiago (BRS) in western-central México. In addition to its dietary use, it has been used as an ornamental and medicinal plant since pre-Hispanic times. It is currently used in traditional medicine for the treatment of stomach discomfort,diarrhea, and dysentery and to disinfect wounds (Universidad Nacional Autónoma de México 2009). However, in regions whereM. mexicanagrows wild or is grown as a marginal crop, the plant has not been fully studied; information on fruit characteristics, nutritional properties and vitamin C content is lacking despite its nutritional and commercial potential (Maldonado et al. 2016). The content of vitamin C inM. mexicanafruits found in previous unpublished study,supported by the University Center of Biological and Livestock Sciences at the University of Guadalajara, surpassed expectations. Despites is wide distribution in the western part of the country, and despite the great natural diversity of the warm-dry tropical forests of the BRS, the species has not been studied in this area.

M. mexicanain Mexico is naturally distributed in deciduous tropical forests (Miranda and Hernández-Xolocotzi 1963; Rzedowski and Mcvaugh 1966), although it can also grow in temperate climates and is cultivated outside its area of distribution (Colegio de la Frontera Sur 2015; Maldonado et al. 2016). Warm-dry tropical forests are the most abundant and widely distributed tropical vegetation in Mexico.They are characterized by the loss of foliage during the dry season, high f loristic diversity and high level of endemic species (Trejo 2005).

Ecological niche models (also referred to as species distribution models; see Peterson et al. ( 2011) for a discussion on conceptual diff erences) are a useful approach for the analysis of such a spatially and temporally complex process as the distribution of species. Even though models do not ref lect with precision the mechanisms that lead to the observed distribution of a species, they are useful to describe the relations between species and environmental variables at wide spatial scales. This characterization of species-environment relations can then be used to generate spatial predictions of species occurrence.

Within this context, the general objective of this work was to determine the current distribution patterns and potential of the manzanitaM. mexicanain Mexico to contribute to the valuation and use of this phytogenetic resource on a national level. The specif ic objectives were to estimate the potential distribution of the species and the climatic factors that shape its potential distribution range and analyze the obtained predictions in the BRS in further detail. We hypothesized that the potential distribution ofM. mexicanacoincides with the distribution of the warm-dry tropical forests of Mexico.

Materials and methods

Study area

This paper examines the potential distribution ofM. mexicanain the warm-dry tropical forests of Mexico, with emphasis on western central Mexico, on a portion of the geological fault strike-slip (Rodriguez-Castañeda and Rodriguez-Torres 1992), known as the Canyon of the Santiago River (Fig. 1). Its location between the intersection of the Sierra Madre Occidental and the Neovolcánica Plateau allows for connectivity with the coastal plains, making it a biological corridor between the tempered ecosystems from the central part of the State of Jalisco and the tropical coastal environments. The 72,338 ha portion considered as a subarea of the present study is located north of the city of Guadalajara, bounded by the coordinates 20° 43′ 00″ and 21° 08′00″ N; 103° 13′ 00″ and 103° 53′ 00″ W. The winding canyon includes territory that overlaps protected natural areas(PNA): most of the surface of the Municipal Hydrological Protection Area, known as Barranca del Rio Santiago, and a portion of the PNA Oblatos-Huentitán Canyon. In addition,the BRS presents connectivity with the polygons of the PNA Nixticuil-Sn. Esteban-El Diente, and the recently decreed PNA Barrancas de los Ríos Santiago y Verde.

Inside the BRS are three variants of tropical climates,according to the Köppen classification modified by García ( 1998a): the warm sub-humid Aw o predominates over 72.4% of the surface, while semi-warm sub-humid climates (A)C(wo) and (A)C(w1) cover 15% and 12.5%respectively. The annual total precipitation in accordance with García ( 1998b) is 600 to 800 mm for 55% of its surface and 800 to 1000 mm for the remaining 45%.Although agriculture has been the main productive activity inside the Canyon, most farmers, especially those with a low socioeconomic profile, practice survival agriculture.

Fig. 1 Location of Barranca del Río Santiago (BRS) study area in México and adjoining protected natural areas by type of jurisdiction

Species distribution

A database was integrated with the species records available at GBIF-org. ( 2018), the records from the collections of the Herbarium of the Botanical Institute of the University of Guadalajara (IBUG) and the Herbarium of the Universidad Autonoma de Guadalajara (GUADA), as well as collected specimens of the BRS recorded by the author.This initial database featured 401 records, then reduced to 207 valid records once the duplicates from field observations without botanical collection information and others with inconsistent information were eliminated.

