Bitter (1911) used a typological species concept for Polylepis in which specimens that were morphologically different to each other were assigned to different species, resulting in a treatment with 33 species. Unfortunately, Bitter had few specimens available and did not know the genus in the field. Consequently, he did not consider the intraspecific morphological variability of the species in his taxonomic study. Based on many more herbarium collections, as well as personal fieldwork, Simpson (1979) had a better idea of the natural variability of the species and used the biological species concept (Mayr 2000) which helped her to reduce the number of species recognized until that time to 15. However, some of the species recognized by her were very variable and, in a subsequent publication (Simpson 1986), she indicated that some of her species would be better treated as “species groups”, although, except for P. tarapacana and P. tomentella, she did not separate them as different species. Later, Kessler and Schmidt-Lebuhn (2006) used a phylogenetic species concept recognizing at species rank, populations with distinct morphology, biogeography and ecology, even if they show hybridization with other species. Based on this narrower delimitation of species, they increased the number of species to 26.

The species concept used here is the general lineage concept of de Queiroz (1998, 2007), in which a species is a segment of an evolutionary lineage at the population level. To separate species, we have used the species criterion of Davis and Heywood (1973) of phenetic similarity, with discontinuities that may reflect geographical, ecological or reproductive isolation. We applied this concept to morphological traits and also to biogeographical and ecological data. Biogeographically, we considered that allopatric populations of morphologically or ecologically different populations are species because they are reproductively isolated from each other and, hence, evolutionarily independent. Ecologically, we considered that morphologically- and biogeographically distinct populations that also have different climatic niches are species because they show different adaptations to the environment. While ecological data are not commonly used to define species, we consider this to be valuable information, especially since, in most cases, there are different ecological conditions within dispersal distance, showing that the ecological differences are not simply due to the different geographical ranges inhabited by the species. For instance, P. simpsoniae and P. weberbaueri, which have previously been considered to be conspecific, grow at different elevations (2500–3800 m and 3500–4970 m, respectively), yet in their distribution ranges, there are mountains that would allow them to grow at lower and higher elevations, respectively. The rather narrow species concept applied here allows us to better identify independent evolutionary entities for the conservation and management of Polylepis species and forests. We do not recognize infraspecific taxa, such as subspecies or varieties as done by Bitter (1911) and Kessler (1995a), because on present knowledge, we are unable to rank the distinctness of population. For this reason, we consider it easier to simply recognize species.

Species delimitation and identification in Polylepis relies on populations instead of single herbarium specimens, because there is phenotypical variability even within a single plant (e.g., leaf size, texture, shape and indumentum). This variability depends on the growing stage of a branch and its exposure (sun or shade). Between individuals of a species, phenotypic variability is even more pronounced depending on genetic background, as well as growing and microhabitat conditions, with individuals from sheltered and humid sites having larger leaves with more leaflets, less dense hair cover and longer inflorescences with more flowers.

We also, for the first time, provide a formal sectional and subsectional infrageneric classification of the genus to facilitate communication. This classification is based on morphological differentiation supported by climatic niches and ploidy levels. In this sectional classification, we clustered species, based on morphological similarity (number of lateral leaflet pairs, indument type of the lower leaflet surfaces, stipule sheaths and/or fruits, leaflet apex shape and fruit shape). This morphologically based separation is broadly supported by biogeography (with most species of a section or subsection replacing each other geographically), similarity in climatic niches and similarity in ploidy levels. In the two larger, more variable sections, we also applied a subsectional classification to group similar species. The orthography of sectional and subsectional names has been designated in accordance with the ‘Shenzhen Code’ (Turland et al. 2018).

The sections defined by us largely correspond to the informal groups as outlined by Simpson (1979, 1986) which are, at least, partly upheld by molecular analyses (see Taxonomic History). However, P. subsericans was placed by Simpson (1979) in her sericea group, although she mentioned that it might be shown to be in an intermediate position between the sericea group and the incana complex. Here, we place it in a separate section Subsericantes. Additionally, Simpson (1979) included P. hieronymi within the sericea group and we are placing it in section Reticulatae, based on morphological similarity with the species in this section.

Our study is based primarily on the examination of the external morphology of herbarium specimens, supplemented with observation of rehydrated material, photographs and living plants in the field. Field studies by TB were conducted in Peru (Ancash, Arequipa, Ayacucho, Cajamarca, Cerro de Pasco, Cusco, Huancavelica, Junín, Lima, Moquegua, Puno and Tacna) and Ecuador (Azuay, Chimborazo, Loja, Napo and Pichincha). MK has studied Polylepis in Venezuela (Mérida), Colombia (Antioquia), Ecuador (Azuay, Napo, Pichincha), Peru (Ancash, Arequipa, Cuzco, Lima and Puno), Bolivia (Chuquisaca, Cochabamba, La Paz, Oruro, Potosí, Tarija) and Argentina (Córdoba, Jujuy, Salta, Tucumán). Together, we have studied all, but three species of Polylepis in the field. Voucher specimens were collected according to standard herbarium techniques. Photographs of plants and associated notes were taken in the field.

All the herbarium specimens representing each species were carefully observed and those spanning the morphological variation and geographical distribution range were chosen for measurements. Characters were measured from corresponding positions on mature, reproductive plants in order to minimize variation due to developmental differences. We have chosen characters for measurements, based in part on those that have been previously used to differentiate species within the genus Polylepis (Bitter 1911; Simpson 1979; Romoleroux 1996; Kessler 1995b; Mendoza and Cano 2012). We examined more than 1400 specimens (including type material) from the Herbaria AAU, COL, CUZ, F, GOET, HUA, LOJA, MEDEL, MERF, MO, NY, QCA, US, USM, VEN and Z/ZT and the material available on Jstor (https://plants.jstor.org/) and other online Herbaria (COL, F, NY, US). The terminology used for describing the morphological characteristics of the species was based on Simpson (1979), Hickey and King (2000) and Stearn (2004).

We combined occurrence records, based on herbarium specimens with field locations and observations from previous studies from multiple sources into a comprehensive, relational database. This database comprises approximately 3,250 quality-checked records of all species of Polylepis. All coordinates were cleansed and georeferenced if needed. We created the distribution maps using QGIS 2.18.14.

We extracted Mean Annual Precipitation (MAP) and Mean Annual Temperature (MAT) as climatic variables from the global climatic model CHELSA version 1.2 (Karger et al. 2017) working with data points from the record database (see above). We used R version 3.0.5 to analyze the climatic data applying the R packages “devtools” and “easyGgplot” to determine and visualize the climatic differences between species. To illustrate the environmental preferences of Polylepis species, we used the function boxplot. We ran ANOVAs to analyze the variance on the climatic variables among the Polylepis species to define their climatic niche differences, followed by Tukey HSD tests to determine the significant differences among the Polylepis species.

We based our conservation assessment on the International Union for Conservation of Nature guidelines (IUCN 2015, see <http://www.iucnredlist.org>). A full assessment following IUCN criteria gives a comprehensive evaluation of extinction risk, based on population size reduction (Criterion A), geographic range (Criterion B), population size (Criteria C and D) and qualitative estimates (Criterion E). We used the package ‘ConR’ (Dauby 2018) to estimate the geographical range parameters for a preliminary assessment of the conservation status following Criterion B. Extinction risks were assessed globally.

In his revision of the genus, Bitter (1911), established the two sections Dendracaena and Gymnopodae. He also recognized 11 informal groups, but he never designated if he created these as subsections or series. In the classification presented by Simpson (1979, 1986), three informal and apparently natural species groups were defined, based on morphological similarity and ecological specialization rather than a phylogenetic concept. These informal groups agreed with Bitter’s sections, except that his section Gymnopodae that was split into two. All species under Bitter’s section Dencracaena were placed in the sericea group, whereas section Gymnopodae was divided into the reticulata group and the incana complex. Simpson’s informal classification has a certain level of agreement with the AFLP phylogeny presented by Schmidt-Lebuhn et al. (2006a) which provided evidence that the sericea group represents a paraphyletic grade that subtends the other two groups (reticulata group and incana group) which are each considered to be monophyletic.

