constituted the geo-referenced scats and giant armadillo burrow locations. ..... species that strongly avoid human disturbance (giant armadillo) or are unlikely to cross ...... A comparison of survey methods for detecting bobcats. ...... (vi) 900μl of each sample was transferred to filter plate and centrifuged for 5-10 min, filter plate.
© Copyright 2010 Carly Vynne
Landscape use by wide-ranging mammals of the Brazilian Cerrado
Carly Vynne
A dissertation submitted in partial fulfillment of the requirements for the degree of
Doctor of Philosophy
University of Washington 2010
Program Authorized to Offer Degree: Department of Biology
In presenting this dissertation in partial fulfillment of the requirements for the doctoral degree at the University of Washington, I agree that the Library shall make its copies freely available for inspection. I further agree that extensive copying of the dissertation is allowable only for scholarly purposes, consistent with “ fair use” as prescribed in the U.S. Copyright Law. Requests for copying or reproduction of this dissertation may be referred to Proquest Information and Learning, 300 North Zeeb Road, Ann Arbor, MI 481061346, 1-800-521-0600, to whom the author has granted “ the right to reproduce and sell (a) copies of the manuscript in microform and/or (b) printed copies of the manuscript made from microform.”
Signature ____________________________ Date ______________________________
University of Washington Abstract Landscape use by wide-ranging mammals of the Brazilian Cerrado Carly Vynne Chairpersons of the Supervisory Committee: Professor Samuel K. Wasser and Professor Martha J. Groom Department of Biology Conserving animals beyond parks is critical since even the largest reserves may be too small to maintain viable populations for many wide-ranging species. Identification of sites that will promote population persistence is a high priority, in particular, for protected areas that reside in regions of otherwise extensive habitat loss. This is the case for Emas National Park, a small but important protected area located in the Brazilian Cerrado. In order to determine the relative importance of resources found within the Park, as well as to identify key sites outside the reserve, I used scat detection dogs to survey for five large mammals of conservation concern: maned wolf (Chrysocyon brachyurus), puma (Puma concolor), jaguar (Panthera onca), giant anteater (Myrmecophaga tridactyla), and giant armadillo (Priodontes maximus). I quantified the effectiveness of dog teams to determine species presence and evaluated how each of the species were distributed within and around Emas National Park. I assessed how measurable sample quality factors influence DNA amplification success as well as measurable hormone quantities and found that amount of odor and moisture (indicating freshness) predicted mtDNA amplification success, as well as mean hormone levels. To determine how each of the species were using resources, I fit resource selection probability models, which show how each species uses sites relative to those available. Finally, to evaluate how ranging behavior may influence physiological health in maned wolves, which are nearly endemic to the Cerrado, I measured fecal glucocorticoids, indicative of stress, thyroid hormone, indicative of nutritional status, and androgens, indicative of reproductive health. Glucocorticoid concentrations increased with distance from natural habitat patches and during times of peak harvest activity. Thyroid hormone levels were higher, indicating good nutritional status, in areas with more cropland, thus i
supporting my hypothesis that maned wolves select agricultural areas due to availability of rodents. Progestin levels in females were higher inside than outside the Park, suggesting that females have higher reproductive success in the Park compared to those residing outside the Park. These analyses illustrate the landscape features that must be maintained if we are to promote persistence of diverse, wide-ranging species.
ii
Table of Contents page List of Figures ..................................................................................................................... v List of Tables .................................................................................................................... vii Acknowledgements………………………………………………… …………………….1 Chapter 1: Effectiveness of scat detection dogs in determining species presence ............ 5 Introduction..................................................................................................................... 6 Methods........................................................................................................................... 8 Results........................................................................................................................... 13 Discussion ..................................................................................................................... 18 Chapter 2: Habitat preferences of wide-ranging mammals in the Brazilian Cerrado..... 24 Introduction................................................................................................................... 24 Methods......................................................................................................................... 28 Results........................................................................................................................... 35 Discussion ..................................................................................................................... 45 Chapter 3: Factors influencing degradation of DNA and hormones in maned wolf scat…56 Introduction................................................................................................................... 56 Study Area .................................................................................................................... 59 Methods......................................................................................................................... 59 Results........................................................................................................................... 65 Discussion ..................................................................................................................... 74 Management Implications……………………………….……………………...……..77 Chapter 4: Physiological implications of landscape matrix use by free-ranging maned wolves in Brazil………..………………………………………………………...…….78 Introduction................................................................................................................... 79 Methods......................................................................................................................... 84 Results........................................................................................................................... 88 Discussion ..................................................................................................................... 98 List of References ........................................................................................................... 104 Appendix A: Survey effort ............................................................................................. 122 Appendix B: DNA results............................................................................................... 123 Appendix C: Number of confirmed species location by type…………………………..124 iii
Appendix D: Resource selection model fit by species………………………………….125 Appendix E: Protocol for DMSO preservation and extraction…………………………128 Appendix F: Mucosal swab DNA extraction protocol…………………………………129 Appendix G: Species controls of sympatric carnivores in the region of Emas National Park, Brazil and their band lengths as analyzed using HSF21 and LTPROB13 primers and fragment length polymorphisms………………………………..…130 Appendix H: Mitochondrial DNA amplification success and species assignments from scat samples collected from putative maned wolf scat................................……131 Appendix I: Classification tree outputs for how sample condition influences DNA amplification…....................................................................................................132 Appendix J: Classification tree outputs showing all variables that split for nuclear DNA amplification of maned wolf scats……………….……………………………..133 Appendix K: Significance of parameters included in model for glucocorticoids in maned wolf scat………………………………………………………………………...134 Appendix L: Significance of parameters included in model for glucocorticoids in maned wolf scat collected in the landscape outside of Emas National Park…………...135 Appendix M: Significance of parameters included in model for thyroid hormone in maned wolf scat……………………………………………………………..….136 Appendix N: Significance of parameters included in model for testosterone in maned wolf scat……………………………………………………...…………………137
iv
List of Figures
Figure number
page
Figure 1. Location of Emas National Park within the Cerrado biome of Brazil……..…..8 Figure 2. Survey quadrats and relation to Emas National Park and land use classes, as well as 5x5 km2 quadrats depicting presence of target species………………….17 Figure 3. Photos of four large mammals included in the study: giant armadillo, giant anteater, maned wolf, and tapir…………………………………………………..31 Figure 4. Giant armadillos select positively for an interaction between distance from roads and proportion of natural habitat within their home range………………...38 Figure 5. Distribution of available habitat with respect to each species’ selection probability for samples found inside versus outside of the Emas National Park, Brazil………………………………………………………….………………….39 Figure 6. Giant anteater selection preference for grasslands and being near forest…….40 Figure 7. Anteaters avoid roads except when in grassland……………………………...41 Figure 8. Maned wolves show a strong preference for open habitat types, increasingly avoiding areas as the proportion of closed-canopy cover within 1km2 of a sample increases………………………………………………………………………….42 Figure 9. Jaguar prefer areas with a greater proportion of closed-canopy habitat and strongly prefer areas within the Park…………………………………………….43 Figure 10. Distribution of available sites in and around Emas National Park with respect to distance to road, distance to forest, and probability of selection by puma……44 Figure 11. A photo of a sugar cane plantation, located just northeast of Emas National Park, Brazil………………………………………………………..……………..48 Figure 12. Proportion of putative maned wolf scat samples that amplified for mtDNA and nDNA based on amount of odor, moisture, and fruit in the sample………...69 Figure 13. Classification trees of the most parsimonious models for mtDNA and nDNA amplification from maned wolf scats collected in the Brazilian Cerrado………..71 v
Figure 14. Median hormone levels of maned wolf scats with respect to varying levels of odor and moisture at time of collection………………………………………….73 Figure 15. Global distribution of the maned wolf and extent of the Cerrado biodiversity hotspot……………………………………………………………………………80 Figure 16. The interaction between distance from patch and crop height predicts levels of glucocorticoids, indicative of stress, in maned wolf scats……………………….92 Figure 17. For samples found inside the Park, the amount of agriculture in the area predicts higher glucocorticoid levels………………………………………...…..93 Figure 18. Maned wolf samples consisting of more protein content have higher levels of thyroid hormone than samples comprised entirely of fruit………………………94 Figure 20. Timing of activities related to maned wolf reproduction, crop harvest, and rain……………………………………………………………………………….96 Figure 21. Progestin levels from female maned wolf samples collected inside and outside of Emas National Park, Brazil……………………………………………….…..97
vi
List of Tables Table number
page
Table 1.1. Number of samples found on roads, wildlife trails, or off of roads and trails, by species………………………………………………………………………...14 Table 1.2. Joint, conditional, and overall detection probabilities of scat dog teams for each of the target species and survey type……………………………………….15 Table 1.3. Most parsimonious combinations of visits per 5x5 km2 quadrat and 3x3 km2 grid and number of sites required in order to reach 0.95, 0.90, and 0.80 certainty of detection for each of the study species………………………………………..18 Table 2.1. Definitions and labels for covariates tested in resource selection models.......34 Table 2.2 Parameter estimates and standard errors in the resource selection models for species surveyed in the Cerrado of Brazil………………………………………..37 Table 3.1. Description of categorical variables used to understand factors influencing quality of maned wolf scat samples collected in the Cerrado of Brazil………….65 Table 3.2. Parameter estimates from statistical model showing factors contributing to probability of DNA amplification from maned wolf scats collected in the Cerrado of Brazil………………………………………………………………………….68 Table 4.1. Influence of sample condition at time of collection on mean hormone levels in maned wolf scat………………………………………………………………….88 Table 4.2. Amount of natural habitat available across the study area as a whole versus in sites where maned wolf scats were deposited……………………………………89 Table 4.3. Covariates included in generalized linear models explaining variation of hormone levels of maned wolf scats and their significance within whole model results……………………………………………………………………...….90-91 Table 4.4. Effect tests of model for progestin levels in scat sample from female maned wolves……………………………………………………………………………94
vii
Acknowledgements
Many kind thanks to Dr. Gustavo Fonseca and Dr. Thomas Lacher for your essential support in launching this project and for investing in my personal development as a conservation biologist. I am very grateful for both the financial as well as inspirational support and look up to you and the work you have done; indeed it is what drove me to get the doctoral training so that I could similarly contribute to international conservation as a productive scientist.
