Population Genetics and Structure of the Sumatran Tiger

Population Genetics and Structure of the Sumatran tiger

Olutolani Smith

A thesis submitted for the degree of Doctor of Philosophy Division of Biology, Imperial College London

May 2012 Declaration

I confirm that the work presented in this thesis is my own. Where information has been derived from other sources I confirm that this has been appropriately referenced.

- 2 - Abstract

The two key determinants of population persistence in fragmented landscapes are population size and connectivity. Populations with high levels of genetic variation and large population size are expected to have a lower risk of extinction. Similarly, populations with high rates of connectivity are expected to persist long-term. For many elusive landscape species it is difficult to obtain direct estimates of these parameters, but genetic sampling can offer powerful indirect assessments. Whilst these techniques have been applied to the study of many wide-ranging carnivores, this study represents the first example in the Sumatran tiger (Panthera tigris sumatrae). Extensive field surveys were conducted to collect faecal samples from several Tiger Conservation Landscapes and protected areas on Sumatra. Samples were then processed according to optimised protocols to obtain reliable results. In order to quantify extinction risk I first estimated genetic variation and effective population size using microsatellite loci. I also determined relative levels of connectivity using estimates of differentiation (FST), gene flow and genetic clustering. Results indicate that Sumatran tigers have high levels of genetic variation and that their effective population size is within the expected range. There is very little population structure and there is no obvious evidence for barriers to dispersal. The Batang Hari/Kerinci Seblat ecosystem emerged as a potential source population and in contrast there was some evidence of isolation affecting the population of Way Kambas NP in the extreme south of the island. Overall, despite high levels of human land cover conversion over the past 20-30 years, few genetic changes have been expressed in the Sumatran tiger. The immediate threat to tigers is not the loss of genetic diversity, but the rapidly declining area of suitable habitat in which they can survive.

- 3 - Acknowledgements

This project would not have been possible without the help and support of a huge number of people both in Indonesia and the UK. My supervisors Tim Coulson, Jinliang Wang and Chris Carbone deserve special mention for believing that a shy, small animal vet could sample one of the world's most elusive cats on the sixth largest island in the world. Their knowledge, expertise and guidance has been invaluable. Special thanks must also go to Dada Gottelli whose yoda-like advice has proved that when the force is with you the scat is not your enemy. Thanks also to the IOZ genetics theme and the lab gurus Rob Pickles, Patricia Brekke, Kate Ciborowski and Geoff Hobbs whose lab advice got me through the early days of trying to appease the PCR gods. UK support was also provided by Sarah Christie, Malcolm Fitzpatrick and BIAZA who were instrumental in encouraging zoos to provide samples for an initial pilot study. Samples were donated by Howletts Wild Animal Park, Wildlife Heritage Foundation, Dublin zoo, Dudley zoo, London zoo, Paignton zoo and Thrigby Hall Wildlife Gardens.

Tiger samples from Sumatra were kindly sent to the Eijkman lab by the following people: Tom Maddox, Dolly Priatna and Citra Novalina (ZSL Indonesia Program), Karmila Parakkasi, Wishnu Sukmantoro and Sunarto (WWF Riau), Matt Linkie, Elva Gemita and Yoan Dinata (FFI Batang Hari), Agung Nugroho (FFI Kerinci), Tony Sumampau, Retno Sudarwati and Hariyo Wibisono (HarimauKita and Taman Safari Indonesia), Iding Haidir and Rahmad Simbolon (Sembilang NP), Matt Linkie and Dolly Priatna (FFI Aceh). I am also grateful to the staff at Way Kambas NP, Kerinci Seblat NP and PT Andalas Merapi Timber for being adventurous enough to participate in the detection dog surveys. Sumianto, Ali Mashuri, Sunarwanto and Nur Alim (Way Kambas NP), Rika Putra Abbas and Hambali (Kerinci Seblat NP), Dwi Adhiasto and Lilik Prastowo (Bukit Barisan Selatan NP) all provided welcome assistance in the field. I also received great support from office and field staff in the Wildlife Conservation Society Indonesia Program. In particular, I am grateful to Nick

- 4 - Brickle, Noviar Andayani and Minda Saanin for readily providing introductions and support letters for numerous permits and visas. Ibu Yani, I hope this project serves as the spark you needed to fire up the genetics revolution in Indonesia!

Elva Gemita, Gigih Suparmano and Jasman were core members of the scat collection A-team. Elva as always had boundless energy and an infectious spirit that was wonderful to work alongside. We had many giggles and bonded over a shared love of Green & Blacks chocolate. Gigih willingly ferried people, equipment and dogs through most of west Sumatra, and Jasman excelled as the master scat spotter. Although not officially part of the field team, Pak Heri and Pak Wawan drove for 18 hours from Java to Sumatra and back again to transport me, Rudi and all our equipment to base camp. Pak Heri also waited patiently in many car parks while I pursued dogs, scats and permits all over Java and south Sumatra. Karmele, Magda and Lia also helped to organise the paperwork and logistics involved with moving Rudi between Java and Sumatra. The detection dog surveys benefitted from the initial guidance of Amy Dickman, and later from the teaching of Richard Clarke and Stanley the wonder dog. I am particularly grateful to Richard for handler training in the UK and for his willingness to come to Indonesia to train Rudi. I'm sure that is something the residents of Bogor will never forget! And of course I must thank Rudi himself for keeping his spirit despite the many trials we put him through. The Eijkman laboratory were kind enough to let me bring smelly samples into their wonderful facility. Prof Sankot Marzuki, Dr Herawati Sudoyo and Dr Helena Surjadi were supportive from the very beginning and now have a growing wildlife forensics service. Thank you also to Dr Safarina Malik for giving up her lab space and for her unfailing positive energy, Eva for her calm acceptance of my endless sequencer runs and Wulan for her support and friendship.

During my time in Indonesia the 'Bogor massive' made me feel very welcome as the newbie on the block and introduced me to the joys of Bintang, G&Ts and pot luck. In particular, heartfelt thanks go to Nick, Anna, Robbie and Alex for welcoming such a

- 5 - bewildered ex-pat into their home. Thank you to my friends and family who have been wonderful over the past few years, even though I'm sure they still don't understand what it is I've been doing. Watch this space - there are more adventures to come! At the IOZ my office mates Lisa Signorile, Martina di Fonzo and Farid Belbachir have had to endure many months of grumbling but we have also had some great conversations about food and the meaning of life. I know that you will do well in your PhDs and go on to great things. And last but not least, thank you to Joe for being a source of inspiration and encouragement since the day we met. I wouldn't be on this crazy journey if it wasn't for you.

I have been helped along the way and have been given advice by so many people that I may have missed one or two. My sincerest apologies if that happens to be you.

This study was funded by a National Environment Research Council (NERC)/CASE Studentship and I also received valuable support from the Panthera-Kaplan Graduate Award scheme and the Zoological Society of London. Indonesian research permits were provided by the Ministry of Research & Technology (Ristek), the Department of Forestry, and the Department for Forest Protection and Nature Conservation (PHKA).

- 6 - Contents

Declaration 2

Abstract 3

Acknowledgements 4

Contents 7

List of Tables 15

List of Figures 16

1. Research Background 18

1.1 Introduction ………………………………………………………………19

1.2 Metapopulation dynamics ………………………………………………20

1.2.1. Patch size & isolation ………………………………………………21

1.2.2. Extinction threshold ………………………………………………23

1.2.3. Time lag ……………………………………………………………23

1.2.4. Stochasticity…………………………………………………………24

1.2.5. Source-sink dynamics ……………………………………………24

1.2.6. Extinction risk ………………………………………………………25

1.3 Landscape ecology ……………………………………………………25

- 7 - 1.3.1. Population viability …………………………………………………26

1.3.2. Matrix quality & patch boundaries ………………………………27

1.3.3. Connectivity & corridors ……………………………………………29

1.3.4. Population persistence ……………………………………………30

1.4 Population genetics ……………………………………………………30

1.4.1. Population size ……………………………………………………31

1.4.2. Isolation ……………………………………………………………33

1.4.3. Extinction risk ………………………………………………………35

1.5 Landscape genetics ……………………………………………………35

1.5.1. Isolation by distance ………………………………………………36

1.5.2. Barriers to dispersal ………………………………………………37

1.5.3. Functional connectivity ……………………………………………38

1.5.4. Population persistence & connectivity……………………………39

1.6 Overall summary ………………………………………………………40

1.7 Case study - the Sumatran tiger (Panthera tigris sumatrae) ………41

1.7.1. Sumatra ……………………………………………………………41

1.7.2. Sumatran tigers ……………………………………………………42

1.7.3. Threats ………………………………………………………………43

1.7.4. Population viability …………………………………………………45

1.7.5. Genetics ……………………………………………………………47

1.7.6. Further study ………………………………………………………50

Thesis outline 51

- 8 - 2. Sampling effects in Conservation Genetics Studies 52

2.1 Abstract 53

2.2 Introduction ………………………………………………………………53

2.3 Methods …………………………………………………………………56

2.3.1. Simulated data………………………………………………………56

2.3.2. Genetic parameters ………………………………………………56

2.4 Results ……………………………………………………………………59

2.4.1. Genetic diversity and sample size ………………………………59

2.4.2. Population structure and sample size ……………………………60

2.5 Discussion ………………………………………………………………63

3. The Inﬂuence of Genotyping Error in Noninvasive Genetics Studies 68

3.1 Abstract …………………………………………………………………69

3.2 Introduction ………………………………………………………………69

3.3 Methods …………………………………………………………………73

3.4 Results ……………………………………………………………………75

3.4.1. Allelic dropout ………………………………………………………75

3.4.2. False alleles …………………………………………………………76

3.4.3. Missing alleles ………………………………………………………81

3.4.4. Combined error ……………………………………………………81

3.5 Discussion ………………………………………………………………87

3.5.1. Genetic variation and differentiation ……………………………87

- 9 - 3.5.2. Population structure ………………………………………………89

3.5.3. Error thresholds ……………………………………………………90

3.5.4. Recommendations to reduce error ………………………………92

3.6 Conclusions………………………………………………………………94

4. Sample Collection and Genetic Analysis of Noninvasive Samples 95

4.1 Abstract …………………………………………………………………96

4.2 Introduction ………………………………………………………………96

4.3 Field methods ……………………………………………………………98

4.3.1. Study sites …………………………………………………………98

4.3.2. Survey method ……………………………………………………100

4.4 Laboratory Methods……………………………………………………102

4.4.1. DNA extraction & Species identification ………………………103

4.4.2. Sex identification …………………………………………………105

4.4.3. PCRs ………………………………………………………………106

4.4.4. Genotyping…………………………………………………………107

4.4.5. Error rates …………………………………………………………108

4.5 Results …………………………………………………………………110

4.5.1. Sample collection …………………………………………………110

4.5.2. PCR success ………………………………………………………112

4.5.3. Error ………………………………………………………………114

4.6 Discussion………………………………………………………………115

- 10 - 4.6.1. Sample collection …………………………………………………115

4.6.2. Sample preservation and PCR success ………………………117

4.7 Conclusions ……………………………………………………………118

5. Population Genetics of the wild Sumatran tiger (Panthera tigris sumatrae) 120

5.1 Abstract …………………………………………………………………121

5.2 Introduction ……………………………………………………………121

5.3 Methods…………………………………………………………………124

5.3.1. Sample collection and genotyping ………………………………124

5.3.2. Estimating genetic diversity………………………………………125

5.3.3. Is there evidence for population differentiation? ………………126

5.3.4. What is the effective population size? …………………………127

5.3.5. Measuring rates of gene flow ……………………………………130

5.4 Results …………………………………………………………………131

5.4.1. Higher genetic variation than expected…………………………132

5.4.2. Heterozygosity and allelic diversity ……………………………133

5.4.3. Genetic differentiation and restricted gene flow in the south…134

5.4.4. Small effective population size …………………………………135

5.5 Discussion………………………………………………………………136

5.5.1. High genetic variation ……………………………………………136

5.5.2. Small effective population size …………………………………137

5.5.3. Population differentiation and gene flow ………………………138

- 11 - 5.5.4. Conservation management………………………………………140

5.6 Conclusions ……………………………………………………………141

6. Population Structure of the Sumatran tiger 142

6.1 Abstract …………………………………………………………………143

6.2 Introduction ……………………………………………………………143

6.3 Methods…………………………………………………………………147

6.3.1. Sample collection and genotyping ………………………………147

6.3.2. Neighbourhood size ………………………………………………147

6.3.3. Barriers to gene flow ……………………………………………148

6.3.4. Connectivity ………………………………………………………149

6.4 Results …………………………………………………………………151

6.4.1. Small neighbourhood size ………………………………………151

6.4.2. Two genetic clusters………………………………………………152

6.4.3. Multiple routes for dispersal ……………………………………153

6.5 Discussion………………………………………………………………155

6.5.1. Isolation by distance………………………………………………155

6.5.2. Spatial genetic structure …………………………………………156

6.5.3. Potential for connectivity …………………………………………159

6.6 Conclusions ……………………………………………………………160

7. General discussion 161

7.1 Key findings ……………………………………………………………162

- 12 - 7.2 Implications for Sumatra ………………………………………………164

7.3 Application to landscape species ……………………………………165

7.4 Conclusions ……………………………………………………………167

Bibliography ………………………………………………………………168

Appendices 210

A. Appendix 1 - Sample Collection & Genetic Analysis 211

A.1 Dog training protocol …………………………………………………211

A.2 DNA Cleanup protocol ………………………………………………212

Table A.1.1 The allele size range and fluorescent tag for each microsatellite Fca locus used in this study. ………………………………213

Table A.1.2 Per-allele and per-genotype error rates for each Fca locus used in this study. …………………………………………………………214

Table A.1.3 Consensus genotypes for the 25 samples used in this study. …………………………………………………………………………215

B. Appendix 2 - Sampling effects 216

Figure A.2.1 Variance in mean number of alleles per locus, expected heterozygosity and pairwise FST with a reduction in sample size. ……217

Figure A.2.2 Variation in the modal value of !K with a reduction in sample size. …………………………………………………………………218

C. Appendix 3 - Inﬂuence of Genotyping Error 219

Table A.3.1 Values of the mean number of alleles per locus (NA), expected heterozygosity (He), pairwise FST and the number of clusters (k) with an increase in error rate (") and a reduction in sample size (n). ………………………………………………………………………219

D. Appendix 4 - Population Genetics 221

- 13 - Table A.4.1 The results of a locus-by-locus test for Hardy-Weinberg equilibrium performed in GENEPOP. ………………………………………221

Table A.4.2 Results of a Fisher's exact test for linkage disequilibrium performed in GENEPOP. ……………………………………………………222

Table A.4.3 Locus-by-locus values of observed and expected heterozygosity. ………………………………………………………………223

Table A.4.4 Results of a test for population structure using AMOVA analysis performed in ARLEQUIN.……………………………………………224

E. Appendix 5 - Population Structure 225

Table A.5.1 Summary settings for the Bayesian clustering programs used in this study. …………………………………………………………225

Figure A.5.1 Resistance map using the human footprint index as a base layer. …………………………………………………………………224

Figure A.5.2 Resistance map using FAO land cover data as a base layer. Resistance values are outlined in Table A.5.2. ……………227

Table A.5.2 The FAO land cover classification system (LCCS) for Sumatra. Resistance values were assigned based on a relative scale of tiger habitat preference from 1 (high preference, low resistance) to 5 (low preference, high resistance). ………………………………………228

Table A.5.3 The settings used for CIRCUITSCAPE analysis. ……………229

- 14 - List of Tables

Table 1.1 Mean levels of heterozygosity for tigers and other felids. ……48

Table 2.1 EASYPOP settings used to simulate the test populations. ……57

Table 2.2 Minimum samples sizes required to determine the correct number of populations using STRUCTURE. ……………………………………60

Table 4.1 Tiger Conservation Landscapes (TCL) and protected areas sampled in this study. …………………………………………………………99

Table 4.2 A list of PCR primers used in this study. ……………………104

Table 4.3 A list of locations from which faecal samples were collected.………………………………………………………………………111

Table 5.1 The number of positive samples collected in each regional group across Sumatra. ………………………………………………………132

Table 5.2 A comparison of genetic diversity measures in subspecies of Panthera tigris. ………………………………………………………………133

Table 5.3 Regional values of observed heterozygosity (Ho), expected heterozygosity (He), unbiased expected heterozygosity (UHe), allelic richness (ag) and private alleles (#g). ………………………………………134

Table 5.4 Regional differentiation computed using pairwise FST.………134

Table 6.1 The number of genetic clusters estimated by STRUCTURE, TESS, BAPS and GENELAND.……………………………………………………152

Table 6.2 Pairwise landscape resistance values for occupied tiger habitat on Sumatra as estimated by CIRCUITSCAPE. ………………………………154

- 15 - List of Figures

Figure 2.1 Variation in the mean number of alleles per locus, expected heterozygosity and pairwise FST with decreasing sample size. …………61

Figure 2.2 Variation in the number of population clusters estimated with STRUCTURE as a function of sample size, using both the ln P(X|K) and Evanno !K methods. …………………………………………………………62

Figure 3.1 Variation in the mean number of alleles per locus, expected heterozygosity and pairwise FST with an increase in the rate of allelic dropout. ………………………………………………………………………77

Figure 3.2 Variation in STRUCTURE output with increasing rates of allelic dropout. The number of population clusters was estimated using the ln P(X|K) and Evanno !K methods with STRUCTURE HARVESTER. …………78

Figure 3.3 Variation in the mean number of alleles per locus, expected heterozygosity and pairwise FST with an increase in the rate of false alleles. ………………………………………………………………………79

Figure 3.4 Variation in STRUCTURE output with increasing rates of false alleles. ………………………………………………………………………80

Figure 3.5 Variation in the mean number of alleles per locus, expected heterozygosity and pairwise FST with an increase in the proportion of missing data. …………………………………………………………………83

Figure 3.6 Variation in STRUCTURE output with an increasing proportion of missing data. ………………………………………………………………84

Figure 3.7 Variation in the mean number of alleles per locus, expected heterozygosity and pairwise FST with an increase in the combined rate of genotyping error. ……………………………………………………………85

Figure 3.8 Variation in STRUCTURE output with increasing rates of combined error. ………………………………………………………………86

Figure 4.1 Detection dog training in Bogor, West Java. A 2 year old, male, Labrador Retriever was given initial training by Richard Clarke of the Barking Mad Dogs Training School in Hertfordshire, UK. …………101

Figure 4.2 Each scat collection team consisted of 2-3 staff and a dog handler. Teams searched along animal trails and ex-logging roads for

- 16 - tiger scats. ……………………………………………………………………101

Figure 4.3 A total of 148 faecal samples were collected from 6 Tiger Conservation Landscapes, 3 protected areas and other human-tiger conflict zones on Sumatra. …………………………………………………112

Figure 5.1 A map of Sumatra showing regional subdivision of sample sites. ………………………………………………………………………131

Figure 6.1 A test for isolation-by distance using linear regression of genetic distance and the logarithm of geographic distance. ……………151

Figure 6.2 CIRCUITSCAPE output showing landscape resistance for tigers on Sumatra. ……………………………………………………………153

- 17 - Chapter 1

Research Background

- 18 - 1.1 Introduction

Habitat loss and fragmentation are recognised as one of the major drivers for the decline of biodiversity and species abundance in many areas. Human disturbance is by far the biggest cause of landscape change, and with the human population set to rise to over 9 billion by 2050, there will certainly be increased pressure on what little native habitat remains (Geist & Lambin 2002; Sanderson et al. 2002b; UN Population Division/DESA 2009). The persistence of animal populations within this context will depend on individual species' ecology and the amount, pattern and rate of landscape change. Many studies have attempted to quantify the relationship between habitat fragmentation and population responses by modelling metapopulation dynamics, landscape dynamics, or a combination of the two. All methods start by defining the state of the landscape as a series of suitable habitat patches separated from each other but subject to movement of individuals (dispersal) between them. It is useful to start with a definition of habitat loss and habitat fragmentation, as although related they result in very different changes to the landscape. In a disturbed landscape, habitat refers to a range of environments suitable for a given species, and in a fragmented landscape suitable habitat may occur as a series of discrete patches. Habitat loss refers to a reduction in the total amount of suitable habitat due to its removal and replacement with less suitable habitat types. Habitat fragmentation describes a change in configuration and patch isolation as single large expanses of habitat are broken up into several smaller patches (Fahrig 2003). Thus, there is an increase in the number of patches, a decrease in the size of individual patches, and an increase in the isolation of each patch. Overall, habitat loss has been shown to have a far greater effect on biodiversity and species abundance than fragmentation (Fahrig 2003; With 2004; Turner 2005; With & Pavuk 2011). Despite these differences, it is difficult in empirical studies to tease apart the effects of loss from fragmentation so they are often combined in metapopulation analysis.