The infrageneric classification adopted here includes sections partly corresponding to the three informal and meaningful groups previously defined, based on morphology and supported by climatic niches and ploidy levels. We propose the five sections Sericeae, Reticulatae, Subsericantes, Australes and Incanaee (Table 1). Section Sericeae includes species with usually sericeous lower leaflet surfaces and/or stipule sheaths, usually with many pairs of lateral leaflets and fruits with variable numbers of spines. Within this section, we recognized four subsections: Lanuginosae, Pauta, Sericeae and Pepea. Species in section Reticulatae have relatively few lateral leaflets pairs, rugose or shiny upper leaflet surfaces and emarginate leaflet apices. Section Subsericantes includes three species with characters intermediate between those of sections Sericeae and Incanaee, with only one pair of lateral leaflets, pilose or strigose hairs on the leaflets and fruits with 3–4 irregular flattened ridges with a series of spines. Section Australes includes two species with numerous, largely glabrous leaflets and very distinctive winged fruits. Finally, species in section Incanaee usually have few lateral leaflet pairs (often only one), frequently glabrous upper leaflet surfaces and fruits with variable number of flattened ridges with a series of spines. Within this section, we recognize three subsections: Racemosae, Besseria and Incanaee. We refrain from using Bitter’s (1911) sectional names because his classification differs in important aspects from ours and we consider that keeping distinct names for different classifications will avoid confusion.

All species of Polylepis are woody plants growing as trees, multi-stemmed trees or shrubs (Fig. 1). As a result of their peculiar branching pattern, young plants of all species display shrub-like growth in the first few years. Adult height is usually between 1 m and 20 m, but trees of up to 32 m have been measured (Hertel and Wesche 2008). Simpson (1979) claimed that P. pepei is the only species of the genus to have a purely shrub-like growth form, but we have seen single-stemmed individuals up to 3 m tall that thus fulfil the criteria of a tree (Miehe et al. 2007). On the other hand, P. microphylla, P. nana and P. rodolfovasquezii mainly also have shrub-like growth, even though occasional trees are found in P. microphylla and P. rodolfovasquezii. Many individuals of P. tarapacana also only reach shrub form at their distributional limit in arid south-western Bolivia. Generally speaking, tree height within species decreases with decreasing temperatures (at high elevations) and precipitation (Hoch and Körner 2005; Kessler et al. 2014; Camel et al. 2019a), but in arid regions, there are complex interactions, so that shady (colder but more humid) habitats have taller trees than sunny (warm) slopes (Kessler et al. 2007). In many places, excessive logging or fires destroy the apical meristem, leading to the formation of numerous new shoots and thus to shrub-like growth. The trunk and branches usually have a gnarled, often twisted habit; occasionally the trunk lies on the ground for several meters (Young 1993). The trunk can be over 1 m thick (Koepcke 1961; Young 1993; Hertel and Wesche 2008) but is usually much slimmer.

Figure 1. 

Growth habit of Polylepis species: tree growth form: A P. microphylla, Chimborazo, Ecuador B P. fjeldsaoi, Lucanas, Peru C P. rugulosa, Moquegua, Perú D P. sacra, Mantanay, Cusco, Peru E P. acomayensis, Paruro, Cusco, Peru F P. pilosissima, Lima, Perú G P. simpsoniae, Cajas, Azuay, Ecuador; shrubby growth form H P. pallidistigma, Azángaro, Puno, Peru I P. tarapacana, Santa Rosa, Puno, Peru J P. microphylla, Chacan, Cusco, Peru. Photographs A E. Bastidas B, C, E E.G. Urquiaga F D, F–I T.E. Boza E.

The bark of Polylepis is one of the characteristic features of the genus. Indeed, the name Polylepis is derived from the Greek words poly (many) and lepis (layers, skins), referring to the shredding, multilayered bark that is common to all species of the genus. The bark can be made up of more than 100 such layers (Miyagawa 1975; Kessler 1995a) and may be up to 3 cm thick, but is usually no thicker than 1 cm (Kessler 1995a). Within the genus, there are differences in the thickness, structure and color of the bark. In section Sericeae, species. such as P. pepei, P. rodolfovasquezii and P. sericea, have a bark that is thin, flakes off in relatively thick, long stripes and, when fresh, has a light brown to orange-brown color (Fig. 2). The bark of members of section Reticulatae, such as P. hieronymi, has a similar structure, but a more grey-brown color (Kessler 1995a). In species of sections Australes, Incanaee and Subsericantes, the bark typically is significantly thicker, flakes off in thin, short pieces and has a deep red-brown color. Despite these general trends, because the bark characteristics also change with tree size and environmental conditions (wind, cover of epiphytes, etc.), there is much variation in bark development between individuals of the same species and bark characters are not of taxonomic value at the species level.

Figure 2. 

The multi-layered, shredding bark characteristic of all species of the genus A P. multijuga Cajamarca, Peru B P. hieronymi cultivated at Zurich Botanical Garden C P. humboldtii Chimborazo, Ecuador D P. pepei La Paz, Bolivia E P. incana Papallacta, Ecuador F P. pauta Ecuador G P. sericea Colombia H P. canoi Junin, Peru. Photographs A, E E.G. Urquiaga F. B, C, F T.E. Boza E. D A. Fuentes G A. Möhl H H.R. Quispe.

The branching of Polylepis is sympodial (Bitter 1911; Simpson 1979, 1986; Kessler 1995a). Polylepis trees tend to have twisted stems and branches which might be related to the windy, cold and arid habitats (Simpson 1979). The branches often show a striking arrangement of the leaves: on young shoots, the leaves are usually all closely clustered at the top, causing a shrub-like growth, whereas the basal internodes stretch rather significantly afterwards, typically for 5–12 cm (Bitter 1911; Simpson 1979; Kessler 1995a) (Fig. 3). This branching mode is a typical feature of the genus but appears not to be useful to distinguish species. Only the degree of compression of the end rosette varies between species, with species of section Sericeae having less compressed shoots (Bitter 1911). However, this character is difficult to quantify, since its expression can vary greatly depending on the location and age between individuals of a species and even on a plant, for example, on shade and sun shoots (Kessler 1995a). In Polylepis, the petioles usually remain on the branches for a long time even after the leaves have fallen off.

Figure 3. 

Branching patterns of Polylepis: petioles remaining on the branches A P. rugulosa Moquegua, Peru F P. incana Napo, Ecuador; leaves closely clustered at the top of the branches B P. pallidistigma Puno, Peru E P. rodolfo-vasquezii Huancavelica, Peru; twisted stems and branches C P. humboldtii Chimborazo, Ecuador D P. sacra Cusco, Peru. Photographs A–D T.E. Boza E. E G. Vargas F E.G. Urquiaga F.

Another characteristic feature of the genus is the growth of the two stipules fused around the branch, forming a sheath. The sheaths are congested at the ends of the shoots and shaped like tubes nested inside each other (Fig. 4). These remain intact for several years even after the leaves have fallen off and provide some taxonomically important features: a) Small spurs can be found on both sides of the petiole at the upper margin. Their size, shape and hairiness are relatively constant within the species; b) The hair type on the outside of the stipule sheaths also varies significantly between the species; c) Long, coarse or woolly hair often extends from the inside of the stipule sheaths beyond the edge of the sheaths. With increasing age, the stipule sheaths become bald; an examination of young shoots is therefore necessary (Simpson 1979; Kessler 1995a). However, short glandular hairs are often easier to find on older stipule sheaths than on young sheaths, since they are then no longer covered by longer wool or felt hairs.

Figure 4. 

Stipule sheaths congested at the ends of the shoots A P. multijuga Cajamarca, Peru B P. reticulata Azuay, Ecuador C P. ochreata Pichincha, Ecuador D, E P. fjeldsaoi Ayacucho, Peru F P. weberbaueri Ancash, Peru G P. subsericans Cusco, Peru. Photographs A, D–F E.G. Urquiaga F. B, C, G T.E. Boza E.