Many organizations and individuals in Brazil helped make this project possible. The Brazil Program of Conservation International, particularly Dr. Ricardo Machado and Dr. Mario Ramos Neto, were instrumental in arranging permits, logistics, providing financial and in-kind support, exporting samples and dealing with just about every imaginable challenge of conducting fieldwork in a remote site and getting samples to another country. The Jaguar Conservation Fund provided an immeasurable amount of in-kind support through their local knowledge, relationships with local landowners, lodging, and general guidance. Thanks in particular to Dr. Leandro Silveira, Dr. Anah Jácomo, Mariana Furtado, and Natália Tôrres. Dr. Jader Marinho-Filho of the University of Brasília kindly acted as my in-country academic advisor and also helped negotiate the permitting process as well as store samples while we awaited export permits.
The staff persons at Emas National Park were tremendously warm and welcoming. I thank you for your introduction to this wonderful place and the care you are providing for the creatures of this special region. The Brazilian Institute of Renewable Natural Resources and CNPq provided project licensing and innumerable landowners kindly allowed access to their private farms so that we could study wildlife use of these areas. Your kind generosity is much appreciated.
1
My dissertation committee provided support and technical advice as needed and thanks to each of the following individuals for your insights, generosity of time, and selflessness in helping to keep this project on track: co-chair Dr. Samuel Wasser, co-chair Dr. Martha Groom, Dr. Ray Hilborn, Dr. Jim Kenagy, Dr. John Marzluff, and Dr. John Skalski. Both Sam and Martha kindly read drafts of each of the chapters and helped improve the work immensely.
The conservation canines are entirely responsible for the richness of data we were able to collect in this study and their tireless efforts are greatly appreciated, not only for the excellent data collection that they enabled but also for the enthusiasm you inspired to go to work every day. I couldn’t imagine better field companions or getting up so early without their genuine and open love of going to the field. Many thanks to Mason, Ally, Marvin, CJ, Gator, and Boon. Barbara Davenport, Bud Marks, Heath Smith, and Chris Zieminski all provided valuable training guidance in preparing our dog-handler teams for fieldwork. Samples for training the dogs were provided by The Smithsonian Conservation and Research Center, Woodland Park Zoo, Parque Zoológico de Goiânia, and Karen DeMatteo.
Raphael Almeida, Matt Baker, Julie Betsch, Claudia Ferro, Cyntia Kashikavura, Mike Price, Samantha Herzog, Jessica Harmon, Laura Anglin, Ashley Ragsdale, Heath Smith, and Chris Zieminski worked as dog handlers or field assistants for extended periods with me in Brazil. I was so fortunate to have teams that were patient, understanding, optimistic, willing to change tires and push trucks, and truly made field camp life a great joy. These are times that would be impossible to recreate but will not soon be forgotten. Thanks to those of you that helped so much with emotional support during challenging times of the project as well; you will always hold a special place in my heart.
Laboratory training and guidance was provided by Katherine Ayres, Rebecca Booth, Celia Mailand, Lynn Erkmann, Robert Livingston, Danielle Mitchel, Siri Nelson, Lisa Hayward, and Kathleen Gobush. Zyanya Breuer, Bishnu Dhaurli, Tyler Mann, Kim 2
Pham, Anny Ngo, Jessica Harmon, Laura Anglin, and other student volunteers and assistants all enabled the processing of samples in the laboratory for DNA and hormone analyses.
My parents Barb and John Vynne provided immeasurable support including a rainy season visit to Brazil to assist in the field and, this last year, Wednesdays with Viera during which much of the writing of this dissertation occurred. Thanks to you both for your unconditional love and support. Matt Baker not only selflessly sacrificed so many aspects of his life to allow me to conduct this project but also became an important project collaborator as I asked his input on nearly every aspect of the work as the results came together. Thanks for being the patient, understanding, and intelligent person and wonderful father that you are. Viera lived and breathed this work (including a trip to Argentina at 6 weeks old) for the first year of her life and her presence has only added joy and enthusiasm to this work. I can’t wait to take you to Emas to see the animals one day soon!
The following individuals were collaborators on each of the dissertation chapters and will be co-authors in forthcoming publications of the manuscripts: Dr. John Skalski, Dr. Ricardo Machado, Dr. Martha Groom, Dr. Anah Jácomo, Dr. Jader Marinho-Filho, Dr. Mario Ramos Neto, Dr. Cristina Pomilla, Dr. Leandro Silveira, Heath Smith, Dr. Samuel Wasser (Chapter 1); Dr. Jonah Keim, Dr. Ricardo Machado, Dr. Jader Marinho-Filho, Dr. Leandro Silveira, Dr. Martha Groom, Dr. Samuel Wasser (Chapter 3), Matt Baker, Zyanya Breuer, Dr. Samuel Wasser (Chapter 3), Dr. Samuel Wasser (Chapter 4). Dr. John (Mike) Kinsella conducted all parasite analyses that are not a part of this dissertation but are an important part of contributing to understanding of maned wolf ecology and this work is being published separately.
Funding was principally provided by the TEAM Network of Conservation International, funded by the Gordon and Betty Moore Foundation, the Morris Animal Foundation, and Conservation International of Brazil. A National Science Foundation Graduate 3
Fellowship provided me the flexibility to spend extensive periods of time in both the field and lab and the richness of this project would not have been possible without that support. A National Security Education Program Boren Graduate Fellowship provided funding for language training that enabled me to attain proficiency in the Portuguese language and this greatly enriched my experience in Brazil. A Moore Foundation grant to S. Wasser provided support for our pilot study. Assistance for field crew was provided via an NIH grant (T37-MD001378-04) to Christian Brothers University. The Kids Art for Earth project provided funds for dog food while in Brazil. The Department of Biology at the University of Washington provided funding for me to attend several conferences and share the work with others during my tenure with the Department as a PhD Candidate. A Department of Biology Kathyrn Hahn Writing Fellowship was instrumental in enabling me to finalize the dissertation as well as get some of the chapters submitted and revised for journals.
4
Chapter 1: Effectiveness of scat detection dogs in determining species presence in a tropical savanna landscape Abstract Most protected areas are too small to sustain populations of wide-ranging mammals and thus identification and conservation of critical habitat outside parks is often a high priority, particularly for protected areas in regions of extensive land conversion. This is the case for Emas National Park, a small but important protected area in the Brazilian Cerrado where conversion of native vegetation to agriculture in the last 40 years has transformed the surrounding landscape, yet the region supports an intact fauna. To assess the contribution of private lands to persistence of wide-ranging mammals, I employed scat detection dogs to survey for five species of conservation concern: maned wolf (Chrysocyon brachyurus), puma (Puma concolor), jaguar (Panthera onca), giant anteater (Myrmecophaga tridactyla), and giant armadillo (Priodontes maximus). Scat detection probabilities varied by species and survey quadrat size, but were consistent across team, season, and year. The probability of occurrence in a randomly selected site within the study area ranged from 0.14 for jaguar, which occur primarily in forested areas of the Park, to 0.91 for maned wolf, the most widely distributed species. Giant armadillo were associated with the open grasslands in the Park, yet when in the agricultural matrix tended to occur in quadrats containing riparian woodlands. At least one target species was detected in every survey quadrat, and giant armadillo, jaguar, and maned wolf were more likely to be present in quadrats located inside than outside the Park. The effort required for detection was highest for the two felids. These results provide some of the first confirmed occurrences for these species outside the Park, while also providing a framework that will enable transfer of the detection dog method to other species and systems of conservation importance.
5
Introduction Conserving animals beyond protected areas is critical because even the largest reserves may be too small to maintain viable populations for many wide-ranging species (Boyd et al. 2008). Therefore, conservation of large mammals will often depend on protecting and restoring linkages through networks of smaller reserves and private lands (Woodroffe 2001). This, in turn, requires effective methods to evaluate how wide-ranging species respond to changing land use practices and to identify conservation priorities within complex landscapes. In particular, survey methods are needed that yield high detection rates independent of species density, enable simultaneous sampling of multiple species, and are efficient at the large scale that is necessary to address questions of how to conserve wide-ranging species.
