Amphicarpic plants: definition, ecology, geographic distribution, systematics, life history, evolution and use in agriculture

ABSTRACT Although most plants produce all of their fruits (seeds) aboveground, amphicarpic species produce fruits (seeds) both above‐ and belowground. Our primary aims were to determine the number of reported amphicarpic species and their taxonomic, geographic, life form and phylogenetic distribution, to evaluate differences in the life history of plants derived from aerial and subterranean seeds, to discuss the ecological and evolutionary significance of amphicarpy, to explore the use of amphicarpic plants in agriculture, and to suggest future research directions for studies on amphicarpy. Amphicarpy occurs in at least 67 herbaceous species (31 in Fabaceae) in 39 genera and 13 families of angiosperms distributed in various geographical regions of the world and in various habitats. Seeds from aerial and subterranean fruits differ in size/mass, degree of dormancy, dispersal and ability to form a persistent seed bank, with aerial seeds generally being smaller, more dormant and more likely to be dispersed and to form a seed bank than subterranean seeds. In addition, plants produced by aerial and subterranean seeds may differ in survival and growth, competitive ability and biomass allocation to reproduction. Amphicarpic plants may exhibit a high degree of plasticity during reproduction. Subterranean fruits are usually formed earlier than aerial ones, and plants may produce only subterranean propagules under stressful environmental conditions. Differences in the life histories of plants from aerial and subterranean seeds may be an adaptive bet‐hedging strategy.


I. INTRODUCTION
Sexual reproduction is a core component of many life-history strategies, and it often serves as the focal point for studies on the evolutionary process (Silvertown & Charlesworth, 2001;Goodwillie, Kalisz, & Eckert, 2005). The persistence of a sexually reproducing species in its natural habitat depends on seed production and recruitment of new individuals to the population . Plant propagation by sexual reproduction can increase the number of individuals and expand the distribution range and increase the genetic variability of the species (Cocks, 1999;Silvertown & Charlesworth, 2001). In flowering plants, reproductive strategies and mating systems vary greatly among species and may involve outcrossing, selfing or some combination of the two (mixed mating) (Goodwillie et al., 2005).
In nature, most plants produce fruits and seeds only aboveground, but some species produce sexual propagules belowground. In the peanut (Arachis hypogaea), for example, the fertilized ovary of the sessile chasmogamous flower penetrates the soil by means of an elongating 'peg', the tip of which enlarges to form the subterranean pod (Darwin, 1888;Kaul, Koul, & Sharma, 2000). However, some species produce both aerial and subterranean fruits and seeds on the same plant, a phenomenon known as amphicarpy (Cheplick, 1987;Kaul et al., 2000;Barker, 2005;Kumar, Lawn, & Bielig, 2012;Koontz et al., 2017). This dual reproductive strategy maximizes fitness, since it combines the advantages of both types of seeds (Cheplick, 1987(Cheplick, , 1994Kaul et al., 2000).
In amphicarpic plants, aerial and subterranean seeds (fruits) differ in size/mass, dispersal ability and/or degree of dormancy. In addition, plants produced by aerial and subterranean seeds often differ in many ways including survival and growth, competitive ability and biomass allocation to reproduction (Table 1). Thus, amphicarpy can be considered a subset of a much larger set of plant species that exhibit reproductive dimorphism (Plitmann, 1995). Consequently, the selective factors responsible for the evolution of amphicarpy may be similar to those proposed to explain the evolution of reproductive dimorphism.
Research by various authors has contributed greatly to our understanding of the differences between aerial and subterranean seeds and the life-history performance of the plants that arise from them. However, the definition of amphicarpy is complex, and its ecological and evolutionary significance has received relatively little attention. The aims of this review are to (i) clarify the definition of amphicarpy; (ii) survey the reported occurrence of amphicarpy in angiosperm families, genera and species; (iii) determine the geographical distribution, habitats and life forms of amphicarpic species; (iv) review the differences in life-history traits of plants derived from aerial and subterranean seeds; (v) discuss the ecological consequences and evolution of amphicarpy; (vi) determine the use of amphicarpic legumes in agriculture; and (vii) suggest future research approaches on amphicarpic species.

Dormancy and germination
Germination response to temperature Subterranean seeds usually germinate to a higher percentage than those of A at a given temperature (Koller & Roth, 1964;McNamara & Quinn, 1977;Weiss, 1980;Walker & Evenson, 1985b;Choo et al., 2015;Zhang et al., 2015; Permeability to water and storage behaviour Aerial seeds are water impermeable and orthodox and S water permeable and recalcitrant (Schnee & Waller, 1986;Zhang et al., 2015) Germination ecology Ability to form a persistent seed bank A > S (Cheplick, 1987;Zhang et al., 2015) Soil burial depth of seedling emergence Greater seed burial depth decreased emergence percentages of A and S. With increase of seed burial depth, plants reared from both A and S seeds allocated more biomass to S seeds (Walker & Evenson, 1985a;Cheplick & Quinn, 1987) Seed germination phenology Usually no difference in timing of seed germination (Cheplick, 1987) Earlier germination of S seeds (Choo et al., 2015) Effect of abiotic and biotic environment on proportion of the two morphs Competition (density) Plants from A seeds had significantly less total growth and seed production than those from S seeds in both high and low competition, and S seed production was less affected than A seed production by density in both A and S plants (Weiss, 1980;Cheplick & Quinn, 1982Ruiz de Clavijo & Jimenez, 1998;Cheplick, 2007;Sadeh et al., 2009;Nam et al., 2017) Fire S seeds (protected from fire) germinated after the fire, whereas A seeds were killed by fire (Cheplick & Quinn, 1987, 1988b Flooding damage during late vegetative growth tended to decrease production of S seeds and total plant mass, while allocation to A seeds did not differ among treatments (Choo et al., 2014) Pathogens Infection by fungi increased pre-reproductive mortality and decreased seed production of small and large plants of Amphicarpaea bracteata, whereas smaller plants were more affected. With an increase in intensity of fungal infection, the proportion of CL seeds (presumably) increased (Parker, 1986) Irradiance Plants from A seeds had significantly less total growth and seed production than those from S seeds under the same irradiance. However, with an increase in irradiance the number of A seeds increased more than that of S seeds for plants from both A and S seeds (Trapp & Hendrix, 1988;Berjano et al., 2014;Nam et al., 2017;Zhang et al., 2017) Litter S seeds on bare soil surface in clay pots were more likely to lose viability and less likely to germinate than seeds protected by litter or by burial in soil (Cheplick & Quinn, 1987) Reciprocal transplanting between populations S seeds placed in the same habitat as the parents produced seedlings of greater vigour and adults of higher reproductive capacity than plants from seeds transplanted to a different habitat far removed from the parents (Cheplick, 1988) Soil moisture Plants from A seeds had significantly less total growth and seed production than those from S seeds at both dry and wet sites (Cheplick & Quinn, 1982) Substrate fertility (nutrients) Plants from A seeds had significantly less total growth and seed production than those from S seeds under the same nutrient availability (Weiss, 1980;Cheplick, 1987Cheplick, , 1989Jiang & Kadono, 2001;Sadeh et al., 2009;Kim et al., 2016) With an increase of nutrient availability, total A/S seed mass of plants derived from both A and S plants increased (Weiss, 1980;Cheplick, 1987Cheplick, , 1989Sadeh et al., 2009;Kim et al., 2016) [Correction added on 26 June 2020, after online publication:  fruits; e, aerial cleistogamous fruits; f, ground-level fruits (basicarps); g, subterranean fruits; h, subterranean and aerial achenes of Catananche lutea (the five fruits of C. lutea from left to right are amphic-1 and amphic-2 produced by ground-level chasmogamous flowers, and peripheral, intermediate and central fruits produced by aerial flowers); i, subterranean stem; j, aerial axillary shoot that arises from the first node that becomes buried in soil. All drawings are by K. Zhang: A modified from Zhang et al. (2015); B modified from Cheplick (1987); C and D modified from Plant illustrations website (http://www. Plantillustrations.org); fruits in D modified from Ruiz de Clavijo (1995); E modified from Ortiz et al. (2009); F modified from Lye (2003); G modified from Plant illustrations website based on the description by Barker (2005); H based on the description by Tindale & Craven (1993) and various online photographs; I modified from Plant illustrations website based on the description by Ernst-Schwarzenbach (1956). Scale bars = 5 cm. found in Glycine arenaria, G. hirticaulis, G. pindanica, G. pullenii and G. tomentella (Tindale & Craven, 1993;Pfeil & Craven, 2002).
Herein, we define amphicarpy as the production of aboveand belowground fruits [amphi-geocarpy of Barker (2005) includes both amphicarpy sensu stricto and amphicarpy sensu lato of ]. Fruits produced near ground level, such as in some Bulbostylis and Schoenoplectiella species of Cyperaceae (Haines, 1971;Bruhl, 1994) and Ceratocarpus arenarius (Amaranthaceae)  are not amphicarpic but amphi-basicarpic. In the aquatic species Blyxa aubertii var. aubertii, B. aubertii var. echinosperma and Ottelia ovalifolia (Fig. 1I), flowers and fruits are produced below the surface of the water and other flowers and fruits above the water surface (Ernst-Schwarzenbach, 1956;Jiang & Kadono, 2001); this pattern of sexual reproduction also is considered to be amphicarpy. Species with opportunistic amphicarpy were not included in our list of amphicarpic species.