The types and density of hairs represent some of the most important taxonomic features within the genus, although there are significant fluctuations especially in the length and density of the hair within species (Fig. 5). Hairs can be found on the stipule sheaths, petioles and rachises, under- and uppersides of the leaflets, flower bracts, sepals, stamens, styles and fruits, often with different hairs on different organs.

Figure 5. 

Lower leaflet surfaces of Polylepis species with different types of hairs. Lanate: A P. lanata B P. serrata. Pannose: C P. besseri D P. rugulosa. Sericeous: E P. rodolfo-vasquezii F P. sericea. Tomentose: G P. microphylla H P. reticulata. Pilose: I P. flavipila J P. pilosissima. Villous: K P. acomayensis. Puberulous: L P. neglecta. Strigose: M P. subericans. Photographs by T. E. Boza E.

Bitter (1911) distinguished two groups of hairs: multicellular, mostly glandular capilli resiniferi or capilli pulverulenti and unicellular, non-glandular hairs as pili. Since there is a great variability in this last group in particular (Simpson 1979; Kessler 1995a; Romoleroux 1996), we here recognized eight hair types, with the terms following Hickey and King (2000).

1. Lanate: woolly, long, interwoven hair that gives a rough, woolly impression.
2. Pannose: felt-like, composed of densely matted woolly hairs.
3. Pilose: softly hairy, with short hairs.
4. Puberulous: slightly hairy (minutely pubescent).
5. Sericeous: silky, short to long, straight, smooth-fitting hair that give a silky impression.
6. Strigose: long, straight, rough or stiff hairs or bristles that give a rough-haired impression.
7. Tomentose: densely covered in soft and very curled hairs.
8. Villous: covered with long, shaggy hairs.

In some species, glandular hairs are intermixed with the longer hairs, often tinting the latter ones yellowish, as in P. flavipila and P. incana. In other species, only glandular hairs are found on some organs. In the extreme case of P. tarapacana, the resin forms a thick, translucent layer on the upperside of the leaflets.

The imparipinnate leaves provide some of the most important taxonomic features within the genus, especially since their features are often correlated with those of the inflorescences and fruits. In addition, they can be determined, based on vegetative material. Important features are:

Figure 6. 

Leaflet sizes in Polylepis: A P. microphylla; 0.3–0.7 × 0.2–0.5 cm B P. rodolfo-vasquezii; 0.9–1.1 × 0.4–0.6 cm C P. tarapacana; 0.7–0.8 × 0.3–0.4 cm D Polylepis fjeldsaoi; 1.2–2.1 × 0.6–0.7 cm E P. incana; (1.4–)1.8–2.7 × 0.4–0.7 cm F P. subsericans; (1.3–)1.7–2.8 × 0.5–0.7 cm G P. canoi; (2.4–)3.4–3.9 × (0.8–)1.1–1.5 cm H P. humboldtii;1.8–2.8 × 0.6–0.9 cm I P. multijuga; 2.9–3.6(–5.4) × 1.1–2.0 cm. Photographs A, D, E, I E.G. Urquiaga F. B G. Vargas F T.E. Boza E. G H. Huaylla H E. Bastidas.

1. Number of lateral leaflets pairs, which can range from 1 to 7 (Fig. 6), but often varies within the species and even on one specimen, for example, on sun and shade branches. Plants growing under more humid conditions tend to have more leaflets, so that even species which normally only have a single pair of lateral leaflets (e.g., P. tomentella) can have a second pair of smaller ones (Kessler 1995a, b).
2. Leaflets size (Fig. 7). For the largest pair of leaflets in a leaf, this ranges from about 0.3–0.7 × 0.2–0.5 cm in P. microphylla to 2.9–3.6(–5.4) × 1.1–2.0 cm in P. multijuga.
3. Leaflet shape (Fig. 8), which can range from elliptic to ovate and obovate.
4. Leaflet apex (Fig. 8), which can range from acute to deeply emarginate.
5. Leaflet margin (Fig. 8), which can range from entire to serrate.
6. Number of teeth/crenations in non-entire leaflets.
7. Upper leaflet surface hair type, density and length.
8. Lower leaflet surface hair type, density and length (Fig. 5).

The inflorescences are simple clusters, rarely branched, generally long and pendulous as in P. multijuga (15.4–36.0 cm; 47–83 flowers), P. ochreata (8.1–17.4 cm; 21–49 flowers) and P. serrata (7.6–17.3 cm; 16–35 flowers). In other species, the inflorescences are more reduced, in the extreme to the axillary region of the leaves, such as in P. microphylla (3.8–5.3 cm; 1–3 flowers), P. pepei (1.2–3.5 cm; 3 flowers) and P. rodolfovasquezii (0.9–1.1 cm; 1 flower) (Fig. 9).

Figure 7. 

Number of lateral leaflets pairs in Polylepis: A P. subsericans (1) B P. serrata (4–7) C P. weberbaueri (2–3) D P. microphylla (3–6). Photographs by T. E. Boza E.

Figure 8. 

Leaves showing leaflet shapes, margins and apices in the sections and subsections of Polylepis: section Sericeae: A P. multijuga; elliptic, serrate, obtuse (subsect. Lanuginosae) B P. ochreata; elliptic, entire to slightly serrate, emarginate (subsect. Sericeae) C P. pauta, elliptic, crenate, emarginate (subsect. Pauta) D P. rodolfo-vaquezii, elliptic, entire, emarginate (subsect. Pepea). Section Reticulatae: E P. microphylla, broadly elliptic, entire, deeply emarginate F P. reticulata, elliptic to obovate, entire or slightly crenate, deeply emarginate. Section Australes: G P. australis, elliptic, serrate, emarginate. Section Incanaee: H P. besseri, obovate, crenate, obtuse or emarginate and I P. pallidistigma, elliptic, crenate, round or emarginate (subsect. Besseria) J P. tarapacana, obovate, entire or very slightly crenate, obtuse or acute and K P. incana, elliptic to obovate, crenate, obtuse to emarginate (subsect. Incanaee) L P. acomayensis, narrowly obovate, crenate, round to emarginate and M P. sacra, obovate, crenate, emarginate (subsect. Racemosae) N P. flavipila, obovate, crenate, acute or emarginate. Section Subsericantes: O P. subsericans, narrowly elliptic, entire to slightly serrate, round or emarginate. Photographs A, E, K, L E.G. Urquiaga F. B–G, M, O T.E. Boza E. H M. Kessler J A. Domic D G. Vargas.

Figure 9. 

Inflorescence length and number of flowers in Polylepis: A P. multijuga (15.4–36.0 cm; 47–83) B P. pepei (1.2–3.5 cm; 3) C P. subsericans (1.9–5.6 cm; 3–6) D P. rodolfo-vasquezii (0.9–1.1 cm; 1). Photographs A E. G. Urquiaga F. B A. Fuentes C, D T.E. Boza E.

The flowers of Polylepis are hermaphroditic and have a number of adaptations to wind pollination: the petals are missing, the 3–4 sepals are mostly green or rarely red, nectar or fragrances are missing, the stamens are exerted and the styles are multilobed. Previous taxonomic work has rarely taken into account flower features. However, Simpson (1979) previously pointed out that the number of stamens differs among and within the species (Fig. 10). Indeed, we have found great variability, fluctuating from P. pepei (5–9 stamens per flower) to P. sacra (23–27) and we use this trait for species delimitation. The anthers have long purple filaments, are always hairy, and colored red or violet, apparently without taxonomically informative variation (Bitter 1911; Simpson 1979; Kessler 1995a). In contrast, the length of the styles differs between species and is here used for the first time to differentiate species in Polylepis. Style length ranges from 0.9–2.0 mm in P. australis to 3.0–4.9 mm in P. pepei.

Figure 10. 