- 19 - Species that occur within fragmented landscapes are assumed to occur as a metapopulation, i.e. the structure of the population mirrors that of the landscape. For many species, the transition from a single large population to a network of smaller subpopulations is considered to lower the probability of persistence. In the following review I summarise the 4 main methods of quantifying extinction risk for fragmented populations: (i) Metapopulation dynamics, (ii) Landscape ecology, (iii) Population genetics, and (iv) Landscape genetics. In section 2, I outline metapopulation dynamics theory, a patch-based approach which considers population persistence to be related to the cycles of local extinction and recolonisation that occur within a patch network. In section 3, I give some examples of how landscape ecology analysis has been used to combine metapopulation dynamics with a more complete model of the landscape (the network of habitat patches and the matrix). In the final sections I describe population genetics, which considers only the genetic properties of the population (e.g. genetic variation and inbreeding), and landscape genetics, a relatively new technique which attempts to combine the principles of landscape ecology with population genetics. Finally, I bring all these concepts together and describe how they can be applied to the conservation of an endangered carnivore, the Sumatra tiger. Sumatra has undergone very high rates of deforestation in the last 20-30 years and tigers now exist in discrete patches of native forest surrounded by varying levels of human disturbance.

1.2 Metapopulation dynamics Metapopulation models were the first to consider population dynamics in a fragmented landscape, and the classic model proposed by Levins is perhaps the most widely used (Fahrig 2007). It is based on a pattern of patch occupancy in which patches exist in one of two states, occupied or unoccupied. Patches experience a series of local extinction and recolonisation, and all patches have the same local extinction and colonisation probabilities. The change in the proportion of occupied patches is given by:

- 20 - dp = mp(1- p) - ep (1.1) dt where p is the proportion of occupied patches, m is colonisation rate, and e is extinction rate for each patch in time t (Hanski 1991). And at equilibrium,

e p = 1! (1.2) m

The classic form of this model is not applicable for many populations as it assumes that all patches are equal and it does not take into consideration local population dynamics that may also affect patch occupancy. Spatially realistic stochastic patch occupancy models (SPOMs) were developed to account for heterogeneity of patches and the change in colonisation and extinction rates with patch size and isolation (Hanski & Ovaskainen 2003). Rather than constant rates for all patches, the probability of colonisation and extinction is patch-specific and dependent on habitat type. Extinction risk or species abundance and distribution can then be predicted given a particular habitat configuration (Hanski 1994; Moilanen 2004). Another way to incorporate spatial heterogeneity in metapopulation models is to first derive habitat suitability maps for a given species to identify the size and location of habitat patches in which the metapopulation exists. Metapopulation dynamics can then be determined based on carrying capacities and vital rates for each local population, and dispersal rates between patches (Akçakaya & Atwood 1997; Akçakaya et al. 2004).

1.2.1 Patch size & isolation In all these models the two fundamental factors determining population persistence are patch area and patch isolation. Patch area is often used as a proxy for population size because it determines resource availability and thus carrying capacity (Hanski 1998). Small patches are expected to support small populations.

- 21 - Small populations are more vulnerable to demographic, environmental and/or genetic stochasticity and so have a higher risk of extinction (Gaggiotti & Hanski 2004). Therefore, extinction risk scales with patch size (Hanski & Ovaskainen 2003). Populations that are isolated or poorly connected also have a much higher risk of extinction (Fahrig & Merriam 1985; Fahrig & Merriam 1994; Fischer & Lindenmayer 2007). High levels of dispersal increase the persistence of populations by mitigating the effects of demographic and genetic stochasticity; adding new genetic material and more individuals to the populations receiving immigrants (Gaggiotti & Hanski 2004; Reed 2004). Patch isolation can be represented as a function of the distance between patches and patch size. An empty patch that is separated by a large distance from occupied patches is expected to receive less immigrants, and small patches will provide a much lower number of colonists to the dispersal pool. The simplest measure of patch isolation uses nearest-neighbour distance (Moilanen & Nieminen 2002):

dNN Ii = c b (1.3) Ai ANN where dNN the distance from patch i to its nearest neighbour, Ai the size of patch i, and ANN the size of its nearest neighbour. Parameters b and c scale emigration and immigration rates to patch size. More robust estimators of patch isolation are given by buffer measures, which consider all occupied patches within a given radius around i, and incidence function models (IFM), which consider connectivity to all occupied patches. An example of the IFM defines the connectivity of patch i as:

b (1.4) Si = # p j exp(!"dij )Aj where pj is the state occupancy of patch j, dij the distance between patches i and j, ! a constant that sets the distance-dependent migration rate, Aj the area of patch j, and parameter b which scales expected population size to patch area (Balkenhol et al. 2009). Other parameters that influence metapopulation dynamics are the total

- 22 - amount of available habitat, the rate and extent of landscape change, demographic/ environmental stochasticity and source-sink dynamics.

1.2.2 Extinction threshold In some cases habitat loss may have little effect on species' distribution until some critical threshold is reached (With & Crist 1995; Fahrig 2003). This is the extinction threshold and it describes the minimum habitat area capable of supporting a given metapopulation (Hanski 1998). Below this threshold the most likely trajectory for the metapopulation is a decline to extinction. The exact level will vary between species, but in general it could be as low as 20-30% of native habitat (With 2004). Allee effects are expected to add to the extinction threshold when patch sizes are small because of low patch occupancy rates, which in turn reduce colonisation rates. For some species this could mean that even if the proportion of available habitat is above the extinction threshold, the metapopulation may still go extinct if the number of occupied patches is too low (Amarasekare 1998). The amount of habitat will also determine the spatial distribution of a population. For example, habitat generalists are expected to form a patchy distribution when their preferred habitat forms a small proportion (<35%) of the landscape. For habitat specialists the threshold is much higher (With & Crist 1995).

1.2.3 Time lag Not all populations will respond quickly to changes in the amount of available habitat. There is often a time lag between habitat disturbance and a change in patch occupancy. This means that some populations could continue to persist long after environmental change has occurred, although the future trajectory will still be one of decline (Hanski 1998; With 2004). This 'extinction debt' is considered to be larger for low abundant species, e.g. large, rare or tropical species (Tilman et al. 1994). Longer delays are also expected in regions that are already fragmented and that have small amounts of suitable habitat. This of course makes it difficult to determine

- 23 - the cause of population decline once it has been detected, and could mean that the opportunity to make any effective management interventions has passed.

1.2.4 Stochasticity Not all natural metapopulation dynamics occurs in a deterministic fashion due to demographic and environmental stochasticity (Gaggiotti & Hanski 2004). Demographic stochasticity refers to a fluctuation in population size due to variation in factors intrinsic to the population e.g. sex ratio, reproductive and mortality rates. Environmental stochasticity refers to a fluctuation in exogenous factors such as food availability and climate. These have an indirect effect on population size and growth rates. Stochasticity may also act at the level of the metapopulation by the degree of spatial autocorrelation or variation in recolonisation rates. Populations that are quite closely linked will experience the same temporal patterns of environmental fluctuation. This could lead to an increased risk of extinction for the entire metapopulation in the face of detrimental environmental changes (Reed 2004). Colonisation stochasticity means that some populations won't be recolonised after extinction (Hanski 1991).

1.2.5 Source-sink dynamics Source-sink metapopulation structure is a particular case in which dispersal is unidirectional. It is commonly described in empirical studies due to local demographic differences between patches. Source populations provide the colonists. They have a reproductive rate greater than mortality rate so produce a surplus of individuals to add to the disperser pool. In contrast, sink populations have a much lower reproductive rate than mortality rate so would become extinct without immigration from nearby source populations (Storfer et al. 2007). Sink populations often occur in low quality habitats or small habitat patches and are associated with much lower probabilities of persistence than source populations. Any threat to the source, e.g. a reduction in adult survival, may result in reduced persistence for the associated sinks (Donovan et al. 1995). Sinks can either increase or decrease the

- 24 - persistence of source populations. They are beneficial if they increase the relative population size of the source or provide extra habitat for mating and foraging (Foppen et al. 2000). They can be detrimental in the sense that individuals from a healthy source are migrating to a poor quality sink. Once in sink habitat there is no further dispersal and low population growth.

1.2.6 Extinction risk Here I have discussed two important determinants of population persistence under metapopulation dynamics theory: patch size and patch isolation. The greater the patch size and the smaller the distance between patches, the greater the probability of persistence as each patch is able to offset the effects of mortality and local extinction. And for the metapopulation as a whole, the extinction threshold determines the minimum habitat amount required for persistence, i.e. below this the metapopulation is most likely to follow a trajectory towards extinction.

1.3 Landscape ecology Landscape ecology is primarily concerned with the relationship between landscape structure and ecological processes. It is a wide and varied field that encompasses many different disciplines, including social sciences, economics, geography and forestry (Turner et al. 2001). Landscape ecology models have been used to study the effects of landscape change on distributions of critical habitat, reserve design, and response to climate change etc. (Carroll et al. 2003a; Opdam & Wascher 2004). The classic models follow the tradition of metapopulation dynamics by assuming a binary landscape with habitat and non-habitat (With 1997). Discrete habitat patches are occupied by the target species and non-habitat forms the matrix in which the habitat patches are located. The matrix is considered to be homogenous and benign such that patch size, inter-patch distance, and patch quality are the only important metrics of connectivity and population persistence. Various measures such as the distribution of patches, fractal geometry, patch shape and boundary effects, are used to quantify landscape structure and the pattern of fragmentation (Turner 1989;

- 25 - Tischendorf & Fahrig 2000). Neutral landscape models have also been used to predict how different patterns of fragmentation affect population dynamics (With 1997).

Alternative approaches consider a more complex, heterogeneous landscape with a blurred distinction between patch habitat and matrix (Fahrig 2003). Instead there is a mosaic of different habitat types or levels of disturbance which affect dispersal rates and dispersal success in different ways (Cushman et al. 2010c). The study area is usually represented as remotely sensed GIS maps which allow quantification of different metrics such as canopy cover, road density and human population density. Because mosaic methods take into account all land cover types they may give a more realistic representation of the landscape and they have been used to determine habitat suitability and distribution for many species (e.g. Carroll & Miquelle 2006; Carroll et al. 2001). Habitat maps can also be linked to demographic data to investigate how changes in landscape structure alter species' distribution, individual movement and population persistence. Recent advances have taken this concept further by attempting to model the landscape as a continuous surface at a scale relative to that at which the target species perceives its environment. In this way standard landscape metrics are represented as continuous variables rather than the traditional method of categorical classification of habitat types (Cushman et al. 2010d).

1.3.1 Population viability More robust estimates of population persistence can be derived by combining complex landscape models with demographic data. In particular, spatially explicit population models (SEPMs) have been developed that link measures of individual survival and fecundity to habitat type (Dunning et al. 1995; Schumaker 1998). Each habitat type is given a relative value based on habitat suitability and/or preference for the target species. A matrix of species' vital rates is also required that is then weighted by habitat quality - for most species it is expected that vital rates will be

- 26 - lower in low quality habitats (Carroll et al. 2003b). Other parameters such as demographic and environmental stochasticity, and dispersal behaviour can also be accounted for. Using the total information on habitat distribution and associated vital rates, SEPMs are then able to track each individual through time to determine population size, distribution and extinction risk, as well as local migration rates (Carroll et al. 2004). Source-sink populations can also be identified by estimating relative reproductive and mortality rates. The landscape can remain static or can be altered to see the effect of changes in land cover composition or configuration, human disturbance or alternative management strategies. This allows critical habitat to be identified, and landscape patterns to be correlated with dispersal success or population persistence (Pulliam et al. 1992; Wiegand et al. 1999; Schadt et al. 2002).

The amount of detailed demographic data required to parameterise these models limits their useful to species for which extensive field data is available. Most studies have been done on birds, butterflies or small mammals as they are easy to track, have short generation times and occur at a manageable landscape scale. Also, because the outcome of SEPMs can be particularly sensitive to input parameters it is recommended that multiple simulations are conducted to account for uncertainty in parameter estimates (Dunning et al. 1995). This is particularly important for dispersal data as this determines how individuals are modelled moving through the system. Also, it is not always possible to survey the entire area of interest so data on habitat preferences and species' distribution is often based on expert opinion or extrapolated from a smaller survey area. It is also important to define the grain (e.g. GIS cell size) and extent (total area) of the landscape so that they are at an appropriate scale for the target species or ecological process that is being measured (Turner 2005; Prugh et al. 2008).

1.3.2 Matrix quality & patch boundaries A second advantage of the landscape ecology approach is that is able to take into account the matrix. Patch size and isolation alone are poor predictors of inter-patch

- 27 - dispersal when the matrix is heterogeneous or if it contains barriers to dispersal e.g. rivers (Schultz & Crone 2001; Tischendorf et al. 2003; Walker et al. 2003; Bender & Fahrig 2005; Prugh et al. 2008). Dispersing individuals have to cross the matrix, and some land types will facilitate dispersal whilst others will prevent it. Population persistence will also be reduced in cases where there are high rates of mortality during dispersal through the matrix (Vandermeer & Carvajal 2001; León-Cortés et al. 2003; Schtickzelle & Baguette 2003; Fahrig 2007). Dispersal mortality rates may be higher for individuals emigrating from smaller patches and vagile species. Smaller patches are expected to have higher emigration rates because dispersing individuals are more likely to encounter the patch boundary (Stamps et al. 1987). And more vagile individuals are more likely to come into contact with roads or other anthropogenic disturbance than less vagile species (Carr & Fahrig 2001).

The boundary between habitat patches and the matrix will also affect population persistence and dispersal rates. Many studies have shown that there are differences in many ecological processes within edge habitat. For example, there may be increased levels of predation or parasitism, and changes in microclimate (e.g. temperature and humidity) or ecological flow (Ries et al. 2004). This edge effect may be evident for varying distances (metres to kilometres) from the edge into the habitat interior (Haddad 1999; Schtickzelle & Baguette 2003; Ries et al. 2004). Patch boundaries will also affect dispersal behaviour depending on their 'permeability'. Soft boundaries are expected to be permeable to dispersing individuals and so subject to much higher emigration rates. They are usually associated with boundaries between habitats that are fairly similar in vegetation type/structure and resource availability. In contrast, hard edges represent an extreme gradient between habitats and may act as a barrier to dispersal (Haddad 1999). For example, some butterfly species have been shown to turn back into a patch at the forest edge rather than enter the matrix (Ries et al. 2004). Hard edges are expected to have the biggest effect on population persistence as they reduce emigration rates and may cause isolation for some populations (Fischer & Lindenmayer 2007). In these cases, dispersal can be

- 28 - increased by improving matrix quality (so improving boundary permeability) or providing corridors or stepping stone habitat between patches (Stamps et al. 1987; Haddad 1999).

1.3.3 Connectivity & corridors Many studies agree that connectivity is crucial for the persistence of a metapopulation as isolation is one of the biggest factors contributing to extinction risk (With 2004). Under metapopulation dynamics theory connectivity is a function of inter-patch distance and patch area. However, it is more difficult to evaluate in complex landscapes where it is also a function of landscape structure, habitat type, fragmentation, habitat preference, dispersal ability, and human influence (e.g. road density and human population density). For most species a reduction in the amount of preferred habitat and an increase in the degree of fragmentation will result in a reduction in dispersal success (With & King 1999). Habitat generalists and highly vagile species may be able to cope with more habitat disturbance and loss of their preferred habitat before local populations become isolated (With & Crist 1995; Crooks 2002; D'Eon et al. 2002; With 2004; de Angelo et al. 2011).

Corridors facilitate movement between habitat patches and as such are important to maintain connectivity. They may be well-defined linear habitat, a diffuse mosaic of different land cover types or an array of stepping stone habitats (Beier & Noss 1998; With 2004; Akçakaya et al. 2006). It is not always clear what constitutes a good corridor from an array of different land types, although it may be possible to track individuals to see which paths they use, which habitats they prefer and which they avoid (Tigas et al. 2002; Cushman et al. 2010a). Other methods include species' distribution maps based on sightings, trapping, tracks, camera trap photographs etc. which can be used to determine habitat suitability (Akçakaya & Atwood 1997; Carroll et al. 2001; Carroll et al. 2003b; Akçakaya et al. 2004; Carroll & Miquelle 2006; de Angelo et al. 2011). Alternatively, corridors can be identified using least-cost path (LCP) analysis using GIS maps in which each pixel is given a relative value based

- 29 - on a particular habitat type or landscape variable (e.g. canopy cover, slope, elevation). The values correspond to the resistance-to-movement for the focal species - a high resistance value implies a higher cost to movement (Adriaensen et al. 2003; Spear et al. 2010). The cost values are set by empirical movement or habitat suitability data, or by expert opinion if empirical data is not available (Wikramanayake et al. 2004; Rabinowitz & Zeller 2010). Various algorithms have been described which are then able to find the path(s) between habitat patches with the lowest cost to the species (Ray 2005). LCP analysis doesn't always suggest a biologically meaningful corridor and is influenced by the landscape variables and cost values included in the model (Sawyer et al. 2011). Improvements to the method include uncertainty analysis or inclusion of multiple corridors to find the best routes (Epps et al. 2007; Beier et al. 2009). The usefulness of corridors can also be assessed by comparing landscapes with corridors versus those without, or patches with connections and those without (Beier & Noss 1998).

1.3.4 Population persistence Landscape ecology is able to extend metapopulation dynamics theory by accounting for landscape heterogeneity. Here, the most important factors for population persistence are considered to be patch quality, matrix quality and connectivity. Patch quality determines population dynamics and vital rates, such that higher quality patches are expected to have higher rates of survival and fecundity. Matrix quality and connectivity influence dispersal rates between patches. For example, a poor quality matrix may result in higher dispersal mortality, and poor connectivity will reduce immigration rates into a given patch. This in turn will increase the extinction risk for that patch; conservation of the population should therefore involve improving patch and matrix quality as well as protecting dispersal corridors between patches.

