## Introduction

Collecting a large number of biological samples from free-ranging animals remains the overarching limitation for mammalian conservation genetics and molecular ecology. While great advances have been made in recent years, the focus has been on how to do more with less: deeper sequencing and fewer samples1,2. This is nowhere more apparent than in primate biology, which often requires obtaining DNA from arboreal, small, fast, cryptic, and nocturnal animals with large home ranges. Combined with the facts that most primates are threatened, and that almost all of them are distributed throughout the Global South—often in difficult to reach jungles and mountainous terrain—international population and conservation genetic projects are routinely deemed impossible before they can begin. For most populations of most species, the cost per sample is simply too high and too time consuming. As a result, the trajectory of decades of genetic research has been to deepen the study of a few, well-known populations, and to write-off the majority of this diverse radiation as unknowable.

## Methods

### Ethics Statement

All methods were carried out in accordance with the Principles for Ethical Treatment of Non-Human Primates set forth by The American Society of Primatologists and the guidelines of the Institutional Animal Care and Use Committee (IACUC) of Washington University in St. Louis. All experimental protocols were approved by the Washington University in St. Louis Animal Studies Committee (A-3381-01 20090204). Permits to collect fecal samples were issued by the Yunnan Forestry Department ( 55 [2009] ).

### Study Sites

We conducted our field work in two forested regions of Yunnan: Wuliangshan National Nature Reserve and Yongde Daxueshan National Nature Reserve. Both reserves are sky island habitats, isolated from neighboring forested areas by continuous lowland agriculture. Wuliangshan National Nature Reserve is located in central Yunnan Province, between the Mekong (Lancang Jiang) and Black (Chuan He) Rivers in the upper reaches of the Wuliang Mountains (Fig. 2). The reserve covers 31, 313 ha in Jingdong, Nanjian, and Zhengyuan counties, with a length of 83 km and a variable width (~5–15 km)34. Yongde Daxueshan (15, 786 ha) is located in Yongde County, Yunnan, about 100 km west of Wuliangshan and near the Burmese border35. The forest structure of both sites is largely similar. Within the suitable primate habitat (~1,800–2,700 m above sea-level), the forest grades from semi-humid evergreen broadleaf forest through a sub-humid evergreen broadleaf zone, above which the forest grades into dwarf rhododendron forest. Inside the reserve, these sites contain a mixture of primary and secondary forest with a greater degree of the former at higher altitudes36,37. Four species of primates are known to be codistributed throughout Wuliangshan (Nomascus concolor, Trachypithecus crepusculus, Macaca arctoides, and M. mulatta)38 and Yongde Daxueshan (N. concolor, T. crepusculus, M. arctoides, and Macaca assamensis, M. leonina). In both mountains, small felids, deer, serows, gorals, flying squirrels, and songbirds are fairly common; bears, leopards, and raptors have also been reported in small numbers, but their present status is unknown.

Wuliangshan and Yongde Daxueshan are protected areas; however, human perturbations of the forest are commonplace, and include encroachment of grazing goats and cattle, small scale logging for house beams and firewood, and foraging for mushrooms. Previously forested areas outside the nature reserves have been almost exhaustively cleared for plantations (e.g. tea, walnuts, corn). Hunting is prohibited, but hunting pressure remains, predominantly for medicinal and commercial purposes. Steel offset-jaw leg traps (“bear traps”), rope snare spring-noose traps, and small caliber bullet shells are present in both reserves, but are more common in Yongde Daxueshan. The Jingdong Management Bureau of Wuliangshan National Nature Reserve has halted primate hunting in Wuliangshan thanks to aggressive conservation efforts, but conservation in Yongde Daxueshan remains challenging, and primates are occasionally discovered deceased.

### Study Animals

Gibbons, langurs, and macaques occupy different ecological niches (Fig. 1). Gibbons are near obligate arborealists, relying on brachiation for locomotion and rarely descending to the forest floor30,39, whereas langurs and macaques are quadrupedal primates occupying a mixed terrestrial/arboreal niche. The gibbons of central Yunnan have a more variable diet than many gibbons, consuming an average of 50% leaves and buds, 43% fruit, and 5% flowers from 83 plant species33,40,41. Indochinese gray langurs in Wuliangshan have been observed to consume a diverse, leaf-heavy diet (148 plant species; 54% leaves and buds, 32% fruit, 6% flowers, and 6% soil)42. Their high level of folivory is also evidenced by morphological adaptations for foregut fermentation32. Macaques have a versatile diet, variably composed by different species at different sites31, but the particular diets of the macaques at our study sites have not been investigated thoroughly.