Number of stamens in Polylepis A P. albicans (7–18) B P. rodolfo-vasquezii (13–15) C P. sacra (23–27). Photographs A, C T.E. Boza E. B G. Vargas.

The fruits are indehiscent achenes that envelop the single carpel with only one ovule. The surface has differently-shaped protuberances including irregular flattened ridges (as in section Subsericantes), flattened spines (in sections Sericeae and Reticulatae), thick wings and ridges (in sections Incanaee and Subsericantes) (Fig. 11) and thin wings (in section Australes). These traits are thus useful for sectional and subsectional delimitation, but, with a single exception (P. crista-galli), do not have taxonomic value for species delimitation. On the other hand, the extent and type of hair on the fruits are of taxonomic value at species level. The seeds are elongated, spindle-shaped, with a thin or subcoriaceous testa (Romoleroux 1996).

Figure 11. 

Fruit type in Polylepis: A P. reticulata: variable numbers and placement of flattened spines B P. pauta: with variable numbers and placement of flattened spines C P. flavipila: irregular flattened ridges with a series of spines D P. subsericans: irregular flattened ridges with a series of spines E, F P. australis: 2–3 irregular and pronounced thin wings. Photographs by T. E. Boza E.

Polylepis is wind pollinated. Earlier studies suggested low dispersal distances in the range of a few tens of meters (Salgado-Laboriau 1979; Salgado-Laboriau et al. 1984; Kessler 1995a). However, pollen dispersal of up to 80 km was later documented in P. australis (Seltmann et al. 2009a) and genetic studies also document extensive gene flow between populations (Schmidt-Lebuhn et al. 2006b; Seltmann et al. 2009a), suggesting extensive long-distance pollen dispersal. Likely, as is typical for wind-dispersed plants (Whitehead 1983; Friedman and Barrett 2009), most pollen is dispersed over short distances, while there is also substantial long-distance pollen dispersal. Selfing can occur in Polylepis, although in P. australis, seed germination rates are lower in selfing than in outcrossed pollination, presumably due to incomplete gametophytic self-incompatibility (Seltmann et al. 2009b).

The fruits of Polylepis are poorly adapted for long-distance dispersal. Some species have long thin spines that appear to be adapted for ectozoochoric dispersal, whereas others have thin wings designed for wind dispersal (Simpson 1979, 1986). Nevertheless, actual fruit dispersal distances appear to be in the range of tens of meters (Enrico et al. 2004; Torres et al. 2008; Quinteros-Casaverde et al. 2012). Dispersal of Polylepis fruits over distances of hundreds of meters to a few kilometers may only occasionally occur in the fur of animals or in mud on the animal’s feet. Accordingly, gene flow in Polylepis is likely to be mainly due to pollen dispersal (Schmidt-Lebuhn et al. 2007).

A further potential unknown complication in Polylepis may be the occurrence of apomixis (asexual seed production). Apomixis leads to taxonomic complications because the scarcity of sexual reproduction leads to the formation of numerous clonal lineages that are evolutionarily partly independent (Campbell and Dickinson 1990; Richards et al. 1996). Apomixis is well known in taxonomically complex genera of Rosaceae such as Alchemilla, Crataegus, Rubus and Sorbus (Nybom 1988; Talent and Dickinson 2007; Gehrke et al. 2008; Pellicer et al. 2012; Samaniego et al. 2018). Kerr (2004) suggested that apomixis might occur in Polylepis, but so far, this has not been confirmed.

Hybridization is widespread in the plant kingdom and is also well known in Rosaceae (Dickinson et al. 2007; Lo et al. 2009; Robertson et al. 2010). Hybridization includes a wide range of phenomena, ranging from occasional hybridization events between evolutionarily independent species where the offspring does not further reproduce, to hybridogenic species formation (Mallet 2007; Rieseberg and Willis 2007; Whitney et al. 2010). Polylepis is no exception in this regard, although this was long overlooked due to the similarity of the species and, accordingly, the difficulty of identifying hybrids.

Simpson (1979, 1986) was the first to suggest hybridization between species of Polylepis, based on morphologically intermediate specimens between P. incana and P. racemosa. Later, based on her morphological taxonomic revision of the Ecuadorean species, Romoleroux (1996) proposed that the following species hybridize where their ranges meet or overlap: P. incana and P. pauta, P. incana and P. ochreata (as P. sericea), P. incana and P. reticulata and P. pauta and P. ochreata (as P. sericea). These hybrids are also recognized in the present treatment.

For Bolivia, also based on morphological intermediacy in mixed species forests, Kessler (1995a, b) identified further putative hybrid individuals and populations. These ranged from occasional individuals found where two species meet, to large hybrid zones up to hundreds of kilometers wide, although some of these are now interpreted differently. Kessler (1995a, b) reported occasional hybrids between P. besseri and P. incanoides, P. besseri and P. tomentella, P. incanoides and P. subtusalbida, P. incarum and P. triacontandra and P. neglecta and P. subtusalbida. All of these are also recognized here. More extensive hybridization was found in an extensive introgression zone at the locality Mojón, Cochabamba, where P. besseri, P. lanata and P. subtusalbida meet and form a large hybrid swarm in which all possible combinations of character traits among the three species can be found in an area of a few square kilometers. At an even larger scale, in southern Bolivia, there is a large region about 200 km wide where P. tarapacana and P. tomentella broadly intergrade. Simpson (1979) used this intergradation to justify treating both as conspecific, but later considered that the species are sufficiently distinct over most of their respective ranges to be treated as distinct species (Simpson 1986). Kessler (1995a, b) followed suit and treated the intermediate populations as a hybridization zone. A further putative intermediate population was recognized by Kessler (1995a, b) between P. lanata and P. triacontandra. However, this population was later studied in more detail and is now recognized as the distinct species P. pacensis (Kessler and Schmidt-Lebuhn 2006).

Besides these hybridization events between extant taxa, there is also indirect evidence for speciation via hybridization. Schmidt-Lebuhn et al. (2006a) proposed that P. crista-galli may have originated from hybridization between an ancestor of P. australis/neglecta and P. subtusalbida. This is based on the fact that, in the small area of geographical overlap between P. neglecta and P. subtusalbida, Kessler (1995a, b) found two hybrid individuals with intermediate characters between the parent species and that are morphologically indistinguishable from P. crista-galli. Genetically, P. crista-galli also appears to be intermediate between the putative parental species (Schmidt-Lebuhn et al. 2006a). Interestingly, the current range of P. crista-galli in southern Bolivia to northernmost Argentina almost perfectly fills the distributional gap between P. neglecta (central Bolivia) and P. australis (Argentina). This suggests that an ancestral form of P. australis/neglecta might have occurred from central Bolivia to Argentina and that the central part of this range was lost due to hybridization with P. subtusalbida or a related species, resulting in the formation of P. crista-galli. Based on morphological intermediacy, in the present study, we also propose further candidates for hybridogenic speciation, such as P. albicans or P. frontinensis, in both cases involving a member of section Sericeae and a member of section Reticulatae. Other species of potential hybridogenic origin include P. incarum and P. racemosa.

Summarizing, based on morphological traits, hybridization in Polylepis appears to be common, ranging from cases where species occur sympatrically, but where no putative hybrids have yet been found (Kessler 1995a) to cases of occasional primary hybridization between two well characterized species, large hybrid swarms between two or three species and, finally, even cases of hybridogenic speciation that may have wiped out part or all of the parent species. Interestingly, the majority of hybridization events known in Polylepis are between species of different sections or subsections. This may partly be because species of a section or subsection are mostly allopatric, so that they cannot easily hybridize, but also because it might be difficult to recognize hybrids between species that are morphologically very similar, as is typical within sections or subsections.

