Detection dog efficacy for collecting faecal samples from the critically endangered Cross River gorilla (Gorilla gorilla diehli) for genetic censusing

Population estimates using genetic capture–recapture methods from non-invasively collected wildlife samples are more accurate and precise than those obtained from traditional methods when detection and resampling rates are high. Recently, detection dogs have been increasingly used to find elusive species and their by-products. Here we compared the effectiveness of dog- and human-directed searches for Cross River gorilla (Gorilla gorilla diehli) faeces at two sites. The critically endangered Cross River gorilla inhabits a region of high biodiversity and endemism on the border between Nigeria and Cameroon. The rugged highland terrain and their cryptic behaviour make them difficult to study and a precise population size for the subspecies is still lacking. Dog-directed surveys located more fresh faeces with less bias than human-directed survey teams. This produced a more reliable population estimate, although of modest precision given the small scale of this pilot study. Unfortunately, the considerable costs associated with use of the United States-based detection dog teams make the use of these teams financially unfeasible for a larger, more comprehensive survey. To realize the full potential of dog-directed surveys and increase cost-effectiveness, we recommend basing dog-detection teams in the countries where they will operate and expanding the targets the dogs are trained to detect.

2. Introduction

Population estimates derived from genetic capture–recapture studies can be more accurate and precise than traditional ape surveys based on nest counts [1–4]. In addition to population estimates, genetic capture–recapture studies may permit long-term population monitoring by revealing ranging patterns and group composition [1,2,5]. However, the accuracy of mark–recapture and population monitoring studies depend entirely upon effective sampling of the surveyed area. To be effective, the number of samples collected should be at least two-and-a-half times the number of individuals suspected to be in the population [1,6,7]. Furthermore, because the entire study area cannot be searched in one day, subdivision of the landscape into daily search areas is necessary. The search effort expended in these sub-areas should be consistent and each zone should be revisited as many times as possible. In order to collect sufficient samples for genetic mark–recapture analysis, researchers have used specially trained dogs to detect faeces from a variety of target species over large geographical areas (e.g. bears: [8,9] wolves: [10], wolves, pumas, jaguars, anteaters and armadillos: [11]). These dogs can detect scat samples from specific species with a high degree of accuracy and at rates 5–15 times higher than humans [12]. The use of detection dogs should result in increased sample detection rates and accordingly larger sample sizes, therefore allowing for higher levels of precision than standard sample collection regimes. In addition to detecting more samples, it is possible that dogs detect samples in a less biased way than humans. In the case of gorillas, human-directed detection teams tend to be biased towards collecting samples from nest groups as they are easier to detect then single faecal remains; similarly, collection tends to be biased towards large adult faecal samples rather than smaller infant and juvenile samples. Dogs may be able to overcome these biases and provide a more homogeneous capture rate, and therefore, more precise population estimate, than currently possible from human-directed surveys.

The Cross River gorilla (Gorilla gorilla diehli) is the most endangered of all the African apes and one of the world's most critically endangered primates [13]. These gorillas are found only in remote and mountainous regions along the Nigeria–Cameroon border and are restricted to approximately 11 localities in areas of forest that range from small fragments of 20 km² to large blocks of over 1000 km² [14]. Estimates based on nest counts suggest the population numbers only 200–300 individuals [14]. This population is Critically Endangered [15] and is under intense threat from bush meat hunting, habitat loss and habitat fragmentation [14]. As a flagship species for the region, their conservation is also critical for the protection of the biodiversity hotspot which they inhabit and the other species that share their habitat.