1.4 Population genetics Genetic studies provide a convenient way to sample individuals where it is difficult to track them by other means. DNA can be obtained from non-invasive samples such

- 30 - as hair or scats which allow populations to be sampled without observing individual animals (e.g. Belant et al. 2005; Taberlet et al. 1997; Adams & Waits 2007; Lucchini et al. 2002). Historically, allozymes and polymorphic microsatellites have been used to ensure enough variation to distinguish between individuals and populations (e.g. Charruau et al. 2011; Dallas et al. 2002; Miththapala et al. 1996). Newer techniques use genomic data to look for differences in both neutral and adaptive variation between populations (Morin et al. 2004; Kohn et al. 2006; Perry et al. 2010). As with metapopulation models extinction risk is linked to small population size and isolation. Population size can be estimated using multi-locus genotypes generated for each sample, and the degree of population isolation can be tested using measures of differentiation and genetic clustering (McRae et al. 2005; Cegelski et al. 2006; Millions & Swanson 2007). Dispersal rates can also be estimated directly using Bayesian methods (e.g. Wilson & Rannala 2003). Many of the assumptions of populations genetics theory, e.g. neutral markers, discrete generations, equal sex ratio, constant population size, often need to be relaxed for natural populations. However, by ensuring large sample sizes and repeated sampling, the effects of these uncertainties can be reduced.

1.4.1 Population size

In genetic terms the effective population size (Ne) is the size of an idealised population that would produce the same genetic properties as that observed for the real population under consideration, i.e. the number of individuals that actually contribute gametes to the next generation (Cabellero 1994; Falconer & MacKay 1996; Hedrick 2005b). In contrast, the total population size (N) is the total number of individuals in a population. The ratio of Ne:N is usually in the order of 0.10-0.11

(Frankham 1995). A variety of different factors act to reduce Ne relative to total population size, e.g. variation in reproductive success, population bottlenecks (Wang & Caballero 1999; Frankham et al. 2002). It is a particularly important measure for isolated populations as it determines the rate of loss of genetic variation due to genetic drift. It can be measured in a number of different ways including temporal

- 31 - changes in gene frequencies, linkage disequilibrium, heterozygosity excess, sibship frequency and Bayesian methods (Valière et al. 2002; Leberg 2005; Wang 2005; Tallmon et al. 2008; Wang 2009). The total population size can also be estimated from genetic data using the cumulative number of unique individuals (Kohn et al. 1999; Eggert et al. 2003) or mark-recapture analysis (Eggert et al. 2003; Bellemain et al. 2005; Mondol et al. 2009a).

Genetic variation The most commonly used measure for genetic variation is heterozygosity, the proportion of heterozygous individuals at a locus within a population. Genetic variation has been linked to adaptability, extinction risk and fitness, with the expectation that a higher level of heterozygosity enhances a population’s probability of persistence over time (Frankham 1998; Brook et al. 2002). Small populations lose genetic variation because of the effect of genetic drift (Cabellero 1994; Falconer & MacKay 1996). Genetic drift refers to the random sampling of genes between generations due to the fact that not all individuals will mate successfully and produce offspring. This leads to a loss of genetic variation at a rate given by:

t " 1 % $1! ' (1.5) # 2Ne & where Ne is the effective population size and t is the number of generations. The amount of drift increases as effective population size decreases, i.e. smaller populations lose genetic variation at a faster rate than larger ones and thus have a higher risk of extinction (Schwartz et al. 2003; Haag et al. 2010).

Inbreeding Small populations are also expected to have lower levels of genetic variation due to inbreeding. Inbreeding occurs because of mating between closely related individuals and is more likely in small populations because of the reduced availability of unrelated individuals. Inbreeding may reduce population fitness through many

- 32 - mechanisms including increased homozygosity, reduced environmental adaptation, and inbreeding depression, which occurs because of the exposure of deleterious recessive genes (Reed & Frankham 2003; Gaggiotti 2004). It has been reported in many different species and manifests as reduced adult/juvenile survival, reduced reproductive output, phenotypic deformities, and increased disease susceptibility (e.g. Coltman et al. 1998; Hale & Briskie 2007; Wildt et al. 1982; Roelke et al. 1993; Crnokrak & Roff 1999; Keller & Waller 2002).

1.4.2 Isolation In genetic studies population isolation is measured via estimates of gene flow. Gene flow refers to the movement of genetic material between local populations i.e. the movement of individuals or gametes and subsequent mating (Avise 1994; Mills & Allendorf 1996; Hedrick 2005b; Allendorf & Luikart 2007; Hamilton 2009). In the same way that dispersal reduces extinction risk, gene flow facilitates genetic rescue by providing new genetic material and by increasing effective population size (Gaggiotti 2004). In this way it counters the effect of genetic drift. In the absence of gene flow, local populations lose genes under the influence of genetic drift. Each population will retain a different set of genes and so they become differentiated. Population isolation can thus be measured by estimating migration rate (gene flow) or the degree of population differentiation.

Population differentiation

The classic measure of differentiation is Wright's FST (Falconer & MacKay 1996;

Futuyma 1998; Hedrick 2005b; Balding et al. 2007). FST provides a measure of population divergence and can be estimated from the heterozygosity or variance in gene frequencies among subpopulations using neutral genetic markers (Hedrick 2005a; Allendorf & Luikart 2007):

2 ! p F = (1.6) ST p(1" p)

- 33 - 2 where " p is the variance of frequencies of allele p among subpopulations and p(1 - p) is the maximum possible variance given the observed mean allele frequencies in the population. Values of FST range from 0 (no differentiation) to 1 (complete differentiation).

The amount of divergence represents a balance between gene flow and genetic drift in isolated populations. Isolated populations with high FST are expected to have a higher risk of extinction because genetic drift is the predominant force acting to reduce genetic variation. Calculation of FST is sufficient for most populations, but there are a number of factors that may reduce its effectiveness, e.g. the use of highly variable markers, alternative mutation models, or multiple alleles. Several alternatives such as RST and $w have been proposed to account for this, although FST is still the most reliable measure when sample sizes are small or few markers are used (Nei 1973; Weir & Cockerham 1984; Slatkin 1995; Michalakis & Excoffier 1996; Goodman 1997; Gaggiotti et al. 1999; Hedrick 2005a; Excoffier 2007; Jost 2008).

Migration rate A direct measure of gene flow also indicates the degree of population isolation. The exact method used will depend on the gene flow model under consideration (Avise 1994; Mills & Allendorf 1996; Hedrick 2005b; Allendorf & Luikart 2007). Under the continent-island model, gene flow occurs between a single large continent and a smaller island. This is similar to the source-sink model in which dispersal is unidirectional from source to sink, and the persistence of the sink is dependent on immigration from the source. Under the infinite island model immigrants may come from any population with equal probability, and migration rate is the same for all populations. This is comparable to Levins' metapopulation model in that all populations are considered to be equal. Finally, under the stepping-stone model gene flow is restricted to neighbouring populations. This is similar to the nearest-

- 34 - neighbour measure of patch isolation in that population divergence is a function of the distance between populations.

FST can be used to estimate migration rate under the infinite island model by:

1 " 1 % (1.7) Ne m = $ ! 1' 4 # FST & where Nem is the effective number of migrants per generation. Migration rates can also be estimated using assignment methods or Bayesian algorithms to identify migrants and asymmetrical migration between populations (Rannala & Mountain 1997; Cegelski et al. 2003; Wilson & Rannala 2003; Paetkau et al. 2004; Dixon et al. 2006).

1.4.3 Extinction risk The key genetic qualities for the persistence of a subdivided population are high genetic variation and migration. Populations with high levels of heterozygosity are more likely to have the necessary environmental adaptations for survival, and are likely to have higher levels of fitness, e.g. better survival, fecundity, and juvenile growth rates (Ralls et al. 1988; Frankham & Ralls 1998; Saccheri et al. 1998; Bijlsma et al. 2000; Reed & Frankham 2003). Migration increases the relative size of a population and counteracts the effects of genetic drift, which acts to reduce variation. In the absence of migration, there is both an increase in inbreeding and differentiation. There is a far greater effect of isolation in small populations. Thus, small populations with a low level of heterozygosity and low migration rates have the highest risk of extinction.

1.5 Landscape genetics The field of landscape genetics emerged out of a desire to combine the landscape- based approach of landscape ecology with empirical genetics data (Manel et al. 2003; Storfer et al. 2007; Holderegger & Wagner 2008). It is most useful for

- 35 - populations in which individuals are distributed across a landscape and their population structure is not immediately obvious. Spatial genetic patterns are assessed at the individual level without defining populations in advance, and these patterns are then correlated to landscape variables. As with previous metapopulation concepts population persistence is dependent on dispersal - both the dispersal ability of the species (e.g. dispersal distance), and landscape structure (barriers or resistance to movement). Landscape genetics allows the study of effective dispersal (animal movement plus successful breeding) in relation to landscape structure. It is particularly beneficial for studies in which movement data is lacking or where direct tracking is not possible.

1.5.1 Isolation by distance The simplest landscape genetics model is the null model of isolation by distance. Under this scenario, genetic differentiation (or genetic distance) is a function of the geographic distance between individuals or populations alone (Hardy & Vekemans 1999). There is no direct effect of landscape structure. The most common tests for a statistically significant relationship are the Mantel test or linear regression of genetic distance versus geographic distance (e.g. Broquet et al. 2006; Ernest et al. 2003; Trizio et al. 2005). Genetic structuring can also be tested for using spatial autocorrelation which tests for a correlation at a much finer scale (Waser & Elliott 1991; Smouse & Peakall 1999; Arnaud 2003; Peakall et al. 2003). The x-intercept of the resulting correlogram can be used to estimate the distance over which the spatial genetic relationship exists, or alternatively it can be interpreted as the diameter of a neighbourhood patch (Epperson 1990; Diniz Filho & De Campos Telles 2002; Epperson 2005; Vignieri 2005). Species with greater dispersal capabilities are expected to have a much wider spatial genetic range, and as a consequence are expected to have a much higher probability of persistence in fragmented habitats (Hoehn et al. 2007).

- 36 - Spatial genetic structure can also be used to estimate the size of a population neighbourhood, a measure equivalent to the effective population size. Under the classic isolation by distance model individuals naturally cluster into mating neighbourhoods, the size of which is a function of dispersal distance; 4#D%2, where D is the population density and %2 the variance in parent-offspring axial dispersal distance (Wright 1946; Crawford 1984; Rousset 2007). There is a much higher probability of individuals mating with other individuals within the same neighbourhood (Falconer & MacKay 1996; Hamilton 2009). A neighbourhood is therefore equivalent to a local population in metapopulation terms, and the size of a neighbourhood (Nb) determines the genetic properties of the population in the same way as effective population size. Certain measures of genetic distance, e.g.

Rousset's a for individuals or FST/(1-FST) for populations, are commonly used to estimate neighbourhood size (Rousset 1997; Rousset 2000; Sumner et al. 2001; Hardy & Vekemans 2002; Watts et al. 2007).

1.5.2 Barriers to dispersal Landscape heterogeneity can affect the null pattern of dispersal and introduce further population structure. Populations that are isolated will receive less immigrants and are expected to experience higher rates of genetic drift and levels of differentiation. Standard differentiation measures (e.g. FST) can be used to estimate migration rates (Nm), but there are newer Bayesian methods which are able to detect genetic structure by grouping individuals into clusters of random mating individuals (Pritchard et al. 2000; Dawson & Belkhir 2001; Corander et al. 2003; Guillot et al. 2005; François et al. 2006). Bayesian methods cluster individuals into populations or genetic groups assuming Hardy-Weinberg and linkage equilibrium, and that the separation between clusters is created by impediments to dispersal. The genetic clusters can then be visualised on a GIS habitat map to try to identify the landscape features that may be reducing dispersal, e.g. water bodies, human developments or roads (e.g. Coulon et al. 2008; Latch et al. 2008; Millions & Swanson 2007; Goossens et al. 2005; Riley et al. 2006). Bayesian methods can also

- 37 - be used to directly infer migration rates (Beerli & Felsenstein 1999; Wilson & Rannala 2003). Analysis and interpretation must be conducted with care because of the time lag between creation of the dispersal barrier and the subsequent genetic effects. For some species there is a short delay, but for others the current genetic structure may be the result of historic land change (Holzhauer et al. 2006; Clark et al. 2010; Landguth et al. 2010).

A significant correlation between genetic distance and landscape variables can be tested for using a partial Mantel statistic (Smouse et al. 1986). This is an extension of the classic Mantel test that is able to look for a relationship in the presence of spatial autocorrelation, i.e. it tests for a correlation between genetic distance and environmental variables whilst controlling for geographic distance. It is also a useful method to test between alternative causal models (Legendre & Troussellier 1988; Cushman et al. 2006; Cushman & Landguth 2010). Many studies have used partial Mantel tests to identify the important landscape elements (e.g. landscape barriers, elevation, roads, slope, land cover class/habitat type, snow cover) that shape dispersal patterns (Keyghobadi et al. 1999; Spear et al. 2005; Cushman et al. 2006; Epps et al. 2007; Schwartz et al. 2009). Some authors have questioned the validity of using partial Mantel tests for landscape genetic analysis and several alternatives have been suggested, such as canonical correspondence analysis and Bayesian inference (Angers et al. 1999; Faubet & Gaggiotti 2008; Balkenhol et al. 2009).

1.5.3 Functional connectivity Another way to model dispersal is to map the paths that the target species may take to traverse the landscape. Similar to landscape ecology methods, the landscape is represented as a resistance map with each feature of the landscape, e.g. habitat type, slope, given a relative value based on its resistance-to-movement for a given species (Adriaensen et al. 2003). A higher value represents a greater cost to individuals and a lower probability of movement. In a GIS framework, the simplest methods find the best path between each pair of source populations that

- 38 - accumulates the lowest cost, i.e. the least-cost path (e.g. Ray 2005; Sawyer et al. 2011). More complex models compute corridors by representing the lowest 10% least-cost paths (LCP) or total resistance between populations by considering the total number and width of all possible paths linking them (McRae 2006; McRae & Beier 2007). The total length/resistance of the path often correlates better with genetic distance than standard straight-line geographic distance (Michels et al. 2001; Arnaud 2003; Spear et al. 2005; Vignieri 2005). The main limitation of these techniques is that they are dependent on accurate representation of landscape resistance. This is often based on expert opinion and is rarely supported by empirical movement/telemetry data (Wikramanayake et al. 2004). The outcome of the models are also dependent on the grain and scale of analysis, and they are particularly sensitive to parameter uncertainty. Ideally, the GIS pixel size should be relative to the species' perception of the landscape, and sensitivity analysis should be conducted with different resistance and habitat values (Beier et al. 2009). Other criticisms of LCP methods are that a long, single (pixel-wide) path is unrealistic and not necessarily biologically relevant, and that they cannot account for individual behaviour such as the reaction to nearby disturbance (Sawyer et al. 2011). Robustness can be improved by incorporating multiple paths and by validating the model outcome with empirical data e.g. resource selection functions or available movement data (Epps et al. 2007). Alternative connectivity methods model dispersal patterns across a heterogeneous landscape given certain migration rates, habitat quality, mating success etc. (e.g. Landguth & Cushman 2010; Currat et al. 2004; Balloux 2001).

1.5.4 Population persistence & connectivity Landscape genetics attempts to explain observed population genetics patterns using landscape variables. As with metapopulation dynamics, landscape ecology and population genetics, connectivity/gene flow is the key to maintaining population structure and reducing extinction risk. Populations that have high levels of connectivity are more likely to persist than those that are isolated. Under the null

- 39 - model, genetic differences between individuals can be explained by geographic distance alone (close individuals are more related than distant ones). More complex models can identify dispersal barriers such as roads or water bodies that are causing fragmentation of a population. Least-cost path and circuit theory analyses use habitat preferences for a given species to determine which populations are connected and which are at risk of isolation. Once this information is known, management units or corridors can be designated and protected.

1.6 Overall summary Although metapopulation dynamics, landscape ecology, population genetics and landscape genetics have slightly different ways of approaching the problem of fragmented populations, there are some areas of overlap. They all agree on 2 key determinants of population persistence: (i) Patch area/population size and (ii) Connectivity/migration rate. Patch size is often used as a proxy for population size such that larger patches are expected to support larger populations. Large populations have a much higher probability of persistence than smaller ones. Small populations are more susceptible to demographic, environmental and/or genetic stochasticity. They are more likely to experience Allee effects and to have low genetic variation or inbreeding. Connectivity is recognised as a key determinant for population persistence as it determines the rate of immigration or colonisation for a given population. Immigration is required to maintain population dynamics, to increase genetic variation and to reduce the effects of stochasticity. Isolated populations with low levels of connectivity are expected to decline to extinction much faster than those that have a consistent flow of immigrants. In the following section, I try to show how the concepts of population size and connectivity have been applied to the conservation of wild carnivores, using the Sumatran tiger as a case study.

- 40 - 1.7 Case study - the Sumatran tiger (Panthera tigris sumatrae) The Sumatran tiger provides a good model to test the use of population genetics and landscape genetics to assess the population viability of a critically endangered species. Sumatra has been subject to very high rates of deforestation in the past 3 decades, and it is likely that this will continue until all lowland and submontane forest has been converted to other land uses. It is expected that wild tigers now exist as a metapopulation that is largely limited to protected areas. This means that the principles of metapopulation dynamics and landscape ecology apply. However, as there is limited demographic and movement data available for wild Sumatran tigers, I have opted to use genetic sampling to determine effective population size, genetic variation and population structure as a measure of extinction risk.

1.7.1 Sumatra Geography and climate Sumatra is the sixth largest island in the world with an area of ~ 470,000km2 and approximately 1790km along its axis (Santiapillai & Ramono 1987). It is home to >50 million people spread across 8 provinces (Aceh, North Sumatra, West Sumatra, Jambi, Riau, South Sumatra, Bengkulu and Lampung). The highest population densities are found in the central parts of the island with an overall growth rate of ~2.3% since 2000 (Sub Directorate of Statistics Indicator 2010). A history of volcanic activity means that most of the island contains very fertile soil that supports a broad range of vegetation types. Up to 17 endemic plants are found here, with Rafﬂesia arnoldii and Amorphophallus titanum the most famous. Overall, there are 5 main bioregions represented on Sumatra: freshwater swamp in the eastern alluvial plains, lowland rainforest (<1000m), montane rainforest (>1000m) along the western Bukit Barisan mountain range, peat swamp along the eastern coast, and tropical pine forest near Lake Toba in the north and along the Bukit Barisan mountain range in the west (Whitten et al. 2000; Wikramanayake et al. 2002). The climate of Sumatra is much like the rest of tropical Asia and is characterised by heavy rainfall and high humidity. Average rainfall ranges from ~1200 to over 3000mm/yr with up to

- 41 - 86% humidity (Sub Directorate of Statistics Indicator 2010). However, wet and dry seasons are not as well defined as in other parts of Indonesia or Asia.

Biodiversity Sumatra is also one of the richest islands in the world for biodiversity. It is second only to New Guinea for the number of mammals (200 species) and birds (580 species); 9 mammal species are endemic to the mainland (e.g. Sumatran rabbit, Aceh squirrel, Sumatran shrew-mouse) and 14 endemic to the Mentawi islands off the west coast of Sumatra (Whitten et al. 2000). Sumatra is also host to 15 other species found only in Indonesia (e.g. the orangutan), and 22 species that are found nowhere else within Indonesia (e.g. Dayak fruit bat, White-handed gibbon, Siamang, Hog badger, Smooth-coasted otter, Asian golden cat, Tapir, mountain goat). In addition, it has populations of several mammals which are virtually extinct elsewhere in Indonesia, such as the Sumatran rhino, Sumatran elephant, Sumatran tiger, and dhole.