Primate populations in both Wuliangshan and Yongde Daxueshan are small and threatened. 435 individual (87 groups) western black crested gibbons are estimated to remain in Wuliangshan43, and a 2011 survey identified only 11 individuals—three groups and one solitary male—in Yongde Daxueshan. 1,960 individual (43 groups) Indochinese gray langurs were estimated to remain in the Wuliang Mountains (Jingdong County) in 201338. Reliable census data are not available for the langurs in Yongde Daxueshan and the macaques in either reserve, but forest rangers and officials have seen both in recent years and believe them to be more abundant than gibbons.

### Detection Dog Training

Over the course of three months during the summer/autumn of 2010, the Kunming Police Dog Training Base of the Chinese Ministry of Security (KPDTB) trained a Belgian Malinois, named “Pinkerton”, to locate and identify scat from three species of primates (N. concolor, T. crepusculus, and M. arctoides). The KPDTB commonly trains Belgian Malinois for scent work, because of their potential for high levels of excitability, durability, adaptability, strong sense of smell, and hunting instincts. Prior to Pinkerton’s training, we collected scat for our species of interest from the Kunming Zoo with gloved hands to avoid scent contamination, deposited it into 50 ml plastic tubes, and desiccated it with an electric fan for preservation. While we were able to collect fresh scats from M. arctoides (four individuals) and T. crepusculus (two individuals), N. concolor and other macaques were not available. We collected feces from the closely related white-cheeked gibbon, N. leucogenys (4 individuals) instead. We then covered the purchase price of Pinkerton from the KPDTB and supplied the head trainer with fecal samples for the three primate species of interest. Pinkerton’s training began at the age of eight months; at this age, dogs can quickly learn and remember new skills. Pinkerton was trained to signal that scat was deposited by one of the three target species by placing his paws on either side of the scat and lying down beside it; he was not trained to differentiate among scats from the three target species. (See Supplemental Methods for detailed training protocols).

Once Pinkerton could reliably locate and identify primate scat, the trainer spent one month in Kunming (October 2010) habituating the field handler (J.D.O) to the dog and teaching him the search commands, visual cues, and other practices necessary to work with Pinkerton in the forest. During fieldwork, the handler practiced these techniques with Pinkerton at least twice per week during downtime in the forest or while waiting for transportation to field sites.

### Field Surveys

During the dry seasons (October–May) of 2010/11 and 2011/12, we visited nine sites in Wuliangshan known or presumed to be inhabited by our species of interest (Bangwai, Huangcaoba, Huangcaoling, Langanqing, Raomalu, Shaniucun, Xiaobahe, Xincun, and Yenshancun) for one to five weeks each. We camped in the forest within gibbon home ranges when the risk of forest fire was low; otherwise, we slept in the homes of our local assistants. At each site in the Wuliang Mountains, we hired one to four local field assistants, depending on their availability and knowledge of the forest. Although it was not always possible to work with a consistent number of field assistants, we split opportunistically into two search groups—a human-dog team and a human-only team—at Bangwai, Huangcaoba, Raomalu, Shaniucun, and Xincun. In such cases, two field assistants would collect fecal samples independently of the two-person human-dog team.

Fieldwork in Yongde Daxueshan occurred during a gibbon population census led by The Kunming Institute of Zoology (KIZ) and Flora and Fauna International (December 2010–January 2011). During this four-week survey, the survey team stopped at six unnamed listening posts in Yongde Daxueshan for two to four days each. The human-dog team was only able to search when the survey team stopped.

### Search Methods

We prioritized the search for gibbons, but we trained our dog to detect primate scat broadly so that we could collect samples from sympatric species with different diets and behaviors in a single effort. In mountainous habitats, gibbons are best located by climbing to high-elevation listening posts before dawn and waiting to identify the geographic origin of their morning songs36,44. When multiple field assistants were available, we occupied separate listening posts in an attempt to flank a gibbon group’s calling site. When we heard gibbons, we walked promptly to the presumed area of the calling tree (usually 1–2 km away). When gibbons did not call, we walked to locations where our assistants had seen primates in the past (usually small river bottoms, cliffs, and feeding trees). When contacting primates, we never approached them with the dog to reduce the risk of disease transmission. Because hunting dogs and cattle occasionally enter these forests, any potential disease risk from Pinkerton would not have constituted a novel threat.