III. SYSTEMATIC/PHYLOGENETIC OCCURRENCE AND LIFE FORM
To identify amphicarpic species, an extensive literature search was conducted using online databases. We defined a species as amphicarpic (or not) based on the descriptions and illustrations (where available) of each amphicarpic species in the literature. A few cases of purported amphicarpy in the literature only provided a name, and we searched for the original reference. If the description in the original reference was vague, we verified the presence of amphicarpy by using online search engines, various regional floras and virtual herbaria to locate additional information on the species in question. Amphicarpic species of Clitoria sp., Neocracca sp., Orobus, Eremiti, and Libyella cyrenaica were difficult to verify; nevertheless, we counted these taxa as being amphicarpic. Plants of Pisum fulvum (Fabaceae) exhibit a gradient from both aerial and subterranean flowers and fruits (amphicarpic plants sensu stricto) to plants with only aerial flowers and fruits (Mattatia, 1977a). Also, some plants are basicarpic with CH flowers near the soil surface, i.e. 'sub-amphicarpic' (Mattatia, 1977a). We included P. fulvum in our list of amphicarpic species (but see footnote d to Table S1). To avoid synonymy, we checked the name of each species using the The Plant List (2013; http://www. theplantlist.org). The nomenclature of species, genera and families was updated to reflect currently accepted names (see online Supporting information, Table S1).
The phylogenetic position of orders that contain plant families with amphicarpic species shows that amphicarpy is found mainly in phylogenetically advanced families (Fig. 2). No families of the ANA grade (Amborellales, Nymphaeles, Austrobaileyales) or magnoliids contain species reported to be amphicarpic. Four families of monocots and nine of eudicots are reported to be amphicarpic; 31 of the 67 species are in the Fabaceae (Table 2; Table S1). The phylogenetically widespread distribution of amphicarpy within angiosperms suggests that this reproductive strategy has evolved repeatedly.
We used geo-referenced data obtained from Global Biodiversity Information Facility (GBIF; https://www.gbif.org) to estimate the geographic distribution of amphicarpic species. Raw data were examined by hand and with automated scripts for obvious mistakes such as collection sites that occurred in the sea or where the sign of the coordinates was inverted (Rubio de Casas et al., 2017). With the help of ArcGIS 10.2, a sighting point map was developed. Annual precipitation data obtained from the WorldClim global climate database (www. worldclim.org) were extracted based on species occurrence records and further analysed for their relationship with the presence of amphicarpy. We found that amphicarpic species are widely distributed and occur in temperate, subtropical, tropical and arid/semiarid regions (Table S1; Fig. 3A). The latitudinal range of amphicarpic species extends from 44.5 S to 68.5 N, with the greatest abundance of individuals and species at approximately 25 S (Fig. 3B). Amphicarpic species occur where annual precipitation ranges from 10 to 5300 mm, and they are most abundant at around 1000 mm (Fig. 3C). Therefore, we can conclude that amphicarpic species are not most abundant in desert regions.
In some amphicarpic species in temperate regions, amphicarpy has been viewed as an adaptation to escape fire (Kaul et al., 2000), for example, Amphicarpum amphicarpon (as Amphicarpon purshii) (Cheplick & Quinn, 1987), Commelina virginica (Kaul et al., 2000) and Vigna minima . In Vigna minima, a colonising species from the Western Ghats in India,  argued that amphicarpy is primarily an adaptation to forest fires. Amphicarpum amphicarpon is an annual grass that grows in fire-prone habitats in disturbed sandy areas on the Coastal Plain of the eastern USA from New Jersey to Georgia. Populations of this species respond positively to fire due to protection of subterranean seeds from high temperatures and post-fire germination in an environment relatively free of competing perennial vegetation in the spring following experimental burns the previous autumn (Cheplick & Quinn, 1988b).
Aerial CH flowers are potentially outcrossed, while subterranean CL flowers are obligately self-pollinated (Gopinathan & Babu, 1986;Schnee & Waller, 1986;Ruiz de Clavijo, 1995;Zhang et al., 2006;Kumar et al., 2012). However, CH flowers can also produce inbred seeds by selfing or by pollen from one CH flower fertilizing the ovules of another CH flower on the same plant (geitonogamous selfing) (Cheplick & Quinn, 1986;Trapp & Hendrix, 1988;Stewart, 1994;Zhang, Yang, & Rao, 2005;Liang et al., 2009;Speroni et al., 2009). Four breeding systems occur in the amphicarpic sensu stricto species Vigna minima: (i) crosspollinated CH flowers; (ii) cross-or self-pollinated aerial pseudocleistogamy (i.e. closed and open flowers are morphologically similar, but those that do not open are self-pollinated); (iii) self-pollinated obligate subterranean true CL; and (iv) self-pollinated subterranean pseudocleistogamy. In the latter breeding system, aerial shoots with flowers that do not open grow into the soil, thus placing the developing fruit underground  Trifolium polymorphum reproduces asexually by stolons and sexually by aerial CH and subterranean CL flowers (Real et al., 2007;Speroni et al., 2009Speroni et al., , 2014. Aerial flowers are papilionaceous, with morphological features that are common to entomophilous flowers (i.e. large, brightly-coloured, scent-emitting, and with nectar at the base of the ovary), while the corolla of subterranean flowers is reduced to three petals and an androecium with only three stamens, whose anthers touch the stigma (Speroni & Izaguirre, 2001Speroni et al., 2014). Both flower types have bisporangiate anthers, secretory cells, and a developed endothecium (Speroni & Izaguirre, 2001).
Ontogenetic studies have shown that the embryo sac does not develop apomictically in aerial and subterranean flowers of T. polymorphum (Speroni & Izaguirre, 2001;Speroni, Izaguirre, & Bernardello, 2010). Anthers of subterranean flowers never open, but the pollen tubes grow through the anther wall. Thus, subterranean seeds are produced via obligate self-pollination. Anatomical studies of embryo sac ontogeny and egg cell development of aerial flowers revealed that zygote formation occurs before anthesis (Speroni & Izaguirre, 2001); therefore, self-pollination also occurs in aerial CH flowers. However, studies of intrafloral phenology of T. polymorphum show that aerial CH flowers exhibit morphological and functional characteristics that promote outcrossing and delay selfing (Speroni et al., 2009). When the anthers dehisce prior to anthesis, the stigma is positioned above the anthers. Moreover, pollen viability of aerial CH flowers was maximal 1-2 days following anthesis, whereas stigmatic receptivity was maximal 3-4 days after anthesis. The authors concluded that T. polymorphum is an allogamous, self-compatible species and that seed production is increased by pollinator visitation (Speroni et al., 2009(Speroni et al., , 2014. In the non-amphicarpic cleistogamous, mixed-mating species Impatiens pallida, average effective selfing in CH flowers was >50% (Stewart, 1994). In the amphicarpic annual herb Amphicarpaea bracteata, the typical outcrossing rate was only about 0.5% or less (Parker, 1988). In the amphicarpic perennial herbs Polygala lewtonii (Swift et al., 2016) and Trifolium polymorphum (Real et al., 2007), self-fertilization ranged from 80 to 93% and 60%, respectively. Mixed-mating in amphicarpic species provides a fitness advantage through production of genetically diverse progeny via CH flowers, while preserving locally adapted alleles via CL flowers (Koontz et al., 2017). More biomass (energy) is required to produce CH than CL flowers, and CL flowers may produce more seeds than CH flowers (Schoen & Lloyd, 1984;Oakley, Moriuchi, & Winn, 2007;Winn & Moriuchi, 2009;Koontz et al., 2017). However, CL flowers increase the susceptibility of a population to genetic drift and inbreeding depression if deleterious alleles cannot be purged (Zeide, 1978;Oakley et al., 2007; see papers on purging by Crnokrak &Barrett, 2002 andDart &Eckert, 2013). Fitness trade-offs between CH and CL flowers help maintain amphicarpy as a mixedmating strategy (see  and references cited therein). Furthermore, the genetic variation generated by CH outcrossing could facilitate adaptation of cleistogamous species to environmental change (Oakley et al., 2007).