1.7.2 Sumatran tigers The Sumatran tiger (Panthera tigris sumatrae) was first described in 1929 and is one of 9 subspecies that were found across Asia (Pocock 1929). It has long been recognised as a distinct subspecies, not only because of its location, but because it is distinguishable by both genetic and morphological analysis (Cracraft et al. 1998; Kitchener 1999; Luo et al. 2004; Kitchener & Yamaguchi 2010). Three island subspecies were once found in Indonesia, but now only the Sumatran tiger remains. Current estimates put the global tiger population at ~3-4000 with up to 500 individuals thought to remain on Sumatra; this includes ~400 individuals in protected areas with ~100 occurring in other forest (Tilson et al. 1993; Linkie et al. 2008c; Chundawat et al. 2011). Tigers can inhabit a broad range of forest types and their distribution is largely influenced by human population density, road density and prey distribution (Miquelle et al. 1999; Seidensticker et al. 1999; Kinnaird et al. 2003;

- 42 - Uphyrkina & O’Brien 2003; Uphyrkina & O’Brien 2003; Carroll & Miquelle 2006; Johnson et al. 2006; Linkie et al. 2008a).

They are wide-ranging carnivores that occur at low densities (1-3 tigers/100km2) in much of their range (Santiapillai & Ramono 1987; O'Brien et al. 2003). Highest densities are found in lowland forest which is far more productive, and can support more ungulate prey, than montane forest or peat swamp. Adult males have much larger home ranges than females with dispersal distances dependent on habitat type (Goodrich et al. 2010; Barlow et al. 2011). Both males and females disperse at ~18 months, but females usually exhibit natal philopatry with mothers often donating some territory to their daughters. In contrast, subadult males have to disperse over much longer distances to find vacant territories (Smith et al. 1987; Smith 1993; Goodrich et al. 2010). First reproduction is at 3-4 years, and in good quality habitat tigers are able to breed successfully up to ~12 years. Females will have litters of 2-3 cubs with an inter-birth interval of 21-24 months (Tilson & Seal 1987; Sunquist et al. 1999; Kerley et al. 2002; Linkie et al. 2006). Population growth rates are linear and range from 0.06 tigers/year in the wild up to 7 tigers/year in captivity (Ballou & Seidensticker 1987; Smirnov & Miquelle 1999).

1.7.3 Threats Habitat loss & fragmentation It is likely that tigers were once distributed across much of mainland Asia and the Sunda region mirroring the distribution of preferred habitat and prey species (Seidensticker et al. 1999; Kitchener & Dugmore 2000). However, high rates of deforestation mean that tigers now occupy just 7% of their former range (Sanderson et al. 2006). As elsewhere in Asia, habitat loss and fragmentation in Sumatra is being driven by human disturbance - agricultural expansion, industrial plantations, logging, road construction and urban development (Geist & Lambin 2002; Kinnaird et al. 2003; Gaveau et al. 2009b; Suyadi 2010). The human population on Sumatra has been growing steadily since the 1970s when the government first introduced a

- 43 - transmigration scheme to relocate people from the overpopulated nearby island of Java (Imbernon 1999). Lampung province was the first place they settled in, and the lowland forest here has largely been removed or converted to other land types. Riau is also a productive area that is now subject to high rates of human migration and land cover conversion (Uryu et al. 2008; Sub Directorate of Statistics Indicator 2010; Broich et al. 2011). Lowland forest is by far the most severely affected by deforestation as it is both accessible and relatively unprotected (Linkie et al. 2010). Rates of up to 2.6% forest loss per year have been reported, with rates of up to 1% within protected areas (Gaveau et al. 2009a). Even protected areas have been affected, although the rates of habitat loss are lower within their boundaries (Gaveau et al. 2009a).

Prey depletion One of the key determinants for tiger abundance is prey availability (Miquelle et al. 1999; Karanth et al. 2004). Tigers will preferentially predate large ungulates such as sambar deer, smaller deer species (e.g. Axis deer, muntjac, hog deer), wild boar, and small primates such as macaques (Stoen & Wegge 1996; Biswas & Sankar 2002; Franklin 2002; Reddy et al. 2004; Maheshwari 2006). Prey abundance is generally low where human population density is high, although some studies suggest that the edge habitat created in the early stages of land conversion may suit some species such as the muntjac or tapir (Sunquist et al. 1999; O'Brien et al. 2003). The effects of prey depletion on tiger population dynamics have not been widely studied, but the correlation between prey availability and tiger abundance suggests that landscapes with low prey numbers will also contain declining tiger populations (Karanth & Stith 1999; Dinerstein et al. 2006).

Human-tiger conﬂict/Poaching Tiger mortality in the wild is largely due to human-tiger conflict and poaching. Common causes for human-tiger conflict are habitat disturbance and low habitat availability, poor livestock husbandry, and the presence of people in tiger habitats

- 44 - (Michalski et al. 2006; Inskip & Zimmermann 2009). Human-tiger conflict is most common at the boundaries of national parks or in disturbed habitats where there is an overlap between tiger and human activities (Sekhar 1998; Nyhus & Tilson 2004a; Nyhus & Tilson 2004b; Johnson et al. 2006). Individual tigers may attack and occasionally kill people that enter into forested areas, or have taken livestock from villages, and this makes them a target for retaliatory killing, e.g. by poisoning or shooting. Encroachment of people into tiger habitat also means that tigers may be killed indirectly by snares set for other species such as muntjac deer (Shepherd & Magnus 2004). Tiger poaching is still prevalent in many parts of central and southeast Asia for skins, bones and other body parts which are worth a lot of money in the illegal wildlife trade (Nowell 2000). Law enforcement is often ineffective or infrequent so there is little incentive for poachers to stop (Linkie et al. 2003). If left unchallenged however, the high mortality rates due to poaching may be the biggest threat to tiger survival, especially in small populations or if adult female survival is low (Linkie et al. 2006; Chapron et al. 2008; Goodrich et al. 2010).

1.7.4 Population viability Metapopulation structure Tiger conservation efforts have traditionally centred around Tiger Conservation Units (TCUs), Tiger Conservation Landscapes (TCLs), and more recently source sites (Wikramanayake et al. 1998; Sanderson et al. 2006; Walston et al. 2010). Sumatra alone holds 12 TCLs and 8 source sites covering up to 88,000km2 (Wibisono & Pusparini 2010). The boundaries of TCUs and TCLs were designated using GIS analysis to identify remaining tiger habitat that had some level of conservation/ protection, and that was capable of supporting reasonable populations (Dinerstein et al. 2006). Today, these conservation units largely correspond to protected area boundaries which are the only regions that still contain native habitat, although tigers are found in other forested areas (Wibisono & Pusparini 2010). Conservation of the metapopulation is aimed at preserving what tiger habitat remains in an attempt to save as many individuals as possible.

- 45 - Population persistence & connectivity Although many studies have estimated abundance and population density from camera trap or other survey data, few have directly estimated extinction risk or population viability (Tilson & Brady 1992). Those that have show that the probability of extinction risk can be linked to poaching pressure, prey availability or human disturbance. However, the inclusion of metapopulation theory or landscape factors often helps to explain some of the variation in tiger abundance. For example, if poaching pressure is high in small populations, neither protection of habitat nor increase in prey numbers is sufficient to prevent population decline (Linkie et al. 2006; Chapron et al. 2008). Even if poaching is then brought under control, some populations will still follow a trajectory of decline to extinction (Kenney et al. 1995). Changes in extinction risk relative to land use and prey dynamics has also been modelled using individual-based models (Ahearn et al. 2001; Cramer & Portier 2001). There is a higher probability of persistence in tropical forest or logging concessions, with the lowest probabilities in human settlements or oil palm plantation (Imron et al. 2011). Population persistence can also be affected by the quality of the matrix surrounding protected areas (Baeza & Estades 2010). This is not surprising given that the matrix can affect patch population dynamics and dispersal mortality within heterogeneous landscapes (Vandermeer & Carvajal 2001; Fahrig 2007). Human population density is increasingly recognised as a key factor affecting extinction risk and should also be considered in any population viability analysis (Cardillo et al. 2004; Carroll & Miquelle 2006). Low tiger abundance is often correlated with areas of high human population density or human disturbance (e.g. Linkie et al. 2006; O'Brien et al. 2003; Johnson et al. 2006).

Most studies of carnivore connectivity rely on least-cost path (LCP) or corridor analysis using GIS maps and information on habitat preference (e.g. Epps et al. 2007; Kautz et al. 2006). Resistance values are set for each habitat type or land use based on expert opinion and previous survey data. This information is then used to

- 46 - determine the most likely path through the landscape which minimises movement costs (e.g. Rabinowitz & Zeller 2010). LCP analysis has been applied to tiger populations (Wikramanayake et al. 2004; Joshi 2010), but not yet to the Sumatran tiger. A newly proposed Tiger Corridor Initiative seeks to create a corridor in central and southeast Asia to link existing TCLs (Panthera). However, this is a challenging concept given the international, governmental and regional co-operation that would be needed to create and protect a corridor of this scale.

1.7.5 Genetics Population size

Small values of effective population size (Ne < 50) and total population size (N < 500) are often quoted as being detrimental to a population's survival. In particular, they are thought to increase the risk of inbreeding depression and extinction due to demographic/environmental stochasticity (Avise 1994; Primack 2002). However, many natural populations are able to persist with much lower population sizes, and the tiger is no exception (Palstra & Ruzzante 2008). The global tiger population has been estimated at ~4000 individuals, with India the only country with a regional population of >500 tigers (Chundawat et al. 2011; GTRP 2011). Effective population size also appears to be low in tigers with previous estimates of <30 individuals for the Bengal and Amur subspecies, and Ne:N ratios as low as 0.03-0.07 (Smith & McDougal 1991; Smith 1993; Henry et al. 2009; Alassad et al. 2011). This compares well to estimates in other carnivore species which vary from Ne = 8-14 in the ocelot up to Ne = 80-100 in brown bears (Miller & Waits 2003; Tallmon et al. 2004; Aspi et al. 2006; Jane&ka et al. 2008; Schwartz et al. 2009). Ne:N ratios are difficult to estimate accurately and have only been published for a few felid species with a range of 0.25-0.50 (Nowell & Jackson 1996). To date there have been no formal estimates of effective population size for the Sumatran tiger, but it is expected that they would be in line with those derived for other tiger subspecies. It also remains to be seen what effect this is likely to have had on their genetic variation. For example, small population sizes and the resultant inbreeding have been linked to reproductive

- 47 - disorders in free-living cats (Wildt et al. 1987; Barone et al. 1994; Munson et al. 1996). However, no such problems have been reported in tigers yet.

Genetic variation The Sumatran tiger may still retain a reasonable amount of genetic variation despite its isolation from mainland Asia (Luo et al. 2004). It has higher levels of heterozygosity compared to the Amur subspecies, although there are signs of genetic drift and population differentiation when compared to the Bengal and Indochinese (Henry et al. 2009; Mondol et al. 2009b). Again, in comparison to other felid species, they do not appear to have experienced any severe loss of genetic variation or inbreeding (Table 1.1).

Table 1.1 Mean levels of heterozygosity for tigers and other felids estimated with domestic cat Fca microsatellite loci. Marker panels were not identical between studies but showed some degree of overlap between the loci used.

Felid Number of Number of Expected Observed Reference individuals loci heterozygosity heterozygosity Sumatran tiger 16 30 0.47 ± 0.02 0.49 ± 0.04 Luo et al. 2004

Bengal tiger 73 5 0.70 ± 0.16 - Mondol et al. 2009b Indochinese tiger 33 30 0.64 ± 0.02 0.67 ± 0.03 Luo et al. 2004

Amur tiger§ 95 8 0.26 ± 0.11 - Henry et al. 2009 Jaguar 42 29 0.74 - Eizirik et al. 2001 Leopard 17 25 - 0.78 ± 0.15 Uphyrkina et al. 2002 Clouded leopard 15 51 0.47 ± 0.02 0.67 ± 0.03 Buckley-Beason et al. 2006 Lion 60 51 0.39 ± 0.01 0.67 ± 0.03 Buckley-Beason et al. 2006 Puma 431 11 0.42-0.45 - Ernest et al. 2003 Snow leopard 9 7 0.51-0.75 0.44-0.58 Jane&ka et al. 2008

§6HDZ microsatellite loci (WIlliamson 2002) were used in this study.

- 48 - Population structure The majority of genetic studies in tigers have attempted to structure the global population into subspecies clusters (e.g. Luo et al. 2004). Other studies have attempted to clarify local population structure by estimating pairwise differentiation values (FST) for neighbouring protected areas (e.g. Joshi 2010; Henry et al. 2009). Analysis of sequence variation in mitochondrial DNA genes has also been used to identify population structure in Indian tigers (Sharma et al. 2009; Joshi 2010; Sharma et al. 2010). Similar genetic studies are lacking for the wild Sumatran tiger population.

Other genetic techniques that could be used are testing for isolation by distance (IBD), Bayesian clustering, and landscape genetics analyses. Many species that occur at low population densities and disperse over large distances show IBD. Under this model, genetic distance is a function of dispersal distance and it is positively correlated with geographic distance, such that individuals that are further apart are also genetically differentiated (e.g. McRae et al. 2005; Sacks et al. 2004; Pope et al. 2006; Ruiz Garcia 2003). Wide-ranging species are also more likely to encounter anthropogenic barriers such as roads or swaths of human disturbance that will interrupt their natural pattern of dispersal (e.g. Cegelski et al. 2003). This causes a discontinuity in the spatial distribution of genes that can be detected by Bayesian methods. Individuals can thus be clustered into discrete groups based on their genotypes (e.g. Millions & Swanson 2007; Manel et al. 2004; McRae et al. 2005). More recent landscape genetics analyses have attempted to correlate genetic differences with landscape features. Using a similar GIS framework to LCP analysis, the landscape is coded as a resistance surface with each habitat variable given a value relative to its cost to movement for a given species. Cumulative resistance between source populations or sampling points can then be used in place of straight- line geographic distance in isolation by resistance tests (e.g. Castilho et al. 2010; Schwartz et al. 2009). Alternative methods, such as distance-based redundancy

- 49 - analysis, have also been used to look for the environmental variables that significantly influence genetic structure (Geffen et al. 2004; Pilot et al. 2006).

1.7.6 Further study It is clear from the summaries above that there is still a great deal that is unknown about the Sumatran tiger and its population viability in the wild. Although limited to designated TCLs, it is unclear if these habitat patch boundaries correspond to actual population boundaries and what the dispersal dynamics between them are. Noninvasive sampling and microsatellite analysis will allow us to determine effective population size, levels of genetic variation and differentiation for each TCL, and the amount of dispersal (gene flow) between them. It is also possible to test for isolation by distance and to use this relationship, if it exists, to estimate neighbourhood size. Bayesian clustering and available satellite imagery from 2000-2008 would allow some analysis of the influence of land cover class on dispersal. Ideally, this landscape genetics analysis would be extended to include different time series maps from the 1970s to the present day to determine which was the crucial period (and which are the critical land cover types) shaping the spatial genetic pattern we see today. Combining this genetic information with other metapopulation and landscape ecology studies would hopefully provide a more complete picture of metapopulation structure and population persistence for Sumatran TCLs.

- 50 - Thesis outline In the following chapters I explore the use of population genetics and landscape ecology to determine extinction risk for the tiger population on Sumatra.

Due to the challenges involved with using faecal DNA I first consider the implications of sampling and genotyping errors on commonly applied conservation genetics analyses. In Chapters 2 and 3 I present simulated datasets to test the effect of sampling error, allelic dropout, false alleles and missing data on measures of allelic diversity (NA), expected heterozygosity, pairwise FST and STRUCTURE analysis.

Having determined potential thresholds for sample size and error I then explore the empirical data in Chapters 4 to 6. I first outline the study sites, sample collection and laboratory methods used in this study. In particular I explain the use of detection dogs, DNA extraction and genotyping methods. I also provide estimates for rates of genotyping error and discuss the utility of noninvasive surveys for conservation. In Chapter 5 I use the multilocus genotypes to estimate genetic variation, effective population size, regional differentiation, and gene flow. These results suggest the potential for genetic source and sink populations on Sumatra. I then use circuit theory to investigate the potential for connectivity between Tiger Conservation Landscapes in Chapter 6. I also test for evidence of population structure using isolation-by-distance and Bayesian clustering techniques such as STRUCTURE. Finally, I use the relationship between genetic distance and geographic distance to estimate neighbourhood size for tigers. In the final chapter I present a summary of the key findings from my thesis and discuss the implications for tigers and other landscape species. I then make some recommendations for future research and expansion of the current study.

- 51 - Chapter 2

Sampling effects in Conservation Genetics Studies

- 52 - 2.1 Abstract Noninvasive samples are a key source of DNA for many wildlife conservation studies. Samples such as hair or faeces are frequently deposited in the environment allowing multiple individuals to be detected. However, in many cases a large number of specimens is reduced down to a small group of individuals during genetic analysis. This may mean that there is insufficient statistical power for researchers to detect significant differences in genetic diversity or population differentiation between groups. Here I investigate the effect of sample size on estimates of genetic variation, population differentiation and population structure. I found that sample size had little effect on measures of expected heterozygosity, pairwise FST and population structure. However, small sample size caused a large downward bias in estimates of allelic diversity. These results suggest that researchers faced with studying low density, wide-ranging species have sufficient power to determine the genetic status of wild populations from a small number of individuals.

2.2 Introduction Noninvasive samples are increasingly used in wildlife studies to determine the conservation status of endangered populations. In particular, they offer the opportunity to study wide-ranging or elusive species without the need to trap and handle individuals. DNA derived from noninvasive sources has allowed the estimation of genetic variation, population structure and population size for many species (e.g. Bellemain et al. 2005; Okello et al. 2005; Mowat & Paetkau 2002; Prigioni et al. 2006; Sloane et al. 2000). In the last few years sample types have expanded from hair and faeces to include a variety of other sources such as urine, feathers, and saliva (Palsbøll et al. 1997; Valière & Taberlet 2000; Segelbacher 2002; Sundqvist et al. 2008). These are frequently shed or deposited into the environment and are relatively easy to collect without specialist equipment. Sample encounter rates can also be increased by using detection dogs, targeting den sites, using hair traps, bait lures, etc (Woods et al. 1999).

- 53 - Despite their many advantages, there are also certain disadvantages that make the use of noninvasive samples particularly challenging. For example, they may be difficult to detect in some environments such as carnivore scats in heavy leaf litter, and surface DNA is quickly degraded in samples that have been exposed for a long period of time (Piggott 2004; Santini et al. 2007). Sample quality is often much lower in tropical climates where high temperates and high humidity encourage the activity of exo- and endonucleases which degrade DNA further (Farrell et al. 2000; Hájková et al. 2006). Whilst having multiple samples from the same individual is beneficial for capture-recapture estimates of population size, genetic recaptures mean that a larger number of specimens is needed to target an adequate proportion of the total population. As many as 49 samples may be collected per individual greatly increasing the survey effort required to enlarge the sample size (Prugh et al. 2005). There may also be the confounding effect of sympatric species. For many mammals such as medium-sized felids, there are not enough differences in sample morphology to be able to visually distinguish between them. Discarding non-target samples will further reduce the proportion of collected specimens that can be used for later analysis. More challenges arise during genetic analysis as not all extracts will contain sufficient quantities of amplifiable DNA to construct a multilocus genotype. Consequently, although the initial number of samples collected in the field may be large, genetic analysis will ultimately be conducted on a much smaller number of unique genotypes (Borthakur et al. 2010; Reddy et al. 2012). In the face of these challenges, what level of power do conservation researchers have with restricted sample sizes?

