Pleistocene origin and colonization history of Lobelia columnaris Hook. f. (Campanulaceae: Lobelioideae) across sky islands of West Central Africa

Abstract We aimed to infer ancestral area and historical colonization of Lobelia columnaris in the sky islands of Bioko and Cameroon through dated phylogeny using chloroplast genomes. Specifically, we aim to answer the following questions: (1) What are the phylogenetic relationships among Bioko Island and Cameroon populations? (2) Are the older populations found in the older sky islands? We assembled novel plastomes from 20 individuals of L. columnaris from 5 mountain systems. The plastome data were explored with phylogenetic analyses using Maximum Likelihood and Bayesian Inference. The populations of L. columnaris have a monophyletic origin, subdivided into three plastomes‐geographic clades. The plastid phylogenomic results and age of the sky islands indicate that L. columnaris colonized first along with the Cameroon Volcanic Line's young sky islands of Bioko. The crown group (1.54 Ma) split the population in Bioko and mainland Cameroon. It is possible that Bioko was the ancestral area and likely isolated during cold and dry conditions in forest refugia. Presumably, the colonization history occurred during the middle‐late Pleistocene from South Bioko's young sky island to North Bioko and the northern old sky islands in Cameroon. Furthermore, the central depression with lowland forest between North and South Bioko is a current geographic barrier that keeps separating the populations of Bioko from each other. Also, the shallow sea channel keeps isolated the populations of Bioko and the mainland populations. The Pleistocene climatic oscillations led to the divergence of the Cameroon and Bioko populations into three clades. L. columnaris colonized the older sky islands in mainland Cameroon after establishment in Bioko's younger sky islands. Contrary to expectations, the biogeography history was an inverse progression with respect to the age of the Afromontane sky islands.

Only three volcanic peaks in the CVL have elevation above 3000 m (Pico Basilé, Mt. Cameroon, and Mt. Oku) ( Table 1). The geological structure of the CVL is a combination of tectonic and volcanic origins with unequal ages ranging from the middle to late Tertiary (Jesus et al., 2005). The oldest mountains are in the north, decreasing with the age of volcanic activity in the southern area (Missoup et al., 2016) and then getting older at the oceanic islands in the Gulf of Guinea (Table 1). The marked geographic separation and isolation of the mountains are analogous to islands in the sky or sky islands.
Moreover, the sky islands and sky island archipelagos of the Gulf of Guinea and West Central Africa (Figure 1) possess an extraordinary diversity of angiosperms (Figueiredo, 1994), small mammals (Missoup et al., 2016), and amphibians (Zimkus & Gvoždík, 2013). This African region is part of the Guinea biodiversity hotspot (Myers et al., 2000) and is critical for conserving endemic species of plants and animals that inhabit the sky islands (Tropek & Konvicka, 2009).

Giant lobelias may have experienced rapid diversification on East
Africa mountains and subsequently dispersed to West Africa (Knox & Li, 2017). Lobelia columnaris Hook f. (Campanulaceae: Lobelioideae) is a giant lobelia listed as a vulnerable species in the IUCN Red List of Threatened Species 2015 (Cheek & Thulin, 2015). L. columnaris and L. barnsii Exell (Mabberley, 1974a) (Popp et al., 2007). The hypothesis of recurrent connections over time between the West and East African mountains provides a framework to investigate the biogeographic origin among close relatives in both bioregions, like the endangered and endemic Mount Oku rat, Lamottemys okuensis, in the CVL (Missoup et al., 2016). However, some taxa have a geographically widespread distribution, from the eastern mainland to the outlying islands of western Africa, like the endangered Prunus africana (Dawson & Powell, 2008).
The present study's objective is to infer the phylogenetic relationships of populations of Lobelia columnaris using chloroplast genomes and estimate the divergence time to reconstruct its historical colonization in the sky islands of Bioko and Cameroon. Specifically,

Sky Island
Elevation ( (Schabetsberger et al., 2004). The vegetation of Bioko is arranged in elevational rings dominated by Guineo-Congolian tropical species with Afromontane elements appearing at higher elevation (Fa et al., 2000).
Mainland Cameroon comprises several Afromontane sky islands. Oku and Bamenda-Banso highlands (2260 m) were uplifted during the Cenozoic (Oligocene-Miocene) (Missoup et al., 2016). Mainland southern volcanoes like Mt. Cameroon are the youngest with origins during the Pliocene-Pleistocene (Zimkus & Gvoždík, 2013). The vegetation is arranged in elevational bands within the montane forest and has been highly disturbed by grazing, fire, and human activities, except on Mt. Cameroon (Ineich et al., 2015).