While recent research has offered a number of insights into the biology of the Cross River gorilla (e.g. [16–20]), several questions critical to their effective conservation remain unanswered, including a precise estimate of the gorillas' current population size [14]. Without this knowledge it is impossible to effectively assess subpopulation viability, prioritize interventions or measure success of conservation activities over time. Despite substantial effort over many years (e.g. [21]), efforts to unambiguously determine population size have been thwarted by the total size of the landscape inhabited by the gorillas (over 2000 km²), the difficult terrain, and the reclusiveness of the gorillas. As a result, no DNA-based estimate of population size has been possible and current estimates are based entirely on nest counts, which are prone to significant error [22]. Another important question to address is the degree of connectivity between gorilla localities. Although analysis of non-invasively collected DNA has offered a number of insights into the population structure, genetic diversity and migration between fragmented habitat sites of the Cross River gorillas [16,23], these findings have been limited in their generalizability by small sample sizes from some sites, and a complete lack of samples from others.

Prior to our study, the use of detection dogs for locating faeces had not been attempted in an African tropical forest, an environment where the detection of samples by humans or dogs can be particularly challenging. Here we present the results of a study piloting the use of dogs in the collection of gorilla faeces for genetic analysis to determine whether this approach is suitable for a population-wide study. Our specific goals were to estimate the number of gorillas at two Cameroonian Cross River sites (the smaller and better-studied Kagwene Gorilla Sanctuary and the more expansive and mountainous Mone River Forest Reserve), estimate grouping patterns of the gorillas, and compare the newly identified individual genotypes with those collected from multiple localities 10 years earlier [16] to check for resampled individuals or any individual movement. We also compare the efficiency with which human and dog-directed teams collected samples and assess the relationship between the number of samples and the number of individuals found and the resultant effect on the population estimate, by the two team types. Finally, we draw some conclusions about the applicability of dog-directed faeces detection in mountainous tropical rainforest environments.

3. Material and methods

3.1 Training of detection dogs

Cross River gorilla faeces collected by field staff in Cameroon and from captive western lowland gorillas (Gorilla gorilla gorilla) were initially used to train three Working Dogs for Conservation (WDC) dogs to identify gorilla faeces using methods previously applied in wildlife research [8]. The dogs were trained using a series of increasingly complex detection scenarios, using procedures developed by WDC for the introduction of a new scent target [8,9]. In December 2011, the dogs and their handlers travelled to Cameroon to complete their training via simulated Cross River gorilla searches conducted over a 4 day acclimatization period in the town of Limbé using additional fresh Cross River gorilla faecal samples. Following the acclimatization period, the dog teams travelled to Kagwene Gorilla Sanctuary where final field training was conducted.

3.2 Detection dog-directed field surveys

Dog-directed searches were conducted in December 2011 and January 2012 at two sites known to be occupied by gorillas: Kagwene Gorilla Sanctuary (19 days) and the northern portion of Mone River Forest Reserve (25 days) (figure 1). Kagwene was selected as the first test site because it is small (19 km²), the ranging behaviour of the estimated 20–25 gorillas there is regularly monitored, and the Wildlife Conservation Society (WCS) has a permanent presence at the site. Mone was chosen as the second site as it is more representative of a typical Cross River gorilla locality with a larger forest area (460 km², of which ca 100 km²is occupied by gorillas [14]), steep terrain and limited knowledge of gorilla numbers and ranging patterns. Initially, a 1.5 km² grid was used to define sampling units which included the estimated range of the gorillas at each site. Survey teams planned to conduct 1.5 km guided reconnaissance walks [22] across each cell, following a bearing from the edge of the cell towards its centre. This approach was adjusted in the field based on logistical constraints (see Results) and a method that attempted to maximize the number of grid cells explored each day was used. Whenever a dog detected a scat, the team followed the dog to the source of the scent. Gorillas routinely construct nests of vegetation each night, and samples were collected at nest sites or on trails. Two field teams operated concurrently, each consisting of a dog, a dog handler, a field assistant and a guide. Faecal samples for genetic analysis were collected using the two-step ethanol-silica procedure [24], stored in the field for two months and at 4°C thereafter. GPS coordinates of samples and survey routes were recorded using a Garmin GPSmap 62CX receiver.