Some authors have already tried to address the issue of sample size. The number of individuals tagged in conservation genetics studies varies widely from hundreds of individuals to less than 20 (e.g. Schwartz et al. 2005; Iyengar et al. 2005; Ernest et al. 2003; Manel et al. 2004). This is less likely to cause problems when the complete batch of samples is considered as a single group. However, frequently they are split into smaller groups to test for regional differences in genetic variation, differentiation,

- 54 - etc (e.g. Lu et al. 2001; Csiki et al. 2003). This could mean insufficient power to detect any significant divergence between groups (Archie 1985; Fitzpatrick 2009). Previous work suggests that large sample sizes are required to accurately measure levels of allelic diversity because there is a tendency to detect only the most frequent alleles when a small sample is taken (Sirkkomaa 1983; Sjögren & Wyoni 1994; Rao 2001; Neel & Cummings 2003). A sample of at least 30-50 individuals is required to detect all alleles including rare alleles (q < 0.05) with a probability of 95%. Thus, the low levels of allelic diversity reported in some conservation studies could be due to sampling alone. In contrast, measures of expected heterozygosity and differentiation do not vary as much with sample size as they are less sensitive to the number of alleles (Nei et al. 1975; Allendorf 1986; Leberg 1992; Beaumont & Nichols 1996; Luikart et al. 1998b; Spencer et al. 2000; Yan & Zhang 2004). Unbiased estimates of these parameters may be possible with as few as 7 individuals depending on the type and number of markers, and the global level of differentiation (Luikart et al. 1998b; Ruzzante 1998; Dyer & Sork 2001; Pruett & Winker 2008; Sinclair & Hobbs 2009). It may also be possible to determine population structure with as few as 10 individuals as long as a sufficient number of polymorphic loci are used (Luikart et al. 1998b; Spencer et al. 2000; Rosenberg et al. 2001; Rosenberg et al. 2001; Cavers et al. 2005; Fogelqvist et al. 2010). In both instances, it may be that sample size is not as important as the number of alleles per marker (Kalinowski 2002), and some authors suggest that it is better to sacrifice the number of individuals rather than the number of markers used (Nei & Roychoudhury 1974; Nei 1978; Kalinowski 2005; Breazel 2008).

Recent genetic analysis of the Sumatran tiger provides a good example of the challenges faced by a typical noninvasive study. The logistics and risks involved with direct sampling mean that noninvasive samples provide a unique opportunity to study the genetic status of this elusive cat. Large felids regularly deposit faecal samples along roads and trails and these are easily collected for later analysis. However, sample numbers are expected to be small. The total population is

- 55 - estimated at just ~500 individuals (Chundawat et al. 2011) and the tropical climate of Indonesia means that each sample may be exposed to high temperatures, high humidity, UV and heavy rain. Any DNA contained in the intestinal cells on the surface of each sample is likely to be degraded. Given that the final number of unique genotypes is likely to be low as in other carnivore studies, can researchers still be confident that they are getting an accurate representation of the population's genetics? In order to address this question, in the following study I investigate the effect of small sample size on estimates of the mean number of alleles per locus

(NA), expected heterozygosity (He), FST ($w) and the number of subpopulations (k).

2.3 Methods 2.3.1 Simulated data

I simulated 3 subpopulations of 100 individuals in EASYPOP (Balloux 2001) using a simple island and 1-dimensional migration model with equal migration rates between each pair of populations (Table 3.1). A step-wise mutation model with 50 allelic states and maximal variability was used to create individual genotypes with 10, 20, 50 and 100 loci (L). Simulations were continued for 10 and 35 generations to achieve a pairwise FST of 0.05 and 0.15 respectively. I used sample sizes of n = 100, 90, 80,

70, 60, 50, 40, 30, 20, 10, 5 for each combination of loci and pairwise FST. The same number of individuals was sampled from each subpopulation. Each dataset consisted of 100 replicates for each combination of n, L and FST. EASYPOP produces files in Genepop (.gen) format which are easily converted using the R Package

"POPGENKIT" or FORMATOMATIC v0.8.1 (Manoukis 2007; Paquette 2011).

2.3.2 Genetic parameters The most commonly used measures of diversity are allelic diversity (e.g. number of alleles per locus, number of effective alleles) and heterozygosity (either observed or expected values under Hardy-Weinberg equilibrium). In this study I will explore the effect of sample size on the most commonly reported measurements of genetic diversity: the mean number of alleles per locus and expected heterozygosity

- 56 - (Frankham et al. 2002; Hedrick 2005b).

Table 2.1 EASYPOP settings used to simulate the test populations.

Setting Value Ploidy level 2 Two sexes N Random mating Y Number of populations 3 Same number of individuals in each population Y Number of individuals 100 Same migration scheme over all simulation? Y Migration model Island model; 1-D stepping stone model Proportion of migration 0.0001 Number of loci 10; 20; 50; 100 Free recombination between loci? Y Do all loci have the same mutation scheme? Y Mutation rate 0.001 Mutation model SMM (single step mutation) Number of possible allelic states 50 Variability in the initial population Maximal (randomly assigned alleles) Number of generations 10; 35

The mean number of alleles per locus (NA) is given by:

r Aj (2.1) N A = ! j=1 r

where Aj is the total number of alleles at the jth locus and r is the total number of loci.

The mean expected heterozygosity per locus (He) is given by:

- 57 - r hj (2.2) H e = ! j=1 r

where hj is the heterozygosity of the jth locus, and r is the total number of loci. The heterozygosity of each locus in this case is given by Nei's unbiased estimator (Nei 1978):

2 2n(1! pi ) h = " (2.3) j 2n ! 1 where n is the number of individuals and pi the frequency of the ith allele of locus j.

The classic measure of population differentiation, FST, estimates allele frequency divergence among subpopulations (Falconer & MacKay 1996; Hedrick 2000). In this study I investigate the effects of sample size on the weighted estimator $w (Weir & Cockerham 1984):

""alu ˆ l u !w = (2.4) ""(alu + blu + clu ) l u where a is the variance in allele frequency between populations, b the variance between individuals within populations and c the variance between gametes within individuals for u alleles at l loci. I used ARLEQUIN v3.5 (Excoffier et al. 2005) to estimate the mean number of alleles per locus (NA), mean expected heterozygosity

(He) and pairwise FST for each dataset. I also used the program ADZE (Szpiech et al. 2008) to compute mean allelic richness and the mean number of private alleles. The summary data was then imported into R (R Development Core Team 2011) to perform a two-way ANOVA and Tukey's multi-comparison test to look for significant differences between parameter values for each sample size.

STRUCTURE is perhaps the most frequently used Bayesian clustering program to determine population structure. It estimates the most likely number of clusters by:

- 58 - P(K | X) ! P(X | K)P(K) (2.5) where K is the number of clusters and X the observed genotypes in the sample. It is difficult to estimate P(K|X) directly, so STRUCTURE uses an estimate of P(X|K) for each value of K to approximate the posterior distribution of P(K|X) (Pritchard et al. 2000). I used STRUCTURE v2.3.3 (Pritchard et al. 2000) to determine the number of subpopulations (k) for each sample size. I chose the following settings: 30,000 burn- in length, 1 x 106 steps, admixture model, correlated allele frequencies model, and 3 runs at k = 1-5. I visualised the output with STRUCTURE HARVESTER (Earl & von Holdt

2011) to determine k using both STRUCTURE's ln P(X|K) and Evanno's !K method.

STRUCTURE estimates the log probability of the data for each value of k, and the mean value of ln P(X|K) is assumed to reach its maximum value at the most likely number of clusters. Above this, ln P(X|K) will plateau and there is an increase in variance (Pritchard et al. 2010). The Evanno method attempts to highlight this break point by computing the second order rate of change of ln P(X|K) with respect to k. This produces a modal peak of !K at the most likely number of clusters, the height of which reflects the strength of the signal detected by STRUCTURE (Evanno et al. 2005).

2.4 Results 2.4.1 Genetic diversity and sample size There was a significant reduction (up to 50%) in estimates of the mean number of alleles per locus, allelic richness and the number of private alleles with a decrease in sample size. This relationship did not vary with the migration model, number of loci used or the level of population differentiation (Fig. 2.1) In contrast, there was no significant change in mean expected heterozygosity or pairwise FST with a decrease in sample size (Fig. 2.1). Overall, there was a reduction in sample variance when a larger number of loci were used (Appendix 2). For the mean number of alleles per locus there was no significant difference in variance with a decrease in sample size.

However, for expected heterozygosity and pairwise FST variance increased significantly at n < 20 and n < 10 respectively.

- 59 - 2.4.2 Population structure and sample size

When the populations were highly differentiated the results of the STRUCTURE and Evanno !K methods were similar. Both methods estimated the correct number of subpopulations down to a sample size of n = 10 with as few as 10 loci. However, for the smallest sample size of n = 5, up to 50 loci were required to obtain the correct result (Fig. 2.2). Overall, larger sample sizes were needed when the populations were poorly differentiated, and twice as many samples were required for the stepping-stone model compared to the simple migration model under this scenario (Table 2.2). When only 5 individuals were sampled in the poorly differentiated populations, STRUCTURE was unable to resolve any population structure and returned a value of k = 1. In contrast, the Evanno method was able to return the correct value for k but with 100 loci. Unsurprisingly, the height of the modal peak for !K decreased steadily with a reduction in sample size (Appendix 2).

Table 2.2 Minimum sample sizes needed to determine the correct number of populations using STRUCTURE with varying numbers of loci.

Minimum sample size (FST = Minimum sample size (FST = 0.15) 0.05) Number of loci Island model Stepping Island model Stepping stone model stone model 10 10 10 20 40 20 10 10 10 20 50 5 5 10 20 100 5 5 (5) 20

- 60 - Figure 2.1 Variation in the mean number of alleles per locus, expected heterozygosity and pairwise FST with samples taken from highly differentiated populations under an island migration model. Estimates were computed for 10, 20, 50 and 100 loci.

- 61 - Figure 2.2 Variation in the number of population clusters estimated with STRUCTURE as a function of sample size in highly differentiated populations, using both the ln P(X|K) and Evanno !K methods under an island migration model.

- 62 - 2.5 Discussion Noninvasive sampling methods are increasingly being used to study elusive or cryptic species. The data they contain have allowed wildlife managers to incorporate genetic information such as genetic variation, population size and gene flow into assessments of a population's conservation status. Studies of wide-ranging species usually involve prolonged field surveys covering vast areas of forest or mosaic landscapes. This frequently yields a large number of specimens but researchers are commonly left with a much smaller number of unique genotypes to work with. The main aim of this study was to investigate the effect of restricted sample sizes on measures of genetic variation and population structure.

Genetic diversity and differentiation Many conservation genetics studies routinely report the mean number of alleles per locus (NA) as a measure of allelic diversity. High levels of diversity are expected to correlate with a low extinction risk and high population fitness (Frankham 1998; Brook et al. 2002). However, the use of this measure is problematic as it is highly dependent on sample size. In fact, allele counts are commonly used as a measure of past population bottlenecks (Luikart et al. 1998b; Spencer et al. 2000). Therefore, the observed variation in allelic diversity between groups may simply reflect the relative number of sampled individuals. A number of methods have been proposed to overcome the sensitivity of NA to sample size, namely extrapolation, rarefaction and repeated random subsampling (Leberg 2002). Extrapolation methods attempt to estimate the maximum number of different alleles contained within a population (Foulley & Ollivier 2006; van Loon et al. 2007). There is however a large amount of uncertainty in the final estimate when extrapolating from small sample sizes (Frantz & Roper 2006) and both rarefaction and resampling offer better alternatives for comparing between groups. With rarefaction, estimates of allelic diversity in each region are scaled down to the size of the smallest sample (El Mousadik & Petit 1996; Kalinowski 2004; Pruett & Winker 2008). This works well in the majority of cases, but consideration should still be given to the size of the final sample, i.e. the rarefied

- 63 - value in a small sample is likely to be much lower than that in a larger sample (El Mousadik & Petit 1996). An alternative are resampling or permutation methods that are designed to test the significance of the observed diversity against a random distribution (Glaubitz et al. 2003; Yan & Zhang 2004; Schwartz et al. 2005). For both approaches the absolute level of diversity is not important as it is the difference between groups that is being tested for significance.

Other secondary factors that may influence the observed amount of allelic diversity in a sample are the diversity of the total population and the polymorphism of the loci used. In this study I simulated populations using microsatellite loci with a maximum of 50 alleles but some authors suggest that microsatellites may be constrained to as few as 10 allelic states (Nauta & Weissing 1996; Neff & Gross 2001; Balloux &

Goudet 2002). The rate at which NA decreases with a reduction in sample size is expected to be slower for markers with fewer alleles (Allendorf 1986; Yan & Zhang 2004). However, small samples are still likely to miss rare alleles and using less polymorphic loci may result in other problems such as the shadow effect in which two individuals share the same genotype (Mills et al. 2000). Thus, if wildlife studies wish to use the number of alleles as a measure of population diversity, they should account for sample sizes by using one of the corrective measures outlined above.

Heterozygosity and differentiation are also commonly measured in conservation studies. Expected heterozygosity is thought to correlate with fitness such that populations with a low level of heterozygosity are expected to suffer from inbreeding depression and its secondary effects (e.g. Coltman et al. 1998; Hale & Briskie 2007; Roelke et al. 1993). Differentiation is used as a measure of population isolation and populations that are connected by high rates of gene flow are expected to be less differentiated than those separated by dispersal barriers (Hedrick 2005b; Hamilton 2009). Both the unbiased estimator of heterozygosity and differentiation used in this study implicitly account for differences in the number of loci and the number of individuals (Nei 1978; Weir & Cockerham 1984). Both showed little variation with a

- 64 - reduction in sample size from 100 down to 5 and this relationship did change with the number of markers used. This makes them particularly useful for studies that have sampled <20 individuals and/or <20 loci as they can still provide a true reflection of total population diversity and differentiation (e.g. Gaggiotti et al. 1999; Hardy et al. 2003; Henry et al. 2009; Castilho et al. 2010; Cegelski et al. 2003; Dixon et al. 2006; Haag et al. 2010; Ernest et al. 2003). Other weighted differentiation measures have been proposed to account for the mutation model of microsatellite loci (e.g. Michalakis & Excoffier 1996; Rousset 1996; Goodman 1997; Slatkin 1995) and in theory they should behave in a similar fashion to FST. However, the few studies to formally look at the relationship between differentiation and sample size suggest that FST is the better estimator when a small number of individuals or a small number of loci are used (Gaggiotti et al. 1999; Balloux & Goudet 2002; Palstra & Ruzzante 2008). Accordingly, this means that conservation projects should perhaps use expected heterozygosity and FST when faced with restricted sample sizes. To increase certainty researchers can test for significant differences between groups using statistical methods such as the G-test, ANOVA or random resampling (Archie 1985; Guo & Thompson 1992; Luikart et al. 1998b; Degen et al. 1999; Glaubitz et al. 2003; Glaubitz et al. 2003; Schwartz et al. 2005). The level of uncertainty can also be reduced by using more markers.

Population structure Spatial genetic structure within a population is expected to reflect the level of connectivity between subpopulations and the dispersal ability of a given species (Hoehn et al. 2007). Bayesian clustering methods are frequently used to determine population structure as they don't require any a priori information about the number of populations; they use the assumptions of Hardy-Weinberg and linkage equilibrium to cluster individuals into random mating groups (Latch et al. 2006; Chen et al. 2007; Safner et al. 2011). However, the ability of Bayesian methods to form the correct clusters depends on the degree of spatial structure within the population and the quality of the dataset. For both high and low levels of differentiation STRUCTURE was

- 65 - able to infer the correct number of subpopulations as long as 10-20 individuals were sampled, and more individuals were needed when only 10 loci were used. Notably, the power to detect population structure was substantially reduced when the smallest sample of 5 individuals was considered. In this case, far more loci were needed to accurately resolve the number of genetic clusters - 100 markers were required when the populations were poorly differentiated, and 50 when they were highly differentiated. This is in agreement with previous work concluding that more loci may be needed if there is a low level of differentiation (Kalinowski 2005). However, it would not be feasible for most noninvasive studies to budget for this many markers, and drastically increasing the number of markers could lead to inflated rates of genotyping error. Therefore, 10 individuals genotyped at 20 loci could be considered as the minimum sample size for assessment of population structure. Alternatively, projects could aim to sample a minimum of 20 individuals in each subpopulation and genotype them at just 10 loci. However, it is preferable to genotype individuals at more loci to reduce the variance of mean estimates. And with such a small number of markers care must be taken to screen for the most polymorphic in order to avoid multiple individuals sharing the same genotype (Rosenberg et al. 2001).

Without prior knowledge of the number of subpopulations it is possible that there will be an unequal number of samples per subpopulation. Disparity in the number of samples could affect both STRUCTURE analysis and estimates of differentiation

(Ruzzante 1998; Han et al. 2010). STRUCTURE is able to infer the correct number of populations as long as the average number of individuals per population exceeds a certain threshold (Ruzzante 1998; Kalinowski 2011). Preliminary analyses in this study suggest that for unequal samples, a reasonable threshold is n > 7 across all populations. Also, whilst Bayesian methods may infer the correct number of clusters, the assignment of individuals to each cluster should be checked when a small sample size is used (Frantz et al. 2012). Any discordance between cluster assignment and geographic location would suggest that no structure exists or that

STRUCTURE has not been able to resolve the genetic pattern (Pritchard et al. 2010).

- 66 - Researchers can test for any uncertainty in population structure analysis by using more than one Bayesian clustering program (e.g. Corander et al. 2008; Guillot et al. 2005; François et al. 2006) or by using alternative methods such as AMOVA or principal coordinate analysis (Excoffier et al. 1992; Piertney et al. 1998; Excoffier et al. 2005; Patterson et al. 2006; Fitzpatrick 2009).

Recommendations Noninvasive genetics studies are a valuable tool for conservation managers looking to understand more about the populations they protect. In particular, they help to quantify population fitness and extinction risk for endangered species. The challenges of sample collection and variable sample quality mean that huge amounts of survey effort may result in only a handful of individuals being sampled. Despite this, the results of this study suggest that researchers can still obtain unbiased estimates of population diversity and differentiation from as few as 5 individuals genotyped at 10 loci. In contrast, accurate estimation of population structure requires a larger sample. In both cases highly polymorphic markers should be chosen, and this can be determined in an initial pilot study. Researchers should also consider using more than one analytical method or statistical test to validate any conclusions about significant differences between regions.

- 67 - Chapter 3

The Inﬂuence of Genotyping Error in Noninvasive Genetics Studies

- 68 - 3.1 Abstract Many conservation genetics studies rely on noninvasive samples because they provide a convenient source of DNA with which to study endangered populations. This has allowed estimates of genetic variation, population structure and population size to be made for many elusive species such as the tiger. However, noninvasive samples are also prone to genotyping error and high rates of error may bias these estimates. In the absence of reference data many methods have been proposed to increase the reliability of noninvasive multilocus genotypes, but few studies have offered guidelines for acceptable levels of error. Here I study the effect of genotyping errors on estimates of the mean number of alleles per locus, expected heterozygosity, pairwise FST and population structure. Allelic dropout and false alleles had a much smaller effect than expected, and all parameters were sensitive to missing data. Overall, reasonable estimates of genetic variation and population subdivision are possible using a small number of noninvasive samples as long as error rates are kept below 20%. For most studies this could be achieved by using methods such as the multi-tubes approach to reduce the incidence of allelic dropout and false alleles, and optimised PCR protocols to increase amplification success.