VI. LIFE-HISTORY DIFFERENCES BETWEEN AERIAL AND SUBTERRANEAN SEEDS AND OF PLANTS DERIVED FROM THEM
(1) Seed morphology and size Generally, amphicarpic plants produce a large number of relatively small aerial seeds and a small number of relatively large subterranean seeds (Walker & Evenson, 1985a,b;Schnee & Waller, 1986;Cheplick, 1987;Kawano et al., 1990;Kaul et al., 2000;Speroni & Izaguirre, 2001;Zhang et al., 2015). Moreover, aerial and subterranean seeds differ in their morphology/structure. Catananche lutea (Asteraceae) produces five types of achenes: subterranean amphic-1 and amphic-2 and peripheral, intermediate and central in an aboveground capitulum. Both types of subterranean achenes are larger than the aerial ones and are not dispersed; amphic-1 achenes do not have a pappus, whereas amphic-2 achenes have a pappus (Ruiz de Clavijo, 1995).
Subterranean fruits of Commelina benghalensis are threeseeded and aerial fruits five-seeded (Maheshwari & Maheshwari, 1955). Kumar et al. (2012) compared aerial and subterranean seeds of three Australian endemic legumes, Flemingia pauciflora, Glycine falcata and Vigna lanceolata, and an exotic species, Centrosema rotundifolium. The aerial dehiscent pods of F. pauciflora were only two-seeded, and those of the other three species were generally multi-seeded and dehiscent. However, the subterranean pods of the four species were thin-walled, non-dehiscent and mostly one-or twoseeded. The aerial seeds were smaller than the subterranean seeds, and in the three endemics they were generally darkcoloured or speckled black. By contrast, the subterranean seeds were uniform in size and more lightly coloured (Kumar et al., 2012).
Aerial and subterranean seeds also may differ in seed coat anatomy. In Amphicarpaea edgeworthii, aerial seeds are dark brown and kidney-shaped, while subterranean seeds are purple-brown and either kidney-shaped or irregular spherical. The aerial seed coat is composed of cuticle, a palisade cell layer (water-impermeable), light line, hourglass cell layer and parenchyma cells and the seed coat of subterranean seeds of a 'pre-palisade' cell layer (water-permeable) and several layers of parenchyma cells. However, subterranean seeds did not have a cuticle or a light line (Zhang et al., 2015).
(2) Seed dispersal Amphicarpy involves more than one dispersal strategy in an individual plant (Table 1). Aerial seeds are generally telechorous, and they are dispersed away from the mother plant. However, burial of subterranean seeds near the mother plant is an effective way of ensuring atelechory, i.e. no dispersal (van der Pijl, 1982). The aerial fruits of Gymnarrhena micrantha are dispersed by wind (anemochory), while subterranean fruits never leave the site of the dead parent plant (Koller & Roth, 1964;Loria & Noy-Meir, 1979/80). These authors stated that functionally subterranean fruits of G. micrantha could be considered as equivalent to dormant vegetative regenerative buds of a perennial.
Peripheral achenes of Catanache lutea have a weakly developed pappus that is subtended by the inner bracts of the capitulum, and they are dispersed when the capitula are released by the dead mother plant (short-range dispersal). Central achenes have a highly developed pappus and are wind dispersed soon after maturity (long-range dispersal). Some intermediate achenes are dispersed like the central ones, while others are dispersed like the peripheral ones. The two subterranean achenes (amphic-1 and amphic-2) are not dispersed (Ruiz de Clavijo, 1995;Ruiz de Clavijo & Jimenez, 1998). In Amphicarpaea bracteata (Trapp, 1988) and A. edgeworthii (Zhang et al., 2015), the dehiscent aerial fruits are dispersed ballistically >2 m, while those of the indehiscent subterranean fruits remain close to the mother plant. A similar phenomenon has also been observed in Lathyrus setifolius and Pisum fulvum from Israel (Mattatia, 1977a,b).
In general, it has been suggested that if there is a high cost to dispersal, such as a high likelihood of seeds moving to an unsuitable site (Bonte et al., 2012), a highly restricted dispersal system could evolve and lead to local adaptation (Cousens, Dytham, & Law, 2008). From the perspective of ecology and evolution, aerial seeds have relatively high dispersal ability, which facilitates their reaching new sites away from the mother plant, thus expanding the geographical area of the population and establishing new populations. On the other hand, subterranean seeds are formed/placed in the vicinity of parental microsites, thus maintaining populations in the safe environment (Koller & Roth, 1964;Ellner & Shmida, 1981;Cheplick, 1987Cheplick, , 1994. Furthermore, sibling competition could arise among seedlings that emerge from subterranean seeds because of the very restricted dispersibility of these seeds (Auld & Rubio de Casas, 2013;Hidalgo, Rubio de Casas, & Muñoz, 2016). The pronounced differences in dispersal of aerial and subterranean seeds may help ensure continued occupation of population sites in unpredictable environments in which seed production may not occur every year. In addition, high dispersal ability of the aerial seeds has the advantage of escaping from seed predators associated with the parent plant and of avoiding pathogens that may become established in dense stands of seedlings near the parent. In a mathematical model, Schoen & Lloyd (1984) showed how selection could favour dispersal dimorphism if CL seeds with limited dispersal are better able to establish within the maternal habitat and CH seeds that are widely dispersed are better able to escape deteriorating conditions of the local habitat; amphicarpic species were used to illustrate some of their arguments.

(3) Seed germination
Aerial and subterranean seeds differ in degree (depth) of dormancy, dormancy-breaking requirements and germination response to light/dark and temperature (Table 1). Subterranean and aerial achenes of the amphicarpic sensu lato desert annual Gymnarrhena micrantha germinated to higher percentages in light than in darkness (Koller & Roth, 1964). In subterranean seeds, the optimal temperature for germination was 15 C in both light and dark, while in aerial seeds 15 C was optimal only in light; germination percentages in darkness increased with a decrease in temperature (Koller & Roth, 1964).
Subterranean seeds of the amphicarpic sensu lato annual Emex spinosa (in Australia) were less dormant than aerial seeds, but scarification followed by treatment with gibberellic acid and kinetin overcame the physiological dormancy of both aerial and subterranean seeds. Germination of subterranean seeds of E. spinosa was less temperature dependent than that of aerial seeds. Aerial seeds required light to germinate to high percentages, but subterranean seeds germinated Biological Reviews 95 (2020)  equally well in both light and dark (Weiss, 1980). However, in another study (in Israel) subterranean seeds of E. spinosa were more dormant than aerial seeds. Scarified (water-permeable) subterranean seeds germinated to higher percentages in darkness than in light at constant temperatures from 5 C to 25 C, whereas scarified (water-permeable) aerial seeds germinated equally well at all temperatures tested (10 C, 15 C, 20 C, 25 C, 30 C, and 35 C) in both light and darkness (Evenari et al., 1977).
In a recent review paper on seed germination of cleistogamous species,   Interestingly, the desiccation-tolerant aerial seeds of Amphicarpaea bracteata and A. edgeworthii have combinational dormancy (water impermeable seed coat plus a physiologically dormant embryo) that is broken by scarification followed by cold stratification, and the desiccation-sensitive subterranean seeds have physiological dormancy that is broken by cold stratification (Zhang et al., 2015). These are the only cases we are aware of in which amphicarpic plants produce recalcitrant seeds.
(4) Seed dispersal/dormancy strategy in (diaspore dimorphic) amphicarpic species does not fit the high-risk/low-risk strategy model In most non-amphicarpic fruit/seed (diaspore) dimorphic species, there are two dispersal-dormancy strategies. One diaspore is easily dispersed and has low (or no) seed dormancy (i.e. high risk, H/H), while the second morph has one low (or no) dispersal ability and high (or relatively high) seed dormancy (i.e. low risk, L/L). Thus, the species has a high-risk-low-risk (H/H-L/L) strategy for seed dispersal/ dormancy-germination (Venable, 1985;Baskin et al., 2013. Theoretically, the H/H morph can colonize new sites away from the mother plant shortly after dispersal, while the L/L morph remains near the mother plant and is slow to germinate due to time/conditions required for breaking dormancy and germination. However, in amphicarpic sensu stricto species Amphicarpaea bracteata (Schnee & Waller, 1986;Trapp & Hendrix, 1988), A. edgeworthii (Zhang et al., 2015) and Amphicarpum amphicarpon (as Amphicarpon purshii) (McNamara & Quinn, 1977) and in the amphicarpic sensu lato species Emex spinosa (Weiss, 1980) and Gymnarrhena micrantha (Koller & Roth, 1964), aerial seeds are dispersed for relatively longer distances (H, high risk) and have higher dormancy (L, low risk), whereas subterranean seeds have low dispersal ability (L) and relatively shallow seed dormancy (H). Thus, this species has a H/L-L/H risk strategy for diaspore dispersal and dormancy, which does not fit the high-risk (H/H)-low-risk (L/L) strategy (Zhang et al., 2015). A conceptual model of dispersal/dormancy strategy is summarized in Fig. 4, and we speculate that most amphicarpic species, at least those in the Fabaceae, fit this model.