Direct molecular evidence of hybridization in Polylepis is still lacking. However, both Kerr (2004) and Schmidt-Lebuhn et al. (2006a), using molecular markers, found low levels of phylogenetic resolution, based on AFLP (amplified fragment length polymorphism) and sequence data, respectively and that geographically separate accessions of widespread species tend to cluster genetically with geographically close individuals of unrelated species rather than with their conspecific samples from geographically more distant locations. Based on incongruence between chloroplast and nuclear markers, Kerr (2004) even found evidence of hybridization between Polylepis and the closely related genus Acaena. All this suggests that gene flow between species of Polylepis may be even more frequent than inferred from morphology alone.

Changes in ploidy level are an important macro- and microevolutionary processes (Adams and Wendel 2005; Jiao et al. 2011). Polyploidy results from either auto- or allopolyploidization, the latter often linked to hybridization, where it allows for the stabilization of genomes of mixed origin (de Wet 1971; Tate et al. 2005). Besides its evolutionary implications (Weiss-Schneeweiss et al. 2013), polyploidy is also of taxonomic relevance, since populations of different ploidy levels are at least partly reproductively isolated, allowing for divergent evolutionary trajectories that can be treated at species level (Soltis et al. 2007). Rosaceae is renowned for the occurrence of polyploid complexes, for example, in Crataegus (Lo et al. 2010) and Sorbus (Robertson et al. 2010; Pellicer et al. 2012).

The following account of ploidy levels in Polylepis is based on Boza Espinoza et al. (2020b). Ploidy levels in Polylepis have been studied by direct chromosome counts, nucleus size measurements via flow cytometry and guard cell measurements (Table 2). Each of these methods has its own advantages and limitations. Direct chromosome counts are difficult in Polylepis due to the small size of the chromosomes and high chromosome numbers and because they require live plant material (Simpson 1979; Kessler 1995b; Quija-Lamina et al. 2010; Schmidt-Lebuhn et al. 2010). Genome size measurements via flow cytometry also require live material and do not provide direct numbers of chromosomes. Guard cell size is well known to be correlated to ploidy level in angiosperms in general (Masterson 1994; Sugimoto-Shirasu and Roberts 2003; Beaulieu et al. 2008) and Rosaceae in particular (Joly and Bruneau 2007) and has been shown to be correlated to genome size in Polylepis (Schmidt-Lebuhn et al. 2010). Still, there is notable variation of guard cell sizes within species, even when only a single ploidy level is known in the species (Schmidt-Lebuhn et al. 2010; Boza Espinoza et al. 2020b). This variation may be due to anatomical plasticity of a species depending on growth conditions, differences in measurement methods or aneuploidy. Thus, guard cell measurements allow inference of major ploidy levels, but not of minor variations in chromosome numbers or genome sizes. Such variations have been found in detailed studies of Ecuadorean species, based on genome measurements and chromosome counts (Quija-Lamina et al. 2010; Quijia Lamiña et al. 2010; Montalvo 2013; Zurita et al. 2013; Segovia-Salcedo 2014; Segovia-Salcedo and Quijia-Lamiña 2014; Caiza et al. 2018). Based on values of dozens of individuals of each species, these studies documented that many species have variable chromosome numbers and genome sizes, suggesting reductions in chromosome numbers that could be the result of aneuploidy (loss of DNA and reduction of chromosome size) and dysploidy (chromosome fusion) (Stebbins 1971; Morgan et al. 1994; Mishima et al. 2002).

Assigning ploidy levels to species of Polylepis can be based on two different base numbers. Segovia-Salcedo (2014) and Zurita et al. (2013) used the base haploid chromosome number of 7 in the family Rosaceae as reference, thus interpreting a chromosome count of 42 as hexaploid (x = 6). Schmidt-Lebuhn et al. (2010) and Kessler et al. (2014) instead used the lowest number in the genus (2n = 42) as baseline, interpreting this as functionally diploid. The evolution of chromosome numbers in angiosperms is highly complex, with repeated polyploidization events commonly followed by reductions in chromosome numbers via aneuploidy and dysploidy (de Wet 1971; Adams and Wendel 2005; Tate et al. 2005; Jiao et al. 2011; Weiss-Schneeweiss et al. 2013). We consider that, ultimately, within a plant group, the crucial factor is the behaviour of the chromosomes, i.e. whether they behave as bivalents so that during chromosome pairing each chromosome has a single counterpart or as polyvalents where they can pair with several other chromosomes. This behaviour is unknown for Polylepis or related genera. For simplicity, we here use a base chromosome number of 2n = 42 as baseline for defining diploids.

Based on this approach, for guard cell length, based on comparison with chromosome counts and genome size measurements, Boza Espinoza et al. (2020b) proposed that diploidy is related to guard cell lengths of 9–15 µm, tetraploidy to 14–20 µm and higher ploidy levels to 18–23 µm, showing that some overlap occurs that may make it difficult to assign a specific measure to a ploidy level. For genome size, the diploid condition is related to 2C values around 1.4–1.7 pg, triploidy to 2.0–2.3, tetraploidy to 2.6–3.4 pg, hexaploidy to 4.6–4.9 pg and octoploidy to 5.7–5.8 pg. Finally, chromosome counts of around 42 correspond to diploids, around 84 to tetraploids and around 126 to hexaploids. However, numerous published counts differ notably from these values. For example, Caiza et al. (2018) and Segovia-Salcedo and Quijia-Lamiña (2014) reported chromosome counts of 59–77 for nine individuals of P. ochreata (as P. sericea) from Yanacocha, Ecuador. In this situation, it is unclear if these numbers reflect the difficulty of fully counting the tiny chromosomes or whether they correspond to real values with would suggest triploidy and other intermediate chromosome levels as a result of aneuploidy and dysploidy.

At present, for the 45 species of Polylepis, combined data on guard cell length, chromosome number and genome size are available for nine (20%) species, on guard cell length and genome size for another nine (20%) species and on guard cell length and chromosome number for three (7%) species, whereas for 24 (53%) species, only guard cell measurements are available (Table 1). Bearing in mind the potential limitations of incomplete data and some uncertainty in the interpretation of the data, we infer that, at present knowledge, 19 (42%) species are purely diploid, 15 (33%) purely tetraploid and one (2%) purely octoploid. The remaining eight (18%) species have mixed ploidy levels, with three (7%) being di- and tetraploid, two (4%) di- and hexaploid, one (2%) tetra- and hexaploid, one (2%) tetra- and octoploid and one (2%) di-, tri-, tetra- and hexaploid. While it is likely that further studies will reveal more cases of mixed ploidy, at least some well-studied species appear to consistently show diploid (P. hieronymi, P. lanuginosa, P. simpsoniae) or tetraploid (P. tarapacana, P. tomentella) conditions.

Placing the ploidy levels in an evolutionary context, in section Sericeae, most species are diploid, but mixed di- and polyploidy is present in P. ochreata and P. pauta. These two species overlap in northern Ecuador where they hybridize extensively and it is conceivable that the polyploid condition stems from this hybridization, as polyploidy is often correlated with hybridization (de Wet 1971; Tate et al. 2005).

In section Reticulatae, again most species are diploid, but polyploidy occurs in P. microphylla and P. reticulata. Interestingly, at least in P. reticulata, polyploidy is only known only from cultivated plants (Segovia-Salcedo 2014), suggesting that this condition may be related to domestication.

In section Australes, P. australis has been well studied and includes di-, tri-, tetra- and hexaploids, most likely due to autopolyploidiation (Kessler et al. 2014). These ploidy levels show a clear geographical distribution pattern, with populations from the northern Argentinean Andes being purely diploid and those from the central Andes tetraploid, whereas in the isolated Sierra de Córdoba, all four ploidy levels co-occur in mixed populations. This suggests that the triploid plants may be hybrids between the di- and tertraploid ones, but whether they are sterile primary hybrids or can reproduce by themselves is unresolved. Additionally, the degree of reproductive isolation and, hence, evolutionary independence between the diploid and tetraploid populations remains unknown.