3.2 Introduction Many conservation genetics studies rely on noninvasive samples to study endangered populations. Noninvasive samples such as hair and faeces offer a readily available source of DNA that does not involve capturing or handling individuals. This is especially advantageous for studies of cryptic or low density species as multiple individuals can be sampled by conducting surveys over a wide area. The success of these studies ultimately depends on the power to distinguish between individuals, i.e. every sampled individual must have a unique genotype. This depends on the number of individuals sampled, the level of variation in the population and the polymorphism of the markers used (Mills et al. 2000; Waits et al. 2001). Once confirmed, the genotypes are then used to derive estimates of

- 69 - population size and genetic differentiation at the population level (e.g. Eggert et al. 2003; Prugh et al. 2005; Lukacs & Burnham 2005), or genetic variation and genetic distance at the individual level (e.g. Mondol et al. 2009b; Paetkau et al. 1998). Combining this information with other demographic or ecology-based data is an effective way to direct management decisions about the species. In order for these final decisions to be robust the observed sample genotypes must as close to the true genotypes as possible as any deviations have the potential to bias subsequent analyses (Waits & Leberg 2000). Researchers rarely have access to reference genotypes, e.g. from blood, and so many methods have been developed to estimate how accurate the observed genotypes are (Zhan et al. 2010). These vary from repeated blind genotyping of a subset of samples (Bonin et al. 2004; Pompanon et al. 2005), to maximum likelihood simulations (Miller et al. 2002; Johnson & Haydon 2007a) and multiple PCRs per sample (Valière et al. 2002; McKelvey & Schwartz 2005). In all these methods the main aim is to identify the types of error present and to quantify how much error exists in the dataset. This provides a measure of uncertainty and also helps to highlight the poorest quality samples (and markers) to be discarded or targeted for further work.

Genotyping errors are caused by a wide variety of factors and can be induced at any stage of the genotyping process from sample preservation to PCR amplification and construction of the consensus genotypes. Errors are thought to occur because, unlike tissue or blood, noninvasive samples contain low quantities of DNA or degraded DNA templates that may prevent primer annealing (Taberlet et al. 1996). Faecal samples are particularly problematic as they also contain PCR inhibitors that are commonly co-purified during DNA extraction (Monteiro et al. 1997; Venturi et al. 1997; Kemp et al. 2006; Bessetti 2007), and they are often reported as having much higher error rates than other sample types (Waits & Paetkau 2005). Increasing sample age and degradation during sample storage can either reduce the amount of DNA template available for amplification or increase the chance of mispriming and reduce the amount of PCR product (Frantzen et al. 1998; Fernando et al. 2003;

- 70 - Roon et al. 2003). Low concentrations of DNA in the sample extract may also reduce the likelihood of the template being present in the pipetted sample and thus lower PCR success (Foucault et al. 1996; Morin et al. 2001). Because of the many opportunities for genotyping error to occur, most genetics studies routinely report their error rates as an indication of sample quality and data reliability. Allelic dropout is perhaps the most common with published rates varying from ~1-30% depending on the sample type, genetic method and markers used (Fernando et al. 2003; Broquet & Petit 2004). It occurs because one of the two alleles in a heterozygote fails to amplify, and rates of dropout are expected to be highest for DNA concentrations <25-50pg and large alleles (Foucault et al. 1996; Taberlet et al. 1996; Gagneux et al. 1997; Morin et al. 2001; van Oosterhout et al. 2004; Buchan et al. 2005; Broquet et al. 2007). False alleles occur less frequently at a rate of ~0-8%, although the rate may be higher with very poor quality samples or with high levels of contamination (Bradley & Vigilant 2002; Bonin et al. 2004). They are caused in part by mispriming or template slippage during PCR amplification (Hauge & Litt 1993; Bovo et al. 1999; Clarke et al. 2001) and are observed as false heterozygotes (true homozygotes scored with an extra allele) or ambiguous heterozygotes (more than 2 alleles). Microsatellite stutter peaks may also be scored as extra alleles if they are of similar intensity to true allele peaks (Smith et al. 1995; Brownstein et al. 1996; Ginot et al. 1996; Magnuson et al. 1996; Kalinowski et al. 2006). Finally, genotypes may be recorded as missing if they are ambiguous, e.g. a heterozygote with 3 allele peaks, or if no alleles are observed for a particular individual at a given locus. No alleles will be observed in the case of PCR failure or null alleles. Null alleles are a particular case caused by mutations in one or both of the microsatellite flanking sequences that prevents primer binding (Callen et al. 1993; van Treuren 1998; Holm et al. 2001).

Despite the fact that some level of error is expected in noninvasive studies, few directly address the impact that an error-prone dataset may have on subsequent parameter estimates. The majority focus on optimising methods to reduce error rates

- 71 - (e.g. Hansen et al. 2008; Arandjelovic et al. 2009; Bellemain & Taberlet 2004), and to assess sample quality (e.g. Miquel et al. 2006). Others have largely concentrated on the bias in estimates of population size and differentiation. Population size may be grossly overestimated if mismatched genotypes are treated as multiple individuals (Creel et al. 2003; Kalinowski et al. 2006; Knapp et al. 2009). Similarly, the shadow effect (different individuals sharing the same genotype) will lead to a downward bias (Mills et al. 2000; Waits & Leberg 2000; Creel et al. 2003; McKelvey & Schwartz 2004; Waits & Paetkau 2005; Knapp et al. 2009). Null alleles generally occur at low frequency (q < 0.40) but they may increase the observed proportion of homozygotes and reduce estimates of population differentiation (Dakin & Avise 2004; Chapuis & Estoup 2007; Kelly et al. 2011). The presence of null alleles in consensus genotypes can also result in higher rates of exclusion in paternity studies (Hoffman & Amos

2005; Legendre & Fortin 2010). A study of FST found a downward bias due to the presence of false alleles and an increase in observed FST up to a threshold frequency of 0.3-0.4 null alleles (Breazel 2008). Most studies are aware of the problems that erroneous genotypes can cause and generally they are able to keep error rates in the region of 0.02 allelic dropout and 0.002 false alleles per locus. However, simulation studies suggest that a lower threshold of '0.01 per locus is necessary to prevent the problems outlined above (Taberlet et al. 1996; Waits & Leberg 2000; Paetkau 2003; Waits & Paetkau 2005). Given that genetics parameters are sensitive to the presence of error within a dataset, what level of uncertainty is introduced by the range of error rates found in many wildlife genetics studies?

The Sumatran tiger provides a good example of the application of noninvasive genetics to the conservation of an endangered species. There are just 4-500 individuals left on Sumatra spread over an area of ~88,000km2 (Wibisono & Pusparini 2010) and traditional conservation efforts have relied on camera trap and occupancy surveys to assess the status of the current population (Wibisono et al. 2011). More recently, noninvasive techniques have made a valuable contribution to

- 72 - tiger conservation by allowing estimates of genetic variation, population structure and effective population size for subspecies in Russia and India (Henry et al. 2009; Mondol et al. 2009a). Similar techniques using faecal DNA are now being employed to study the Sumatran tiger. It is likely that genotyping error will be an issue for the Sumatran study for two reasons: (i) Faecal samples are particularly error prone compared to other noninvasive sample types, and (ii) The high temperatures and high humidity of a tropical environment are known to increase the amount of DNA degradation. Given that this study is likely to encounter high rates of genotyping error, in the following chapter I test the influence of error on measures of genetic variation (the mean number of alleles per locus and expected heterozygosity), and population subdivision (pairwise FST and the number of population clusters estimated by STRUCTURE).

3.3 Methods I first simulated baseline datasets containing 3 subpopulations of 100 individuals in

EASYPOP (Balloux 2001) using a simple island migration model with equal migration rates between each pair of populations. Settings were the same as those outlined in Chapter 2. A step-wise mutation model with 50 allelic states and maximal variability was used to create individual genotypes with 20 loci. Simulations were continued for

35 generations to achieve a pairwise FST of 0.15. I used sample sizes of n = 100, 10, and 5 individuals for each combination of error rates, and the same number of individuals was sampled from each subpopulation. Each dataset consisted of 100 replicates for each combination of n and error rate.

I then applied three types of error to each dataset: allelic dropout (ADO), false alleles (FA) and missing genotypes. Error was applied at a constant rate and in an independent manner to each allele at a locus, i.e. error applied to Locus 1 Allele 1 did not affect the rate of error applied to Locus 1 Allele 2 (Wang 2004). Similarly, error applied to Locus 1 did not affect the rate of error applied to Locus 2. Error files were generated using a simple algorithm implemented in ERRORGEN (Wang 2012).

- 73 - Whether or not a particular error was applied to an allele was determined by a random number generator and the specified error rate. The relevant error was applied when the random number ' error rate ("). Where ADO occurred a heterozygote genotype (AiAj) was changed to either AiAi or AjAj with an equal probability. False alleles were assumed to affect all loci with an equal probability, and where this error occurred the affected allele was substituted for an allele taken at random from the 50 possible alleles. Each of the 50 alleles could be observed with equal probability. In reality error rates may be higher for larger alleles, and stutter (with a change of ±1 or ±2 bp change) may be more common for microsatellite loci. Missing data was assumed to result from PCR amplification failure and no genotype was recorded for the affected locus. The genotype was replaced by 00/00 and thus no other error could be applied:

True genotype Allelic dropout False allele Missing 24/15 24/24 or 15/15 24/32 00/00

I used a wide range of error rates to span the range of possible values noted in empirical field studies: "1 = 0.1 - 0.9, "2 = 0.01 - 0.20, and "3 = 10 - 90%. The rate of allelic dropout is represented by "1, the rate of false alleles by "2, and the proportion of missing data by "3. Errors were applied to each baseline dataset individually and in combination. The combined rates were chosen to encompass the low, medium and high rates possible with noninvasive samples: 0.1/0.01/10%, 0.2/0.02/20%,

0.5/0.05/50%, 0.7/0.07/70% and 0.9/0.09/90% representing the ratio of "1/"2/"3. In this case I considered that errors were more likely to occur together and that the relative rates would be similar, i.e. low rates of ADO would occur with low rates of FA, and high rates of ADO were more likely to occur with higher rates of FA.

For each baseline dataset and error file I computed mean values for the number of alleles per locus (NA), expected heterozygosity (He), and pairwise FST in ARLEQUIN v3.5 (Excoffier et al. 2005). Here NA was computed as the total number of alleles at each locus divided by the total number of loci. He and FST were calculated using

- 74 - unbiased estimators weighted across all loci and individuals (Nei 1978; Weir & Cockerham 1984). The summary data was then imported into R (R Development Core Team 2011) to perform a two-way ANOVA and Tukey's multi-comparison test to look for significant differences between parameter estimates for each error value.

Each dataset was also subjected to STRUCTURE analysis (Pritchard et al. 2000) with the following settings: 30,000 burn-in length, 1 x 106 steps, admixture model, correlated allele frequencies model, and 3 runs at k = 1-5. In the absence of genotyping error, STRUCTURE was not able to infer the correct value of k with 5 individuals so analysis was only performed for n = 100 and n = 10. I used STRUCTURE

HARVESTER (Earl & von Holdt 2011) to visualise the output files in order to determine k using both the STRUCTURE ln P(X|K) and Evanno !K methods.

3.4 Results Genotyping error had minimal effect on the parameter estimates when all individuals (n = 100) were sampled. However, significant changes in the estimated values were introduced by sampling a small number of individuals (n = 5 and n = 10) from the population. Initial values in the baseline datasets were NA = 10.6 alleles, He = 0.83 and FST = 0.15. To allow comparisons between the response of parameter estimates and STRUCTURE analysis to changes in error rate, results are reported for n = 100 and n = 10 only. However, full results are noted in the figures that follow and Appendix 3.

3.4.1 Allelic dropout

With the total population there was no significant change in estimates of NA, He or FST with an increase in the ADO error rate. STRUCTURE was also able to infer the correct number of clusters regardless of error rate. With a smaller sample of 10 individuals there was a significant reduction in estimates of NA and He, and a significant increase in the estimated value of FST with an increase in "1 (Fig. 3.1). However, the size of the difference between "1 = 0 and "1 = 0.9 was small for all parameters; mean NA decreased by 1.6 alleles, mean He decreased by 0.04, and FST increased by 0.04.

STRUCTURE was also able to correctly infer k = 3 for all error rates if the Evanno !K

- 75 - method was used (Fig. 3.2). The standard ln P(X|K) method only returned correct values up to "1 = 0.5. These results suggest that allelic dropout is unlikely to significantly alter any conclusions of genetic variation or differentiation made from the data.

3.4.2 False alleles The increasing incidence of false alleles appeared to have a very different effect to that of ADO. When the whole population was sampled there was a large and significant increase in NA (21.6 alleles) with a smaller increase in He (0.05) and a corresponding decrease in FST as "2 increased from 0 to 0.20 (Fig. 3.3). This relationship did not change with a reduction in sample size although the increase in

NA was much smaller (2.8 alleles). Again STRUCTURE was able to infer the correct number of clusters with both the full sample and subsamples of 10 individuals (Fig. 3.4). The presence of false alleles has a slightly greater effect than that of allelic dropout, but again the overall conclusions of genetic diversity and differentiation are unlikely to change. However, the larger the sample the more drastic the effect on the mean number of alleles noted in the data, and thus NA cannot be used as a reliable indicator of allelic diversity.

- 76 - Figure 3.1 Variation in the mean number of alleles per locus, expected heterozygosity and pairwise FST with an increase in the rate of allelic dropout. Estimates were computed for the total population and for sample sizes of 10 and 5 individuals.

- 77 - Figure 3.2 Variation in STRUCTURE output with increasing rates of allelic dropout. The number of population clusters was estimated using the ln P(X|K) and Evanno !K methods with STRUCTURE HARVESTER.

- 78 - Figure 3.3 Variation in the mean number of alleles per locus, expected heterozygosity and pairwise FST with an increase in the rate of false alleles.

- 79 - Figure 3.4 Variation in STRUCTURE output with increasing rates of false alleles. The number of population clusters was estimated using the ln P(X|K) and Evanno !K methods.

- 80 - 3.4.3 Missing alleles

In the total population there was a significant reduction in NA but little change in values of He or FST with an increase in "3. STRUCTURE was able to infer the correct value for k up to 80% missing data. However, this may also be dependent on the distribution of error. In this simulated study error was distributed randomly across all populations and loci, but this may not always be the case in empirical studies. With a smaller sample of 10 individuals there was a significant decrease in He above 60% missing data, and a significant decrease in FST above 40% (Fig. 3.6). The highest missing rate ("3 = 0.9) represented a large reduction in He of 0.3 and in FST of 0.5.

This resulted in negative values for pairwise FST, i.e. no detectable differentiation.

STRUCTURE output was also sensitive to missing data when only 10 individuals were sampled. Both the standard ln P(X|K) and Evanno !K methods could only estimate k

= 3 when "3 was ' 20% (Fig. 3.7. Missing data perhaps has a greater effect than either allelic dropout or false alleles above a certain threshold. If too many genotypes are missing, e.g. above 50%, there is a marked reduction in estimates of both heterozygosity and differentiation, which is most likely due to a loss of genetic information and not a simple reduction in sample size. This would most certainly result in an underestimate of the amount of diversity within the sampled population and the attenuation of any signal of population structure.

3.4.4 Combined error

For the total population there were only minor differences in values of NA, He and FST with an increase in the level of error. In contrast, with the subsample of 10 individuals there was a noticeable decrease in parameter values with combined error values > "1 = 0.2, "2 = 0.02 and "3 = 20% (Fig. 3.8). He was most affected with a reduction of 0.15 from its starting value which would result in an underestimate of heterozygosity in the given data set. In both cases the STRUCTURE was able to infer the correct number of clusters when the combined error rate was <50% (Fig. 3.9).

These results suggest that it is still possible to obtain reasonable estimates of NA, He

- 81 - and FST if allelic dropout, false alleles and missing data can be kept at a rate of "1

< 0.5, "2 < 0.05 and "3 < 50% respectively.

- 82 - Figure 3.5 Variation in the mean number of alleles per locus, expected heterozygosity and pairwise FST with an increase in the proportion of missing genotypes.

- 83 - Figure 3.6 Variation in STRUCTURE output with an increasing proportion of missing data.

- 84 - Figure 3.7 Variation in the mean number of alleles per locus, expected heterozygosity and pairwise FST with an increase in the combined rate of genotyping error.

- 85 - Figure 3.8 Variation in STRUCTURE output with increasing rates of combined error.

- 86 - 3.5 Discussion 3.5.1 Genetic variation and differentiation Many wildlife genetics studies rely on noninvasive samples to determine the level of genetic variation and spatial structure in their study population. Given that these measures are increasingly being used to support management decisions, it is vital that the genetic data is as robust as possible. However, noninvasive samples are particularly prone to genotyping error and this may have an impact on downstream analyses. The main aim of this study was to investigate the effect of genotyping error on commonly used population genetics measures: NA, He and FST.

There appears to be much less bias in measures of genetic variation and differentiation if researchers are able to sample the majority of individuals within a population. For all types of error (allelic dropout, false alleles and missing data), there was a minimal change in estimates of heterozygosity and differentiation when the total population of 100 individuals was sampled. This is a surprising result given that very high rates of error will have affected almost all the genotypes within a dataset. In contrast, estimates of allelic diversity (NA) were most sensitive to the presence of false alleles. Increasing the false error rate up to a frequency of 0.20 produced an increase in NA of ~200%. This relationship is expected given that this type of error introduces novel alleles into the dataset. Many studies routinely report the mean number of alleles per locus (NA) as a measure of genetic variation, and these results suggest that this measure should be used with care if false error rates are high. However, because many noninvasive studies report multilocus false error rates of ≪0.1 this relationship is unlikely to have much influence on the final result.

In reality, researchers will rarely have the opportunity to sample all individuals within a population and are more likely to begin analysis with a small subset of samples. The results of this study suggest that under this scenario, researchers need to take far more care to reduce error rates. All parameters showed an increased sensitivity to error when the number of sampled individuals was reduced to 10, although in

- 87 - general parameter values changed as expected. There was some variation in measures of genetic variation and differentiation with the introduction of allelic dropout and false alleles. Both allelic diversity and heterozygosity decreased with allelic dropout and increased with false alleles. In contrast, FST showed an increase with allelic dropout and a decrease with false alleles. As in the total population, values of He and FST did not change drastically (± 0.05 units) even at the very highest rates of error. Using a constrained allele model with 10 allelic states did not appear to alter this relationship (Nauta & Weissing 1996; Neff & Gross 2001; Balloux & Goudet 2002). One alternative explanation is that both allelic dropout and false alleles mimic the action of inbreeding and outbreeding respectively. Under this scenario, error would act primarily on genotype proportions without affecting allele frequencies and so levels of expected heterozygosity and differentiation would stay the same (Hedrick 2005b; Hamilton 2009). The main implication of this result is that unbiased estimators of He and FST are robust to changes in both sample size and genotyping error (allelic dropout and false alleles). Researchers should therefore use similar parameters when assessing their populations.

Many conservation genetics studies routinely report NA as one measure of population diversity, however it is not an ideal measure for this purpose. Although allelic diversity showed a similar pattern of change with rates of allelic dropout and false alleles, the absolute level of diversity varied with sample size. Values of NA fell to 69% and 51% of its initial value in the subsamples of 10 and 5 individuals respectively. As noted in Chapter 2, allele counts are particularly sensitive to sample size and a large sample is needed to detect all alleles (including rare alleles) within a population (Sjögren & Wyoni 1994; Rao 2001). For this reason, although NA provides some indication of allelic diversity within a sample, it is not an ideal measure of the total diversity within a population. In particular, care should be taken when comparing between regions as differences in sample size can produce false conclusions about relative diversity (El Mousadik & Petit 1996).