(5) Seedlings
Seed mass affects various aspects of plant life history, especially seedling ecology . Differences in seedlings from aerial and subterranean seeds often are associated with differences in seed mass. For a few amphicarpic species, a positive relationship between seed size and seedling survival has been found (Loria & Noy-Meir, 1979/80;Cheplick & Quinn, 1987;Cheplick, 1994). Plants from subterranean seeds often have a competitive advantage (Weiss, 1980;Cheplick & Quinn, 1982) and are more resistant to stress (Koller & Roth, 1964;Evenari et al., 1977;Loria & Noy-Meir, 1979/80;Weiss, 1980;Cheplick & Quinn, 1982;Cheplick, 1983;Trapp & Hendrix, 1988;Ruiz de Clavijo, 1995) than those from aerial seeds. Seedlings of Gymnarrhena micrantha (Koller & Roth, 1964;Evenari, Shanan, & Tadmore, 1971;Loria & Noy-Meir, 1979/80) and Cardamine chenopodifolia (Cheplick, 1983) from subterranean seeds are much larger and more tolerant of low soil moisture and drought stress than those from aerial seeds; however, seedlings from subterranean seeds grow much slower than those from aerial seeds. Seedlings of Polygonum thunbergii (as Persicaria thunbergii) from subterranean seeds had 37% more mass and greater stem length, leaf number and root length than those from aerial seeds. Moreover, stem length of seedlings from subterranean seeds increased more rapidly than that of seedlings from aerial seeds, which allowed them to escape flooding conditions (Choo et al., 2015). By contrast, for another population of P. thunbergii, size of aerial and subterranean seeds was similar, and relative stem length growth rate, biomass allocation and biomass of seedlings from the two seed types did not differ between mother plants grown under different nutrient availability conditions (Kim, Nam, & Kim, 2016).

(6) Development of fruiting structures
Although all amphicarpic species produce subterranean fruits/seeds, the way in which they are formed varies. The underground fruiting structures of the rhizomatous perennials Flemingia pauciflora and Vigna lanceolata develop on rhizomes that originate at underground cotyledonary nodes, whereas in Glycine falcata the rhizomes originate at the junction between the fleshy taproot and the stem. Rhizomes that emerged at the soil surface or from water-drainage holes in the bottom of pots produced new ramets (Kumar et al., 2012). In Cardamine chenopodifolia (Cheplick, 1983(Cheplick, , 1987 Fig. 1B), Commelina benghalensis (Walker & Evenson, 1985a,b), C. forskalaei (Maheshwari & Maheshwari, 1955) and Polygala polygama (Fig. 1C), all subterranean flowers are formed on subterranean stems. However, some of the subterranean flowers of Amphicarpaea bracteata (Darwin, 1888;Schnee & Waller, 1986) and A. edgeworthii Biology of amphicarpy apices of short cotyledonary shoots and on aerial axillary shoots that arise from the first node and become buried in soil. By contrast, subterranean flowers/seeds of Amphicarpum amphicarpon (as Amphicarpum purshii) are formed on the tip of downward-growing tillers. Each tiller ends in a much reduced inflorescence consisting of a single closed spikelet that matures only one seed (technically a caryopsis). The flowers of Catananche lutea (Ruiz de Clavijo, 1995; Fig. 1D), Emex spinosa (Evenari et al., 1977;Fig. 1E) and Gymnarrhena micrantha (Zamski, Ucko, & Koller, 1983) that give rise to subterranean fruits are formed aboveground, and contractile roots pull the basal part of the plant belowground after the basal fruits start to develop following insect pollination.
The Vigna lanceolata complex includes fibrous-rooted annuals and tuberous-rooted perennials as well as erect bushes and semi-erect to prostrate vines and both aerial and amphicarpic reproduction. Amphicarpy has been observed in six morphotypes of this complex, among which two are fibrous-rooted annuals and four tuberous-rooted rhizomatous perennials (Lawn & Holland, 2003). In the four amphicarpic perennials, amphicarpy is habitual, and leafless underground stems or rhizomes give rise to fruiting structures. In the two annuals, specialized underground axillary geotropic stems originate along prostrate aerial stems, penetrate the soil and then produce seeds. One of the annuals exhibited habitual amphicarpy and the other opportunistic amphicarpy, i.e. underground fruiting structures developed only where aerial stems become covered with soil or leaf litter.
(7) Post-seedling growth/survival and seed production Seed size also influences post-seedling plant growth and reproductive output of amphicarpic species (Weiss, 1980;Cheplick & Quinn, 1982;Schnee & Waller, 1986). In general, plants from aerial and subterranean seeds differ in phenology, subterranean/aerial seed production ratio, seed mass, seed number and response to environmental change (Cheplick, 1987;Kawano et al., 1990;Sadeh et al., 2009;Choo et al., 2014;Zhang et al., 2017). Only relatively large plants growing under favourable conditions invest in aerial flowers and fruits, in which case plants growing in stressful conditions may produce only subterranean seeds. Plants of Amphicarpaea bracteata from the large subterranean seeds were much larger than those from the small aerial seeds, which made them more likely to produce aerial seeds (Trapp & Hendrix, 1988). Moreover, larger plants of A. bracteata are less subject to pathogen attack (Parker, 1986).
Plants from aerial seeds of Amphicarpum amphicarpon (as Amphicarpum purshii) (McNamara & Quinn, 1977;Cheplick & Quinn, 1982) and Cardamine chenopodifolia (Cheplick, 1983) flowered later than those from subterranean seeds, while plants from subterranean seeds of Commelina benghalensis flowered later than those from aerial seeds (Walker & Evenson, 1985a). Plants from the relatively small aerial seeds of A. amphicarpon (as Amphicarpum purshii) were more sensitive to competition than those from the large subterranean seeds. At a density of 15 plants per pot (11.4 cm diameter), plants from aerial seeds did not produce aerial spikelets, whereas at densities of 5, 3 and 1 plant(s) per pot 8, 85 and 100% of the plants produced aerial spikelets (Cheplick & Quinn, 1983). When subterranean seeds were planted at a density of 30, 15, 5 or 1 per pot 24, 43, 90 and 100% of the resulting plants, respectively, produced aerial spikelets. Unpublished data from the Cheplick & Quinn (1983) study on numbers of the two seed types in relation to density are shown in Fig. 5A. Aerial seed production was greatly reduced compared to that of subterranean seeds as density was increased in monocultures of plants from both aerial and subterranean seeds. In Polygonum thunbergii, plants from subterranean seeds also produced more seeds of both types than plants from aerial seeds; however, seedling planting density did not significantly affect aerial seed production (Nam, Kim, & Kim, 2017 ; Fig. 5B).
The abiotic and biotic environment, geographic location and plant size can influence the ratio of aerial and subterranean seeds produced by amphicarpic plants. With a decrease in amount of light, water and nutrients, the number of aerial seeds decreased significantly more than that of subterranean seeds in several amphicarpic species (Weiss, 1980;Schnee & Waller, 1986;Cheplick, 1988Cheplick, , 1994Cheplick & Quinn, 1988b;Nam et al., 2017;Zhang et al., 2017). With an increase in nutrient availability, total number of spikelets and seed mass of Amphicarpum amphicarpon (as Amphicarpum purshii) increased, but changes were greater for aerial than for subterranean reproduction. Biomass allocation to subterranean reproduction of this species was not affected by nutrient availability (Cheplick, 1989). Because A. amphicarpon (as Amphicarpon purshii) size also was reduced significantly with density, the allometry of CH (aerial)/CL (subterranean) ratio to vegetative dry mass tended to increase up to a point, but then levelled off to where biomass allocation to CH was about 60% of that allocated to CL in the largest plants (Cheplick, 2007). It was also noted that CH allocation was more variable than CL allocation at low, medium and especially high density, as quantified by coefficients of variation.
In species that produce both CH and CL, the ratio of CH/CL flowers can also be affected by biotic and abiotic factors such as light level, nutrient availability and herbivory (Fukui & Takahashi, 1975;Culley & Klooster, 2007). Production of CL flowers is often favoured under stressful growth conditions (Cheplick, 1987;Trapp & Hendrix, 1988), apparently because they are energetically less costly to produce than CH flowers (Culley & Klooster, 2007; see . With an increase in density in A. amphicarpon (as Amphicarpum purshii), dry mass allocated to CL was remarkably constant, varying from only 13 to 16%. By contrast, CH allocation declined from 8% in the control to 0.6% at the highest density. Thus, the CH/CL ratio decreased precipitously with increasing density (Cheplick, 2007).
Changes in flower type during development are due to alterations in the initial primordial buds (Culley & Klooster, 2007). In Amphicarpaea edgeworthii, aerial CH flowers are papilionaceous, consisting of five sepals, five petals, 10 (9 + 1) stamens and a solitary carpel in the centre. Aerial CL and subterranean CL flowers are composed of sepals, five rudimentary petals, two stamens that produce anthers and a single carpel. Using scanning electron microscopy (SEM), Zhang et al. (2006) found little difference in organ initiation among the three kinds of flowers (i.e. aerial CH, aerial CL and subterranean CL). At the mid-to late-development stage, development of the carpel and the petals and stamens in the aerial and subterranean flowers diverged, at which time aerial CH flowers become papilionaceous but CL flowers do not (Zhang et al., 2006).
In Polygonum thunbergii, aerial seed mass decreased along a gradient of decreasing light availability, but subterranean  (Cheplick & Quinn, 1983) and (B) Polygonum thunbergii (Nam et al., 2017) were grown from subterranean or aerial seeds in monocultures. Density was per 11.4-cm-diameter pot in A and per 30-cm-diameter pot in B. Filled circles with different lowercase letters (subterranean seeds) and open circles with different uppercase letters (aerial seeds) indicate statistically significant differences (P < 0.05) among densities within a group. seeds were not affected (Kawano et al., 1990). Experimental shading also reduced number and mass of aerial but not of subterranean seeds in P. thunbergii (Nam et al., 2017). Production of subterranean seeds of Emex spinosa was less affected by density than that of aerial seeds. In mixtures of plants reared from aerial and subterranean seeds, plants from subterranean seeds had a larger leaf area and larger stems and produced more aerial seeds than plants from aerial seeds; these differences were not apparent in monocultures of plants from subterranean seeds (Weiss, 1980). However, with an increase in nitrogen availability, allocation to aerial seed production in E. spinosa increased from 7 to 46%, whereas allocation to subterranean seeds decreased from 38 to 3% (Weiss, 1980). Thus, production of both aerial and subterranean seeds in this species is quite plastic and exhibits considerable adjustment to abiotic conditions (Cheplick, 1994).
Under limited resources (abiotic stress), amphicarpic plants produce subterranean seeds first (Weiss, 1980;Cheplick & Quinn, 1982;Kawano et al., 1990), and the aerial/subterranean seed ratio is low. However, under favourable conditions the aerial/subterranean seed ratio is high (Koller & Roth, 1964;Cheplick & Quinn, 1983;Kaul, Sharma, & Koul, 2002;Sadeh et al., 2009). Cheplick (1994) concluded that production of aerial seeds is more plastic than that of subterranean seeds. The ability of amphicarpic species to shift the aerial:subterranean seed production ratio may increase fitness over generations by reducing variance in number of offspring left per generation (R o ), a requirement for bet-hedging (Simons, 2011).