Scat detection dogs are a promising tool for monitoring animal populations because they can detect scats over large distances (Smith et al. 2003; Wasser et al. 2004; Long et al. 2007a). Trained to locate feces of target species, detection dogs are more effective at detecting carnivores than traditional methods such as hair snares, scent stations, and camera traps (Wasser et al. 2004; Harrison 2006; Long et al. 2007b) and they are particularly promising for studies of wide-ranging, elusive, or rare species (Long et al. 2007a; Vynne et al. 2009). To design effective detection dog surveys, however, it is important to quantify the ability of the dogs to detect target species and to evaluate how this ability is influenced by survey design and other variables such as team and season.
In this chapter, I present results from a 4-year study that employed scat detection dogs to evaluate landscape use by five large mammals of conservation concern, maned wolf (Chrysocyon brachyurus), jaguar (Panthera onca), puma (Puma concolor), giant anteater (Myrmecophaga tridactyla), and giant armadillo (Priodontes maximus), in and around Emas National Park, in the Brazilian Cerrado. While more than 55% of the Cerrado has been cleared for agriculture and livestock grazing in the last 50 years (Klink & Machado 2005), federal law requires landowners to leave between 20 and 30% of the original land cover intact (Brannstrom et al. 2008). Thus, the landscape mosaic outside of Emas 6
National Park is dominated by large-scale agriculture and cattle pasture, interspersed with forested riparian corridors and woodland fragments. The region still maintains its full complement of large (>20 kg) mammals (Morrison et. al. 2007), yet the Park, by itself, is unlikely to support these populations over the long term (e.g. Silveira et al. 2009a). Because the contribution of the surrounding landscape to large mammal persistence was unknown, we sought to determine the distribution of the target species within and around the Park, as well as evaluate the effectiveness of scat detection dogs in large-scale studies of wide-ranging mammals of varying degrees of rarity.
Knowing the level of effort required to detect presence of wide-ranging or rare species with a known level of certainty will enable effective survey design. Such estimation requires consideration of the error sources that contribute to the overall variance in an observation (Skalski & Robson 1992). Estimates of the error variance can then be used to determine the level of replication needed to detect species presence. I address these issues by developing and applying models that determine the detection probabilities by dog teams for each of our target species. More specifically, I develop likelihood profiles for detecting a species for a given sampling quadrat size and level of effort. These analyses allow me to address five objectives. First, I evaluate if the dogs are able to locate target species and distinguish them from non-target species. Second, I present a model to determine both conditional and joint detection probabilities by the dog teams, as well as the proportion of grids in our study area occupied by each species. Third, I consider the consistency of detection rates across study variables and evaluate how survey design and target species influence detection probabilities. Next, I apply the model to see how probability of occurrence varies with location relative to a protected area. Finally, I determine the level of effort required to assess species presence-absence status.
7
Methods Study area The 1320 km2 Emas National Park and surrounding farms (for a total study area of 4000 km2) are located in the tri-state region of Goiás, Mato Grosso, and Mato Grosso do Sul States (18°S, 52°W), Brazil (Fig. 1). Emas National Park is considered one of the most important protected areas of the Cerrado biome, which comprises 21% of Brazil and is the world’s largest, richest, and most threatened tropical savanna (Silva & Bates 2002). The Park protects large tracts of grassland plains (97%), small patches of shrublands (1%), and marshes and riparian forest (2%). The surrounding area is comprised of agriculture (44%, predominately soy, cotton, corn, and sugar cane), cattle pasture (25%), and remnant vegetation (31%, Fig. 2a).
Fig. 1. Location of Emas National Park within the Cerrado biome of Brazil.
8
Dog selection and training Dogs were trained and worked in the field according to methods described in Wasser et al. (2004). Briefly, dogs were selected from animal shelter facilities in the Pacific Northwest, USA, on the basis of their obsessive drive to play fetch with a tennis ball. Professional training of dogs was done at Packleader Dog Training (2004, Gig Harbor, WA) and at University of Washington Conservation Canine facilities (2006-2007, Seattle and Eatonville, WA). Dogs were taught to associate odor with a play reward. Once this association was engrained, dogs were taught to search for, locate, and indicate (by sitting) scats of target species.
Dogs were initially trained with scats from 6-8 captive and wild individuals of maned wolves, jaguar, and puma. Training dogs to detect samples from multiple known individuals of a given target species allowed them to generalize detection to any sample from that species, regardless of its sex or reproductive status. In 2006, I added giant armadillo and giant anteater to the dogs’ repertoire while in Brazil with scats from visually-identified wild individuals (n=4 each).
Field surveys One to 3 detection teams (comprised of a dog, dog-handler, and field assistant) conducted surveys at 70 sites between August of 2004 and April of 2008. Surveys occurred during three dry season sessions of 10 weeks each (May-August 2004, 2006, and 2007) and one rainy season (January-April 2008). Sampling sites comprised 5x5 km2 search areas (5x5), which were randomly spaced across the study area (n=57), and 3x3 km2 grids (3x3, n=13) that were contiguous (Fig. 2a). The spatially separated 5x5 sampling scheme was designed to maximize search area, with a different 2.5 x 2.5 km2 quadrant of the plot sampled during each visit; the 3x3 grids were sampled completely each visit. Safety concerns prevented us from sampling additional quadrats placed in the agricultural expanse to the west of the Park.
9
Teams typically surveyed each quadrat 3 to 5 times per field season. Teams walked freely and dogs searched off-leash within pre-designated areas. This allowed the handler to guide the dog according to wind direction as well as to follow a dog pursuing a distant scent. Each team searched a single quadrat on a given day and every team surveyed all quadrats at least once per season. Vegetation types were searched in proportion to their occurrence in a quadrat by monitoring survey routes using GPS units. When a dog located a scat, the handler rewarded the dog, recorded the GPS position, and collected the sample. Samples that were both odorless and formless were not collected. Otherwise, a portion of the sample was preserved in a vial with 25-mL of 20% dimethyl sulfoxide buffer and frozen until DNA extraction (Frantzen et al. 1998; Chapter 3).
Establishing detections Samples of putative maned wolf, puma, and jaguar origin were subjected to DNA analysis since these could be confused either between target species or with non-target species such as ocelot (Leopardus pardalis) and fox (Cerdocyon thous). Giant anteater scats were highly distinctive due to their unique shape, large size, and contents (Chame 2003). I thus did not subject giant anteater scats to genetic analysis for species confirmation. Giant armadillo scats could potentially, though not easily, be confused with other sympatric species of armadillo. As putative giant armadillo scats had high coincidence with other signs (tracks, paths, diggings, burrows, or the actual animal) and were not found in areas of other armadillo species activity or sightings, I am confident that species identification of these samples is accurate.
I processed scat samples for DNA analysis at the Center for Conservation Biology, University of Washington (Seattle, USA) using the Qiagen QiaAmp Stool and Blood/Tissue kits (Qiagen, Valencia, CA) with modified protocols (Chapter 3: Appendix E & F). The species test consisted of fragment analysis from PCR amplification of the mitochondrial control region (D-loop) (Wasser et al. 1997). A subset of the samples was also confirmed by sequencing at the Sackler Institute of Comparative Genomics with primers adapted to amplify the 16S ribosomal RNA gene (Pomilla et al. 2009). I 10
excluded samples that did not yield DNA for species identification or were from a nontarget or unidentified species.
Deriving detection probabilities and evaluating factors influencing detection The conditional probability of a species being detected (i.e., event A), given it was present (i.e., event B), p( A B ) , was derived by summing across quadrats and pooling by team and year (when not significantly different according to chi-square test) and then dividing the number of detections (1’s) by the number of opportunities to detect (1’s and 0’s). I based this approach on the Manly-Parr (1968) technique of estimating detection probabilities. I assumed species were never falsely detected at a quadrat when absent and that an animal may or may not have been detected in a quadrat when present. Once presence was confirmed in a quadrat, subsequent detections or lack of detections represented dog team successes or failures to detect the species. I calculated p( A B ) using data from all sites visited two or more times that had one or more detections and at least one subsequent visit to that site following a confirmed detection.
I used data from all sites to calculate the joint probability of a species being present and detected p( AB ) by:
n
p( AB) = pˆ =
xi i =1 mi n
where:
xi = the number of successes at the ith quadrat (i=1….n) mi = the number of visits to the ith quadrat (i=1…n) n = the number of unique quadrats
I next determined the likelihood of occurrence for each of the target species in a randomly selected quadrat in our study area by calculating the proportion of quadrats 11
p(B ) in which each of the species occurred. By definition, the joint probability p( AB ) of a species being present (B ) and detected ( A) is equal to the probability of the species being detected, given it is present p( A B ) times the probability of the species being present, p(B ) :
p( AB ) = p( A B ) • p(B )
The probability of a species being present, p(B ) , was calculated by dividing the joint probability of being present and detected, p( AB ) , by the conditional probability of an animal being detected when present, p( A B ) . All measured and predicted probabilities were averaged for the study area as a whole, and thus do not reflect vegetation-specific rates of occurrence.