Variables considered

The climatic variables of WorldClim 2.0 (Fick and Hijmans 2017) were used for the modeling at the national level. Among the 19 bioclimatic variables available, we used average annual temperature, hottest month temperature, minimum temperature in the coldest month, temperature annual range, annual precipitation, precipitation in the wettest month, precipitation in the driest month and precipitation seasonality. We chose these variables since they express average and extreme conditions and variability in temperature and precipitation.

For modeling the BRS and its surroundings, we added topographic variables altitude, exposure, terrain ruggedness index (Riley et al. 1999), soil type, and soil texture to the climatic variables.

Debugging records

Various sources ofM. mexicanarecords were used in this study, so the representativeness of the climatic conditions where the species inhabits is likely biased, and taxonomic identif ication errors may exist. Although there is no consensus on how to debug the records to be used in the modeling process, correcting for sampling bias has been widely demonstrated to provide a better description of species’ relations with the environment and therefore better predictive accuracy (Kramer-Schadt et al. 2013; Boria et al. 2014; Fourcade et al. 2014). For the present work, we decided to take an environmental f iltering approach similar to that developed by Varela et al. ( 2014). The records were plotted in a scatter plot, with the axes being the variables of average annual accumulated precipitation and average annual precipitation.Records outside the usual conditions forM. mexicana, particularly those with high rainfall and low temperatures were discarded. Records packed under similar climatic conditions,not allowing more than two records to overlap in the same conditions were also discarded. Of the 207 records remaining, 67 were removed, leaving 140.

Determination of the environment for Maxent

Maxent is a “presence-background” method; it compares the environmental conditions at the pixels with a species with the conditions of the entire study area, taking a sample of 10,000 background pixels (although the number is modif iable) to estimate the average environmental conditions of the study area as a basis for the aforementioned comparison (Merow et al. 2013). This is relevant since the conditions of the background points are interpreted as the total conditions that the studied species can inhabit. The ideal study area from where the background points would be obtained should include all those areas within the range of environmental tolerances and dispersal range of the species and exclude those that are not (Elith et al. 2011). The overlapping of records to be used on the layer of terrestrial ecoregions of Mexico (INEGI et al. 2008) and a satellite image of a GIS server (ArcMap) enabled us to observe that the vast majority of records occurred on warm-dry tropical forests ecoregions, some in the temperate highlands,and the fewest records were very close to warm-wet tropical forests. Given that the ecoregions map was originally developed on a very large scale and that the tropical forests of Mexico are widely variable, where the warm-dry tropical forest is the most widely distributed tropical vegetation in Mexico and its f lora is a mixture of elements of diverse provenance and diff erent entry routes (Rzedowski and Calderón 2013), we decided to incorporate the polygons of these three ecoregions to def ine the extent of the background for modeling ofM. mexicanapotential distribution(Fig. S1 in Supplementary material). Subsequently, based on f ield experience and satellite imagery, the edges of this polygon were adjusted to incorporate those records outside margins but close to a warm-dry tropical forest.Inversely, regions in the northeast and northwest of the country corresponding to the temperate mountain ranges were excluded because they are far from the ecosystems typically inhabited byM. mexicana. The resulting polygon was used to def ine the background extent in the analysis of potential distribution with Maxent.

Determination of the Maxent conf iguration using the ENMeval tool

One relevant aspect of ecological niche modeling is the balance between complexity and generalization. Ideally,the aim is to f ind models that satisfy the imposed restrictions by the input data but not too strictly so that their predictive capacity is maintained when contrasted with independent data (Peterson et al. 2011). Maxent regulates this balance in two ways: by means of feature classes (FC)to be allowed during the elaboration of the model and by means of the multiplier value of the regularization multiplier (RM). FCs are mathematical transformations of the values of each one of the variables and are used to address the complex responses of the species to the climatic variables.