(8) Survival strategies
Early production of subterranean CL seeds provides reproductive assurance (i.e. obligate self-pollination in CL flowers results in high seed set), whereas later production of aerial seeds allows plants to increase reproductive output near the end of the growing period according to the availability of resources (Cheplick, 1989). Because there is no guarantee of survival of the plant in an unpredictable environment, subterranean seeds are produced as soon as possible: a 'pessimistic strategy'. On the other hand, if there is little or no risk to future survival of the plant, yield is maximized in the late vegetative stage by production of aerial seeds: an 'optimistic strategy' (Zeide, 1978). For example, formation of subterranean fruits in Gymnarrhena micrantha follows a pessimistic strategy, which ensures that some seeds mature; the number of fruits produced is independent of plant mass (Zeide, 1978). However, production of aerial fruits in G. micrantha follows an optimistic strategy, i.e. number of fruits increases with increase in plant mass. In Polygonum thunbergii, flooding changed biomass allocation to earlier aerial flowering, with potential impacts on the population (Choo et al., 2014). In this species, aerial flowering starts earlier than subterranean flowering (Kim et al., 2016) in contrast to other amphicarpic species. The authors suggested that there were different evolutionary drivers (e.g. flooding) for seed allocation in P. thunbergii, a wetland species, than for amphicarpic species of arid habitats.

(9) Soil seed bank
A persistent soil seed bank may be crucial for adaptation of plant species to unpredictable environments, and it plays an important role in persistence of populations and contributes to future genetic variability . Cheplick & Quinn (1988b) showed that fire killed aerial seeds, but not subterranean seeds, of the amphicarpic species Amphicarpum amphicarpon (as Amphicarpum purshii). Thus, after fire all emerging seedlings were from subterranean seeds (Cheplick & Quinn, 1988b). In Amphicarpaea bracteata and A. edgeworthii, subterranean seeds form a transient seed bank and aerial seeds a persistent seed bank (Zhang et al., 2015). In most amphicarpic species, the seed bank has not been well described, but the limited available evidence suggests that aerial seeds have greater dormancy and perhaps contribute to a persistent seed bank, while subterranean seeds have low dormancy and contribute to the transient seed bank (Cheplick, 1994).

VII. GENETICS AND QUANTITATIVE GENETIC VARIATION
Plants with breeding systems that are a mixture of selfing and outcrossing have reduced levels of genetic diversity within and greater levels among populations (Hamrick & Godt, 1997). Using isozyme electrophoresis, Marshall & Weiss (1982) demonstrated that Australian Emex spinosa is genetically homogeneous within populations and genetically different among populations. The genetic uniformity of local populations of E. spinosa suggests that this species predominantly is self-pollinated. Analysis of 11 microsatellite loci in the amphicarpic species Polygala lewtonii also revealed high levels of inbreeding (Swift et al., 2016). Using simple sequence repeats (SSRs), Zhang et al. (2005) demonstrated that selfing is more common than outcrossing in populations of A. edgeworthii. Liang, Yang, & Rao (2007) used random amplified polymorphic DNA (RAPD) to evaluate the level and pattern of genetic variation in 15 populations of A. edgeworthii. The results revealed a high level of genetic differentiation among, but not within, populations. Thus, the pattern of genetic structure in A. edgeworthii matches that of an inbreeding species.
Populations of A. bracteata vary substantially in disease resistance, enzyme variants and leaf morphology throughout its range. Such differences are an expected consequence of the restricted recombination associated with the mating system of A. bracteata (Parker, 1991). An isozyme analysis of 978 plants from 33 populations of A. bracteata in seven states of the USA (greatest distance between sites was 1000 km) revealed that three cryptic taxa designated as lineages I a , I b , and II could be distinguished. At seven of 18 electrophoretic loci, lineages I and II did not have any alleles in common, while lineages I a and I b differed at one locus (Parker, 1996). When the two dominant lineages are mixed by outcrossing, the progeny have inferior performance relative to both parental types. The reproductive success of F 3 hybrid plants used in a common garden study was about half that of the parental biotype average, which implies that severe outbreeding depression outweighed any positive contribution of heterosis to the fitness of hybrid plants (Parker, 1991). Kartzinel et al. (2016) characterized single nucleotide polymorphism (SNP) variation in 128 individuals of A. bracteata from southern Wisconsin (USA) and assessed the withinand among-population variation at 3928 SNPs. They found three strongly divergent and highly inbred genetic groups showing little relation to site location. However, estimates of among-group migration were low, and <2% of the individuals were hybrids. In fact, the U-shaped distribution of pairwise among-population genetic differentiation coefficient (F ST ) values seemed to indicate centres ('islands') of genomic divergence. These islands may be associated with hybrid incompatibility loci that arose via allopatry (Kartzinel et al., 2016).
Although selfing is an important component of the breeding system of many amphicarpic species, considerable phenotypic variability in quantitative traits exists in populations in which occasional outcrossing occurs and/or individuals are phenotypically plastic (Cheplick & Quinn, 1986, 1988a. In 10 maternal families (i.e. progeny derived from the same maternal parent) of Amphicarpum amphicarpon (as Amphicarpum purshii), 60% of the phenotypic variation was found within families for all characters examined in plants from seeds of either bagged (self-pollination) or open (potentially cross-pollination) panicles (Cheplick & Quinn, 1986). Thus, the mixed CH/CL breeding system of amphicarpic species that allows for some outcrossing coupled with a highly plastic phenotypic response to environmental variables explains the maintenance of considerable genotypic and phenotypic variation in these species. Some theoretical considerations for maintenance of a mixed CH/CL breeding system in cleistogamous species can be found in .
Subterranean and aerial CL flowers offer reproductive assurance when pollinators are rare or absent. Consistent selfing can eliminate (purge) deleterious recessive alleles within populations over time, which could lead to a decrease in the level of inbreeding depression (Oakley et al., 2007). In Amphicarpum amphicarpon (as Amphicarpon purshii), selfing is a major component of the breeding system due to the subterranean CL flowers and to the potential for selfing in windpollinated aerial CH flowers. Seed viability, germination and fitness of progeny did not differ between seeds produced on open versus bagged aerial panicles (Cheplick & Quinn, 1983). Similar results were found by Speroni et al. (2014) from a total of 210 hand-pollinated aerial flowers of the amphicarpic sensu stricto species Trifolium polymorphum. Pollen tube growth rate and seed production of self-pollinated CH flowers were higher than they were for cross-pollinated CH flowers, but the seeds produced after self-and cross-pollination did not differ in size or germination. Subterranean flowers may increase seed production in the event that aerial flowers are not pollinated, while recombination of new alleles in outcrossed aerial progeny may cause them to exhibit heterosis (Culley & Klooster, 2007).
Some components of the amphicarpic reproductive strategy show significant quantitative genetic variation among families (Cheplick & Quinn, 1986), but these are not the traits most closely allied to fitness (Cheplick & Quinn, 1988a). Low variation in fitness-related traits could reflect that genetic fixation and/or developmental canalization has occurred for some traits such as those related to subterranean seed production. Among 60 quantitative characters of 12 maternal families of Amphicarpum amphicarpon (as Amphicarpum purshii) reared from aerial seeds, only 19 characters differed significantly in between-family phenotypic variation. Total variation due to (maternal) family was 14.9%. The highest narrow-sense heritabilities were for biomass allocation and vegetative characters, while subterranean seed set and mass of both seed types had the lowest genetic variation (Cheplick & Quinn, 1988a). Since number and mass of CL seeds correlate with shoot mass across maternal families in A. amphicarpon (as Amphicarpum purshii) (Cheplick, 1994), directional selection on shoot mass may indirectly select for increased subterranean (but not aerial) seed output.