Sections Subsericantes and Incanaee mainly includes tetraploid species. These sections on average occur at higher elevations and in more arid environments than the other sections, which corresponds well with the polyploid condition, since polyploids are well known to be over-represented at high latitudes and elevations (Masterson 1994; Brochmann et al. 2004), possibly because the different paralogs offer more adaptive potential (Chung et al. 2010). In contrast to the dominant tetraploidy in these sections, both P. fjeldsaoi and P. incana are diploid. Polylepis incana has been considered to be one of the most derived species in section Incanaee (Simpson 1979, 1986; Schmidt-Lebuhn et al. 2006a), so that a diploid condition is surprising if one assumes that P. incana is nested within a tetraploid clade. This suggests that the assumption that P. incana is a derived member of this section is wrong and that the evolution of this section is more complex than previously assumed. In any case, assuming that the sections recognized here, based on morphological and ecological similarity, are evolutionary units, we now can deduce that polyploidy evolved at least five times in the genus, possibly more often. However, considering that the evolution of Polylepis is probably reticulate (Kerr 2004; Schmidt-Lebuhn et al. 2006a), inferring the origins of polyploidy in Polylepis may be very difficult.

Finally, focusing on the taxonomic implications of ploidy levels in the genus, we found that, in some cases, closely related species have different ploidy levels, supporting their treatment as distinct species. For example, P. fjeldsaoi has previously been identified as P. tomentella (Mendoza and Cano 2012), but whereas the first species is diploid, the latter is consistently tetraploid. On the other hand, at least eight species include individuals of different ploidy levels. At least in P. australis, this is clearly a natural condition (Kessler et al. 2014), which raises the question as to how to treat the different ploidy levels taxonomically. It has been suggested that different ploidy levels within a “species” should be treated at species level if there is morphological, ecological or biogeographical evidence that they are evolutionarily largely independent units (Soltis et al. 2007). This approach has been taken in polyploid-apomict species complexes of other genera of Rosaceae, such as Crataegus (Talent and Dickinson 2007) and Sorbus (Robertson et al. 2010), but more information is needed before this approach can be applied to Polylepis. On the other hand, in several species, polyploidization is apparently linked to cultivation, as in P. incana, P. racemosa and P. reticulata. Polyploidization of cultivated plants is a common phenomenon either via auto- or allopolyploidization where higher ploidy levels are often associated with higher plant vigour and adaptive potential (Paterson 2005; Matsuoka 2011; Sattler et al. 2016). Polylepis has long been planted by Andean inhabitants as a source of building material, firewood and as fences (Kessler 1995d) and it is conceivable that natural or artificial hybrids have been favoured. For a more detailed discussion of this situation in P. racemosa, see under that species.

Polylepis belongs to the tribe Sanguisorbeae DC., which is characterized by cup-shaped hypanthium that entirely encloses the carpel(s), resulting in a perigynous position of the flower (Focke 1894; Robertson 1974; Eriksson et al. 2003; Potter et al. 2007). Polylepis has long been placed in close taxonomic proximity to Acaena, a genus of some 100 species of mainly evergreen, creeping herbaceous perennial plants and subshrubs found mostly in New Zealand, Australia and South America, with a few species extending to Hawai’i and California. Previously, Bitter (1911) argued for a derivation of Polylepis from a species group in Acaena that includes A. elongata because this group is characterized by multipinnate leaves and long dense racemes similar to those of Polylepis. Bitter (1911) also implicitly considered that the most ancestral species of Polylepis may be P. multijuga because it has long dense racemes and spine-covered fruits reminiscent of those of Acaena.

Simpson (1979, 1986) provided the first explicit ideas about the evolution of Polylepis, placing the species in three informal groups. The first of these groups, the sericea group (here section Sericeae), includes species with putatively ancestral traits such as large, multipinnate leaves with thin texture, and long inflorescences with numerous flowers. Most species of this group grow at relatively low and often humid conditions at the upper margin of the cloud forest, where Acaena also occurs. The second group is the reticulata group (our section Reticulatae), characterized by nitid, emarginated leaflets with pannose undersides of the leaflets. The species of this group occur at higher elevations than those of the first group and often also in more arid habitats. Finally, the large incana complex (our sections Australes, Subsericantes and Incanaee) is morphologically quite variable (which is why we placed the species in three sections), but includes many species with just one leaflet pair, densely pannose or even glandular lower leaflet surfaces and highly reduced inflorescences. While some species occur at low elevations and in humid regions, many are found in rather arid habitats and often at very high elevations. Simpson (1986) concluded that Polylepis likely originated from Acaena under humid cloud forest conditions and from there, gradually expanded its ecological amplitude to reach highly arid and cold environments of the high Andes, not accessible to any other native tree genus. This evolutionary trend was paralleled by morphological changes including reductions in growth height, leaflet size, number of leaflets, inflorescence length and number of flowers and an increase in leaf thickness. In the next study commenting on the evolution of the genus, Kessler (1995a) largely followed Simpson’s views.

The first attempt to study the evolutionary history of Polylepis using molecular methods was undertaken by Malin Kerr in an unfortunately largely unpublished PhD thesis (Kerr 2004). Using the chloroplast markers trnL/F and Ahd1, as well as a nuclear ITS locus of over 50 accessions of Polylepis, 26 accessions of Acaena and numerous other Rosaceae, she was able to confirm that Polylepis is nested within Acaena, rendering the latter genus paraphyletic. Based on her limited sampling, Kerr (2004) hypothesized that Polylepis, as a whole originated, from a hybridization event between ancestral members of the Acaena elongata and A. cylindristachya species groups. She hypothesized that this hybridization event was probably allopolyploid, as suggested by her understanding that Acaena has a predominant ploidy level of 2n = 42 in Acaena, whereas Polylepis in her view has 2n = 84. However, we now know that 2n = 42 is also common and presumably ancestral in Polylepis (see Ploidy levels). Still, morphologically Polylepis combines traits of both the A. elongata group (spiny fruits) and the A. cylindristachya group (racemose inflorescences), supporting the idea of a hybrid origin of Polylepis. However, to complicate matters, Kerr (2004) also found indication that P. quadrijuga later hybridized with a member of Acaena section Ancistrum (to which neither A. elongata nor A. cylindristachya belong). However interesting, the conclusions of Kerr (2004) were based on limited sampling and the use of the nuclear marker ITS which is notorious for having multiple copies, especially in polyploidy plants, potentially hampering phylogenetic reconstructions (Baldwin et al. 1995).

The second attempt at a molecular phylogenetic reconstruction of Polylepis was undertaken by Schmidt-Lebuhn et al. (2006a), based on 46 accessions of Polylepis and two accessions of Acaena using Amplified Fragment Length Polymorphisms (AFLP) of primarily the nuclear genome. Their results were quite similar to those obtained by Kerr (2004): very limited resolution within the genus and strong geographical clustering of the accessions, to the degree that samples of the same species would cluster with geographically proximate samples of other species rather than with their conspecifics. Combining this data with a morphological data matrix, they nevertheless obtained a phylogenetic hypothesis that clustered similar species in meaningful groups. These groups largely corresponded to Simpson’s (1979, 1986) classification: a basal grade corresponding to the sericea group (our section Sericeae), a monophyletic reticulata group (section Reticulatae), and a monophyletic incana group (sections Australes, Subsericantes and Incanaee) with P. subsericans as sister to the other species in the group. The latter species was placed by Simpson (1979) in her sericea group, but she commented on the intermediacy of the species between the sericea and incana groups and we here place it in a section of its own, together with two other species.

The latest molecular study of the phylogeny of Polylepis was conducted by M. Claudia Segovia S. in a PhD thesis that also remains largely unpublished (Segovia-Salcedo 2014). She used Next Generation Sequencing (Hyb-Seq) to analyze 256 nuclear genes and chloroplastic genomes of 25 Polylepis accessions. Her phylogenetic reconstruction showed a strong geographical signal and recovered groups of morphologically very different species that did not correspond to the groups proposed by Simpson (1979, 1986) and Schmidt-Lebuhn et al. (2006a). As Segovia-Salcedo (2014) did not include replicate samples of widespread species, it is impossible to assess whether samples of the same species from different geographical areas would be recovered close to each other in a phylogenetic reconstruction.