| Study species, DNA extraction, and sequencing
Lobelia columnaris Hook f.  (Bartos et al., 2012). The flowers produce sugared nectar that attracts birds such as the Orange-tufted sunbird (Cinnyris bouvieri) (Janecek et al., 2012;Riegert et al., 2011) and Hymenoptera insects. The flowers are considered bee-pollinated (Givnish et al., 2009), but birds also can be secondary pollen dispersers. The fruit is a capsule with thousands of winged seeds (Mabberley, 1975) dispersed by wind (Knox & Palmer, 1998;Mabberley, 1975 DNA was extracted from silica-dried leaf samples of 20 individuals of L. columnaris using a modified cetyltrimethylammonium bromide (CTAB) protocol (Doyle & Doyle, 1987). Library construction, sequence generation, and bioinformatics processing were done at the Indiana University Center for Genomics and Bioinformatics.
A total of 20 individual plastid genomes of L. columnaris were newly sequenced and assembled (Table 2). Eleven plastomes represent populations from three sky islands in Bioko (Equatorial Guinea), and nine plastid genomes are from two sky islands in Cameroon ( Figure 1). Moreover, the phylogenetic tree included two more plastome sequences of L. columnaris from South Bioko (Perez Perez 3103, MF061188) and another from Mt. Cameroon (Muasya 2085, MF061187) obtained from GenBank (Knox & Li, 2017

| Assembling and annotation of the chloroplast genomes
The raw paired-end reads were filtered and de novo assembled using the GetOrganelle toolkit (Jin et al., 2020). The filtered reads were assembled using the SPAdes version 3.9 using k-mer 21, 33, 45, 65, and 85 (Bankevich et al., 2012). To retain pure chloroplast contigs, the final "fastg" files were filtered using the "slim" script of GetOrganelle toolkit. The filtered De Brujin graph was viewed and the final sequence exported using Bandage (Wick et al., 2015). The chloroplast genome was automatically annotated using CpGAVAS (Liu et al., 2012), then adjusted using Geneious version 9.0 (Kearse et al., 2012).

| Phylogenetic analyses
The whole chloroplast genome matrix was aligned using MAFFT version 7.1 (Yamada et al., 2016). Only one inverted repeat region was used in the phylogenetic analyses. Two matrixes were prepared, one including all gaps, and another removed gaps using trimAl (Capella-  (Hillis & Bull, 1993). The BI analyses were performed using MrBayes version 3.2.6 (Ronquist & Huelsenbeck, 2003), with DNA substitution models selected for each gene partition by the Bayesian information criterion (BIC) using jModeltest version 2.1.10 (Darriba et al., 2012;Guindon & Gascuel, 2003). Markov Chain Monte Carlo (MCMC) analyses were run in MrBayes for 10,000,000 generations, with two simultaneous runs, and each run comprising four incrementally heated chains. The BI analyses were started with a random tree and sampled every 1000 generations. The number of generations for the three datasets was sufficient because the average standard deviation of split frequencies for the datasets was <0.005, and Potential Scale Reduction Factor (PSRF) of Convergence Diagnostic (Gelman & Rubin, 1992) for the datasets was 1.00. The first 25% of the trees were discarded as burn-in, and the remaining trees were used to generate a majority-rule consensus tree. Posterior probability values (PP) ≥ 0.95 were considered as well supported (Alfaro et al., 2003;Erixon et al., 2003;Kolaczkowski & Thornton, 2007).
Both ML and BI analyses, as well as jModeltest, were performed at the CIPRES Science Gateway (http://www.phylo.org).

| Estimation of divergent times and phylogeographic history
Dating analyses were conducted using Markov Chain Monte Carlo (MCMC) methods in BEAST version 2.4 (Bouckaert et al., 2014), which was performed at the CIPRES Science Gateway (http:// www.phylo.org). The setting parameters in BEAUti included "BEAST model test" for "Site model," "Relaxed Clock Log Normal" for "Clock model," and "Yule Model" for speciation. Meanwhile, we selected two crown nodes for calibrations from published data using the CladeAge package (Matschiner et al., 2016).

| Characteristics of plastomes
The plastome of L. columnaris is circular and quadripartite by having a large single copy (LSC), a small single copy (SSC), and two in-  (Table 2).

| Phylogenetic relationships
The phylogenies from the maximum likelihood (ML) and Bayesian inference (BI) analysis had identical topologies. As expected, all populations of L. columnaris sampled from Bioko and Cameroon shared the same ancestor. However, they were arranged in three highly supported clades. Our plastome phylogenetic analysis shows an unexpected re-

| Events of divergence times
Three   columnaris is South Bioko. Therefore, for this analysis, the populations of Bioko were coded like one region and mainland a second region. The clade of South Bioko had the highest probability (44%). The most probable ancestral area for North Bioko was 37%, and for the clade of Cameroon, the probability was the lowest (33%) (Figure 7).

