Sample collection complied with the laws of Ethiopia and Germany and with the guidelines of the International Primatological Society. During sampling of faecal material, no animals were harmed or disturbed.

We non-invasively collected fresh faecal samples at 26 sites in the range of C. guereza (Fig. 1 and Table S1 in the Supplement). Geographic coordinates of each sample were assigned at the time of collection by GPS. Faecal samples were collected and stored following the two-step protocol of Nsubuga et al. (2004) and Roeder et al. (2004). Samples were stored at ambient temperature for up to 3 months in the field and at -20 ∘C upon arrival in the laboratory of the German Primate Center (DPZ).

We extracted total genomic DNA using the First DNA all tissue kit (Gen-Ial) according to the manufacturer's protocols, with minor modifications as outlined in Kalbitzer et al. (2016). After extraction, DNA concentration was measured with a NanoDrop ND-1000 spectrophotometer (Peqlab) and extracts were stored at -20 ∘C until further processing.

We amplified and sequenced two mitochondrial fragments, the complete cytochrome b gene (cytb; 1140 bp) and a region spanning a portion of the NADH dehydrogenase subunit 3 gene, the tRNA for arginine, the complete NADH dehydrogenase subunit 4L gene, and a portion of the NADH dehydrogenase subunit 4 gene (NADH; 790 bp). We focused on these two markers because for both, several orthologous sequences from Colobus are available in GenBank for comparisons.

Cytb was amplified via two overlapping PCR products with sizes of 727 and 663 bp, while NADH was amplified via a single PCR product with a size of 873 bp (for primers see Table S2). To minimize the risk of amplifying nuclear mitochondrial-like sequences (numts), primers were specifically designed for C. guereza on the basis of available sequence data in Genbank. We conducted PCR reactions in a total volume of 30 µL containing a final concentration of 0.33 µmol of each primer, 3 mmol MgCl2, 0166 mmol dNTPs, 1× buffer, 1 U Biotherm Taq DNA polymerase (Genecraft) and 100 ng total genomic DNA. Cycling conditions consisted of pre-denaturation for 2 min at 96 ∘C, followed by 40 cycles, each with denaturation for 1 min at 96 ∘C, annealing for 1 min at 50 ∘C for both cytb fragments and 60 ∘C for NADH, and extension for 1 min at 72 ∘C. At the end, a final extension step for 5 min at 72 ∘C was added. To check for PCR performance, aliquots of the PCR products were run on 1 % agarose gels. PCR products were cleaned with ExoSAP-IT^™ Express PCR Product Cleanup Reagent from ThermoFisher and subjected to sequencing using the amplification primers. Sanger sequencing was conducted at GATC Biotech. Sequence electropherograms were checked by eye with 4Peaks 1.8 (https://www.nucleobytes.com, last access: 18 December 2018) and sequences were assembled and manually edited in SeaView 4.4.0 (Gouy et al., 2010). SeaView was also used to check for the correct translation of protein-coding sequences into amino acid sequences.

To reduce cross-sample contamination, all working steps (DNA extraction, PCR set-up, PCR amplification, gel electrophoresis, and PCR product purification) were conducted in separate laboratories. Work benches were cleaned with 10 % bleach and new gloves were used for each sample. Further, PCR controls (without template DNA) were routinely conducted and procedures were repeated for 10 % of randomly selected samples.

We expanded our dataset with orthologous Colobus sequences from GenBank (Table S1). We generated three datasets: NADH (103 sequences), cytb (79 sequences), and the concatenated NADH and cytb (59 sequences). The number of sequences for each marker differed because (1) for the out-group and most of the non-Ethiopian C. guereza taxa, orthologues sequences were not available in GenBank and (2) we were not able to sequence NADH and cytb from every sample.

For phylogenetic tree reconstructions, identical haplotypes were removed, resulting in datasets with 53 (NADH), 34 (cytb), and 33 (concatenated) sequences. Phylogenetic trees were constructed with maximum-likelihood (ML) methods using IQ-Tree 1.5.2 (Nguyen et al., 2015) and Bayesian inferences with MrBayes 3.2.6 (Ronquist et al., 2012). The best-fit models for each dataset were determined with ModelFinder (Chernomor et al., 2016; Kalyaanamoorthy et al., 2017) in IQ-Tree under the Bayesian information criterion (BIC). The ML tree was reconstructed with 10 000 ultra-fast bootstrap (BS) replicates (Minh et al., 2013), while the Bayesian tree was obtained from a Markov chain Monte Carlo (MCMC) run with 10 million generations, sampling every 1000 generations. For the Bayesian reconstruction, we checked convergence of all parameters and the adequacy of a 25 % burn-in by assessing the uncorrected potential scale reduction factor (PSRF) (Gelman and Rubin, 1992). Posterior probabilities (PPs) for nodes and a phylogram with mean branch lengths were calculated from the posterior density of trees using MrBayes. Phylogenetic trees were visualized in FigTree 1.4.2 (http://tree.bio.ed.ac.uk/software/figtree/, last access: 18 December 2018).