In conclusion, our understanding of the evolutionary history of Polylepis is still very incomplete. While morphological traits point to a plausible story of diversification and adaptation from humid cloud forests to arid high-elevation habitats, molecular data suggest a complex, reticulate evolutionary history. Additionally, while there is evidence that Polylepis is nested within Acaena, we refrain from merging both genera until a clearer picture of their evolutionary relationships emerges.

Based on all the above, we can conclude that there is ample gene flow between populations of Polylepis assigned to different species. Although species can be distinguished on morphological, biogeographical and ecological grounds, it is likely that gene flow between populations of different species in close proximity have more gene flow between them than geographically remote populations of the same species. At the same time, the presence of different ploidy levels in at least eight species of the genus suggests that there may be barriers to gene flow within species.

In such a situation, the classical biological species concept of species being reproductively and evolutionarily independent units is hardly applicable (Mayr 2000). Polylepis may, thus, be a case where a genic species concept, rather than a genomic species concept is more appropriate. The genic species concept (Wu 2001) assumes that species identity is based on relatively few gene regions that determine the fundamental physiological or morphological traits that define a species and that the remainder of the genome can be exchanged between species without compromising species identity. In this view, species identity is not dependent on full reproductive isolation, but rather is determined by selection on relatively few genes. In contrast, the genomic species concept is based on the traditional view of reproductive isolation between species, leading to genetic divergence across the entire genome (Lexer and Widmer 2008). For Polylepis, all of this remains speculative and detailed genomic studies are needed to confirm the speciation mode in the genus.

Polylepis typically forms the uppermost forest belt in the tropical Andes, although some species also grow at lower elevations in mixed forest stands with other tree genera. Due to the high elevations and accordingly low temperatures at which these trees occur, they have been the focus of a number of ecophysiological studies aiming to understand the adaptations to low temperatures. However, since the 45 species inhabit a wide range of habitats, ranging from moderate to extremely high elevations and arid to superhumid environments, the different species express a wide range of physiological adaptations that go beyond only adaptations to low temperatures.

The ability to tolerate the low nocturnal temperatures that are typical of tropical mountains is, in some species, achieved by daily osmotic adjustments and supercooling down to -9 °C as in P. sericea or freezing tolerance as in P. australis and P. tarapacana (Rada et al. 1996, 2001, 2009; Azócar et al. 2007). In the latter two species, leaf tissue freezes at temperatures between -3.5 °C and -9.2 °C, but frost injury is only observed at temperatures between -18 °C and -24 °C.

Adaptations to drought conditions are also frequent in Polylepis and include small, thick leaflets with wax layers and sunken stomata to reduce transpirational water loss (Macek et al. 2009). This is most clearly seen in P. tarapacana (García-Núñez et al. 2004; Toivonen et al. 2014). In arid areas at high elevations, solar radiation is very intense, leading to the development of protective pigments (González et al. 2007).

Photosynthetic rates of Polylepis range from 3 μmol·m^-2·s^-1 in P. tarapacana to 9 μmol·m^-2·s^-1 in P. australis (Azócar et al. 2007). Species inhabiting relatively low elevations with higher temperatures have higher mass-based maximum photosynthesis, stomatal conductance and leaf area than species from the higher and colder habitats (Toivonen et al. 2014). Conversely, species from humid habitats, where water stress is low, but where clouds and fog often reduce insolation, have higher light use efficiency and lower light saturation and compensation points (Toivonen et al. 2014).

Generally speaking, physiological traits in Polylepis are related either to the temperature or precipitation conditions at which they grow, revealing evolutionary specialization and adaptation of the species along environmental gradients.

Polylepis is wind-pollinated. Although most pollen is deposited at close distances to the trees (Kuentz et al. 2007), pollen flight of up to 80 km has been documented (Seltmann et al. 2009a), suggesting extensive gene flow between populations (see Pollination biology and seed dispersal). In P. australis, there is increased mortality in offspring resulting from short-distance crosses and increased vigour (N-metabolism capacity) in long-distance crosses, providing evidence for inbreeding depression (Seltmann et al. 2009b).

The seeds of Polylepis are not well adapted for long distance dispersal. Species in section Sericeae have spines that allow them to be transported in the fur of animals, but nothing is known about dispersal distances. In many other species, the nutlets have spiny ridges that do not appear to be adapted to any specific dispersal type (Simpson 1986) and which mostly accumulate close to the parent trees. Only P. australis and P. neglecta (section Australes) have thin wings apparently adapted to wind dispersal, but flight distances are only up to a few dozen meters (Renison et al. 2004).

Polylepis seeds typically have low germination rates, possibly associated with dormancy (Cuyckens et al. 2021) and seed germination is temperature-dependent, with maximum germination at about 20 °C in P. besseri (Gareca et al. 2012), which is relatively high for high mountain environments. Accordingly, regeneration by seeds often decreases with low temperatures at high elevations (Byers 2000; Hoch and Körner 2005; Gareca et al. 2007; Hertel and Wesche 2008; Wesche et al. 2008). This may be a result of decreasing germination, but also of reduced seed production (Cierjacks et al. 2008). In arid habitats, such as northern Chile, successful reproduction may only occur in exceptionally favourable (humid and warm) years, leading to the formation of age cohorts (Rundel et al. 2003).

The species of Polylepis are generally rather slow-growing, with radial growth rates typically in the order of 1 mm per year, although much variation exists (Domic and Capriles 2009; Jomelli et al. 2012; Montalvo et al. 2018). On the other hand, under optimal conditions, young trees can achieve height increments of up to 50 cm per year, especially in P. racemosa, which is the most commonly cultivated species because of its fast growth (Rosero et al. 2018). Tree height in Polylepis commonly decreases with elevation, but, in arid valleys, can also decrease at low elevations due to drought (Cierjacks et al. 2008; Hertel and Wesche 2008; Kessler et al. 2014). However, tree height is also influenced by more local conditions, so that, in arid regions, shaded (and hence more humid slopes) have taller trees (Kessler et al. 2007). In many forests, tree height is also limited by human extraction of tall trees (Toivonen et al. 2011; Kessler et al. 2014).

Dendrochronological studies on P. tarapacana in Bolivia suggest tree longevity of at least 700 years, with precipitation being positively and high summer temperatures (which increase drought stress) negatively related to radial growth (Argollo et al. 2004; Morales et al. 2004). Growth rates in this species decrease with elevation, but are higher in sheltered microhabitats (Domic and Capriles 2009). In P. besseri, radial growth is limited by the accumulation of reserves the year before ring formation, with a warm period before the growing season (humid period) increasing growth (Gareca et al. 2010). In Polylepis australis, tree vitality and growth are highest in the middle of the elevational distribution of the species, being limited at low elevations by high temperatures and drought, and at high elevations by low temperatures (Marcora et al. 2008). This species has lower growth when surrounded by rocks (Suarez et al. 2008). Such local control of growth has also been observed in P. pepei in Bolivia, where it is influenced by slope and substrate (moraine or scree slope) (Jomelli et al. 2012).

One the most conspicuous patterns in the distribution of Polylepis forests is that, frequently, they occur as isolated forest patches isolated from the closed treeline (Herzog 1923; Troll 1959; Hueck 1962; Vareschi 1970; Kessler 2002). It has long been debated whether this is a natural or human-induced pattern. Especially, earlier authors have argued that these patches occur in microclimatically favoured habitats, especially boulder slopes (Raimondi 1874; Weberbauer 1907, 1930; Herzog 1923; Troll 1929; Herrera 1943; Barreda 1951; Koepcke 1961; Cerrate 1979; Arnal 1983; Salgado-Laboriau et al. 1984; Ferreyra 1986). Walter and Medina (1969) proposed that boulder slopes allow warm air to penetrate deeper into the soil, allowing for better root development. However, soil temperature measurement by Kessler and Hohnwald (1998) have shown that boulder slopes are actually colder than adjacent fine-grained soils, as also known from other mountain regions (Wakonigg 1996). As an alternative explanation, Ellenberg (1958) proposed that rocky ground prevents the spread of fires and thus allows the survival of forests. The negative influence of fires has been documented by subsequent studies (Lægaard 1992; Kessler 2000; Kessler et al. 2014). Often, this also leads to a concentration of Polylepis stands to sheltered ravines and along watercourses (Weberbauer 1911; Troll 1959; Salgado-Laboriau et al. 1984; Rauh 1988).