Upon reaching a search area, the handler commanded Pinkerton to take the lead and search for primate feces along a path of the dog’s choosing. The handler followed five meters behind the dog with assistants five meters behind him. Because detection dogs are motivated by human affection, walking at some distance behind Pinkerton minimized distractions. When Pinkerton identified target feces, the handler rewarded him with a tennis ball and collected the sample. When a human team member identified a sample in disagreement with the dog it was collected and coded as human-identified. In cases where multiple field assistants were available we opportunistically separated into a dog-human team and a human-only team to broaden our search for primate feces.

### Fecal Sample Collection

Fecal samples for genetic analysis were collected following the two-step ethanol and silica method45. All fecal samples were collected with gloved hands and sticks from the forest floor to minimize human contamination. In order to avoid cross-contamination of fecal samples from different individuals, fragmentary samples were placed in separate collection tubes whenever it was not obvious that the scats originated from the same bolus. Samples were kept at room temperature for one week to two months until they could be frozen at −20 °C in Jingdong City, Yunnan, then frozen at −80 °C at KIZ.

### Laboratory Analysis

Genetic work was conducted at KIZ. DNA was extracted using the 2CTAB/PCI method46 under a UV sterilized fume hood with unidirectional airflow. We wore facemasks and lab coats at all times during extraction. To determine species or genus of origin, we attempted to PCR amplify taxonomically informative genetic markers for all collected fecal samples. Although DNA barcoding is typically done with the CO1 locus47, we sought to facilitate future phylogenetic and conservation genetic research on these endangered primates by first amplifying mitochondrial d-loop sequences where possible. Our approach of amplifying a series of taxonomically informative markers until the genus (or species) of origin could be identified also minimized lab costs. Given the presumption that most fecal samples would have been from Nomascus concolor, we began by using primers that we designed to amplify ~600 bp of the d-loop in Nomascus (Table S1). For samples that did not amplify or provided unintelligible sequencing reads, we then used a d-loop primer from which we had successfully amplified both Trachypithecus and Macaca mtDNA (Table S1). For the remaining samples, it was possible that they were 1) too degraded to amplify a ~600 bp mtDNA region, and thus of limited future use; 2) from another organism; or 3) a combination of both factors. To resolve the identity of these remaining samples, we attempted to amplify a vertebrate-specific 220 bp CO1 mini-barcode locus48 that would yield taxonomically informative information, albeit of limited future use.

Each PCR was performed with negative controls in a 25 μl reaction containing 2.5 μl of 10x buffer, 1 μl dNTP, 1 μl BSA, 1 μl of each primer, 0.5 μl of DNA, 0.125 μl of TaKaRa Hot Start taq polymerase, and 17.875 μl of millipure water. Thermocycling conditions were as follows: an initial denaturation of 3 min at 94 °C followed by 10 cycles of 45 s at 94 °C, 1 min at 55 °C with a reduction of 0.5 °C each cycle, and 1 min at 72 °C; this was followed by 25 cycles of 45 s at 94 °C, 1 min at 50 °C, and 1 min at 72 °C and a final annealing stage of 10 min at 72 °C. For samples that did not amplify, a 25 μl secondary PCR was amplified using a 1:10 dilution of the initial PCR product in place of template DNA. PCR purification and Sanger sequencing were conducted at The Beijing Genomics Institute on an ABI 3730xl cycle sequencer.

### Taxonomic Assignment

Geneious R749 was used to trim reads of low quality bases and assemble contigs from the bidirectional reads from each sequenced PCR amplification. Each contig was aligned to the NCBI database using nBLAST with default parameters. Tabular nBLAST output was downloaded to Geneious R7, and the genus of origin for each sample was determined identifying the nBLAST alignment first with the lowest E value, and second by the Geneious “Grade,” which is the weighted average of the E score, pairwise identity, and coverage, if multiple alignments had equivalent E values. Potential human- or cross-contamination of the samples was screened by the presence of multiple traces in sequencing chromatograms. All unique sequences have been submitted to GenBank with accession codes available in Table S2.

### Assessment of Detection Dog Success Rate

All collected samples were coded first by whether they yielded DNA that amplified with our primer set, and secondly whether we could taxonomically identify them to one of our target animals. For both the dog-human team and the human-only team, detection accuracy was defined as the proportion of samples from which target animal DNA could be identified relative to the total number of samples that yielded amplifiable DNA. Because of the hierarchical structure of our sampling methods (two mountains, each with multiple sites), we applied a linear mixed effects model to discern if the dog-team had a significantly greater accuracy and collection rates than the human-only teams. Sampling sites were nested within mountains as random effects with sample collector (dog or human) as a fixed effect. Models were generated in R with the package nlme. P-values were adjusted for significance using Benjamin-Hochberg false discovery rates.