On the other hand, RM regulates the f itting of the generated predictions to the input data. The R package (R Core Team 2015) ENMeval (Muscarella et al. 2014) was used in this study to determine Maxent’s conf iguration in terms of the aforementioned elements. This tool was used to facilitate the conf iguration of the parameters of the algorithm to better adjust them to the input data (species occurrence records and environmental variables) and to our objectives.ENMeval automates Maxent runs in a range of FC and RM conf igurations def ined by the user and evaluates them using six evaluation metrics, including the Akaike information criterion corrected for small samples (AICc). Although there is no consensus on the ideal evaluation metric, the AICc has been used by several authors to compare several models with each other (Warren and Seifert 2011; Fitzpatrick et al.2013), since, in comparison with other evaluation metrics,it consistently allows f inding those models that better balance the complexity of parameters and the goodness of f it to the input data. Test models were run with 0.5 to 4 RM in 0.5 intervals and with the following combinations of FC: L,LQ, H, LQH, LQHP, LQHPT. The results from the multiple models run with the mentioned combinations of FC and RM showed that the conf iguration of RM value of 3.5 and all the feature classes had the lowest values of delta AICc (Fig. S2 in Supplementary Material).

Modeling

The models were elaborated using the Maxent Java interface version 3.4.1 with the conf iguration that was determined to be optimal (RM of 3.5 and all FCs allowed). The models were run using four cross-validation runs. Crossvalidation allows all records to be used at least once to train models and a second one to evaluate them. The chosen type of output that Maxent provides is the most recent one, called cloglog. This new type of output allows the results to be directly interpreted as probability of occurrence (Phillips et al. 2017). The models were f itted using the def ined background polygon and then projected onto the entire Mexican territory.

Reclassif ication of continuous probability predictions in four levels of climatic suitability

To more clearly delineate the regions of climate adaptation by establishing discrete zones, without losing as much information as lost with binary reclassif ication, the original continuous prediction was reclassif ied into four levels of probability: regions with values ranging from 0 to 0.25 were considered as marginal probability, 0.26 to 0.5 as low probability, 0.51 to 0.75 as average probability, and above 0.75 as high probability.

The 0.25 value was used a reference for the following reasons: the common way of transforming continuous probability results in a binary probability, that is, the presence or absence is predicted from the original continuous values through the use of threshold values. Among the results in the Maxent output are threshold values determined under various criteria so that the user can choose among them.One of the laxest threshold criteria is the so-called minimum training presence, which corresponds to the minimum given value of probability in some pixel with a presence record.Another threshold value, which is fairly strict, is the 10 percentile training presence and corresponds to the minimum probability value that 90% of the records with the highest probability values, i.e., probability values are discarded from 10 records with lower probability, and the value which is taken as threshold is the one at the upper limit of that 10%.For the f irst threshold criterion, Maxent delivered a 0.067 value, in the second criterion: 0.361. Given the above it was decided that the 0.25 threshold value was convenient since it would allow reclassifying the continuous probabilities (with continuous values from 0 to 1) in four categories of equal amplitude, while its value stands between a lax criterion and a moderately strict one.

The final models were projected onto the Universal Transverse Mercator coordinates system with Datum WGS 1984 to obtain km 2 areas.

Modeling process at the BRS level

For modeling at the BRS level and its surroundings, as already mentioned, the same climatic variables were used with the additional topographic variables; both groups of variables were bounded to a rectangular window of 3649.3 km 2 , which includes the BRS study area(723.38 km 2 ) and its immediate surroundings. The same modeling process used for the climatic variables in the national model was followed. The topographic variables were obtained and processed as follows: From the global altitude model GTOPO30 (USGS 1996) with a resolution of 30 arc s, altitude, exposure, and terrain irregularity (i.e.,surface heterogeneity) were obtained. Soil type and texture were obtained from the edaphological layer at the 1:50,000 scale of the Institute of Statistics and Geography of the State of Jalisco (IIEG 2019), then they were converted to raster format. The number of records for the species was 20.

Results

Mexican territory

The Maxent-generated model had an area under the receiver operative curve (AUC-ROC) value of 0.89, and the variables that most helped to explain the distribution ofM. mexicanawere annual precipitation and precipitation of the driest month. On the response curves of the most relevant variables,M. mexicanacan be seen to have affi nity for conditions relatively arid: the highest presence probability occurs where annual precipitation is approximately 300 mm. For precipitation of the driest month, the maximum probability occurs around 5 mm.