| Historical colonization and sky island's age
The Hawaiian archipelago is well documented for the correlation between the age of the islands and their colonization and radiation of various plants' lineages such as the Hawaiian lobeliads (Givnish et al., 2009). In contrast, the West Central Africa sky archipelago does not have a simple chronological age from east to west (Suh et al., 2008), and the whole CVL cannot be compared to the chronological age of the Hawaiian archipelago. Nevertheless, we found that the six sky islands that we studied for L. columnaris have a clear progression of age. Their estimated ages run from old (West-Central Cameroon) to young mountains (Bioko and Mt. Cameroon). Our plastome phylogeographic interpretation indicated that L. columnaris colonized the older sky islands in mainland Cameroon after establishing South Bioko's younger sky islands. We recognized with the DEC results using RASP that Bioko was the ancestral area. After this biogeographic analysis, we can infer that Bioko was the cradle for speciation and subsequent dispersal to mainland Cameroon.
Overall, this result contradicts our sky island age hypothesis because Cameroon's sky islands with estimated ages of 3.0, 22.0, and 31.0 Ma are older than Bioko's sky islands (ca. 1.3 Ma) (Table 1).
Our results suggest a biogeography history with an inverse correlation with the age of the Afromontane sky islands. There is no simple answer as to why this giant lobelia dispersal history occurred from young peaks or sky islands to old mountains. It is possible that this plant species found a niche at South Bioko that facilitated its survival during critical volcanic and environmental changes in the region.
Phylogeography is fascinating because we have much to learn before we can understand the evolutionary biology of organisms, especially those in regions such as West Central Africa, where the climatic and geologic history is very complex.
Perhaps, the South Bioko populations are the sister clade on the ML/BI/BEAST trees because the plastomes provide a skewed view of the biogeographic history of L. columnaris, or there is sampling error because we sequenced only one to two individuals per population. Furthermore, we do not have population data from Nigeria, and we did not find more populations in Cameroon, possibly because of local extinction events caused by the intense human transformation of the Afromontane habitat in the sky islands.
It is possible that the incorporation of population-plastome data from Nigeria will not change the biogeographic patterns found in this study. Because our data show that the ancestral area is in South Bioko, F I G U R E 5 MCC tree with 95% highest probability density (HPD) confidence intervals for phylogenomic relationships and estimation of divergence times obtained in BEAST. The two calibrations are represented by star symbols on the respective internal nodes 0 0 0 0 5 5 5 5 1 1 1 1 0 0 0 0 1 1 1 1 5 5 5 5 2 2 2 2 0 0 0 0 2 2 2 2 5 5 5 5 we can predict that populations of L. columnaris in Nigeria will be phylogenomically closer to Cameroon with recent colonization history.
Our historical colonization interpretation using plastid genomes and the DEC analysis might be overestimated because the two calibrations used in the estimation of divergence of populations of L. columnaris are a little older than the age of the sky islands of our study.
However, the colonization history might have underestimated divergence times, but this result will not change the historical dispersal from Bioko to the mainland.

| Afromontane forest
Eastern giant lobelias evolved in Afromontane forests at elevation ranges from 1000 to 2500 m. Through frost tolerant adaptations, some taxa colonized the inhospitable Afroalpine elevation from 3000 to 5000 m (Hedberg, 1969). We found that middle elevation is the most likely habitat for L. columnaris. Only Pico Basilé's (North Bioko) population extends its range from middle elevation up to >3000 m. At this elevation, the habitat is composed of shrubs and subalpine meadows.
The paleoclimatic scenarios of the Afromontane vegetation during the Pleistocene in West Central Africa are not well known.
However, it is known that climatic fluctuations changed the past vegetation patterns dramatically. With pollen analysis it is possible to reconstruct historical processes (Kadu et al., 2011). Moreover, two broad phylogeographic hypotheses might help understand the complex vegetation structure of Africa's mountains. The first is referred to as the "mountain-forest bridges" hypothesis (Kebede et al., 2007); this assumes short-range dispersal between adjacent sky island ranges that facilitated more wide-ranging forest coverage during warm and humid interglacial periods. The second hypothesis proposes that "long-distance dispersal" events among isolated mountains during the Pleistocene glacial periods are responsible for the observed phylogeographic patterns of organisms (Mairal et al., 2017). The interglacial and glacial period dynamic allowed species to disperse and be isolated in adjacent mountains (Zimkus & Gvoždík, 2013). Likewise, these environmental fluctuations could cause local and regional extinctions of organisms in alterations (Gao et al., 2015). Therefore, the current disjunction of L. columnaris in sky islands probably was a Pleistocene product of

