The original microsatellite dataset consisted of 11 loci from 85 red colobus living in KNP. Locus D17S1290 was the only locus with a significant FIS after Bonferroni correction and it was significantly positive (indicative of a heterozygote deficit). This locus was also the only locus with both a significant number of null alleles and a homozygote excess due to scoring errors according to MICRO-CHECKER. Due to these discrepancies, this locus was excluded from all further analyses, including the following summary statistics (Table 1). An average of 7.70 alleles was scored from each locus, and heterozygosities averaged 0.70 for H0 and 0.72 for He. None of the 45 pairs of loci were in genotypic linkage disequilibrium after Bonferroni correction. In turn, only one locus (D20S206) deviated significantly from Hardy–Weinberg expectations, but this departure was not due to either a heterozygote deficit or excess according to its non-significant FIS. This result of a significant Hardy–Weinberg test, but insignificant FIS was due to the existence of a heterozygote excess for the genotypes of some alleles, but a heterozygote deficit for others.
Number of populations
All three approaches for examining the number of populations of red colobus from KNP supported a single population. First, the RST and 4Nm estimates across all six groups were negative according to RSTCALC (−0.01 and −25.28, respectively), with the former not deviating significantly from zero according to its permutation test (P > 0.10). None of the RST for the 15 pairwise comparisons between groups were significant (after Bonferroni correction). Thus, the RSTCALC results reveal no significant genetic differences among the six groups, and indicate that the six groups can be considered as members of one panmictic population.
Second, mean ln L followed a U-shaped pattern as K (number of populations) was increased from 1 to 6 in the STRUCTURE analysis (Fig. 3). This pattern of elevated mean ln L for larger values of K has been reported by the authors of STRUCTURE (Pritchard and Wen 2003) as well as by others (Evanno et al. 2005). It has been attributed to inflated support for values of K greater than the true number of populations. Thus, of greater importance is the failure of the STRUCTURE analysis with K = 6 to clearly assign the 85 individuals of red colobus to separate populations (Fig. 3). This failure, coupled with the earlier warnings of others, indicates that one population is more likely than two or more for the red colobus of KNP.
Figure 3. (Above) Mean ln L (circles) and their standard deviations (vertical lines) for K = 1 to 6 populations as obtained with 10 independent runs per K with STRUCTURE. (Below) Relative probabilities of assigning the 85 individuals of red colobus to six (differently colored) populations. These STRUCTURE results clearly show that little to no support exists for the assignment of the 85 individuals to separate distinct populations.
Download figure to PowerPoint
Finally, the longest internal branch in the unrooted neighbor-joining tree with ∆μ2 distances divided the six groups into two larger groups of Dura, K30, and Large Mikana versus Mainaro, Sebatoli, and Small Camp (see Fig. 2 for group names). The ln mL for the one-population model was −8,196.98 according to MIGRATE. Conversely, the ln mL for the two-population model (with its two groups as defined above by the neighbor-joining tree) was –26,261.91. Thus, the two models differed by a ln BF of 18,064.93. In the statistics literature, a ln BF of >5 is generally considered to be “decisive” for one model over another (Kass and Raftery 1995). Following this literature, the one-population model is thereby taken as strongly favored over the two-population model with mean Θ and symmetrical migration. Given this selection of the simpler one-population model, we stopped there. Otherwise, we would have continued with an evaluation of a two-population/three parameter model (e.g., one with asymmetrical migration and mean Θ). Taken together, the RSTCALC, STRUCTURE, and MIGRATE results suggest high levels of gene flow throughout the park indicating that the red colobus of KNP is one large interbreeding population. These results fit with observational data from the field. For example, we have observed three female dispersals and one male dispersal in a year between the Large Mikana and Small Camp groups. On the basis of the genetic and field evidence, all groups were analyzed as a single population in the calculation of the summary statistics (Table 1) and in the following analyses of population fluctuation.
The results from BOTTLENECK and M_P_VAL showed little evidence of a recent bottleneck for the red colobus in KNP. The one-tailed Wilcoxon test for heterozygote excess was not significant (P = 0.410) and the allele frequency distribution was L-shaped according to BOTTLENECK. The M ratio was calculated as 0.839, which was not significant (P = 0.303) according to its simulations in M_P_VAL. This large M ratio agrees with the empirical estimates of other species with stable population sizes (Garza and Williamson 2001).