Chi-squared analyses were used to test for significant variation in detection probabilities on the basis of Manly and Parr (1968) capture data. This was done by summing the number of detections (1’s) and failures to detect each species (0’s), given it was known to be present at a site. Contingency tables were used to test for homogeneity across five variables: session, team, number of target species, year, and season. For all tests, significance was tested at the 0.05 level.
Determining species distributions relative to protected area To evaluate whether probability of occurrence of our study species was affected by quadrat location relative to the Park, and hence the amount of human disturbance, we grouped sampling quadrats into one of three categories: entirely inside, entirely outside, or on the border of Emas National Park. The probability of an animal being present p(B ) was then derived as described above for each of these groups.
Calculating sampling effort requirements to determine species distributions 12
I applied the method of estimating detection probabilities in order to calculate the optimal study design for determining species presence in a region. To determine the number of quadrats and visits required to reach a given probability of detection (P), we used a binomial sampling model that incorporated four parameters: p( A B ) = the conditional probability of detection, if present, with one visit to a quadrat; n = the number of visits per quadrat; p(B ) the probability of an animal being present; and m = the number of quadrats. The model is described by the probability distribution:
[
][
P = 1 − (1 − p( A B )) • 1 − (1 − p(B )) n
m
]
where 1- p( A B ) is the probability of not detecting the species on a given visit 1- p( A B ) is the probability of not detecting the species on n visits n
[1 − (1 − p(A B)) ] is the probability of detecting a species on n visits, and n
[1 − (1 − p(B )) ] is the probability of a species occurring in m quadrats. m
This equation was set equal to the level of certainty desired (P=0.95, 0.90, and 0.80) and then solved for the required number of visits and quadrats. For the 3x3 km2 grids, minimum effort requirements are only reported for maned wolf, puma, and giant anteater since jaguar and giant armadillo were likely under and over-represented, respectively, by the placement of the 3x3 cells.
Results Between August 2004 and April 2008, detection teams surveyed 70 unique quadrats a total of 407 times. The number of target species, teams, survey quadrats, visits per site, and area searched varied by year (Appendix A). The mean daily survey distance was 7.7 km/team and the total distance walked was 3175 km. The mean density of scats found in quadrats where at least one sample was detected varied from 0.09 scats/km for jaguar, the least frequently encountered species, to 0.3 scats/km for maned wolf.
13
Raw detections We detected a total of 2683 putative target scats during our 4 sampling seasons, resulting in an overall average of 6.6 scats per team per search day. We did not collect 433 samples because they were without odor and form (i.e. very old scats) and 650 samples did not amplify DNA for one of the three target species. There were more maned wolf (n=936) and giant anteater (n=505) than jaguar (n=33), puma (n=70), or giant armadillo (n=56) detections. The proportion of DNA-confirmed scats from putative target species versus a non-target varied annually, from 60% of all confirmed samples in the 2004 pilot study to 85% in 2008 (Appendix B). Seventy-one % (1105 of 1566) of samples were found off of roads and thus would have been extremely unlikely to have been found by human observers alone (Table 1.1).
Table 1.1. Number of samples found on roads, wildlife trails (path), or off of roads and trails (neither), by species. Species Maned wolf Puma Jaguar Giant armadillo Giant anteater Total
road* 425 10 9 12 5 461
path 44 21 13 7 6 91
neither 467 36 11 31 469 1014
no data 0 3 0 6 25 34
total 936 70 33 56 505 1600
* Roads encompass roads on farmlands outside of Park (with few vehicle trips per day) and Park roads, (which are essentially wide trails) and experience on average 10 km from agriculture was in the National Park, I interpret this result to mean that maned wolves are just as likely to select habitats within the core of the Park as they are to select for agriculture. When maned wolves are near croplands, however, they prefer these areas for scat deposition. Maned wolf scat locations outside of the Park tend to have a higher selection probability than those found inside the Park (Fig. 5); this pattern is driven by the availability of croplands outside Park borders. 41
Maned wolves also prefer to be in areas dominated by open-canopy habitat types and areas were increasingly and strongly avoided as the proportion of closed-canopy habitat (forest, cerrado) within 1 km2 of a scat reached 30% or more (Fig. 8). Maned wolves also show preference for areas near natural water springs, while they avoid both forest and ranchland habitat (Table 2.2).
Fig. 8. Maned wolves show a strong preference for open habitat types, increasingly avoiding areas as the proportion of closed-canopy cover within 1km2 of a sample increases.
Habitat preference of jaguar The model for jaguar resource selection indicated a strong association with the amount of closed-canopy habitat within 1.4 km2 and adding the park covariate to the model significantly improved model fit (Table 2.2). Jaguar require a much higher percentage of closed-canopy habitat when outside the Park than when inside the Park (Fig. 9). As demonstrated by a previous study based on camera-trap and radio-telemetry locations of jaguar in ENP (Silveira 2004), jaguar were mainly found along rivers in the forested areas of the Park. Eight samples found outside of the Park were clustered on a private ranchland that borders a forested valley in the northeast of the Park. 42
0.8 0.6 0.4 0.2 0.0
Jaguar Selection Probability
Within Park Outside Park
0.0
0.2
0.4
0.6
0.8
1.0
Proportion of Closed Habitats / 1.4km^2
Fig. 9. Jaguar prefer areas with a greater proportion of closed-canopy habitat and strongly prefer areas within the Park.
Habitat preference of puma Puma locations were nearly entirely restricted to within 500m of closed-canopy vegetation-types, and use decreased with distance from closed-canopy, natural habitat (Table 2.2). Puma scats were found in sites that are both close to forest and far from roads (Fig.10). While their resource selection was similar to jaguar, in contrast they showed no selection preference for the Park (Fig. 5), and adding Park to the model did not improve fit for puma. Puma were found in all habitat types, though they were most commonly located in cerrado (31% of locations), open cerrado (24% of locations), and ranchland (16% of locations).
43
0km 1km 2km 3km 4km >4km
0.6 0.4 0.0
0.2
Puma Selection Probability
0.8
1.0
Forest Distance
0
2
4
6
8
Distance to Road (km)
Fig. 10. Distribution of available sites in and around Emas National Park with respect distance to road, distance to forest, and probability of selection by puma.
Habitat preference of tapir We were unable to find a good-fitting model for the tapir data and so we examined the covariates in isolation and examined how each of these appeared to influence tapir distribution. Tapir appeared to be driven by security and proximity to water springs. They were strongly and negatively associated with distance from closed-canopy habitat types (cerrado woodlands and forest), and prefer to be further from roads and close to springs. They are associated with agricultural areas but only when there is closed-canopy habitat available nearby. They avoid areas without closed habitats, being entirely absent from areas that are more than 30% open-canopy, and their occurrence peaks in areas that are comprised of 70-80% closed-canopy vegetation cover.
44
Discussion Use of indirect animal observations in resource selection studies I assumed that a greater prevalence of sign indicates greater activity by the species. Scat deposition by carnivores, and maned wolves in particular, may be used as territorial markings and are therefore placed within a short distance of evidence of conspecifics, such as near the edges of a territory (Dietz 1984; Wolf & Ale 2009). Thus, it is likely that not all of the samples were truly independent, potentially leading to an overestimation of the importance of some of the habitat variables quantified. The strength of our best-fit models, however, suggests that our findings are robust.
Advancing our ability to understand resource use of animals from sign such as scat is important since virtually all of the world’s large carnivores, for example, are rare, live in low densities, and occupy large home ranges (Sunquist & Sunquist 2002). While surveys based on sign have been subject to criticisms for their accuracy, sign data are generally recommended for surveying at large spatial scales, particularly for monitoring programs (Choate et al. 2006; Barea-Azcón et al. 2007). Furthermore, scat detection dogs have been shown to be superior at detecting species presence compared with other methods such as camera-traps, hair snare stations, or track-plate surveys (Harrison 2006; Long et al. 2007b) The inability to fit a good model to the tapir data may be due to high sampling error since scat detection dogs were not used for this species. While a potential for behavioral bias exists, these results are corroborated where habitat use information is available from other studies. Another study looking at habitat selection by maned wolves collected data via GPS collar on three individuals, for example, and also showed a preference for open habitats, strong avoidance of forests, and preference for sites near water (Melo et al. 2007). These results also support those of another study that employed radio-collars and camera-traps to study the ecology of jaguar and puma inside ENP (Silveira 2004).
Impacts of human disturbance on wide-ranging species 45
Croplands The agricultural landscape surrounding Emas National Park is expansive and, to the casual observer, appears as though it would be hostile to use by native species. It is thus somewhat surprising that the croplands are a preferred habitat for maned wolf, and are only moderately avoided by giant anteater. Tolerance of the agricultural landscape by these species is probably due to the network of habitat fragments conserved on private lands that provide protection, resting areas, and opportunities for foraging for individuals that use the agricultural areas; the species being adapted to use of open landscapes (grasslands); and the relatively frequent (biannual) crop cycle harvest and low-growing crops, such as soybeans, have been a suitable habitat replacement to grasslands for species that generally prefer open habitats.