Figure S3 (in Supplementary material) shows the regions with marginal, low, medium and high probability of occurrence ofM. mexicanafor the whole Mexican territory. Most of the areas with medium (0.51-0.75) to high (> 0.75) occurrence probability (which altogether correspond to 10.41% of the Mexican territory) are coincident with the distribution of warm-dry tropical forest in Mexico, except for its coastline.Some regions with non-marginal probability of occurrence are also found in the ecotones of warm-dry tropical forests with temperate mountain ranges.

The physiographic region with the most extensive areas of medium and high (0.51-0.75 and > 0.75 respectively)probability of occurrence is the Balsas River basin in southern Mexico. The high probability area extends toward the south to the region of the central valleys of Oaxaca and toward the west to the southern region of the State of Jalisco where areas with warm-dry tropical forests converge with oaks and coniferous forests in the Neovolcanic Transverse Axis. The Balsas River basin, the central valleys of Oaxacaand western Mexico support high probability areas intercalated with patches of medium probability such that when taken as a unit they would represent a single continuous,encompassing almost the entire south and western regions of Mexico (excluding the coastline), which can be assumed as the estimated distribution area forM. mexicana.Figure 2 shows a closer view of the larger continuous areas of higher occurrence probability.

Fig. 2 Close-up of the largest continuous areas of medium to high occurrence probability of Malpighia mexicana Juss in the south and western of Mexico

Another region with an important extension of medium and high probability areas occurs toward northeastern Mexico, in eastern-central part of the State of San Luis Potosí, northeast of Guanajuato, the center of Queretaro and some patches that extend to the central zone of the State of Hidalgo. In those areas, temperate forests and warm-dry tropical forests prevail. Another zone with high occurrence probability ofM. mexicanais the western coast and southern end of the Baja California peninsula where arid vegetation and warm-dry tropical forest, respectively,are the dominant vegetation types, although there are no reference specimens for these areas. Other regions with medium and high occurrence probability but of reduced extension are in southern Mexico, the Central Depression of Chiapas, which is occupied by warm-dry tropical forests; toward western Mexico in the area of canyons shared between the northern part of the State of Jalisco and the southern part of the State of Zacatecas; and in the southernmost portion of the Sierra Madre Occidental at its conf luence with the Neovolcanic Transverse Axis, which is also occupied by warm-dry tropical forests.

Considering the entire Mexican territory, the areas of marginal and low probability of occurrence ofM. mexicanaoccupy the largest area, approximately 90% of the country.

It is worth emphasizing that we identif ied a marked tendency for the zones of greater probability to coincide with the warm-dry tropical forests of Mexico. Of the total warmdry tropical forests ecoregion (318,313.54 km 2 ) (INEGI et al. 2008), 166,602.16 km 2 (52.64%) correspond to areaswith more than 0.25 occurrence probability ofM. mexicanapresence. Table 1 shows the areas in each category of occurrence probability for the nationwide analysis with the BRS analysis (see below).

Table 1 Surface categories of probability in Mexico and Barranca del Río Santiago

Area of study in Barranca del Río Santiago (BRS)

The Maxent-generated model had an AUC-ROC value of 0.833. The variable that most helped to explain the distribution ofM. mexicanain BRS was the average annual temperature (61.1%); this variable and altitude together can predict the distribution of the species; the variables that did not contribute to the model were annual precipitation, precipitation in the wettest month and precipitation in the driest month.

In the independent model developed for the focal area,BRS, the largest proportion of its surface (37.8%) corresponds to areas with high probability over 0.75. Figure 3 and Table 1 show that 36.2% of the surface presents medium probability (0.51-0.75), 23.93% presents low probability(0.26-0.5), and only 2.78% of the study area corresponds to marginal probability (0-0.25). The records of the species coincide with the areas of high probability.

Discussion

Our results lead us to accept our initial hypothesis that the potential distribution for the species coincides with the warm-dry tropical forest distribution in Mexico, according to the map of terrestrial ecoregions of INEGI et al. ( 2008).In Mexico, the study and conservation of phytogenetic resources for food has focused on crops with high economic value, neglecting marginalized crops and wild species that supply rural populations with important sources of food.

Compared to Acerola species,M. mexicanahas a better taste, size, and fruit quality for fresh eating. These characteristics give the species a high alimentary and commercial potential, but more research is needed on its nutritional properties, domestication and large-scale cultivation (Maldonado-Peralta et al. 2016). It is important to mention thatM. yucatanea, one of the 12 endemic species ofMalpighiain Mexico is considered a rare species and possibly extinct in the wild (Avilés-Peraza 2016).