VIII. ECOLOGICAL AND EVOLUTIONARY CONSEQUENCES OF AMPHICARPY
For amphicarpy to evolve and be maintained, subterranean fruits must exhibit fitness advantages distinct from those of aerial fruits. That is to say, for the two morphs to evolve and be maintained in the population or species, the fitness advantage to a plant producing both morphs must be greater than that of it producing only the aerial morph or only the subterranean morph (Lloyd, 1984;Venable, 1985). Various suggestions that have been made from an ecological perspective to help explain the ecological (thus fitness) advantages of plants that produce subterranean fruits (Fig. 6), in addition to aerial fruits, are discussed below.
(1) Burial of subterranean seeds affords partial protection from extremes of heat/cold and predators at the soil surface. Compared with aerial seeds, which often remain exposed on the soil surface, subterranean seeds can remain viable in relatively moist soil near the mother plant in arid environments, which protects the germinated seedlings from rapid dehydration and thus is beneficial to their successful establishment (Cheplick, 1987(Cheplick, , 1994. In Amphicarpum amphicarpon (as Amphicarpum purshii), few aerial seeds produce seedlings successfully because the seedlings on the soil surface often lose water, leading to death. However, subterranean seeds, which are produced 3.5 cm belowground, may remain viable and established seedlings to a high percentage (Cheplick & Quinn, 1987;Kaul et al., 2000). In addition, seeds buried in subterranean microsites may escape death from high temperatures generated on the soil surface by fast-moving fires (Cheplick & Quinn, 1988b;Kaul et al., 2000). At the same time, fire changes the original habitat and growth conditions, creating a suitable environment for seed germination and seedling growth for some species.
(2) Subterranean fruits buried in the vicinity of the mother plant allow for a reduced investment in the energy cost of dispersal (Ellner & Shmida, 1981;Kaul et al., 2000;Bonte et al., 2012). Regardless of the habitat, subterranean seeds remain near the mother plant, where conditions are favourable for seed germination and seedling recruitment Trapp, 1988), and thus presence of the species can be maintained in the maternal habitat. On the other hand, aerial seeds often are dispersed away from the maternal plant and may reach additional habitats suitable for seedling establishment (Cheplick, 1994). (3) Referring to amphicarpic sensu stricto species, CL flowers are energetically less costly to produce in terms of biomass than CH flowers (Oakley et al., 2007;Winn & Moriuchi, 2009; see , resulting in more resources available for seed production. Further, in addition to providing reproductive assurance, the obligately selfing subterranean (CL) flowers have a selfing advantage over CH flowers.
In CL flowers both sets of genes (pollen and ovules) are passed on to the progeny, whereas in outcrossing CH flowers only one set of genes (ovule) is passed on to offspring (Culley & Klooster, 2007). At the individual plant level, then, amphicarpic sensu lato species have a 3:2 (50%) gene transmission advantage over a purely outcrossing plant with CH flowers. Thus, an amphicarpic sensu lato plant contributes two sets of genes to its own offspring and one set, via pollen, to the offspring of others, while an outcrossing CH plant contributes one set of genes to its own progeny and one to the progeny of others. (4) Amphicarpic plants exhibit a high degree of plasticity during aerial and subterranean reproduction. Subterranean fruits usually are formed earlier than aerial ones, and plants can change from production of aerial to subterranean propagules based on the current environmental conditions (Zeide, 1978;Cheplick, 1994;Sadeh et al., 2009). Disadvantages of subterranean CL seed production include a decrease in genetic variation and genetic drift if the population is small, potentially high levels of inbreeding depression and an increase in sibling competition among seedlings from subterranean seeds that remain close to the mother plant (Cheplick, 1994). Disadvantages of CH aerial flower production include a high energetic cost in some species in which showy petals and copious pollen and nectar are produced, reliance on pollinators for pollination/fertilization and a possible cost of dispersal, e.g. moving to sites unsuitable for germination and establishment, or the mother plant spending energy on expensive dispersal features like high levels of sugars in a fruit to attract animal dispersers (Bonte et al., 2012).

IX. EVOLUTION OF AMPHICARPY
Subterranean seed production has evolved independently in phylogenetically widespread taxa, and thus the relevant selection pressure is likely to differ from species to species (Cheplick, 1987;Kaul et al., 2000). Severe stress from herbivores and the physical environment (fire, drought) may be strong evolutionary forces that promote the evolution of amphicarpy in various habitats (Ellner & Shmida, 1981;Cheplick, 1987). Cheplick (1994) hypothesized that Amphicarpum amphicarpon (as Amphicarpum purshii) originated from a subtropical rhizomatous perennial clonal herb such as A. muhlenbergianum. As the perennial grass species migrated northward, it encountered a shorter growing season and restricted resources, and there was a reduction in the number of ramets and seeds produced. Consequently, A. amphicarpon evolved into an annual with short, subterranean tillers. With further reduction in ramet growth in response to a shorter growing season and adverse environmental conditions such as fire and dry soil, A. amphicarpon eventually evolved into an amphicarpic annual. This occurred because fire (or another disturbance) or desiccating conditions destroyed seeds produced above the soil surface. Thus any genotypes with a tendency to produce seed-bearing tillers belowground had a selective advantage in the fire-prone habitats where this species occurs (Cheplick, 1994).
From a perspective of evolution, if one phenotype (e.g. aerial diaspore) has the highest fitness in alternative environments, then constant expression of this phenotype will be favoured over alternative phenotypes (subterranean diaspore) (West-Eberhard, 2003). However, under this scenario, if environmental extremes exceed the tolerances of the phenotype, then it will inevitably go extinct (Moran, 1992;West-Eberhard, 2003;Auld & Rubio de Casas, 2013;Hidalgo et al., 2016). Thus, subterranean seed production in amphicarpic species may be symbolized as an adaptive trait with alternative fitness peaks.
Various mathematical models and empirical data have suggested that natural selection might often favour a dispersal dimorphism and mixed mating system under highly unpredictable environments. Using mathematical and computational modelling, Hidalgo et al. (2016) showed that high dispersal ability is adaptive when unpredictability of the environment is low and inbreeding depression high and that a dimorphic dispersal system and a mixed mating system, i.e. amphicarpy, are adaptive under high environmental unpredictability, especially if inbreeding depression is low. Furthermore, they found that populations with a single dispersal and mating system inevitably go extinct under high environmental stress and high inbreeding depression, whereas populations with mixed strategies were maintained under extreme environmental conditions (Hidalgo et al., 2016).
In amphicarpic species, aerial and subterranean seed morphs differ in dispersal and mating systems, and they have high fitness at some time or place. Thus, production of aerial and subterranean seeds represents the optimal strategy in a wide variety of contexts by ensuring that at least some offspring can function appropriately in a variety of environmental conditions (Venable, 1985;Auld & Rubio de Casas, 2013;Hidalgo et al., 2016). Amphicarpy decreases variance in the number of offspring produced per year and thus increases the geometric mean of the number of offspring across generations, i.e. by bet-hedging (Venable, 1985;Simons, 2011).