The plight of the maned wolf under increasing agricultural expansion is particularly concerning since nearly its entire global distribution is undergoing transformation and the impacts of different agricultural practices on their populations is unknown (Rodden et al. 2008). Our results corroborate reports that the maned wolf is tolerant of modified landscapes. We interpret their selection for agriculture, as demonstrated by our analyses, to be related to foraging. The diet of individuals living the landscape mosaic is greatly simplified (consistently nearly entirely of Solanum lycocarpum and rodents) and, since the maned wolves rarely consume the crops themselves (Vynne, unpublished data), we suspect that there must be high rodent availability in the croplands that is attracting the maned wolves to use these fields. It is also probable that maned wolves select native grasslands at the landscape scale but use human-modified habitats at the scale of their home-range to take advantage of increased rodent availability in agricultural areas (eg. Vanak & Gompper 2010). Regardless, maned wolves are doubtlessly using the landscape as a whole to a great degree as we were nearly twice as likely to find a sample in the matrix as expected based on random sampling.
While use of agricultural fields gives us optimism for these species in the face of rapid change, we cannot uncouple use of the matrix from availability of habitat fragments that
46
provide protection, particularly since much of the matrix must be entirely inhospitable during certain times of year (such during harvest and planting periods). Additionally, it is crucial that follow-up monitoring be done for these species to ensure populations are stable and not simply in decline following the relatively recent conversion and use intensification of the region. This is particularly important given that the croplands are currently undergoing another major shift from low-growing soy plantations to sugar cane.
Whereas soy, corn, and cotton are harvested biannually, sugar cane (Saccharum spp. L.) ratoons are harvested annually from a seeded stand for a period of 5-7 years (Rudorff et al. 2010). Thus, transformation to a denser, darker canopy is once again dramatically altering this otherwise open landscape (Fig. 11). While this conversion may enable dispersal of jaguar that are adverse to moving through open habitats, we suspect that the more labor-intensive requirements of growing the sugar cane and increased human presence in the region may make the landscape mosaic more hostile as a place of residence for species that are particularly sensitive to disturbance or hunting, particularly giant armadillo, giant anteater, and tapir. Sugar cane development is also likely to have adverse implications for maned wolves, which are well-adapted for foraging in grassland habitats of the Cerrado, have adapted to forage in the low-growing soy plantations, and strongly avoid closed-canopy habitats. A follow-up research program to monitor the impact of conversion on maned wolves and other species in this area will help direct conservation efforts in this dynamic landscape. Prior to declaring any of the species safeguarded outside of nature reserves, it is imperative to consider the long-term outlook for these species and threats from cumulative impacts of disease-exposure, pesticide use, and susceptibility to impacts from vehicle collision.
47
Fig. 11. Sugar cane plantations such as this, located just northeast of Emas National Park, Brazil, are expanding into the Cerrado as demand for biofuels is increasing.
Cattle pasture Given the positive selection for agricultural lands by maned wolves, their avoidance of cattle pasture was particularly interesting. Because pastures are generally devoid of standing vegetation and thus lack cover for sustaining rodent populations, we interpret this result as avoidance due to lack of food availability. Also, a study in another region of the Cerrado showed that maned wolves that lived in or near areas dominated by pasture had higher levels of tick infestation and significantly different hematological and blood chemistry than maned wolves living either in a protected area or in plantation areas (May-Júnior et al. 2009). Maned wolves may thus be less associated with pastures due to their increased likelihood of contracting disease from interactions with livestock or domestic dogs.
48
Farmers will sometimes leave termite mounds and anthills on their ranches (but not in their agricultural fields), which likely explains why giant anteaters are less likely to avoid ranchlands than agriculture. While human-wildlife conflict due to jaguar and puma killing cattle on ranchlands is not considered a major source of cat mortality in this region (Silveira, personal observation), their occurrence on ranchlands suggests that there is likely underreported depredation and hence potential for conflict in these areas (Palmeira et al. 2007; Silveira et al. 2008). Besides the potential availability of cattle as prey, jaguar and puma (as well as anteater) may be more likely to frequent ranchlands due their decreased level of fragmentation as compared with croplands (Carvalho et al. 2009).
Roads Roads and vehicles affect wildlife in several important ways and studies of ungulates and large carnivores concur that buffer areas around roads are generally avoided (Forman et al. 2003). Whereas predators generally avoid frequented roads and trails, especially in areas where hunting or harassment is common (Whittington et al. 2005; Linkie et al. 2006), lightly traveled roads are more commonly used as wildlife routes (Forman et al. 2003) Heavily used roads and high-speed traffic, which occur along the road surrounding much of Emas National Park, can be major sources of mortality for some populations (Sunquist & Sunquist 2001) and indeed appear to be a significant cause of death for several of our study species (Vynne, personal observation). With the exception of the main road bordering the Park, however, the majority of roads within the Park or on farms in our study area see very few daily vehicle trips.
The influence of these roads on habitat selection by the species in our study was varied. While maned wolves didn’t show a particular selection for roads, they also appeared to use secondary roads frequently, as evidenced by frequent scat deposition, tracks, and sightings of them walking on roads. Giant armadillo, giant anteater, tapir, and puma strongly avoided areas near roads. The use of roads by maned wolves is not surprising, since canids have been previously shown to travel along low-use, linear features (Whittington et al. 2005). While giant armadillos strongly avoid burrowing near roads,
49
this avoidance is consistent with a tendency to avoid use of any human altered habitat. They may be more likely to burrow near roads in areas where there is very little natural habitat intact (i.e. croplands) either because roads tend to follow edges where there is some natural, less-disturbed habitat available or because roads are relatively stable compared to active agricultural fields.
The role of protected areas and Brazil’s Forest Code Law in large mammal conservation Emas National Park is likely the most critical component enabling the continued persistence of the region’s full complement of large mammals in the face of the landscape transformation outside of the Park’s boundaries. While the Park is directly selected by species that strongly avoid human disturbance (giant armadillo) or are unlikely to cross large tracts of agricultural land (jaguar), it is also likely to provide a population-level buffer against threats, even for species that readily use the landscape matrix. Maned wolves, for example, are quite susceptible to disease from domestic carnivores that are prevalent outside of the reserve (Deem et al. 2005). Also, the matrix is not stable, and land use intensification and clearing of native vegetation in the Cerrado continues: 18,980 km2 of new deforestation occurred between 2003 and 2007 (Ferreira et al. 2007). Expanding the protected area network to unoccupied lands, particularly in the north of the biome where there is still quite amount of intact vegetation (Diniz-Filho et al. 2009), will help ensure the long-term persistence of species dependent on reserves.
Under current Brazilian Federal Law (Law 2771 of the Forest Code), landowners in the Cerrado must set aside a minimum of 20% of their land as protected, in addition to vegetation at the sources of rivers, which is automatically required to be set aside as permanent preserves (Brannstrom 2008). Depending on topographic and drainage conditions, these two mandates require that >35% of most of the Cerrado be protected outside of reserves. In comparison, the total amount of public protected areas established in the Cerrado covers less than 3% of the biome (Klink & Machado 2005). These results show the importance of these habitat fragments for large mammal conservation in the Cerrado. Giant armadillo will use cerrado remnants when habitat patches are large
50
enough such >40% of the area within their home range is natural habitat, and their strong avoidance of non-natural habitats suggests that without these protected fragments on private lands armadillos will cease to persist outside of the National Park. Also, the giant anteater used non-preferred habitats only when these were near riparian forests that run throughout the landscape matrix.
Behavioral observations have previously shown that anteaters tend to do active feeding in open areas (where food resources are concentrated) and select forested sites for resting and temperature buffering (Paiva et al. 2006; Mourão & Medri 2007). This is likely why we find that although anteater are associated with open habitats (where food resources are concentrated), they also prefer to be near habitat edges. Also, giant anteater use rivers for drinking or even bathing (Emmons et al. 2004), and the forested river corridors in the region, in particular, appear to provide important shelter for them and for other species as well (Redford and Fonseca 1986).
While these methods were unable to capture jaguar using corridors, riparian forests are likely important for enabling rare dispersal events and indeed jaguar appear to increasingly tolerate being out of the Park as the amount of forest fragments increases. While puma were quite tolerant of their use of the landscape mosaic, this adaptability appears closely tied with availability of stalking cover. I saw evidence of several puma kills made in an open field then dragged into a nearby patch of woodland. Even maned wolves, which typically avoid areas dominated by dense canopy, are known to rest in dense vegetation during the day (Coelho et al. 2007) and availability of cerrado vegetation patches across the landscape is thus likely enabling maned wolves to use other heavily modified areas. Finally, tapir in the area also make use of croplands for foraging (Furtado et al. 2010; this study), but our data show they will not use these areas unless they are near to closed-canopy, natural vegetation-types such as those that are conserved on private lands throughout the region. The remnant vegetation that the Forest Law requires be preserved on private lands appear crucial to enabling the persistence of wide-
51
ranging mammals outside Parks. Thus, increased enforcement of existing laws would likely go a long way toward enabling large mammal conservation in the Cerrado.