The MSVAR results were less clear. The averages and 95% highest posterior densities (HPD) for log10 N0 (current Ne), N1 (ancestral Ne), r (N0/N1), and t (starting time of the population size change) were (3.22, 2.60–3.75), (3.92, 3.24–4.67), (−0.70, −1.58–0.16), and (4.18, 2.54–5.97) according to the 36,000 MCMC samples of MSVAR. Thus, the 95% HPD for log10 r included zero, which is the critical point for N0 = N1. This critical point fell within the right tail of the posterior distribution at P(two-tailed) = 0.092 (Fig. 4). Conversely, the transformed linear means and 95% HPD for r and t were (0.20, 0.03–1.45) and (15,136, 347–993,254 years ago), which suggests a decline in Ne of ~80% over the last ~15,000 years. We also note that Storz and Beaumont (2002) described a ln BF test, which compares how well a model for population contraction fits the data relative to one for expansion. Nevertheless, because our 95% HPD for the log10 r included zero, we cannot reject the hypothesis of a constant population. Furthermore, the estimated mean and median for the start of this decline (~15,000 years ago) falls within a period when tropical forest regions were expanding as earth entered its current interglacial (Castañeda et al. 2009). Thus, when coupled with the non-significant deviation of log10 r from zero, we cannot exclude a constant population hypothesis.
Figure 4. Distribution of log10 r (log10 N0/N1) summarizing the 36,000 MCMC samples from MSVAR. N1 represents the past population size and N0 represents the size of the current population. The upper boundary of our 95% credibility interval overlaps log10 r = 0 as shown by the vertical gray line. This is where N0 = N1, which is indicative of no population size change over time. The log10 r estimates of this histogram support a model of population decline over growth by a ln BF of 20.518 (Storz & Beaumont 2002). Conversely, our study concludes in favor of a constant population given that log10 r = 0 falls within the 95% credibility interval of these estimates.
Download figure to PowerPoint
The EBSP for all models show a flat line for Ne since mean TCOAL (Fig. 5). The posterior probabilities of the demographic.popSizeChange was the highest when it was 0 and all 95% credible intervals included 0 for all models suggesting no population size change in the past. Of these 12 models for microsatellite evolution, the proportional-length, two-step (PU2) model had the best fit to our dataset (Table 2). This model was supported over its closest competitor, the equal-length, constant-size, two-step (EC2) model, by a non-decisive ln BF of 4.15 (Kass and Raftery 1995). The poorest model was the equal-rate, linear-size, single-step model (EL1; ln BF = 80.92 vs. PU2; Table 2). The mean TCOAL was estimated between 98,000 and 197,000 years ago among the 12 models, with the best model (PU2) predicting 98,000 years ago. Similarly, MIGRATE, using the Brownian approximation of the SMM and the one-population model, calculated a mean TCOAL of 112,848 years ago (standard deviation = 71,414). The ln mL for the constant population size analyses with the PU2 and EU1 models were −943.44 and −988.68, respectively, and differed from those of their EBSP counterparts with PU2 and EU1 by ln BF of 3.68 and 2.08, respectively (Table 2). These small ln BF are non-decisive according to Kass and Raftery (1995). Thus, the near-equal fits of both constant population size models confirm that the EBSP graphs are indicative of a stable population for the red colobus in KNP.
Figure 5. Extended Bayesian Skyline Plots for the 10 microsatellite loci using the PU2 (proportional-length, unbiased, two-step) model (a) and EU1 (equal-rate, unbiased, one-step) model (b). These skyline plots are extended to their mean coalescent times (TCOAL) on the right end of the graph. The gray shading corresponds to the 95% HPD around the mean Ne. PU2 exhibits the best fit to the data, whereas EU1 approximates the SMM (Table 2). Both models (as do the other ten) show similar skyline plots with flat lines that correspond to a stable population size over time coalescing between ~100,000 and 200,000 years ago.
Download figure to PowerPoint
The EBSP approach is powerful and flexible, because it is not constrained a priori to a specific model of historical population size change (Heled and Drummond 2008). Thus, the EBSP recovery of a constant population for the red colobus in KNP is derived from the data rather than from the a priori selection of a particular growth model. This support for a constant population is based on the maximum clade credibility (MCC) genealogies of each microsatellite locus (Appendix A2), which show a standard coalescent pattern of increasing waiting times between coalescences as one works backward in time (Wakeley 2008). Thus, relatively few internodes of each MCC gene genealogy occur more than >10,000 years ago. It is important to recognize that this paucity of old internodes does not reflect a lack of support for a constant population with mean and median TCOAL of ~100,000 years ago. Instead, this paucity of old internodes contributes to the signal for a constant population, rather than for some other model such as for an expanding population, where the coalescences are expected to cluster closer to the base of each gene genealogy (Page and Holmes 1998; Hein et al. 2005). However, in keeping with our conservative approach, we limit our estimate of the population stability in red colobus to at least ~40,000 years ago (i.e., to the lower boundary of the 95% HPD of TCOAL for the PU2 microsatellite model; Table 2).This conservative treatment of the duration of the population stability is also warranted given that TCOAL is an estimate of the coalescent time of gene genealogies and not necessarily the most recent common ancestor of the population (Hein et al. 2005).
