*4.1. Recoverability of DNA Barcode Used*

The two DNA barcodes *rbcLa* and *matK* used in this study have long been recognized having sufficient variation to discriminate among land plant species [11,25,26]. Among the three barcodes, *matK* had the lowest rate of recovery (79%), consistent with prior studies [18,27,28]. In contrast, *rbcLa* and *trnH-psbA* had higher rates of recovery (above 95%). However, it is worth pointing out that the rates of recovery were in general higher than in prior studies, probably due to the efficiency of the Canadian Centre for DNA Barcoding that has optimized protocols for higher rates of recovery. For example, recovery rates around 70% have been reported for *matK* in several studies [8,27,29], while sequencing and amplification success for *rbcLa* and *trnH-psbA* is often below 94% e.g., [8,27,30].

### *4.2. Tree Species Identification Using DNA Barcode in Ngel Nyaki Montane Forest*

The morphological identification of the trees in the Ngel Nyaki plot was almost entirely performed by non-professional taxonomists who however accurately identified to species 69% of all tree species occurring in the plot. Only four species were wrongly identified. The DNA barcode was instrumental in updating the identification of 12% of the species in the plot for which prior sequences were available in Genbank. Due to the lack of adequate library in Genbank, 21% of the species in the plot for which good quality barcode sequences were generated could still not be identified to species level. Hence, molecular techniques such as DNA barcode may not replace traditional taxonomic techniques as suggested by some studies [31], but can only supplement it.

This study showed the efficiency of the two barcode loci *rbcLa* and *matK* in accurately assigning Afromontane forest tree species to a correct species or genera. When

used alone, best results for species identification were obtained with *matK* (98%) compared to *rbcLa* (94%). These values are slightly higher than those reported in most lowland forests [8,18,27,30]. The combination of the two markers *matK* + *rbcLa* improved the barcoding success to 99%, a result consistent with those in most lowland forests. Barcoding success was even better at genus level, *rbcLa* alone identifying 98% of all genera, while *matK* and the combination *matK* + *rbcLa* accurately identified all the samples to genus.

The genetic distance method that we used did not allow us to test the accuracy of the intergenic spacer *trnH-psbA*. This locus, easy to amplify and short, is known to be very variable among angiosperms and thus is widely used in plant species identification [32]. In general, *trnH-psbA* locus is more variable than *matK* and *rbcL* and we assume its performance in the identification of montane forest species would even be greater. *matK* and *rbcLa* were variable enough that their combination to *trnH-psbA* was no more relevant.

### *4.3. The Efficiency of DNA Barcoding in the Context of the Afromontane Flora*

DNA barcode is a powerful tool for identifying tree species to genus level. However the identification to species level is not always reliable, especially in plant communities with speciose genera [18]. For example, the identification of tree species (with dbh ≥ 1 cm) in a 50-ha plot in the highly diverse Korup National Park, Cameroon using three DNA barcode markers showed a significant decrease in their performance with increasing number of species per clade (genus) [18]. In fact, the five most speciose genera in the Korup plot *Cola* Schott & Endl., *Diospyros* L., *Psychotrya* L., *Rinorea* Aubl. and *Garcinia* L. have 23, 14, 13, 13 and 10 species respectively [33]. Such closely related species are more likely to hybridize, have incomplete lineage sorting and share haplotypes, all of which can lessen the ability of barcode loci to discriminate among them. At the other end of the spectrum, 165 (33%) species in Korup are represented by a single species.

The Ngel Nyaki plot had a relatively low diversity, with only 105 species in 92 genera. The most speciose genera here are *Ficus* L. and *Psychotria* L., each having three species. Five other genera have two species each, while the remaining 85 species (81%) are represented each by a single species. This species-to-genus (S/G) ratio is not specific to the Ngel Nyaki montane forest. In fact, most Afromontane forests are characterized by a low diversity of trees and low S/G ratio. For example, in Woodbush–De Hoek montane forest in South Africa, 50 species of trees with dbh > 5 cm and dbh > 10 cm in 46 genera (S/G = 1.09) were recorded within 1.5 ha circular plots [34]. Similarly, [35] in a study on trees with dbh ≥ 5 cm in dry Afromontane forests of Awi Zone, northwestern Ethiopia, recorded 18 species in 18 genera, 21 species in 21 genera, 20 species in 20 genera, 16 species in 16 genera and 23 species in 23 genera in 0.6 ha of Bari, Apini, Dabkuli, Tsahare Kan, and Kahtasa forests respectively.

We further explored the relationship between the S/G ratio and elevation, by comparing the Ngel Nyaki data other African forest sites for trees with dbh ≥ 10 cm (Table A1). The S/G ratio decreases with increasing elevation, with a correlation coefficient of −0.722 (Figure 1A). The Lambi 2 and Ngovayang mid-elevation plots in Cameroon had the highest S/G ratio (1.55 and 1.51 respectively) while higher elevation plots Bwindi 1 and Bwindi 4 had the lowest. The Lambi the Ngovayang plots seem to be outliers in our dataset. In fact, a stronger relationship with *r* = −0.80 is shown when these plots are removed. Higher S/G ratio of 2.6 and 3 have been reported elsewhere in the Manu forest (Peru) and Yasuni forest (Ecuador) respectively for trees with the same diameter cutoff [36]. The S/G ratio increases when smaller diameter size classes are considered and the correlation with elevation is stronger (*r* = −0.84, *p*-value = 0.007). A highest S/G ratio of 1.64 is observed for the lowland Rabi plot and 1.15 for the Ngel Nyaki plot for all trees with dbh ≥ 1 cm were measured (Figure 1B). In fact, the understory of most African forests are stocked with speciose genera of small-statured trees that never attain large size diameter classes [37,38]. Several studies have shown the decrease in tree species diversity with elevation, e.g., [39,40]. Our data also shows a decrease of generic diversity with increasing elevation (*r* = −0.84). This means that the low diversity in higher elevations is also due to the decrease in the number of genera,

but coupled with the decrease in the number of species per genera. This result is consistent with Jaccard's observations in the Alps [41], who noted that "with increasing altitude, the number of genera decreases less rapidly than the number of species".

**Figure 1.** Correlation between the species-to-genus (S/G) ratio and elevation, (**A**) for trees with dbh > 10 cm in forty three 1-ha African forest plots, The correlation coefficient *r* = −0.722, *p*-value = 0.00000004635; (**B**) for trees with dbh > 1 cm in seven large (10–50-a) census plots, correlation coefficient *r* = −0.88, *p*-value = 0.007.