In addition to enforcing existing laws, it may be important to consider underrepresented vegetation-types in future conservation planning or mitigation efforts. Open grasslands were once the dominant habitat type in the region but are now underrepresented in the system of private reserves because they are the most desirable landscapes to farm. Currently, less than 10% of the remaining habitat fragments in our study area outside of the Park include natural grasslands. Since giant armadillo, giant anteater, and maned wolf are highly associated with open, grassland vegetation-types, and indeed some show clear avoidance of closed-canopy vegetation-types, there is a strong need to conserve remaining grasslands in the protected area network. Conservation of grasslands will become all the more imperative as current schemes to incentivize biofuels production in the region are changing the agricultural landscape to sugar cane (see above discussion on agriculture).
In light of this change, we urge that the Forest Code Law be extended to ensure that the private preserves be representative of the original landscape of the farm holding, thus ensuring that remaining grasslands on private farms be conserved or restored. Furthermore, using government or private schemes for purchasing or paying for easements of any privately held, remaining grasslands will help conserve the grasslandadapted species of the Cerrado. Finally, there is evidence that voluntary, individual landholder responses to conservation incentives may result in viable reserve networks (Chomitz et al. 2006). The Emas Park landscape, where private lands are already acting as a network enabling persistence of large mammals in the region, would be a valuable and likely very successful place to further incentivize and realize conservation outcomes on private lands.
Extending resource selection to understand corridor use and functional connectivity of the Emas landscape
52
The ability of individual animals to move across complex landscapes is critical for maintaining regional populations (Fahrig 2003; Cushman et al. 2006). Given the mosaic of habitat types that naturally occurs in the woodland savannas of the Cerrado, it is likely that the open-habitat adapted, wide-ranging species experience their surroundings as gradients of differential quality rather than as categorical mosaics (Cushman et al. 2009). When connectivity is considered from this perspective, the focus on movement among discrete habitat patches via narrow linear corridors between agricultural fields is subsumed in a more general question regarding the ability of organisms to traverse a landscape of various resistance levels (Cushman 2006). While scientists and managers have applied a variety of methods to identify and design conservation corridors, these methods typically ignore processes of habitat selection by animals (Chetkiewicz et al. 2006; Beier et al. 2008). Resource selection functions developed for focal species in a spatially-explicit landscape make the approach to corridor design more rigorous, defensible, and transparent (Chetkiewicz et al. 2006). The apparent suitability of matrix habitat for occupancy by maned wolves and puma implies that any subdivided population with continuing interchange between populations will also receive a substantial demographic contribution from matrix habitat (e.g. Carroll 2004). Conversely, populations that become isolated are subject to extirpation if they fail to receive immigration of individuals following a major disturbance, such as the major wildfires that caused substantial giant anteater die-offs (Silveira et al. 1999). Isolation and subsequent increased pressure for mate access has also been credited with infanticide by jaguars in the Park (Soares et al. 2006).
Connectivity planning in this region should focus on ensuring functional connectivity in a broader landscape matrix rather than just on linear corridors (though these also would be important to ensure occupancy, not just dispersal, of jaguar in the region). Also, while fragmentation index analyses suggest that pastures are more likely to benefit conservation than agricultural lands in this region (Carvalho et al. 2009), the results of our study looking at actual use by species suggests that maintaining a network of habitat within the agricultural complex is currently enabling species to persist in these areas as well. 53
It is important to consider that use of habitats does not necessarily mean that the habitats are productive ones and, in the worse case, used habitats might be sinks or traps (Pulliam 1988; Kristan 2003). Where source-sink dynamics are present, resource selection models may predict a high probability of occurrence, but those locations may negatively affect population productivity. Nevertheless, corridors or patches that allow even occasional dispersal may be linked to population-level reproductive success since corridors may sometimes represent poor-quality habitats that still facilitate movement (Haddad & Tewksbury 2005). Monitoring relative densities of the species over time as well as conducting in-depth studies to understand how individual movement and physiological health are impacted by where individuals spend time in the matrix (see Chapter 4) will provide further insights into understanding how habitat use correlates with expected longterm conservation outlook.
Conclusions Since much of the Cerrado biome is degraded, and most reserves are too small to alone ensure the preservation of their large mammalian fauna, understanding the role of the landscape mosaic and managing private lands for conservation will be critical to avoiding extinction debts in the Cerrado’s protected areas. Our data support previous claims that if existing laws were applied efficiently, the resulting habitat fragments could support many Cerrado species (Calvalcanti & Joly 2002). Furthermore, our analyses show that the continued presence of this suite of large, wide-ranging mammals in the Emas Park region is likely due to a combination of a well-managed reserve and an extensive network of habitat remnants in the form of forested river corridors and patches of cerrado woodland. The varied habitat preferences of this particular suite of species demonstrate the multifaceted approach that will be required to achieve comprehensive conservation outcomes. The design of functionally connected landscapes will likely require physically connected movement corridors for some species, whereas functionally connected corridors in which human activities are moderated will be sufficient for others.
54
Together, this information will allow for the identification of strategic conservation areas within our study area. Because the probability quantified in terms of the resource selection probability function can be used to predict the locations of each of the target species for any region for which we can map the model covariates, it is possible to extrapolate our results to help advance research, recovery, and conservation planning of these species in areas beyond our study site. The results may be used to design optimal follow-up sampling protocols for subsequent inventory or monitoring of the species in landscapes of importance throughout the Cerrado biome.
55
Chapter 3: Factors Influencing Degradation of DNA and Hormones in Maned Wolf Scat Abstract The ability to non-invasively detect the presence of species and assess their physiological health by DNA and hormone analysis makes scat a valuable tool for ecology and conservation. I assessed factors associated with DNA and hormone degradation in a four-season study that employed detection dogs to collect scats from maned wolf (Chrysocyon brachyurus) in the Brazilian Cerrado, a tropical savanna landscape mosaic. Fecal DNA sample viability was assessed by attempting PCR amplification of a mitochondrial DNA (mtDNA) locus (~246bp) and a nuclear DNA (nDNA) microsatellite locus (~195bp). I assessed how extraction method, environmental exposure, and amount of odor, moisture, and diet items in the sample influenced DNA amplification and allelic dropout rates. Samples that amplified mtDNA were assayed for glucocorticoids and thyroid hormone. Amount of odor and moisture (indicating freshness) predicted mtDNA amplification success, as well as mean hormone levels. While factors related to sample condition were negatively correlated with lower mean hormone levels, samples comprised mainly of fruit had higher levels of glucocorticoids and lower levels of thyroid hormone, and we thus interpret this result as biologically meaningful. In summary, DNA and hormone degradation are predicted by measures of sample freshness, making the assessment of sample quality an important criterion for sample collection as well to manage measurement error in analyses of hormone concentration associated with environmental disturbance.
Introduction The advent of improved efficiency in molecular methods makes the use of scat samples increasingly feasible for presence-absence, demographic, and physiological studies of otherwise difficult to study species (Foran et al. 1997; Wasser et al. 2004; Ball et al. 2007). Fecal DNA sampling has been extremely useful for identifying species and individuals in an area, evaluating distribution, determining sex ratio, and estimating
56
population size (Kohn et al. 1999; Bellemain et al. 2005; Solberg et al. 2006) and fecal hormone metabolites may be used to assess physiological health and the disturbance response of populations (Creel et al. 2002; Rolland et. al. 2006; Gobush et. al. 2008). Scat is particularly enticing for studies of species of conservation concern because it may be collected non-invasively, without having to capture, handle, or observe animals. When scats are subject to DNA and/or hormone extraction and analysis, however, poor sample quality can increase wasted laboratory costs on extraction and amplification of degraded DNA or hormones. Degradation of hormones in samples is particularly problematic for hormone analyses, since these are quantitative measures and thus necessitate the removal or separation of variation due to hormone degradation from that resulting from the environmental variables of interest (e.g., noise disturbance, habitat degradation). Factors that influence sample quality can be considered to minimize effort collecting and analyzing samples unlikely to yield high quality DNA, as well as to account for variation due to sample condition when analyzing sample hormone levels.
Researchers have previously demonstrated the importance of laboratory factors, including methods related to fecal sample preservation (Wasser et al. 1997; Murphy et al. 2002; Piggot and Taylor 2003; Rutledge et al. 2008), DNA extraction (Frantz et al. 2003; Wehausen et al. 2004), and amplification (Bellemain and Taberlet 2004; Piggott et al. 2004). Variables influencing condition of sample in the field have also been shown to influence DNA amplification success and include age of sample (Lucchini et al. 2002; Piggot 2004; Santini et al. 2007), weather conditions (Farrell et al. 2000; Lucchini et al. 2002; Piggot 2004), diet (Murphy et al. 2003; Maudet et al. 2004), and intestinal slough rate, which may vary among species and within species as diets vary by individual or season (Farrell et al. 2000; Maudet et al. 2004). These studies have suggested that success rates will be highest when samples are as fresh as possible and climatic conditions are either dry (Farrell et al. 2000; Piggot 2004) or very cold (Lucchini et al. 2002). In addition to influencing DNA amplification success, old or poor-quality samples may influence the amount of measurable hormone from a sample (Millspaugh and Washburn 2004). Other studies have suggested that by only including samples that
57
amplify nDNA in hormone analyses, it is possible to exclude variation due to poorquality samples (Wasser et al. 2004), yet this had not been explicitly tested. While there has been work done showing the influence of sample preservation on hormones (Hunt & Wasser 2003; Wieke et al. 2004), there have been few studies examining the relationship of sample quality in field-collected samples and hormone-levels (Washburn & Millspaugh 2002).