F I G U R E 6 Geographic and phylogenomic relationship of plastomes of Lobelia columnaris
Lobelia thuliniana KY354227

| Pleistocene refugia
The entire Cameroon line is considered a Pleistocene forest refuge for its distinctive flora. Forest refugia is also supported because of the high genetic diversity detected in different trees (Pineiro et al., 2017), such as the genus Greenwayodendron (Migliore et al., 2018), and other flowering plants. For example, Arabis alpina (L.) survived the Pleistocene oscillations of temperature and drought in refugia.
Once the environmental conditions changed during the interglacial periods, A. alpina colonized or recolonized new sky islands in East Central Africa (Assefa et al., 2007).
It is possible that the populations of L. columnaris were dynamically isolated and expanded by short-range dispersal within and between the sky islands of West Central Africa. However, Pico Biao-Moka in South Bioko might be a Pleistocene forest refuge to Afromontane plant species. Additional landscape changes occurred in the region at the end of the last glacial period (ca. 10,000 years ago) when a rise in sea level isolated Bioko from the African mainland (Jones, 1994). A shallow channel separates Bioko from the Cameroon coast by 32 km (Schabetsberger et al., 2004). The expansion, colonization, and recolonization of L. columnaris was possible mainly by wind dispersal and possibly by birds. The wind played an essential role in dispersing tiny seeds across sky islands. In Cameroon, high ridges act as a natural forest corridor connecting sky islands and facilitating dispersal and gene flow among contemporaneous populations (Smith et al., 2000).

| Ecology and conservation
The type of ecological habitat may have a possible effect on the morphology of Lobelia columnaris (Mabberley, 1974b). This observation has to be developed in future studies. Our study observed that populations grow in a mosaic of ecological habitats and elevational gradients. Indeed, we observed phenotypic variation in some traits, for example, in plant height, size and number of inflorescences, leaves, and flower color. L. columnaris is smaller in height, inflorescence, and leaf length at higher (approx. 3000 m) and lower elevation (1000 m).
At mid elevation, between 2000 and 2600 m, the morphological variation is spectacular with greater height, inflorescence size and number, flower color, and leaf length.
The populations of L. columnaris in mainland Cameroon are at high risk of local extirpation because of excessive anthropogenic pressure on montane forest fragments. Only the populations on Mt. Cameroon, which is a National Park, have conservation protection plans. The scenario on Bioko for this giant lobelia and the Afromontane forest is better than the mainland. Bioko has an active conservation procedure for two of the three sky islands (Müller & Pócs, 2007). Moreover, South Bioko is undisturbed because of low human population density and supports the highest numbers of plant and animal species on the island (see Jones, 1994).

| Study limits
Several factors limit our results. First, this preliminary study was conducted with few individuals (one to two) for each population, and as such, it should be considered a baseline for future studies.
The available sample size for every population is 10 individuals. We hope to conduct future studies with all the individual samples and include samples from Nigeria to better interpret the colonization history of L. columnaris in its whole geographic distribution. Second, social instability was the main reason no samples of L. columnaris were collected in Nigeria. Future fieldwork in Nigeria to find and collect populations will expand the sampling effort and allow a more robust phylogeographic study. Third, populations from Mt. Cameroon were under-collected because it was challenging to get the permits to work in the National Park. Finally, this study presents a partial history with the analysis of a nonrecombinant marker. A more robust analysis should include nuclear genomic sequences.

| CON CLUS IONS
Our phylogenetic tree based on plastomes enriched our understand-

ACK N OWLED G M ENTS
We express special thanks to Dr. Eric Knox for advice, assistance in the laboratory, and discussions. We thank Dr. Tina J Ayers for her comments that greatly improved the manuscript. We also thank Maximilano Ferro, the Universidad Nacional de Guinea Ecuatorial, and the Urecanos helped with fieldwork in Bioko Island. Collections were made possible and transported with the respective permits from Cameroon, Equatorial Guinea, and the United States. Finally, we thank the two anonymous reviewers who provided insightful comments on an earlier draft.

CO N FLI C T O F I NTE R E S T
None declared. The 20 new annotated plastomes were submitted to GenBank (for accession numbers see Table 2). Voucher specimens were deposited in the Philadelphia Herbarium (PH). Voucher duplicates were deposited in the National Herbarium of Cameroon (YA) and the Real Jardín Botánico Madrid (MA).