3.3 Human-directed field surveys and comparison with dog-directed surveys

Human-directed searches for faecal samples were conducted at Kagwene for 17 days in December 2011. A grid-based survey was not employed and teams consisting of trackers and field assistants searched areas based on individual knowledge of historical gorilla ranging patterns, and predictions of where gorilla foods would be available. When signs of gorilla presence were observed (e.g. feeding signs, signs of passage), teams would attempt to follow the signs to a nest site to obtain samples. GPS coordinates of samples were recorded using a Garmin GPSmap 62CX receiver. The probability of dogs and humans detecting samples found in nest groups versus those found alone on a trail were compared using a Fisher's exact test.

Results from the 2011 human-directed field surveys may be unrepresentative as the samples were collected from the Kagwene gorilla groups that are under study and followed regularly. It was not possible to conduct human-directed surveys in Mone during the 2012 season. Therefore, sample detection rates from dog-directed searches were compared with sample collection rates of human-directed searches at Mone and Kagwene conducted previously between 2002 and 2004 [16]. Both human-directed and dog-directed surveys were conducted during the dry season to maximize the chance of finding usable samples (i.e. those not destroyed by rain) and to control for environmental differences between the two search methods. Sample detection rate was calculated as the number of fresh samples collected per team day (i.e. two teams collecting samples on the same day is counted as two team days). Only faeces that were 1–3 days old, as determined by field staff with years of experience, were collected as older samples tend not to yield usable genotype results [25]. An approximation of gorilla age was not estimated based on faecal bolus size.

3.4 Microsatellite genotyping and analysis

DNA was extracted from faecal samples approximately five months after collection using the QIAmp Stool kit (QIAGEN) with slight modifications [24]. DNA quantification was performed using a 5^′-nuclease assay targeting a highly conserved 81 bp portion of the c-myc proto-oncogene [26].

At least three independent amplifications from each DNA extract along with a minimum of three negative controls were performed at 13 microsatellite loci using a two-step multiplex polymerase chain reaction (PCR) method [27] with the following modifications: (i) in the initial multiplexing step, 16 loci [27] were multiplexed using the Type-it Microsatellite PCR kit (Qiagen) in 20 μl reaction volumes (including 5 μl template DNA); (ii) PCR thermocycling was performed in a PTC-200 thermocycler (MJ Research) with the following parameters: initial denaturation for 5 min at 95°C, 30 cycles of 20 s at 94°C, 90 s at 57°C and 30 s at 72°C, and a final extension of 30 min at 72°C; and (iii) primers were also multiplexed in the second step using the Type-it Microsatellite PCR kit (Qiagen) in 10 μl reaction volumes (including 2.5 μl of 1.5 : 100 diluted 1st-step multiplex PCR product) in the following combinations: (d10s1432 (0.5 μM)-d14s306(0.5 μM)-d5s1457(0.1 μM); d5s1470(0.6 μM)-d4s1627(0.1 μM)-d2s1326(0.2 μM); vWF(0.4 μM)-d7s2204(0.6 μM)-d16s2624(0.1 μM); d7s817(0.2 μM)-d1s550(0.2 μM)-d8s 1106(0.2 μM)-d6s1056(0.2 μM)). Second-step thermocycling conditions were as above except primer-combination-specific annealing temperatures were used (varying from 55°C to 60°C) and the annealing step lasted 3 min. The sex of each individual was determined by amplifying a segment of the x–y homologous amelogenin gene in a one-step PCR [28].

PCR products were electrophoresed on an ABI PRISM 3100 Genetic Analyser and alleles were sized relative to an internal size standard (ROX labelled HD400) using GeneMapper software v. 3.7 (Applied Biosystems). Heterozygous genotypes were confirmed by observing each allele twice in two or more independent reactions. Apparent homozygous genotypes were confirmed by observing only a single allele in at least three to five independent observations depending on DNA quantity [27].