In addition to sample quality, dietary intake may influence fecal excretion of hormone metabolites independent of stress or nutritional status (von der Ohe & Servheen 2002; von der Ohe et al. 2004). The amount of food consumed and diets high in fiber, for example, may increase passage rate and decrease pooling time of steroids (Lewis and Heaton 1997; Goldin et al. 1981). While diet is typically controlled by removing water from the sample (Wasser et al. 1993), von der Ohe and Servheen (2002) postulated that large dietary differences will not be completely adjusted for during lyophilization since dietary intake may impact the degree of reabsorption of metabolites, time of pooling, and exogenous augmentation of glucocorticoid levels.
In this chapter, I present the results of a four-year study that employed scat detection dogs (Wasser et al. 2004) to survey scats of maned wolf (Chrysocyon brachyurus) in and around Emas National Park, in the Brazilian Cerrado. Scat detection dogs enable efficient sampling of wildlife populations without the biases traditionally associated with other methods (Wasser et al. 2004; Harrison 2006; Long et al. 2007b), yet the effectiveness of this sampling method means that a large number of old and degraded samples are encountered. Because factors influencing ability to extract DNA from maned wolf scat had not been studied, and because samples collected in the tropics may be particularly susceptible to degradation of DNA and hormones, we sought to quantify factors affecting sample quality of field-collected scat. Specifically, we examined the influence of extraction method, environment (habitat and season in which the sample was collected, sample exposure (e.g., whether or not the sample was found on a road), sample condition (moisture level, strength of odor, presence of mold or invertebrates on the sample) and
58
food items (amount of fruit) on the ability to amplify DNA. To determine how sample quality may influence hormone levels, we analyzed mean glucocorticoids (predominately cortisol metabolites, which may include small circulating levels of corticosterone) and thyroid hormone levels in samples based on whether or not they amplified nDNA and the amount of odor, moisture, and fruit in the sample.
Study area We surveyed for scat samples in the region of Emas National Park (18°S, 52°W), Brazil. Emas National Park is considered one of the most important protected areas in the Cerrado biome, a tropical savanna-woodland, which comprises 21% of Brazil and is the world’s largest, richest, and possibly most threatened tropical savanna (Silva and Bates 2002). In addition to the 1320 km2 Park, which supports large tracts of grassland plains and open shrublands (81%), woodlands and riparian forest (17%), and marshlands (1%), our study area included 2500 km2 of agricultural farms, ranchlands, and habitat remnants. Average annual precipitation was approximately 1500 mm during the wet season (September-May) and virtually no rain fell in the dry season (June-August), when temperatures reached 39° during the day and dropped to -1° at night (Silveira et al. 1999). The search area spanned 700-900 m elevation.
Methods Sample Collection Specially-trained detection dogs were employed to locate scat from maned wolves (Chrysocyon brachyurus), the largest canid in South America, between August of 2004 and April of 2008 (Chapter 1). When a sample was located, we recorded the GPS position and data on the habitat and sample condition, including moisture level, odor strength, presence of mold, and presence of invertebrates. We did not collect samples that were both odorless (to human observer) and formless (e.g. consisted only of a scattering of undigested animal material or seeds). Otherwise, we preserved a portion of the sample for fecal DNA extraction in a 40-mL vial with 25-mL of 20% dimethyl sulfoxide buffer (DMSO) (Frantzen et al. 1998). When the scat was intact, we collected
59
from the outside of the sample (Stenglein et al. 2010). We then mixed the remainder of the sample with a gloved hand in a plastic bag (Wasser et al. 1996) and retained this portion for subsequent diet, hormone, and DNA analysis. Samples were kept on ice packs in cooler bags and then froze them upon returning from the field until shipment (on dry ice) to the United States. Samples from 2006 experienced a significant delay (3 weeks) in shipment and thus were subjected to a complete thaw.
DNA Extraction, Purification, and Amplification I conducted DNA extractions in a laboratory dedicated to noninvasive DNA samples and spatially separated from polymerase chain reaction (PCR) products. All samples were extracted in either duplicate or triplicate to control for uneven distribution of DNA in scat (Wasser et al. 1997; Fernando et al. 2003). Fecal DNA preserved with DMSO (approximately 0.5 g) was extracted and processed using the Qiagen QiaAmp® stool mini kit & Dneasy® 96 blood and tissue kit (Qiagen, Inc., Valencia, CA) and DNA extracts were then purified using the Geneclean® III kit (Q-BIOgene Inc., Carlsbad, Calif.), both with modified protocols (Appendix E, F). To test the efficacy of extracting DNA from the mucosal layer of the feces, I swabbed a subset of 143 maned wolf samples from the 2007 and 2008 field seasons to collect the epithelial and immune cells (Ball et al. 2007; Rutledge et al. 2008). In both methods, I included one negative control (no scat material added to the extraction) for every 11 extraction tubes to check for laboratory contamination. The species test consisted of fragment analysis from PCR amplification of the mitochondrial control region (D-loop) using unlabeled HSF21 (GTACATGCTTATATGCATGGG) and 5 6-FAM-labeled LTPROB13 (CCACTATTAACACCCAAAGC) primers (Wasser et al. 1997). To ascertain the taxonomic identity of the fecal samples, I compared band sizes with known control samples from maned wolf and other sympatric carnivores (Appendix G). I noted samples that did not yield DNA for species identification, or were of a non-target or unknown species, and excluded these from further analysis.
60
I screened samples with at least one of the extract pairs yielding mtDNA for maned wolf to determine gender. I used a primer set that amplifies a short (195bp) fragment of the zink finger (Zfx and Zfy) protein genes. This primer was described for kit foxes (Vulpes macrotis mutica) and preliminarily tested on five additional canid species, including maned wolves (Ortega et al. 2004). Modifying the Ortega et al. (2004) PCR conditions from 35 to 40 cycles improved our amplification success. I extracted and amplified all samples twice; those loci that were heterozygous (indicating male) in both replicates were scored as reliable and genotypes were recorded; all homozygous and uncertain genotypes (due amplification failure or to allelic dropout) were additionally replicated four times; I discarded all samples that could be not reliably typed after six amplifications. To ensure accurate calling given the potential for allelic dropout of the Y band, I only called a sample as female once we had seen 3 female (and no male) bands. Known maned wolf female, male, and negative controls were run in all PCR-amplified assays and band sizes were as reported for kit foxes (Ortega et al. 2004).
Hormone Extraction and Assays I prepared samples for hormone assays by freeze-drying for 48 hours or until all moisture was removed from the sample and then sifting the sample through a steel-mesh colander to remove non-fecal material such as seeds, fruit pulp, bones, and hair. I recorded preand post-freeze dry weights to determine the percent moisture in a sample and we expressed hormone concentrations per gram dry mass since desiccation was shown to control for most dietary effects on hormone excretion rates (Wasser et al. 1993). For extraction of glucocorticoids, I added 2.0 mL of 90% ethanol to 0.2 g freeze-dried and thoroughly homogenized fecal powder (Wasser et al. unpublished data) and for thyroid extractions 15 ml of 70% ethanol to 0.1g of freeze-dried and thoroughly homogenized fecal powder (Wasser et al. 2010). I then vortexed, centrifuged, and pipetted the supernatant to remove it from the tube containing the fecal pellet and stored the supernatant in an airtight tube at -20°C (Wasser et al. 2010).
61
Fecal hormones were assayed with commercial [125I] assay kits (MP Biomedicals, Costa Mesa, Calif.), which we validated for recovery and accuracy via standard curves and parallelism (Wasser et al. 2000, 2010). All samples were assayed in duplicate, with NSBs and blanks in quadruplicate, a full standard curve in duplicate, and both manufacturer and study low and high controls in duplicate. Any samples falling outside the range of 15%-85% bound or >10% CV between duplicates were re-assayed (interassay CV for glucorticoids 8%, thyroid 12%)
Predictors of DNA amplification and mean hormone levels Exposure to humidity and ultraviolet light were expected to adversely affect sample quality (Murphy et al. 2007). I therefore looked at the influence of season, road deposition, and vegetation type in which the sample was found on DNA amplification success. To understand if vegetation type influenced amplification success, I considered four broad classifications: open-canopy natural vegetation types (grassland and open cerrado), closed-canopy natural vegetation (forest, cerrado), marshlands (inundated grasslands), and converted vegetation types (pasture and croplands).
I used a qualitative classification to define sample condition at time of collection by assessing the sample odor strength, moisture level, presence of mold, and amount of invertebrates present on the sample prior to collection. These categories had previously been demonstrated to influence genotyping rates (Piggott 2004; Wasser et al. 2004; Gebhardt et al. 2009). I recorded the amount of odor in 1 of 6 categories: very strong, strong, moderate, weak, none, or earthy. I assigned level of moisture to 1 of 4 categories ranging from fresh/moist to dry throughout unless the sample was wet due to rain, in which case this was noted. The project leader assured inter-observer reliability by periodically cross-checking all teams for scoring consistency. For data analysis purposes and to help minimize potential scoring differences between observers, however, I collapsed categories into 3 levels each for odor and moisture (Table 3.1).