X. USE OF AMPHICARPIC LEGUMES IN AGRICULTURE
Worldwide, 90% of human caloric needs are supplied by only 20 species, with 60% of the global crop output coming from wheat (Triticum aestivum), maize (Zea mays), rice (Oryza sativa) and soya (Glycine max) (Massawe, Mayes, & Cheng, 2016). The integration of nutrient-rich new or orphan (understudied/under-utilized) crops into food systems is recognized to have an important role in agriculture (Tadele, 2014(Tadele, , 2019. Legumes can fix atmospheric nitrogen and convert it to ammonium (Varshney et al., 2012;Tadele, 2019), and many amphicarpic legumes have excellent nutritional profiles. For example, the chemical composition of both aerial and subterranean seeds of Amphicarpaea bracteata (Peña et al., 1999) and A. edgeworthii (Jiang, Xu, & Ma, 2006) is similar to that reported for other food legumes. Moreover, the seeds of these two species have a higher protein content than either Bambara groundnut (Vigna subterranea) or groundbean (Macrotyloma geocarpum) (Duke, 1981). In comparison with FAO guidelines, seeds of A. breacteata are deficient in sulfur amino acids but not in tryptophan, which is higher than the FAO requirement (Peña et al., 1999). In Vigna minima, levels of essential amino acids are markedly higher than the FAO/-WHO standard, except for sulfur amino acids . Therefore, amphicarpic legumes could help to protect world food supplies, particularly as challenges such as increasing world population, and climate change, diseases and pests threaten global food crop monocultures (Tadele, 2014).
In the tropics, legumes have a high potential for use as forage plants to improve pastures, prevent soil erosion and improve soil fertility (Cocks, 1999). However, the potential impact of tropical pasture legumes on livestock production is limited by lack of persistence when grown together with grasses and subjected to grazing (Cocks, 1999). Thus, maintenance of a soil seed bank plays a major role in their use . It has been suggested that the production of subterranean seeds has the potential to keep the grazingindependent soil seed bank replenished (Schultze-Kraft, Schmidt, & Hohn, 1997). For example, production of subterranean seeds is a key trait in the ability of the geocarpic species Trifolium subterraneum to persist in sheep-production areas of southern Australia. However, subterranean seeds are difficult to harvest, as in T. subterraneum, which limits the wide use of this species (Morley, 1961). The potential value of amphicarpy in a species includes the benefits of geocarpy and the production of aerial seeds that can be easily harvested.
The main agronomic value of amphicarpy is its role in promoting establishment and persistence of forage species in stressful environments (Cocks, 1999). In heavily grazed, drought-prone regions of temperate west Asia and north Africa, production of subterranean seeds by Vicia sativa ssp. amphicarpa make the species less vulnerable to being grazed out, while the aerial seeds allow it to spread to new areas (Cocks, 1999;. The tuberous roots of Macroptilium panduratum have meristems from which new growth can occur after the aboveground biomass has been destroyed (Schultze-Kraft et al., 1997).
Trifolium polymorphum is a southern South America stoloniferous, amphicarpic legume (Speroni & Izaguirre, 2003;Speroni et al., 2009Speroni et al., , 2014. The species is an important component of natural pastures during winter because it produces good quality fodder, particularly for sheep (Speroni & Izaguirre, 2003). In addition to the soil seed bank formed by subterranean seeds, the meristems on the stolons are below the level of grazing by herbivores, which promotes persistence during the time of vegetative development. Therefore, although the species can be heavily grazed, production of subterranean seeds and accumulation of reserves in the stolons allows it to survive even in soils with a high risk of drought and with continuous, intensive and selective grazing by sheep (Coll & Zarza, 1992;Speroni & Izaguirre, 2003). Moreover, amphicarpy enables the plants to adapt to stressful environmental conditions by changing the allocation of resources from underground to aboveground reproduction and vice versa (Section VI. (7)). Thus, as with other traits that facilitate plant persistence (e.g. water-impermeable seeds) amphicarpy might conceivably be an undesirable trait in systems where persistence is not wanted.
Amphicarpy also may be a useful trait for ensuring the persistence of legumes used as cover crops in association with tree crops or plantation crops. In the Australian sugar industry, the geocarpic species Arachis pintoi is a useful companion crop under trees and in sugarcane fallows, because its seedlings can emerge through the layer of litter left on the soil surface after sugarcane is harvested (Garside, Bell, & Magarey, 2001). Byth, Clements, & Syme (1980) suggested that the introduction of genes for amphicarpy from Macroptilium panduratum into the tropical pasture legume M. atropurpureum might be an innovative way to improve persistence of the latter species in tropical pastures.
If species with amphicarpy are shown not to be suitable for direct use, they may serve as donors of genes for plant breeding programs in other economically important closely related species. Of the Australian amphicarpic legumes, the endemic Vigna lanceolata is the most diverse (Redden et al., 2015) and includes fibrous-rooted annuals, tuberous-rooted perennials, erect bushes and semi-erect to prostrate vines, and it has both aerial (only) and amphicarpic reproduction . Of the 56 parental combinations in a diallelic cross with eight accessions of seven perennial morphotypes, amphicarpy was routinely expressed in the hybrids, indicating that development of perennial accessions with potential for use as forages should be relatively straightforward . Many of the F 1 hybrids were self-sterile. However, hybrids that were at least partially fertile produced many more subterranean than aerial seeds. Thus, amphicarpy plays an important role in perpetuating hybrids that might otherwise not set seeds and thus not survive. F 1 hybrids between the two annual morphotypes exhibited habitual amphicarpy, indicating that it should be relatively straightforward to transfer genes for habitual amphicarpy into the more widely adapted and generally more vegetatively vigorous morphotypes .
Transfer of these characters into the gene pool of cultivated soybean would greatly enhance the diversity of the soybean germplasm. However, the genetic diversity of amphicarpic Glycine species has not been fully explored by soybean breeders. Newell & Hymowitz (1983) performed hybridization studies within and between wild perennial Glycine species and found that pod set was 11% and 8% for intraspecific and interspecific crosses, respectively. The derived fertile diploid hybrid from soybean × G. tomentella (2n = 78) was resistant to the soybean cyst nematode Heterodera glycines (Riggs et al., 1998). A colchicine-doubled F 1 hybrid (2n = 118) between soybean (2n = 40) and G. tomentella (2n = 78) produced approximately 100 F 2 seeds, and most F 2 plants had a high level of fertility, although two of 24 plants had low pollen viability and a large number of fleshy pods (Shoemaker et al., 1990). These results suggest that transfer of economically beneficial genes into soybean from G. tomentella is possible.