3.5 Discrimination of individuals

We used Cervus v. 3.0 to identify samples with matching genotypes. We determined the minimum number of loci necessary to obtain PID_sibs≤0.001 to ensure with 99.9% confidence that two matching samples originated from the same individual [29]. Consensus names and genotypes were attributed to matching samples. The consensus genotype was used in all subsequent analyses. Genotypes from different samples mismatching at three or fewer loci were re-examined for possible genotyping errors and additional genotyping was undertaken to resolve any ambiguities.

3.6 Gorilla group determination

Individual genotypes derived from samples collected by dog- and human-directed searches were pooled and the number and minimum composition of gorilla groups at Mone and Kagwene were estimated using three possible schemes. First, we assumed that samples from individuals collected together on the same day at the same GPS location (same nest site or multiple fresh faecal remains found together) belong to individuals from the same group (grouping scheme 1). Second, because unique GPS coordinates were taken for most samples even when collected as close as 1 m apart, samples were assumed to come from the same group if they met the following criteria: (i) they were found on the same day, (ii) they were judged as equally old by the field team, and (iii) they were considered by the field team to belong to a single nest group (grouping scheme 2). Third, based on the fact that nest sites contained samples up to 54 m apart (±3 m GPS estimated position error), samples were considered to come from the same group if: (i) they were found on the same day, (ii) they were judged as equally old by the field team, and (iii) they were found within 57 m of each other (grouping scheme 3). Individuals were then further linked under the assumption that if individuals A and B were found at sampling event one and individuals A and C were found at sampling event two then individuals A, B and C are all part of one group [1]. Therefore, group attribution could not be assigned to individuals whose faeces were sampled only once and not together with faeces of other individuals. Minimum home range was determined in QGIS v. 2.2.0-Valmiera through the creation of minimum convex polygons using the GPS locations of individuals based on pooled samples under grouping scheme 3 [1].

3.7 Population estimation by genetic analysis

We calculated genetic capture–recapture estimates using the maximum-likelihood two innate rates model (ML-TIRM) estimator implemented in the software Capwire (www.cnr.uidaho.edu/lecg; [6]). In a previous study on western lowland gorillas it was found that the ML-TIRM estimator is the most reliable of the available published estimators, while other methods appear to underestimate the population size [1]. Furthermore, other studies have shown that although the confidence intervals of the ML-TIRM estimator are larger than those from other estimators, they always capture the true population size within their limits [30]. The approach assumes a closed population (i.e. no births, deaths or migration in the sampling interval) and a recapture probability equal to the capture probability. It also accounts for capture heterogeneity by classifying individuals as having either low or high capture probabilities [6]. As migration of individuals between the Mone and Kagwene areas is not considered probable [16], we calculated population estimates for each locality separately. We grouped each set of samples into a single-sampling session scheme and used consensus names to identify the number of times each individual was captured. Individuals found at the same location on the same day are false recaptures, so only one sample representing that individual at that location was kept in the dataset. Therefore, population estimates were calculated following grouping schemes 1, 2 and 3 as described above (i.e. under grouping scheme 1, only individuals identified more than once at the same GPS location were considered false recaptures, whereas under grouping scheme 3, individuals identified more than once within 57 m of each other on the same day were considered false recaptures). Furthermore, we calculated a population estimate employing all of the Kagwene samples as well as using samples from dog- and human-directed searches separately.

3.8 Detection of migrants

Using the same parameters as in the ‘Discrimination of individuals’ section above, we used Cervus v. 3.0 to check for samples from this study with genotypes matching those from a previous study of Cross River gorillas from multiple localities [16]. As that study [16] was able to infer the presence of migrant gorillas between some Cross River gorilla localities, we also wanted to determine if any of our newly identified individuals from Mone and Kagwene were migrants from another Cross River locality. To do so, we ran a Structure v. 2.1 analysis [31] with the original 2007 dataset and original parameters but with the addition of our newly identified individuals from this study.