Current understanding of this so-called “Polylepis-problem” suggests that the natural vegetation pattern would be a grassland-forest mosaic, with increasing grassland contribution towards higher elevations (Toivonen et al. 2011, 2018; Sylvester et al. 2014a, 2017). Nevertheless, because of the paucity of undisturbed habitats, it is not known which conditions determine the development of forest or grassland in these mosaics. Factors may involve water saturation of soils, cold air ponding or strong winds (Kessler 1995d; Sparacino et al. 2020). It is known, however, that unless fires further push the forests back (Kessler 2000), the expansion of forests into grasslands is very slow due to the low ability of Polylepis to colonize dense grasslands (Cierjacks et al. 2007a), suggesting that the forest-grassland mosaics are fairly stable over time.

Other habitat associations of Polylepis include a preference for cloud condensation belts in arid regions (Troll 1959; Koepcke 1961; Lauer 1982; Cabido and Acosta 1985) and avoidance of salty soils (Kessler 1995d). Beyond these general patterns, there are many species-specific habitat preferences. For instance, whereas most species of Polylepis form the canopy layer of the forests, P. hieronymi is a successional species that often colonizes raw soils on landslides and road banks and is then displaced by taller trees, such as Podocarpus parlatorei (Kessler 1995b).

The upper limit of the growth of Polylepis forests has been a matter of some debate. Jordan (1980) claimed that P. tarapacana reaches 5200 m on Volcán Sajama, but subsequent studies have highest trees (3 m tall) at 4820 m, with only shrubby plants reaching 5000 m (Hoch and Körner 2005). Similar elevations are reached by P. subsericans in Cuzco, Peru (S. Sylvester, pers. comm.) and P. weberbaueri in Ancash, Peru (T. Boza, pers. obs.). Along with Juniperus tibetica forests at 4900 m in Tibet, these are the world’s highest forests (Miehe and Miehe 1994).

It is generally accepted that the genus Polylepis evolved in the Andes from the genus Acaena (Simpson 1986). The age of the genus is unknown, but the oldest available pollen records show that it was already present 370,000 bp (years before present) in Peru and Bolivia (Hanselman et al. 2011) and 124,000 bp on the Bogotá Plateau, Colombia (van der Hammen and Hooghiemstra 2003). These and other more recent pollen records show that the pollen abundances of Polylepis varied substantially over time, suggesting that forest cover changed over time even prior to the arrival of humans about 12,000 bp. These fluctuations have been linked to climatic changes, with warm and wet periods being correlated with maximum Polylepis abundance (Gosling et al. 2009). Opposed to this, natural fires, especially during dry periods, reduced Polylepis cover. Polylepis forest cover was also patchy due to landscape heterogeneity with, for instance, flat valley bottoms with water-logged soils being devoid of tree cover (Gosling et al. 2009; Valencia et al. 2018). During arid and cold climatic phases, forests were, thus, restricted to climatic microrefugia, such as sheltered valleys (Valencia et al. 2016). Similar forest-grassland mosaics can currently be found in topographic refugia inaccessible to humans and their livestock (Sylvester et al. 2014a, 2017). That some Polylepis species had a natural fragmented distribution is also evidenced by geographic patterns of genetic diversity, as, for example, in P. tarapacana (Peng et al. 2015).

The arrival of humans increased fire frequencies, first by hunter-gatherers and later by agropastoralists with their livestock, leading to widespread reductions in Polylepis abundance (e.g., Graf 1979, 1981, 1983, 1992; Hansen et al. 1984; Ybert and Miranda 1984; Markgraf 1985; Hansen 1992; Baied and Wheeler 1993; Chepstow-Lusty et al. 2005; Urrego et al. 2011; Williams et al. 2011). For example, in Cochabamba, Bolivia, at a site where today P. lanata and P. pepei occur in small patches above the closed treeline, Polylepis forests were more widespread some 18,000–14,500 a bp, declining towards some 10,000 a bp, followed by a period almost without Polylepis until 6,400 a bp, followed by slight increase to current levels (Williams et al. 2011). These changes in the abundance of Polylepis were linked to changes in fire frequency, partly due to dryer climatic conditions and partly due to human activities.

In southernmost Ecuador, at the border with Peru, Polylepis populations are currently restricted to tiny populations, but pollen evidence shows that Polylepis was much more common in the early to mid-holocene (about 11,000–4,000 bp) (Rodríguez and Behling 2012; Villota and Behling 2013). Later, as climatic conditions became more humid and human impact led to increasing fire frequencies, Polylepis declined massively, being replaced by mixed humid montane forest and grassland.

The development of human cultures in the Andes took an important step forward some 6,000 bp with the domestication of camelids (Wing 1986; Wheeler 1985, 1988). The development of large-scale pastoralism led to the widespread ecosystem degradation (van der Hammen and Noldus 1985; Thompson et al. 1988). During the Incan period, up to 30 million people may have inhabited the central Andes (Earls 1976). An important component of this land use was the development of agroforestry (Chepstow-Lusty and Jonsson 2000; Chepstow-Lusty et al. 2009; Chepstow-Lusty 2011; Mosblech et al. 2012), including reforestation with Alnus in the last 1000 years (Chepstow-Lusty et al. 1998). After the Spanish conquista, forest destruction picked up again and continues until today in many regions (Jameson and Ramsay 2007).

Polylepis forests harbour unique biodiversity, including a number of highly specialized, often range-restricted and threatened bird species (Fjeldså 1993; Cahill and Matthysen 2007; Lloyd 2008a; Astudillo et al. 2020; Quispe-Melgar et al. 2020). However, these birds generally do not influence the trees themselves, except for the Thick-billed Siskin (Spinus crassirostris), which eats the seeds of Polylepis without dispersing them (Herzog et al. 2003).

Polylepis forests also provide habitats for many plant species, including the world’s highest vascular epiphytes (Sylvester et al. 2014b) and a diverse liverwort flora (Gradstein and León-Yañez 2020; Gradstein and Pócs 2021). Nothing is known of competitive or facilitative interactions with other plant species in the germination, seedling or growth stages. However, many populations of Polylepis species are parasitized by Tristerix chodatianus (Loranthaceae), a specialist hemiparasite that infects trees of the genus Polylepis (Amico et al. 2007). In P. flavipila forests, up to 48% of the trees can be infected, which leads to the death of branches about 15 years after infestation and, in heavily infested trees, eventually to the death of the whole tree (Arizapana et al. 2016; Camel et al. 2019b).

Fungal interactions are also important in Polylepis (Robledo and Renison 2010). Polylepis australis has vesicular arbuscular mycorrhizal symbionts (Menoyo et al. 2007, 2009; Soteras et al. 2013). The same is presumably true for all other species, but this remains to be studied. Polylepis tarapacana is parasitized by the fungus Paraleptosphaeria (= Leptosphaeria) polylepidis (Leptosphaeriaceae, Pleosporales), which appears to lead to increased tree mortality (Macía et al. 2005; Coca-Morante 2012). Interestingly, this fungus, in turn, is infected by the mycoparasite Sajamaea mycophila (Dictyosporiaceae, Pleosporales), which may perhaps be used as a biocontrol of P. polylepidis (Piątek et al. 2020). In P. tomentella, species of three genera of Ascomycota have been found in stigmas and styles, which appears to negatively affect the germination of pollen grains by inhibiting pollen tube growth, although the fungi apparently are not able to penetrate the ovary (Domic et al. 2017).