Divergence times were estimated with a Bayesian approach as implemented in BEAST 2.5.0 (Bouckaert et al., 2014). We performed two independent analyses for each dataset, each with 25 million generations and tree and parameter sampling occurring every 1000 generations. For the analyses, we assumed a relaxed log-normal clock model and applied a coalescent constant population prior for branching rates. As no reliable information from the fossil record is available to calibrate the molecular clock, we used estimated divergence times based on molecular data (Ting, 2008). For both datasets, we set age constraints using a normal prior. For the NADH dataset, we constrained the split between C. satanas and the other Colobus species at a mean of 3.5 with sigma 0.36, translating into a median of 3.5 Ma and a 95 % highest probability density (HPD) of 2.8–4.2 Ma. For the cytb and concatenated datasets, we constrained the split between C. angolensis and C. guereza at a mean of 2.1 with sigma 0.26, translating into a median of 2.1 Ma and a 95 % HPD of 1.6–2.6 Ma. The adequacy of a 10 % burn-in and convergence of all parameters was assessed by inspecting the trace of the parameters across generations using Tracer 1.6 (http://beast.bio.ed.ac.uk/Tracer, last access: 18 December 2018). Sampling distributions of the run were combined with LogCombiner 2.5.0, and the tree with mean node heights was summarized with TreeAnnotator 2.5.0 using a burn-in of 10 %. The trees were visualized in FigTree.

In primates, as in other taxonomic groups, taxon delimitation based on mitochondrial sequence information often remains unreliable if the respective divergences or radiations are relatively recent. Often mitochondrial clades mirror biogeographic patterns rather than taxonomy (e.g. baboons; Zinner et al., 2009, 2015, 2018b). Since our study was based on mitochondrial markers, we thus have to be cautious when discussing possible taxonomic inferences.

However, results of our study principally support the two-taxa hypothesis for Ethiopian black-and-white colobus. Notably the two Ethiopian mitochondrial lineages' membership of to two distinct C. guereza clades that separated at around 0.7 Ma somehow indicates independent evolutionary histories of the putative subspecies. This divergence age is even more remarkable as the divergence ages among C. g. guereza, C. g. caudatus, and C. g. kikuyuensis are much younger (<0.3 Ma). Also, the phylogenetic relationships among the three taxa, in particular between C. g. caudatus and C. g. kikuyuensis are not well resolved and only weakly supported (Fig. S1a and S1b). Therefore, mitochondrial sequence information also does not provide evidence for C. g. caudatus being the most distinctive form of black-and-white colobus subspecies and thus, do not support the proposal by Butynski and de Jong (2018) to elevate C. g. caudatus to a separate species (C. caudatus). However, at the moment our taxonomic inferences regarding non-Ethiopian C. guereza have to be preliminary since the geographic provenance of several mtDNA sequences is from zoo animals, and we have to rely on the taxonomic classification provided by the authors of the respective sequences.

Since hybridization and introgression is common among various primate taxa (Zinner et al., 2011; Cortes-Ortiz et al., 2019), the Ethiopian guerezas might also be affected, which would partially explain the different views regarding their taxonomy and delineation of subspecies (Yalden et al., 1977; Fashing and Oates, 2013). A thorough study based on genomic data and a reassessment of their morphology would help to shed light on the internal phylogeny of C. guereza.

The close phylogenetic relationship of the eastern Ethioipian C. g. gallarum with the central African lowland C. g. occidentalis was unexpected. C. guereza is ecologically relatively flexible (Fashing, 1999; Fashing and Oates, 2013; Harris and Chapman, 2007) but, as an arboreal species, depends on forest or woodland. Given the historical changes in climate and forest cover in Africa (deMenocal, 2004; Trauth et al., 2009; Hoag and Svenning, 2017), one can assume that ancestral C. guereza populations became isolated in forest refugia under unfavourable conditions, whereas forests and C. guereza populations became reconnected under favourable conditions.

Assuming an origin of C. guereza in the northern forest belt of central Africa, a possible phylogegeographic scenario (Fig. 3) could have started with a range expansion into the forests of north-east Africa and Ethiopia, followed by a first divide into the more central and northern ancestral population of the C. g. occidentalis + C. g. matschiei + C. g. gallarum clade and more southern population of C. g. guereza + C. g. kikuyuensis + C. g. caudatus clade. Both populations further radiated into the respective subspecies. Under unfavourable conditions, C. g. gallarum became isolated in the extreme eastern part of the range. The gap between the eastern and more western and central distribution was then probably filled by a range expansion of the southern clade. In order to find support for such a scenario, genetic analyses of the missing populations of C. g. dodingae and C. g. percivali would be helpful. One could expect clustering of C. g. dodingae with C. g. occidentalis or C. g. matschiei, which would narrow the gap towards C. g. gallarum and clustering of C. g. percivali with C. g. kikuyuensis or C. g. guereza, supporting a south–north expansion into the gap.

If the distinctiveness of the populations carrying C. g. gallarum haplotypes is supported by morphological and nuclear characters, C. g. gallarum is most likely the most threatened subspecies of C. guereza. At the moment the C. g. gallarum haplotypes are only found at two sites where small populations may subsist. Since the forests in the supposed former range of C. g. gallarum have been heavily reduced, the outlook for this taxon is not promising (Gippoliti and Butynski, 2008). In contrast, C. g. guereza is still widely distributed in Ethiopia. Since both Ethiopian guereza taxa are closely related, there is a certain danger of hybridization if they meet. It would probably be interesting, whether genetic traces of earlier hybridization can be found in C. g. gallarum. At the moment, the C. g. gallarum populations seem to be geographically isolated from the closest C. g. guereza population, but care should be taken to prevent the restocking of the small C. g. gallarum population with C. g. guereza individuals, even for conservation reasons (Gippoliti et al., 2018).

It also becomes obvious that, if genetic diversity and local adaptation across the complete range are to be preserved (Harcourt, 2006; Harcourt et al., 2005), several subspecies of C. guereza need particular conservation measures, including C. g. gallarum.