Recently, the public availability of botanical and climatic data and the development of modeling algorithms that correlate species occurrence records with environmental coverage have fostered the development and application of ecological niche models and species distribution models (Soberon et al. 2017), providing an approach to increase knowledge on species biogeography regarding its current, potential and future distribution when considering the Effects of climate change.

Fig. 3 Probability of occurrence of Malpighia mexicana in Barranca del Río Santiago (Jalisco, México)

However, to our knowledge, no studies have focused on the distribution or a niche analysis ofM. mexicana. The present study helps def ine its potential distribution, with the largest continuous areas of high probability of occurrence of the species coinciding with the distribution of the drier zones of the warm-dry tropical forests of Mexico.

Apart from the ecotones with temperate mountain ranges,other areas of high probability are found in arid ecoregions of the country such as the Deserts of North America and Southern Semiarid Elevations (sensu INEGI et al. 2008),which were not included in the background used as the calibration area, so that, despite the affi nity shown byM.mexicanatoward areas with low rainfall (300-500 mm), it is necessary to cautiously interpret the predictions for these regions.

We should also bear in mind that the records of presence came from heterogeneous sampling eff orts, so the records could be environmentally biased. Systematic botanical surveys in these regions could conf irm or dismiss the presence ofM. mexicana.

Since this species is only grown as a backyard crop or for ornamental and medicinal purposes, studies are needed to determine whether plants are a new cultivar and to improve known cultivars, developed from species related toM.mexicana.

As for the BRS focal area, the use of distribution records of the area and the incorporation of topographic variables gave a more accurate prediction for the area than the national model. The fact thatM. mexicanais a representative species of the tropical forest, which prevails as dominant vegetation inside the canyon is coincident with the prediction for the species; more than two-thirds of the ravine (74%) have medium or high conditions (more or less in the same proportion) so that the species is present.

That the most important variable in the canyon model is the average annual temperature and not the annual precipitation can be understood based on the greater climate heterogeneity in the area of distribution of the species nationwide and in comparison with the Santiago River Canyon.

The analysis of the foliar structure ofM. mexicanareveals this species as a mesomorphic species with some xeromorphic traits that allow adaptation to variations in temperature and precipitation. At the same time, the stem has specialized tracheids for adapting to dry environments (Bárcenas-Lopez 2018; Barcenas-López et al. 2019), which help it tolerate climatic changes. Our estimates of its potential distribution and description of the area response curves agree with the past studies, as they show that the species tends to prefer the dryer areas in the warm-dry tropical forests and in the other ecosystems it inhabits.

This work represents a f irst approximation to the potential distribution ofM. mexicana. We consider this species is worth promoting in Mexico due to its economic potential(because of its high vitamin C content) and particularly in the BRS as, in addition to potential economic benef its, it provides other environmental services such as conserving water and soil, fostering biodiversity, preserving aesthetic and cultural values and providing traditional foods. This potential can only be realized if communities in areas suitable for its cultivation work together to develop agricultural and commercial schemes.

However, and as mentioned above, most of the plants observed or collected in the BRS had serious plant health problems, unlike 40 years ago. The problems also need to be evaluated to determine whether they are related to climatic or ecological changes.

Conclusions

Besides being a f irst approximation of the distribution ofM.mexicanain Mexico, the present study also shows its presence in the BRS. In other parts of Mexico as well as in the BRS, the typical association ofM. mexicanawith warm-dry tropical forests was verif ied, which provides support to its potential use as a viable crop for regions with this ecosystem. Within warm-dry tropical forests,M. mexicanahas a higher affi nity for drier areas, so we can be assume that it can persist in drier environments than in those in which it is currently found in the event of climate change. Understanding the distribution potential ofM. mexicanais of great benef it in furthering our knowledge of the species in generating optimal cultivars for the ascorbic acid industry, for fruits destined to be eaten fresh or processed, for elaborating functional foods. Due to its traditional presence and consumption by inhabitants of the BRS and its nutritional benef its, a revaluation of the species is necessary along with research and characterization of its populations for its reincorporation into the diet of the BRS dwellers. We should also consider its use in ecological restoration programs and develop suitable agroecological practices for commercial cultivation.

Publisher’s NoteSpringer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affi liations.