The evolution of microsatellites is complex because it is characterized by various mutational biases (Goldstein and Pollock 1997). A common property is the positive relationship between allele length and mutation rate, which is evident in our data (Fig. 6). Such complexities call for the development and use of more realistic evolutionary models for microsatellites, such as the PU2 model (Table 2). Nevertheless, the fact that even simple models (e.g., EU1 which approximates the SMM) produce skyline plots similar to those of complex models (Fig. 5) suggests that simpler models can also be of value, particularly when dealing with large microsatellite datasets that are computationally intensive. The maximum clade credibility trees scaled to years for the 10 loci for models PU2 and EU1 are in Appendix A2.
Figure 6. Significant positive correlation between Ho and average allele length for seven tetranucleotide loci (Spearman Rank Correlation, P = 0.033). The average allele lengths for each locus are weighted by their allele frequencies. This comparison does not include the three dinucleotide loci of the full dataset.
Download figure to PowerPoint
Effective and census population sizes
Mean Ne varied among the EBSP graphs for the 12 microsatellite models from ~2,500 for the two best models (PU2 and EC2) to ~4,000 (Fig. 5). The BEAST analyses with a constant population size model supported mean Ne of 2,517 (95% HPD of 1,347–3,710) and 3,677 (2,561–4,857) for the PU2 and EU1 models, respectively. In turn, the one-population analysis with MIGRATE (which also assumes a constant population size) supported mean Ne of 4,055 (2,978–5,426). Collectively, these different analyses of population size agree on a mean Ne of ~2,500 to 4,000 for the red colobus in Kibale.
Using this range of 2,500–4,000, we estimate the Ne/Nc ratio to be ~0.15 to 0.25 for this population. These numbers are lower than the previous Ne/Nc estimate of ~0.35 for the red colobus population in KNP (Struhsaker and Pope 1991). However, this previous Ne/Nc estimate is based on older census techniques and calculations of male and female reproductive success, and is currently thought to be an overestimate (T. Struhsaker pers com.). Importantly, a number of factors are thought to reduce Ne with respect to census estimates: skewed sex ratio, overlapping generations, and variation in adult reproductive success (Frankham 1995). The red colobus at KNP are characterized by all of these demographic factors. They exhibit a skewed sex ratio of one breeding male to every 1.3 breeding females (Struhsaker and Pope 1991). Red colobus also have overlapping generations, their overall lifespan is thought to be around 9.3 years (Pope 1996) with an overall reproductive lifespan of 8.8 years, and adults average one birth every 2 years (Struhsaker and Pope 1991). Combined, these factors likely explain the low Ne/Nc ratio found in this population of red colobus.
In conservation biology, a critical Ne of ~500 to 1,000 is often cited as necessary for the genetic security and long-term survival of a population (Franklin and Frankham 1998). However, other studies have argued from the perspective of the nearly neutral theory that a minimal viable population size of ~1,000 to 5,000 is more appropriate (Lynch and Lande 1998). Our Ne estimates for the red colobus in Kibale (~2,500 to 4,000) overlap with the latter values, and thereby suggest that these larger Ne are important for the genetic integrity and long-term viability of populations, similar to other recent studies (Flather et al. 2011).
History of red colobus and Kibale forest
Red colobus (Procolobus[Piliocolobus]) are found throughout most of equatorial Africa in small, mostly declining populations (Fig. 2). The morphotype, P. [Piliocolobus] rufomitratus belongs to a central/eastern African clade of red colobus (Struhsaker 1981b; Grubb et al. 2003) that radiated around 1.4 million years ago (Ting 2008). Procolobus [Piliocolobus] rufomitratus likely split from other Piliocolobus sometime after 600,000 years ago (Ting 2008), and our results suggest that the KNP population of red colobus has remained stable for ~40,000 years or more.
Red colobus are an important indicator species of their habitat, such that evidence of a population size change can be interpreted as a change in their habitat (Struhsaker 2005). Our results suggest that these monkeys have been in a forest approximately the same size as KNP during at least the last ~40,000 years as changes in the available forest environment, or population movement, would have led to a corresponding change in population size. Furthermore, in recent years, no population size change in red colobus has been detected from 1970 to 2006 suggesting the forest is at carrying capacity for the red colobus (Chapman et al. 2010a,b). Thus, the most parsimonious hypothesis is that the forest of Kibale National Park has been persistently occupied by this population of red colobus for the duration of their population stability. This suggests that the forest has served as a refuge for tropical rainforest species during the last ~40,000 years and possibly even longer.