- 88 - Small sample sizes also increased the sensitivity to missing data. The variation in DNA template quality and DNA concentration between samples means that some will amplify consistently while others will intermittently fail to produce any PCR product. Amplification failure is usually recorded as a null genotype (00/00 or ?) to reflect the fact that there were no observed alleles for a given locus. Whilst this does not introduce error into the genotypes per se, it is likely to reduce the power of the data to correctly estimate variation and differentiation. Both He and FST seemed robust to changes in missing data until a particular threshold was reached. This threshold was slightly higher for He (60%) than FST (40%), but above this threshold there was a significant reduction in estimated values. At the highest rate of data loss heterozygosity dropped by 0.3 units and FST became negative. This would result in a gross underestimation of both genetic variation and population structure, and it suggests that missing data is an extremely important factor for noninvasive studies dealing with small samples sizes. Above a certain threshold it is possible that missing data exacerbates the effect of a small sample by reducing genetic information still further and this makes it difficult to estimate any genetic parameters (Kang et al. 2004; Amos 2006; Schlüter & Harris 2006). Therefore, researchers need to pay particular attention to PCR success rates to ensure that as many genotypes as possible are recorded for each individual.

3.5.2 Population structure

The program STRUCTURE is commonly used to test the presence of population structure within a study area as spatial differences in allele frequencies allows the program to cluster individuals into discrete groups. Because it appeared that allelic dropout and false alleles had little effect on allele frequencies it was expected that

STRUCTURE would be able to estimate the correct number of clusters for a wide range of error rates. The program output was robust to both error types, although there was some disagreement between the standard ln P(X|K) and Evanno !K methods with the smaller sample size. This could be due to the reduced support (low estimated log probability of the data) seen at high error rates. Similarly, STRUCTURE was not able to

- 89 - estimate the correct number of subpopulations above a certain value of missing data. This threshold was much higher for the total population (70-80%) than for the subsample (20%). STRUCTURE uses a departure from Hardy-Weinberg equilibrium to partition individuals into clusters and in theory this method is not affected by the presence of missing data as long as it is distributed randomly (Pritchard et al. 2010). However, the results presented here suggest that this assumption does not hold above a certain threshold, particularly if the sample size is small. Although the

STRUCTURE algorithm is not as powerful at small sample sizes (Kalinowski 2011), the sensitivity to missing data is notable. Here with >20% of genotypes missing, there was some disagreement between the ln P(X|K) and Evanno !K methods, and it was not possible to detect full population structure. This again supports the suggestion that researchers should focus efforts on increasing PCR success in order to observe as many complete multilocus genotypes as possible. In the presence of low ln

P(X|K) values or disagreement between the raw STRUCTURE and Evanno outputs, it is also advisable to pay close attention to geographic data and individual membership coefficients to determine which is the most likely pattern of spatial structure (Evanno et al. 2005; Pritchard et al. 2010).

3.5.3 Error thresholds With the increasing use of noninvasive techniques in conservation there has been more recognition of the role that genotyping error plays in genetics analyses. So far error types have been considered in isolation but the effect of concurrent errors provides a good basis for determining empirical thresholds for acceptable rates of allelic dropout, false alleles and missing data. Also, because researchers are often limited to a subset of samples when a proportion fail or are discarded, I have chosen to use results from the analysis of n = 10 to provide a guide to the lower bound of tolerable error. In the total population increasing the amount of the three types of error had little effect on the genetics parameters. However, when the sample was limited to 10 individuals there was a significant drop in estimates of genetic variation and differentiation with error rates of allelic dropout = 0.2, false alleles = 0.02 and

- 90 - missing data = 20%. In contrast, STRUCTURE was able to tolerate error rates up to allelic dropout = 0.5, false alleles = 0.05 and missing data = 50%. Taking into account the results of each error applied individually, this threshold effect is likely to be due to the increasing effect of missing data.

Consequently, it appears that a reasonable threshold for genotyping error is 20% allelic dropout, 2% false alleles and 20% missing data. This is much higher than many studies report and it suggests that estimates of variation and differentiation may be less biased than individual-based estimates, e.g. population size or paternity. In reality it is unlikely that all loci will have equal error rates and so it is better to consider overall error rates across the entire marker panel. The error rates in this study were determined for a panel of 20 loci and so give per locus error rates of 1% allelic dropout, 0.1% false alleles and 1% missing data. This is in agreement with previous work recommending maximum rates of 0.01 per locus (Taberlet et al. 1996; Waits & Leberg 2000; Waits & Paetkau 2005). This may seem particularly stringent when PCR success rates will vary between loci and between samples (Gagneux et al. 1997; Goossens et al. 2000; Lucchini et al. 2002; Hansen et al. 2008). Application of such a low threshold may result in prolonged analysis if many markers or samples have to be discarded. However, conducting a pilot study to screen a panel of markers may be sufficient to reduce error rates to an acceptable level (e.g. McKelvey & Schwartz 2005; Miller et al. 2002; Schwartz et al. 2006). Researchers may also need to design noninvasive studies to ensure sufficient time and resources to sample as extensively as possible. A larger initial collection of samples would allow the problematic ones to be identified and discarded without adversely affecting later analysis (Miquel et al. 2006).

Here I have used a panel of 20 microsatellites but it may be possible to reduce the overall level of error by reducing the number of markers. The size of the marker panel often represents a trade-off between reducing overall error rate and reducing both sampling variance and shadow effects (Creel et al. 2003). Using fewer loci

- 91 - reduces the number of PCR reactions needed to construct multilocus genotypes and also reduces the opportunity for error. However, a small number of loci increases the probability of individuals sharing the same genotype (Mills et al. 2000; Waits & Leberg 2000). Thus, although many studies report that as few as 5 loci may be sufficient to determine genetic variation and genetic structure (Luikart et al. 1998b; Rosenberg et al. 2001; Cavers et al. 2005; Palstra & Ruzzante 2008), in reality the final number depends on marker polymorphism, the relatedness between individuals, and the type of analysis to be conducted. Highly polymorphic markers with multiple alleles may have more power to distinguish between individuals and so fewer markers are needed (Kalinowski 2002). Similarly, where there are a large proportion of closely related individuals more markers are needed to differentiate between individuals (Waits & Paetkau 2005). For most noninvasive studies the probability of identity (PI) provides a convenient way to determine the minimum number of markers that should be used (Woods et al. 1999). At the threshold PI there are enough markers to distinguish between individuals and adding markers above this does not add any further power but may increase the opportunity for error (Waits & Leberg 2000).

3.5.4 Recommendations to reduce error Given that missing data may have the biggest influence on genetics analysis the principle focus of noninvasive studies should be to improve PCR success. This may include relieving PCR inhibition if using faecal samples, adjusting primer sequences and modifying PCR conditions. The first key step is to improve DNA template quality during sample preservation and DNA extraction. Preservation methods have long been recognised as a key determinant of DNA template quality and PCR success (Wasser et al. 1997; Frantzen et al. 1998; Murphy et al. 2002). Although different studies have tested a variety of reagents and conditions, the majority seem to agree that preservation in ethanol or complete drying with silica offer the highest DNA yield and PCR amplification rates (Wasser et al. 1997; Frantz et al. 2003; Roon et al. 2003; Smith et al. 2003; Nsubuga et al. 2004; Roeder et al. 2004). DNA yield can

- 92 - also be increased by using silica-based extraction methods (Rohland & Hofreiter 2007; Nechvatal et al. 2008; Ariefdjohan et al. 2010) or by performing multiple extracts per sample (Goossens et al. 2000). In particular, repeated extractions may reduce the concentration of PCR inhibitors in faecal DNA extracts (Kemp et al. 2006; Nechvatal et al. 2008; Fitzpatrick et al. 2010). PCR amplification is much less likely at starting DNA concentrations <25-35pg (Taberlet et al. 1996; Arandjelovic et al. 2009) and where possible attempts should be made to measure how much DNA is present in the final extract (Morin et al. 2001).

PCR success may also be increased by using 2-step amplification protocols (Bellemain & Taberlet 2004; Piggott et al. 2004; Hedmark & Ellegren 2006; Lampa et al. 2008; Arandjelovic et al. 2009). This usually involves an initial reaction to co- amplify the entire marker panel followed by singleplex reactions to amplify each marker individually (Bellemain & Taberlet 2004; Römpler et al. 2006). Care must be taken to monitor results of the first PCR as samples that do not amplify here are likely to fail again in the second step (Frantzen et al. 1998; Arandjelovic et al. 2009). In some cases the addition of bovine serum albumin (BSA) and dilution of the DNA extract may help to relieve PCR inhibition (Forbes & Hicks 1996; Kreader 1996; Al- soud & Rådström 2000; Rohland & Hofreiter 2007; King et al. 2009). PCR amplification rates may also be affected by the microsatellite primers used. Most wildlife studies rely on the cross-reactivity of primers designed for domestic species such as the cat or dog (e.g. Lucchini et al. 2002; Kohn et al. 1999; Ernest et al. 2003; Bhagavatula & Singh 2006). However, mutational differences often result in changes within the microsatellite flanking sequences which increases the potential for non- specific primer binding, particularly if repeat sequences are introduced (Dakin & Avise 2004; Kelly et al. 2011). Studies with high dropout rates could therefore benefit from redesigning primers to match the microsatellite sequences of the target species (Culver et al. 2001; Jane&ka et al. 2008). Researchers should also try to target short microsatellite sequences as small allele sizes are reportedly less likely to dropout than larger ones (Frantzen et al. 1998; Buchan et al. 2005; Broquet et al. 2007;

- 93 - Waits et al. 2007). Finally, in order to increase certainty in the number of unique individuals all consensus genotypes should be screened for mismatched genotypes that may have come from the same individuals e.g. with SHAZA or ALLELEMATCH (Macbeth et al. 2010; Galpern et al. 2012).

3.6 Conclusions Noninvasive samples provide a great opportunity for researchers to study elusive and cryptic species such as the Sumatran tiger. Estimates of genetic variation and differentiation are important measures of the conservation status of a population. However, studies of large carnivores commonly rely on either hair or faeces for genetic analysis and as such are prone to error. The bias associated with these errors can be minimised by reducing combined error rates to <1% allelic dropout, 0.1% false alleles and 1% missing data per locus. These thresholds increase markedly if each error type occurs alone. Unbiased measures of expected heterozygosity, weighted measures of differentiation, and STRUCTURE analysis perform well with high rates of error. However, these measures become more sensitive when sample sizes are small and the amount of missing data exceeds 40-60%. Therefore, researchers should attempt to increase PCR success rates so that as many individuals as possible have complete genotypes. For example, faecal samples could be preserved in ethanol, DNA could be extracted using silica-based techniques, and primers could be designed to target species-specific sequences.

- 94 - Chapter 4

Sample Collection and Genetic Analysis of Noninvasive Samples

- 95 - 4.1 Abstract Noninvasive samples have been used to study many wildlife populations. They are relatively easy to collect and can yield sufficient DNA for analysis of genetic variation, population subdivision and population size. Wild felids frequently deposit faecal samples along animal trails or small roads, and these can be used for genetic analysis. Whilst faecal DNA has been used to study tiger subspecies in Russia, India and central Asia, noninvasive methods have not yet been applied to studies of the Sumatran tiger. This study represents the first noninvasive genetics assessment of the wild Sumatran tiger population. Faecal samples were collected from Tiger Conservation Landscapes and protected areas for subsequent analysis in an Indonesian laboratory. DNA extraction and PCR protocols were optimised to account for the poor DNA quality, and consensus genotypes were derived using microsatellite loci. Overall, success rates were in line with similar noninvasive studies in other carnivore species.

4.2 Introduction The Sumatran tiger is one of only 6 extant tiger subspecies and is the only surviving island endemic since the Javan and Bali tigers were extirpated in the 1970s and 1930s respectively (Seidensticker 1987). The main threats to the Sumatran tiger are deforestation, human-tiger conflict and poaching. Sumatra has suffered some of the highest rates of deforestation in southeast Asia and primary forest is now mostly limited to submontane and montane regions. Lowland areas have been targeted as they are the most productive and suitable for conversion to plantations or settlements (Miettinen & Liew 2010; Broich et al. 2011). Human-tiger conflict is prevalent at the edges of national parks and tigers are still frequently killed for valuable body parts (Nowell 2000; Shepherd & Magnus 2004; Johnson et al. 2006; Linkie et al. 2008c). Recent estimates put the total Sumatran tiger population at just 500 individuals and they are thought to be largely restricted to Tiger Conservation Landscapes and some smaller protected areas (Wibisono & Pusparini 2010;

- 96 - Chundawat et al. 2011). Tiger Conservation Landscapes (TCLs) were identified as priority areas to safeguard sufficient connected tiger habitat with adequate protection for at least 10 years and Sumatra holds 12 of the 76 sites identified across the entire tiger's range (Dinerstein et al. 2006; Sanderson et al. 2006). The larger Indonesian TCLs were also highlighted for inclusion in the Tigers Forever Initiative that set out to increase tiger numbers by 50% in key sites (Dalton 2006). These concerted efforts to save the tiger have expanded in recent years to include genetic management. Genetic management is now recognised as one of the priorities for increasing tiger numbers alongside the protection of habitat, protection of prey species and anti- poaching measures (Sunarto et al. 2012). Genetic data adds to more traditional ecology and demographic methods by providing information on population size, relatedness, sex ratios, migration rates, population structure and levels of genetic variation (Taberlet et al. 1999). Genetic techniques can also be used as a forensic tool to reinforce law enforcement (Wasser et al. 2007; Ogden et al. 2009). Despite these advantages genetic studies have not yet been conducted on Sumatra. Many papers have outlined patterns of abundance and occupancy (e.g. Wibisono & Pusparini 2010; Wibisono et al. 2011; Linkie et al. 2008b; Sunarto et al. 2012) and genetic analysis could help to provide another layer to conservation management decisions. Noninvasive methods have been successfully applied to studies of the Amur and Bengal tiger to determine population structure and population size (e.g. Joshi 2010; Henry et al. 2009; Mondol et al. 2009a; Bhagavatula & Singh 2006; Sharma et al. 2009; Sharma et al. 2010; Reddy et al. 2012). Similar studies in the Sumatran subspecies would allow us to determine population boundaries and the rates of gene flow between them. Connectivity is an important consideration for the landscape-scale management that has recently been proposed for tigers (Sanderson et al. 2002b; Wikramanayake et al. 2011). Genetic analysis would also allow genetic diversity to be compared to the mainland. Previous studies suggest that Sumatran tigers may be genetically impoverished compared to the larger populations in India and central Asia (Luo et al. 2004; Mondol et al. 2009b).

- 97 - Noninvasive samples such as faeces are a convenient source of DNA for endangered populations. Sampling does not require the trapping or handling of individuals and they are suited to detecting multiple individuals within a given site (Kohn et al. 1999; Prugh et al. 2005). They are particularly useful for cryptic and wide-ranging species such as the tiger as it is relatively straightforward to collect scats from the trails that individuals use to traverse their territories. Faecal DNA methods make use of the intestinal epithelial cells lying on the surface of the sample and these cells may be well preserved in cold or frozen environments. However, they are at risk of damage in tropical environments if the samples are old or if they have been exposed to sunlight, high temperatures or humidity (Reed et al. 1997; Bhagavatula & Singh 2006). For this reason, faecal DNA extracts usually contain both low quantities of DNA and low quality DNA templates. Noninvasive methods must account for this by incorporating optimised techniques for sample preservation, DNA extraction and PCR amplification. Multiple extractions per sample and multiple PCRs per extract are often needed to reduce the effect of genotyping errors such as allelic dropout and false alleles on the construction of multilocus genotypes (Frantzen et al. 1998; Goossens et al. 2000; Hansen et al. 2008). Here I present the sampling and laboratory methods for a recent noninvasive genetics survey of the Sumatran tiger. Faecal samples were collected from key TCLs and protected areas on Sumatra and genetic analysis was performed using optimised laboratory techniques. The genetic information generated from this study will be used to complement previous demographic and ecological data for the management of this critically endangered species.

4.3 Field methods

4.3.1 Study sites Sample collection was conducted from July 2009 - October 2010 in 6 TCLs and other protected areas on Sumatra (Table 4.1). Surveys were done under permission from the Ministry of Research and Technology (Permit No. 0183/FRP/SM/VIII/09 and

- 98 - 07/TKPIPA/FRP/SM/VII/2010), the Department for Forest Protection and Nature Conservation (Permit No. SK.199/IV-SET/2009 and SK.96/IV-SET/2010) and local national park offices. The sites for sample collection were chosen to target those areas with established high tiger density or frequent encounters of tiger signs. Habitat type ranged from montane primary forest to a logging concession and lowland forest. Some samples were collected during dedicated scat collection surveys and some were collected opportunistically prior to this study by NGOs conducting other surveys, e.g. for camera trap placement or footprint surveys. Not all field samples could be included in this study due to concern about the validity of Department of Forestry permits at the time of collection. This also applied to attempts to obtain samples from wild-caught tigers held in Indonesian zoos.

Table 4.1 Tiger Conservation Landscapes (TCLs) and protected areas sampled in this study.

Name Province TCL Class & Size (ha) Habitat type Priority Way Kambas NP Lampung - 130,000 Lowland forest

Batang Hari West Sumatra - 27,800 Hill dipterocarp forest Kerinci Seblat NP West Sumatra I 1,400,000 Lowland/ Global submontane forest Ulu Masen ecosystem Aceh IV 750,000 Montane forest - Tesso Nilo NP Riau III 38,000 Lowland forest Long-term Bukit Tigapuluh NP South Riau I 140,000 Lowland forest Global Berbak NP Jambi IV 160,000 Peat swamp forest - Sembilang NP South Sumatra -

Bukit Barisan Selatan Lampung III 360,000 Lowland forest NP Long-term Class I - habitat to support at least 100 tigers, evidence of breeding, minimal-moderate levels of threat, effective conservation measures in place; Class III - habitat to support some tigers, moderate- high threat and minimum conservation investment; Class IV - lack sufficient information for

- 99 - classification. Global priority - highest probability of persistence of tiger populations over the long- term. Best representatives of tiger habitats; Long-term priority - require sustained long-term effort to restore to Class I status. In near term, important areas for national tiger conservation strategies.

4.3.2 Survey method Previous studies have shown that large carnivore scats are more likely to be found on trails and logging roads that are used for scent marking or traversing territory (Sunquist 1981; Hájková et al. 2006; Kerley & Borisenko 2007). To increase capture probability further, I chose to target sampling effort to those areas that NGOs reported as having a high abundance of tiger sign or a high number of camera trap photos. Survey teams followed the trails most likely to be used by tigers and collected all the scats they found during each transect. Each team covered one transect per day and each route was sampled just once. Each sampling block lasted for an average of 2 weeks. For 3 surveys I also tested the use of a detection dog for scat collection. Detection dogs are better at differentiating between scats from different species and are capable of finding more scats than experienced personnel searching visually (Kerley 2002; Meadows 2002; Smith et al. 2003; Wasser et al. 2004). The highest rates of success are achieved through the careful selection of dogs (e.g. individuals with good concentration and motivation), good training and handler experience (Hurt et al. 2000; Meadows 2002; Smith et al. 2003). A 2-year old, male Labrador Retriever from Bogor, West Java was trained over 3 weeks to recognise the scent of tiger scats (Fig. 4.1). Indonesians are generally nervous around dogs, but Rudi's breed and good temperament made it much easier to integrate him into the field teams. Dog surveys were conducted in Way Kambas NP, Kerinci Seblat NP and Batang Hari forest (Fig. 4.2). Although Rudi was sourced from Indonesia and already acclimatised to the tropical conditions, the effectiveness of scent detection is affected by temperature and how hot the dog gets during periods of work. To avoid any problems with loss of concentration or overheating, Rudi was worked intermittently from early morning to early afternoon with 20 minute periods of work and 5-10 minute rest breaks.