62
I also expected DNA amplification to be affected by the presence of fruit in the samples, since fiber decreases gut passage time, bacterial degradation and sloughing of cells in a sample (Burrows et al. 1982; Wasser et al 1993; Murphy et al. 2007) and may also contain DNA inhibitors. For data analysis purposes, therefore, I grouped samples according to amount of fruit in the sample (Table 3.1).
I used a standard method (Creel et al. 2003; Broquet & Petit 2004) to compute allelic dropout, dividing the number of errors detected by the total number of cases in which an error could have been detected. I used samples that were confirmed as males and took the number of times we saw only a female band and divided this by the number of amplifications that returned female or male calls.
To detect the influence of sample condition on mean hormone levels, I included only samples that successfully amplified for mtDNA. I compared mean values of sample hormone levels across categories for a set of significant predictor variables of DNA amplification success (amount of odor and moisture in sample at time of collection, and presence of fruit in the sample), as well as for percent water weight lost in drying and the overall success or failure of nDNA amplification for the sample.
Statistical Analyses To understand the relative influence and contribution of suspected important covariates related to sample condition and environment of collection and their relationship to DNA amplification success, I used classification trees, computationally intensive methods that facilitate data inspection and selection of explanatory variables (Breiman et al. 1984; Crawley 2002). Classification trees estimate a regression relationship through binary recursive partitioning of responses (amplification success or failure) to develop decision rules for predicting the categorical response (De'Ath and Fabricius 2000). The trees are grown by repeatedly splitting the data, defined by a simple rule based on a single explanatory variable. As the data is partitioned into mutually exclusive groups, the
63
procedure is then applied to each group separately and the split that maximizes the reduction in impurity is chosen (De'Ath and Fabricius 2000).
While amount of odor, moisture, presence of mold, and amount of fruit in sample were categorized into scalar variables ranging across three incremental categories, other categories (invertebrates, season, location relative to road) were binary, and habitat was broken into four separate habitat types (Table 3.1). Classification tree partitioning allowed me to distinguish break points for scalar variables as well as to distinguish and group habitats for a more parsimonious analysis without running each habitat type as a separate variable. From all possible splits of explanatory variables, I selected the one that maximized the homogeneity of the two resulting groups (mostly amplified, mostly failed to amplify). Splits minimized the sums of squares within groups and, when a node was strongly homogenous, it was not further subdivided. I used these results to inform a generalized linear model (GLM) that predicted amplification success based on the suite of categorical variables assessed via classification tree analysis.
I analyzed amplification success for mtDNA and nDNA (separately) through a generalized linear model (GLM) using binary logistic regression (Crawley 2002). I predicted amplification success as a function of the categorical variables listed above to determine the variance explained and the relative importance of each factor. Model fit was determined through maximum likelihood estimation (Burnham and Anderson 1998) after transforming the dependent into a logit variable (the natural log of the odds of the dependent occurring or not), such that the GLM estimated the probability of amplification, given a certain set of independent variables related to sample condition and environment. We ran GLMs for amplification success including all variables (Table 3.1). The response variables for candidate models were, for mtDNA, successful amplification from at least one of duplicate extracts (binary response of yes or no), and for sex identification, successful determination of gender identification of the sample based on the minimum requirements as per our gender calling protocol. We then used a step function to sequentially remove variables and analyze the resultant change in model performance, evaluated on the basis of the Akaike Information Criterion (AIC). 64
Table 3.1. Description of categorical variables used to understand factors influencing quality of maned wolf scat samples collected in the Cerrado of Brazil. The character of variables is denoted by C = condition, S=season, L = location, F = food items. Variable Odor Moisture Mold Invertebrates Road Habitat Season Fruit diet
Character C C C C L L S F
Values 1 (strong), 2 (some), 3 (none) 1 (moist/fresh), 2 (some moisture), 3 (completely dry) 1 (throughout), 2 (some), 3 (none) 1 (throughout, >20), 2 (some/none) on road, not on road closed-canopy, open-canopy, marshland, converted wet, dry 1 (all fruit), 2 (some), 3 (none)
To test for influence of sample quality on mean hormone levels, I used linear regression with amount of hormone measured as log10 ng/g as the dependent variable and 4 suspected influential covariates as independent variables. The percent water weight lost in freeze-drying of sample was also tested via regression for correlation with amount of hormone since we expected this might be influential, in particular, for samples that were fresh and subjected to thaw during shipment. I used Student’s t-tests to compare means of all levels within a variable and, for all statistical tests, significance was measured at the p = 0.05 level (Zar 1999).
Results Collection Effort, Species Determination, and Gender Assignment The dog teams collected a total of 1536 putative maned wolf samples that were extracted and screened for DNA analysis. Over the four seasons of study, an average of 84% of samples amplified mtDNA (61% for maned wolf, Appendix H). Other species identified via DNA analysis included fox (Cerdocyon thous), puma (Puma concolor), and ocelot (Leopardus pardalis) (Appendix G, H).
Of the 1536 putative maned wolf samples analyzed for mtDNA, 713 were excluded from nDNA analysis for gender in maned wolves because the sample failed to amplify for mtDNA, amplified for a non-target species, or could not be reliably genotyped after multiple amplifications. Of the samples analyzed for gender, 61% (n = 500 of 823) 65
yielded positive results (amplified male loci at least twice or female at least three times) and the sex ratio was not different from 50:50 (
1
2
= 0.1650, p = 0.6846), as expected
(Jácomo et al. 2009).
Influence of Extraction Method on Amplification Success Fisher’s exact test showed a significant difference in mtDNA amplification between the two extraction methods; swabbing yielded better amplification for at least one (p = 0.001) or both duplicate pairs (p = 0.001). DMSO would not have returned a species identification when swabbing would have 17% of the time (24 of 143 extracts), whereas the reverse was true only 3% of the time (5 of 143 extracts). Only 3% (4 of 118) of samples extracted a third time (a “ C” replicate) resulted in a positive species identification not previously confirmed by the duplicate pair extracts (“ A” & “ B”).
To determine if extraction method affected nDNA amplification success, I attempted to determine gender in 100 samples that were extracted using both methods and were successfully amplified for mtDNA. Of these, 49 samples yielded an identical result, 34 samples yielded a result for swabbing only, and 21 samples yielded a result for DMSO only. Thus, even in samples that passed screening for mtDNA, the mucosal swab method of extraction improved nDNA amplification success (p = 0.023).
Blood samples of a captive female and male run in 20 PCRs amplified correctly in 100% of 20 trials with no allelic dropout. Freshly collected scat samples from captive and wildtrapped animals amplified in18 of 20 (90%) PCRs, and one of the male failures was allelic dropout (dropout for that sample was 5%). For field-collected samples, allelic dropout of male bands averaged across years was 40% when the total number of female bands seen was divided by the total number of female and male bands seen in samples called as males. When dropout rate was averaged across samples by the rate per sample (rather than calculated as the total number of bands), the dropout rate was 32%.
66
Ninety percent of confirmed males (n = 219 of 242) had dropout (where only a female band was amplified) at least once. The approach of using multiple extractions and repeat amplification of homozygote alleles, however, results in an expected accuracy of 93.6% for males from the data, as males will erroneously be called as females 6.4% of the time (the probability of amplifying only the female band in a true male sample three times is 0.43, or 0.064). A sample extracted via swabbing was significantly less likely to result in allelic dropout compared to the DMSO method ( 1² = 6.572, p = 0.010). Influence of Sample Condition, Exposure, and Contents on Amplification Success The variables that split in the classification tree analyses (i.e. there was a significant break between the factors of the variable) for mtDNA were odor, moisture, and habitat (Appendix I). Significant break points in amplification of nDNA for samples that had previously amplified for mtDNA were if the sample had been rained on, habitat, and diet (Appendix J).
Significant contributing factors to GLM model fit for amplification of mtDNA from putative maned wolf scat were amount of odor and moisture in the sample, as well as habitat where the sample was found (Table 3.2). Based on stepwise removal of variables in succession and analysis of the corresponding AIC values for simplified models, the most parsimonious GLM for predicting amplification success for mtDNA would include season, habitat, odor, moisture, mold, and season only (i.e. exclude road, diet, and invertebrates). Amplification success was positively correlated with the amount of odor and moisture in the sample (Figure 12), and was negatively correlated with closedcanopy habitats (Table 3.2). The classification output for the most parsimonious model for mtDNA was predicted by sample odor and moisture (Figure 13a).
67
Table 3.2. Parameter estimates from statistical model showing factors contributing to probability of mtDNA and nDNA amplification of maned wolf scats collected in the Cerrado of Brazil between August 2004 and April 2008.
mtDNA Term (Intercept) odor moisture mold invertebrates habitat road season fruit diet
nDNA
parameter estimate 1.135
SE
z-value
0.260
4.356
0.326 0.539 -0.501 -0.391 -0.699 -0.081 0.738 -0.089
0.154 0.170 0.310 0.372 0.312 0.205 0.544 0.152
2.118 3.178 -1.615 -1.053 -2.243 -0.393 1.357 -0.588
SE
z-value