The alternative hypothesis, that this population of P. rufomitratus occupied a forest other than KNP during this time and then moved en masse to this area, is less likely for several reasons. The lack of a dramatic drop in population size does not support a recent founder event or the possibility that these red colobus are the remnants of a larger and more widely dispersed population. If these events occurred, they must have done so earlier than ~40,000 years ago. It is also unlikely that this population of red colobus migrated en masse from another forest similar in size and environment sometime during the Late Pleistocene without any significant change in effective population size (e.g., bottleneck or founder event). Therefore, we conclude that the population stability of red colobus in KNP during the Late Pleistocene reflects similar long-term stability of the Kibale forest environment.
East Africa has experienced many paleoenvironmental and anthropogenic changes over the last 100,000 years. This geographic region has undergone many long periods of extreme drought during the Late Pleistocene, first from ~135,000 to 70,000 years ago (Scholz et al. 2007), and again during the last glacial maximum ~18,000 years ago when Lake Victoria completely dried up (Stager and Johnson 2008) and regional vegetation changed to scrub and montane grasslands (Jolly et al. 1997). During these periods, a major Central Refuge for tropical forest species existed to the west of Kibale forest over the Rwenzori Mountains in the neighboring Democratic Republic of the Congo (Hamilton 1976; Jolly et al. 1997; Wronski and Hausdorf 2008). Similar periods of environmental variability existed during the Holocene and are evidenced by a population crash of Ugandan and Kenyan buffalo indicative of a drought ~4,500 years ago (Heller et al. 2008). There is further evidence of anthropogenic impacts to the forests surrounding Kibale forest starting with the arrival of the Bantu-speaking people ~2,300 years ago and through subsequent shifts in their settlement patterns (Taylor et al. 1999).
Despite this variability, there is evidence that upper montane forests may have persisted during these climatic cycles (Jolly et al. 1997). Critically, Kibale forest is considered a separate habitat patch that persists due to its favorable mid-elevation topography and humidity (Struhsaker 1981a; Jolly et al. 1997; Moore 1998; Wronski and Hausdorf 2008). Our results for a constant red colobus population in KNP corroborate the hypothesis of a stable Kibale forest, despite the known environmental and human changes in East Africa during the Late Pleistocene and Holocene.
Our results further suggest that a tropical rainforest patch of only ~795 km2 (i.e., the size of KNP) is sufficient to support a long-term viable population of red colobus. This conclusion is important given the rapid rate at which their habitat is disappearing and the lack of knowledge of what makes a suitably sized conservation area. However, this estimate of a minimal habitat size refers specifically to red colobus because other species differ in their ecology and behavior and will have other habitat requirements (e.g., it is known that a territorial top predator such as a leopard needs a much larger habitat for the long-term survival of its population).
We now call for critical tests of our hypotheses for an old and stable KNP population of red colobus and, thus, Kibale forest. Palynological studies from throughout KNP are needed to critically test the hypothesis of a continuous old tropical rainforest of approximately the same size as that currently found in the park (Jolly et al. 1997). There has been no known connectivity between the red colobus of KNP and any other group since the 1950s. However unlikely, population genetic studies are now needed to test whether gene flow is occurring or has occurred with these other populations in East Africa. These studies are particularly important given that tropical rainforest in East Africa was much more widespread in the past (e.g., during the Holocene Climatic Optimum ~6,000 to 10,000 years ago; Jolly et al. 1997). They are also significant in light of recent simulation studies (e.g., Chikhi et al. 1991; Peter et al. 2010), which show that immigration can result in a false signal of a population decline (i.e., the ~80% decrease at ~15,000 years ago as suggested by MSVAR). Additional simulations are also now needed to validate the efficiency and robustness of EBSP to recover older population size changes with microsatellite data (Wu and Drummond 2011). Such critical tests and experiments are necessary for the final acceptance of our conclusions for an old and stable population of red colobus and Kibale forest.
Biodiversity is lost at an astonishing rate each year, particularly in the tropics where deforestation is rapidly extinguishing species habitats. In Uganda, it is estimated that 85% of the forests have disappeared (Howard et al. 2000). Wise conservation efforts over the last 40 years in Kibale National Park have ensured that this area has persisted as a protected National Park (Box et al. 2008). Our research suggests that this forest has been stable for at least the last ~40,000 years, during which it likely served as an important sanctuary for tropical rainforest species in East Africa. Furthermore, our findings indicate that species with effective population sizes of ~2,500 to ~4,000 may be able to persist long term in stable environments, thereby providing hope for other species in decline or with patchy distributions.
Kibale forest offers a rich source of information about Pleistocene refugia, which increases its importance as a major center for conservation and scientific research. With continued wise management, KNP can likely continue to serve as a critical refuge for these species well into the future.