XI. IMPLICATIONS FOR FUTURE RESEARCH
Amphicarpy is a special kind of reproductive strategy, and a better understanding of it will contribute to knowledge of the ecology and evolution of plant life history and plant sexual reproduction, given that amphicarpic species share several life-history attributes with non-amphicarpic flowerand diaspore-dimorphic species (Table 3). For example, studies of amphicarpic plants could shed light on the adaptive significance of mixed mating systems (Goodwillie et al., 2005) and dimorphic seed production (Imbert, 2002). For the most part, research on amphicarpy has focused mainly on comparison of the morphology of aerial and subterranean seeds, seed dormancy and germination, seed dispersal and biomass and reproduction allocation in mature plants derived from aerial and subterranean seeds. Several important aspects of the ecological functions of amphicarpy have been overlooked or given little attention, such as seed development and maturation and maintenance of a soil seed bank. Further research on these aspects of amphicarpy should provide additional insight into this plant reproductive strategy.
Previous studies provided information on seed dormancy and germination of amphicarpic plants at the whole-seed level, but little is known about sub-seed physiology and biochemistry or molecular biology of plants with this reproductive strategy. Thus, it is necessary to combine additional physiological and genetic techniques (genomics) to explain the mechanisms underlying the ecological functions of amphicarpy. For example, plants of Gymnarrhena micrantha produce both aerial and subterranean fruits under short days in the field. However, only aerial fruits developed under short days in a greenhouse and only subterranean fruits under long days outdoors (Evenari & Gutterman, 1966). What are the physiological processes involved in the accumulation of different metabolic and structural materials during aerial and subterranean seed development? For example, are there also hormonal interactions between aerial and subterranean seeds that influence their anatomical structure and germination? Why are aerial seeds of Amphicarpaea bracteata and A. edgeworthii physically dormant and orthodox (desiccation tolerant) and subterranean seeds physiologically dormant and recalcitrant (desiccation intolerant)? To answer these questions, a combined study is needed on the germination behaviour of aerial and subterranean seeds that takes into consideration the anatomical, sub-whole-seed physiological and possible genetic differences between aerial and subterranean seeds.
Empirical data also are needed to help explore the selective pressures and ecological factors that favour the evolution of amphicarpy. Cheplick (1994) hypothesized that the southerly-distributed (subtropical) rhizomatous perennial Amphicarpum muhlenbergianum might be ancestral to the northerly distributed annual A. amphicarpon. Genetic markers could be used to compare the two species, and comparative life-history data could be obtained to help support (or refute) the hypothesis. Genetic and ecological comparisons between amphicarpic species and closely related non-amphicarpic congeners might shed light on the types of selective factors that contribute to the evolution of amphicarpy.
In addition, given that aerial and subterranean seeds are formed in different environments (above-and belowground) and at different positions on maternal plants, is the cue for their development in the external environment and/or is it due to some internal factor such as location of the meristem on the plant or some more general condition of the maternal plant? Does the position of the meristem on the maternal plant have any effect on external cues? An Evo-Devo Biology of amphicarpy approach (Cheplick, 2015) to amphicarpy may provide important clues into genetic, environmental and developmental controls on the phenotypic plasticity of this reproductive system. Thus far, little is known about the environmental and genetic determinants of the amphicarpic life history. By estimating the contribution of genetics to life-history variation, insight can be gained as to how natural selection has led to the evolution of amphicarpy, a life-history adaptation that has clearly been successful for some species in some habitats. Quantitative genetics techniques can be useful for estimating the heritability of reproductive traits, determining the consequences of amphicarpy for ecological genetics, studying natural selection of phenotypic traits important to fitness and determining the microevolutionary consequences of phenotypic variation in amphicarpic traits for adaptation to a specific habitat. A mixed mating system has been reported for several amphicarpic species (e.g. Schnee & Waller, 1986;Trapp & Hendrix, 1988;Cheplick & Quinn, 1988a;Zhang et al., 2005). However, the extent to which aerial CH flowers of CL species actually outcross is not known. Also, almost all references to amphicarpy state that subterranean seeds are formed by self-pollination of CL flowers, but there is no experimental evidence that this is the case. Thus, do subterranean flowers produce seeds by selfing or by parthenogenesis, apogamy or apospory? In addition, population genetic structure of amphicarpic sensu lato species is strongly affected by inbreeding via CL flowers (Cheplick, 2007;Swift et al., 2016). Although within-population molecular genetic diversity may be low, high levels of genetic differentiation can occur between populations (Zhang et al., 2005;Liang et al., 2009). Common garden studies can reveal genetic differentiation (or its absence) among populations for phenotypic traits, whereas reciprocal transplant experiments can be used to document adaptation to local habitat conditions (Cheplick, 2015).
Investigations also would benefit greatly if genes responsible for the production of aerial and subterranean flowers in different species could be identified. Various techniques are now available, such as in situ hybridization to detect differences in gene expression patterns in aerial and subterranean seeds, real-time reverse transcription-polymerase chain reaction (RT-PCR) analysis of gene expression in aerial and subterranean seeds, and isolating amphicarpic mutants by γ-ray radiation, that can be used to help in gene identification. In addition, detailed micromorphological characterization and molecular analysis (comparative analysis of transcription and genome levels) of mutants need to be undertaken to elucidate the molecular characteristics of amphicarpy.
It seems likely that additional species with amphicarpy will be discovered in the future. As more amphicarpic species are identified, they can be used to test evolutionary hypotheses about the origin of amphicarpy and phylogenetic relationships among amphicarpic taxa. For example, there are five species in Amphicarpaea. Amphicarpaea africana (tropical Africa) and A. ferruginea (endemic to China) are perennial, and A. bracteata (North America), A. edgeworthii (E. Asia) and A. linearis (endemic to China) are annual. Amphicarpaea africana, A. bracteata, and A. edgeworthii produce aerial CH and CL flowers and subterranean CL flowers, whereas A. linearis and A. ferruginea produce only aerial CH flowers (Turner & Fearing, 1964;Zhang et al., 2015). Molecular genetic and phylogenetic research make it possible to estimate the approximate divergence times and number of times amphicarpy has evolved by comparing amphicarpic species with non-amphicarpic congeners.
Most research on amphicarpy has been carried out on annual species. Thus, information is needed on the life history of perennial species with regard to trade-offs between traits such as clonal reproduction versus sexual reproduction, CH versus CL breeding systems, and seed mass versus seed number of the two morphs. Comparisons of closely related amphicarpic species that differ in life form (e.g. annual Amphicarpum amphicarpon versus perennial A. muhlenbergianum) and possibly habitat requirements would provide insight into why the amphicarpic lifestyle evolved and how it may trade off with clonal (vegetative) growth.
Most amphicarpic plants, especially legumes, are underexploited or underutilized in their agriculture context. Legumes serve as a cheap source of protein for human and animals, and due to their ability to fix atmospheric nitrogen, they contribute to soil improvement (Tadele, 2014). There are 31 amphicarpic legumes. However, due to lack of genetic improvement, most amphicarpic plants are less productive in terms of both quality and quantity than domesticated legumes and would benefit from basic research. To boost productivity and diversify the food system, amphicarpic plants should be given due attention. Major improvements by conventional breeding technologies, such as selection of high-yielding strains, hybridization and use of modern techniques developed for major crops, such as genome resequencing, selective sweep mapping, marker-assisted selection (MAS) of desired allele(s) and allele combination(s), and genome-editing technologies [such as clustered regularly interspaced short palindromic repeats/CRISPR-associated protein 9 (CRISPR/CAS9)], also need to be applied to amphicarpic plants (Lenser & Theißen, 2013). The application of these techniques to amphicarpic plants would not only improve the quality of local diets but could also provide farmers with additional income as the seeds attract a market price well above the average for similar legumes (Tadele, 2009).
Furthermore, 67 species are reported to be amphicarpic, but the ecological significance of the production of aerial and subterranean diaspores has been studied for only a few of them. Whether the conclusions about amphicarpy based on the few taxa that have been studied can be generalized is open to question, and further studies on more species are needed.

XII. CONCLUSIONS
(1) Amphicarpy, the production of aerial and subterranean fruits/seeds on the same plant, maximizes fitness by combining the advantages of two reproductive strategies and is considered to be a bet-hedging strategy.
(2) Amphicarpy occurs in at least 67 herbaceous species belonging to 39 genera and 13 families of angiosperms that are phylogenetically widespread, suggesting that this life-history strategy has evolved multiple times. Forty-six per cent of the 67 known amphicarpic species are in the Fabaceae. (3) Amphicarpic species are less common in arid regions than in temperate and subtropical regions. In some temperate regions, amphicarpy may be an adaptation to escape fire. (4) Seeds from aerial and subterranean fruits differ in size, mass and/or degree of dormancy, dispersal mechanism and distance and ability to form a persistent seed bank, with aerial seeds generally being smaller, more dormant and more likely to be dispersed and to form a (persistent) seed bank than subterranean seeds. Furthermore, plants from subterranean seeds generally are larger, and seedlings are more tolerant of competition and stress and produce more aerial and subterranean seeds than those from aerial seeds. (5) Under limited resources (stress), plants produce subterranean seeds first and then shift to production of aerial seeds if there are sufficient resources. Thus, reproduction is assured (i.e. self-pollination of subterranean CL flowers), and later in the growing season there may be increased reproductive output via aerial flowers. (6) Relatively little is known about the sub-seed physiology of amphicarpic species, including biochemistry and molecular biology. Studies on genetics and proteomics would contribute to a better understanding of the adaptive ecological functions of amphicarpy and the selective forces favouring the evolution of this lifehistory strategy.