- 100 - Figure 4.1 Detection dog training in Bogor, West Java. A 2 year old, male, Labrador Retriever was given initial training by Richard Clarke of the Barking Mad Dogs Training School in Hertfordshire, UK.

Figure 4.2 Each scat collection team consisted of 2-3 staff and a dog handler. Teams searched along animal trails and ex-logging roads for tiger scats.

- 101 - Scats less than 7 days old are best for DNA extraction as the quality of DNA on the surface of the scat decreases with age, and the risk of genotyping error increases (Farrell et al. 2000; Piggott 2004; Prugh et al. 2005; Hájková et al. 2006; Santini et al. 2007). However, as ancient DNA studies have shown that it is still possible to obtain DNA from much older material, I decided to collect every scat that was encountered on a trail. A larger number of samples is also preferable as it is extremely difficult to obtain good quality samples from a tropical forest environment. High humidity, high temperatures or heavy rain all act to wash away or destroy the colon cells on the surface of each scat. Each sample was given a unique code and photographs were taken with a measurement marker. Samples that were indicated by the detection dog were also noted. Other information such as environmental conditions, age and condition of faeces, and substrate were also recorded. All samples were preserved by drying with silica gel beads or in 96-99% ethanol to ensure complete dehydration of the sample (Wasser et al. 1997; Hansen & Jacobsen 1999; Taberlet & Luikart 1999; Beebee & Rowe 2004; Kurose et al. 2005; Bhagavatula & Singh 2006; Smith et al. 2006). Samples were preserved as soon as possible on return to base camp or if posted, as soon they were received in the laboratory.

4.4 Laboratory Methods Although the transport of faeces is not restricted under CITES, it was not possible to obtain permits to export the faecal samples from Indonesia to the UK. All faecal samples were therefore stored and processed at the Eijkman Institute for Molecular Biology in Jakarta, Indonesia. Each sample was subjected to DNA extraction and confirmation of the presence of tiger DNA before progressing to genotyping. All extractions were conducted in a dedicated room in a HESKA UV hood with equipment cleaned in 1% sodium hypochlorite (bleach) between samples. To avoid the presence of PCR inhibitors in the DNA eluate I took 2-3mm scrapings from the outer surface and tip of each scat using a clean scalpel blade (Stenglein et al. 2010). In

- 102 - the case of scats consisting of loose material, I took samples of the substrate as well.

4.4.1 DNA extraction & species identiﬁcation I used the QIAamp DNA stool kit (Qiagen) for all extractions. It is based on a guanidine thiocyanate method which has been shown to be very good for DNA extraction from faeces (Boom et al. 1990; Höss & Pääbo 1993; Nechvatal et al. 2008; Fitzpatrick et al. 2010). Cell lysis occurs in the presence of guanidine thiocyanate which allows nucleic acids to bind to a silica matrix. This silica-nucleic acid complex is then washed in a washing buffer before elution of the purified DNA. The addition of carrier RNA and prolonged digestion with proteinase K have also been shown to increase the DNA yield from a variety of samples (Kishore et al. 2006; Rohland & Hofreiter 2007; Shaw et al. 2009). Thus, I used the Qiagen 'Isolation of DNA from stool for human DNA analysis protocol' with the following modifications:

(i) At Step 2, 25µl proteinase K was added to the faecal sample in ASL buffer and left in an incubator shaker overnight to increase homogenisation of the faecal material. (ii) At Step 10, 4µl carrier RNA was added to each 600µl aliquot of Buffer AL before adding to the proteinase K/supernatant mixture. (iii) At Step 11, the sample was placed in an incubator shaker for 2 hours after addition of Buffer AL and proteinase K to the supernatant. (iv) 150µl of Buffer AE was used to perform the elution of DNA in the final step.

There are 3 other medium-sized felids that occur on Sumatra, the clouded leopard, Asiatic golden cat, and the marbled cat, so I used a mitochondrial DNA marker to identify tiger scats. I initially tested three mitochondrial primers in a pilot study using captive tiger DNA: cytochrome Bb, NADH 2c (Driscoll et al. 2009), and Tig 117F/

- 103 - 231R (Wetton et al. 2004). The Cytbb and ND2c primers were designed for use with ancient DNA giving a PCR product of <200bp, which is also ideal for faecal DNA. Tig 117F/231R is an alternative tiger-specific cytochrome b primer which produces a PCR product of ~165bp. Primer dimer is expected with negative samples or if tiger DNA is <0.01ng. The cytochrome Bb and NADH 2c primers worked well with captive tiger samples but gave ambiguous results with some of the field samples. The Tig 117F/231R primer produced more consistent results and so was used for all subsequent screening (Table 4.2). At least 2 PCRs were performed for each sample to confirm a positive result. A positive result was indicated by a single PCR product of the expected size, whilst negative samples produced no band or multiple stutter bands indicating non-specific primer binding. PCRs were performed in a 10µl volume containing 4µl Qiagen multiplex PCR mix (Qiagen Inc.), 1.5µl forward primer (10pmol/µl), 1.5µl reverse primer (10pmol/l), 0.2µl BSA, and 1.2µl faecal DNA. PCR cycling conditions were as described by Driscoll et. al. (2009) and the PCR product was visualised on 2% agarose gel with 1% ethidium bromide.

Table 4.2 A list of PCR primers used in this study.

Test Primer Primer sequence (5'-3') Expected product size Species ID Tig 117F TTTGGCTCCTTACTAGGGGTG 165bp Tig 231R GGCATGTAGATATCGGATAATC Sex ID Zn finger F AAGTTTACACAACCACCTGG 160, 163bp Zn finger R CACAGAATTTACACTTGTGCA Sex ID ZFX-PF TACCGAGCGATATAGCTCCAG 205bp ZFX-PR GTGTTCCTACGTTAAGCTATTG DBY7-PF CTCATGAAGCCCTATTTTTGGTTG 156bp DBY7-PR ACGGCGTCCGTATCTTCCA

- 104 - 4.4.2 Sex identiﬁcation Historically the SRY gene has been used for sex determination as it is specific to the Y chromosome and so only produces a PCR product in males (Kohn et al. 1999; Beebee & Rowe 2004). This can be problematic when using faecal samples as failure to amplify could mean either a female sample, poor DNA template or PCR inhibition. Alternatives such as the zinc-finger gene are found on both sex chromosomes and produce a different PCR product for each sex. I tested 2 methods for sex identification: 1. Felid-specific primers ZFX/ZFY (Pilgrim et al. 2005) and 2. Tiger-specific primers Zfx/DBY (Sugimoto et al. 2006). The ZFX/ZFY primers amplify a portion of the zinc finger region that has a 3bp deletion in males (Table 4.2) and this results in the formation of two PCR products in males (160bp and 163bp) compared to a single PCR product in females (163bp). I fluorescently labelled the forward primer to allow the PCR products to be run on an Applied Biosystems genetic analyser. PCR reactions were performed using a 10µl reaction volume containing 4µl Qiagen multiplex PCR mix, 1.5µl forward and 1.5µl reverse zinc finger primer (10pmol/µl), 0.5µl BSA, and 3µl of faecal DNA. PCR cycling conditions were as follows: 95°C for 15 min, 45 cycles of 94°C for 30s, 56°C for 1 min and 72°C for 30s, followed by 72°C for 10 min. To account for allelic dropout, which is common with poor quality DNA, sexing was done on the basis of 3 positive PCRs for females and 2 positive PCRs for males. The Zfx/DBY primers amplify part of the zinc finger gene and Y-chromosome specific DEAD box polypeptide gene. This produces fragments of 205 bp and 156 bp respectively (Table 4.2). Samples from males are expected to produce 2 bands (205bp and 156bp), whilst samples from females will produce a single band (205bp) when visualised on 3% agarose gel. Allelic dropout is also expected with these markers due to the low DNA quality and/or quantity found in faecal DNA samples. The recommended PCR conditions are as follows: 95°C for 15 min followed by 50 cycles of 94°C for 30s, 56°C for 30s and 72°C for 45s, followed by a final extension step of 72°C for 10 min. I also used 3 positive PCRs to confirm female samples and 2 positive PCRs for males.

- 105 - 4.4.3 PCRs Panels of microsatellite loci are commonly used to provide unique genotypes for each sampled individual. Microsatellites are short runs (<300bp) of di/tri/tetranucleotide repeats that are scattered throughout the genome and they are assumed to be selectively neutral. They are very polymorphic because they experience high mutation rates that change their repeat array length. Almost 600 microsatellite loci with cross-species polymorphism have been described for the domestic cat (Menotti- Raymond & O'Brien 1995; Menotti-Raymond et al. 1999; Menotti-Raymond et al. 2003), and many have been used in studies of wild felids with good success (e.g. Ernest et al. 2000; Eizirik et al. 2001; Jane&ka et al. 2008; Luo et al. 2004; Uphyrkina et al. 2001). To allow comparisons with other studies (e.g. Luo et al. 2004; Mondol et al. 2009a; Xu et al. 2005; Bhagavatula & Singh 2006), I used the following fluorescently labelled microsatellite loci: Fca-5, Fca-8, Fca-32, Fca-44, Fca-69, Fca-77, Fca-90, Fca-91, Fca-94, Fca-96, Fca-105, Fca-123, Fca-126, Fca-129, Fca-139, Fca-161, Fca-176, Fca-201, Fca-211, Fca-212, Fca-220, Fca-229,

Fca-242, Fca-290, Fca-293, Fca-304, Fca-310, Fca-391, Fca-441. I used MULTIPLEX

MANAGER v1.0 (Holleley & Geerts 2009) to visualise and help construct multiplex PCR groups based on allele sizes and minimal potential for interaction between primers.

Because of the difference in DNA quality between the captive and field samples, I used different PCR protocols for each group. For all samples I adopted a three-step PCR protocol (Smith et al. 1995) to try to reduce the incidence of stutter peaks and ambiguous alleles that is common with faecal DNA. PCRs with captive DNA were performed in a 10µl reaction volume containing 4µl Qiagen multiplex PCR master mix, 1µl (0.2µM) multiplex primer mix and 0.5µl BSA with 2µl faecal DNA. PCR conditions were as follows: 95°C for 15 min, 20 cycles of 94°C for 15s, 55°C for 15s and 72°C for 30s, followed by 35 cycles of 89°C for 15s, 55°C for 15s and 72°C for 30s, then a final extension step of 60°C for 90 min. PCRs with field DNA were performed using 4µl Qiagen multiplex PCR master mix, 2µl (0.4µM) multiplex primer mix and 1µl BSA with 3µl faecal DNA. High levels of BSA are recommended to

- 106 - minimise PCR inhibition particularly when using poor quality DNA (Kreader 1996; Rohland & Hofreiter 2007). The PCR protocol was modified to improve PCR success by increasing the number of cycles and reducing the annealing temperatures slightly: 95°C for 15 min, then 45 cycles of 94°C for 15s, 53°C for 15s and 72°C for 30s, followed by 45 cycles of 89°C for 15s, 53°C for 15s and 72°C for 30s, then a final extension step of 60°C for 90 min. Allele sizes were read on an ABI 3130 capillary sequencer using Genescan Liz 500 size standard (Applied Biosystems) and GeneMarker software (SoftGenetics LLC).

4.4.4 Genotyping Genotyping errors are relatively common when using poor quality DNA such as that obtained from scats (Taberlet et al. 1996; Waits & Leberg 2000). With good quality samples it is sufficient to rely on one or two PCRs to construct a genotype. However, with faecal samples a multi-tubes approach is required to ensure that the consensus genotype has minimal errors (Navidi et al. 1992). Sources of error include (i) allelic dropout (false homozygote), which results from the failure of one allele to amplify, (ii) false alleles, which occur due to PCR slippage during the amplification process, and (iii) sporadic contamination from other DNA sources (Taberlet & Luikart 1999; Bonin et al. 2004). The multi-tubes approach uses multiple PCRs to account for allelic dropout and false alleles and thus improve accuracy (Taberlet et al. 1996; Goossens et al. 2000; Frantz et al. 2003). An allele must appear twice to be accepted as a true allele; a heterozygote genotype may be accepted after 2 positive PCRs and a homozygote can be provisionally accepted after 3 positive PCRs. Single PCRs are added one at a time up to a maximum of 7 until a consensus genotype is determined (Frantz et al. 2003; Hansen et al. 2008). I used the multi-tubes method with a negative control to detect contamination. Some samples required multiple extractions and up to 10 PCRs before a consensus genotype could be constructed.

As with other genotyping software, GeneMarker provides allele sizes with multiple decimal places. This occurs because the comparison between the size standard and

- 107 - migration of the PCR product in the capillaries is imperfect and varies between runs. Variation in allele sizes also occurs if there is non-templated addition of adenine to the end of the amplified fragment by Taq polymerase, which leads to the formation of stutter peaks that are usually 1-2 repeat units after the true allele (Smith et al. 1995; Brownstein et al. 1996; Magnuson et al. 1996). Also, many dinucleotide alleles do not have the expected repeat length of 2bp and instead range from ~1.7-2.3bp (Amos et al. 2007). In order to construct a consensus genotype each allele has to be categorised into an allele bin. Each bin represents the mean length in base pairs for that allele, rounded to the nearest integer (Amos et al. 2007). It is difficult to assign alleles to the right reference bin if the correct allele size distribution is not well documented or if there is variation in allele sizes between PCRs. However, this job is simplified by the many automated allele binning methods that have been described to standardise allele sizes across all samples and PCRs (e.g. Alberto 2009; Idury & Cardon 1997; Matsumoto et al. 2004; Johansson et al. 2003; Morin et al. 2009). The most common method is one of least squares minimisation between observed allele length and mean allele bin length. I chose the method implemented in the program

TANDEM v1.08 (Matschiner & Salzburger 2009) for ease of use and compatibility with the conversion program CONVERT v1.31 (Glaubitz 2004). TANDEM uses a power function to round the raw allele sizes to the nearest integer and gives an estimate of rounding error for each locus. Outliers are also identified so that they can be checked and/or removed from the dataset. To reduce the incidence of outliers I only considered allele peaks with the expected morphology and with fluorescent intensity of >100 for final analysis. I discarded any ambiguous peaks, e.g. multiple stutter peaks, >2 heterozygote peaks, abnormal peak morphology, etc. from the final data set.

4.4.5 Error rates Error rates are relatively straightforward to estimate by comparing consensus (faecal) genotypes to true reference (blood) genotypes (e.g. Bhagavatula & Singh 2006; Mondol et al. 2009a; Adams & Waits 2007), or by taking 5% of samples for

- 108 - repeated blind genotyping and comparing these to the first reference genotype (Bonin et al. 2004; Pompanon et al. 2005). Blood samples were not available in this study, so I used automated methods (GIMLET v1.3.3, MICRO-CHECKER v2.3.3, and

PEDANT v1.0) to estimate rates of allelic dropout, false alleles and null alleles (Valière et al. 2002; van Oosterhout et al. 2004; Johnson & Haydon 2007b). GIMLET estimates error by comparing all PCR repeats to the consensus genotype. MICRO-CHECKER identifies null alleles, large allele dropout and stutter by comparing the observed frequencies of homozygote and heterozygote size classes to a random allele distribution derived from the data set. PEDANT uses a maximum likelihood approach to compare duplicate genotypes for each individual. It then estimates per-allele error rates from the frequency of mismatches due to allelic dropout or false alleles.

Another factor that influences the ability to construct accurate genotypes is the power of the chosen microsatellite panel to distinguish between individuals. The probability of identity (PI) is the probability that two unrelated individuals will have the same multilocus genotype (Taberlet & Luikart 1999; Waits et al. 2001; Allendorf & Luikart 2007). The higher the polymorphism of the loci used or the larger the number of loci, the lower this probability will be (Mills et al. 2000; Waits et al. 2001). Ideally genetic profiles should have enough loci to distinguish between individuals with 99% certainty and this usually corresponds to a threshold PI of 1 x 10-3 - 1 x 10-5 with 5-10 loci (Boecklen & Howard 1997; Shriver et al. 1997; Waits et al. 2001). However, small or endangered populations may have a high proportion of closely related individuals, non-random mating or association between loci due to population structure. In these cases the probability of identity for siblings (PISIB) is more appropriate as it represents the upper limit on the possible range of PIs in a population:

2 1 1 1 " % 1 2 2 4 (4.1) PISIB = + ! pi + $ ! pi ' ( ! pi 4 2 i 2 # i & 4 i

- 109 - where pi is the frequency of the ith allele. For this study I used the best 24 microsatellite loci and estimated the multi-locus value of PISIB using GIMLET (Taberlet & Luikart 1999; Waits et al. 2001; Valière 2002). It is also important to determine how many unique individuals are represented in the data set. The most common methods compare genotypes at each locus allowing for mismatches, and a new method implemented in SHAZA is able to account for genotyping error (Macbeth et al. 2010).

SHAZA uses a likelihood test to distinguish between 3 different types of genotype match - (i) False matches in which different individuals have the same genotype (shadows), (ii) False non-matches that represent the same individual with different genotypes due to genotyping error, and (iii) Phantoms that are true matches rejected because of insufficient power. Phantoms are more likely to occur in data sets with incomplete or partial genotypes, and it is these individuals that SHAZA highlights as potential recaptures of the same individual.

4.5 Results 4.5.1 Sample collection A total of 148 scats were collected over the 15 months of sampling (Table 4.3). Samples were collected from 9 different field sites and one captive facility housing tigers that had been removed from areas of human-tiger conflict (Fig. 4.4). Transect length varied from ~2.5 - 10km, and the number of scats encountered at each site varied from 0 to 81. The variation in the number of scats per site mainly reflects differences in sampling effort (e.g. total survey time) and terrain, with lowland sites (e.g. Batang Hari) yielding far more samples than submontane regions (e.g. Ulu Masen). Sampling effort also varied depending on the survey type, i.e. some surveys were conducted specifically to search for scats, and others were conducted primarily for camera trap or tiger presence surveys. Detection rate would also have been affected by the relative experience of the field teams. More scats were noticed on open trails and logging roads compared to forest animal trails, and this is probably because of heavy leaf litter and increased decomposition on the forest floor. Scats on open trails were usually dry and present in the middle of the trail so easy to spot.

- 110 - The age of each scat could only be estimated subjectively and was judged to vary from relatively fresh (up to 1 week) to old and desiccated (up to 1 month). The time between collection and extraction varied from 1 month - 2.5 years and was biased by those samples that had been collected prior to this study. Scat contents included hair, bone fragments, body parts (claws, quills), soil and vegetation. The limited availability of reagents meant that scat quality varied from dry to wet and in some cases mouldy once they reached the lab. None of the scats that Rudi indicated appeared to contain tiger DNA when tested with the forensic primers. This could be for a variety of reasons, such as poor detection or praise and inadvertent reinforcement for any scat found regardless if it was tiger or not. Ultimately, because of the added cost and logistical implications of transporting the dog (e.g. food, grooming, regular exercise, ongoing training), it was far more effective to rely on teams searching visually.

Table 4.3 A list of locations from which faecal samples were collected.

Survey site Region Total no. of No. of positive samples samples Ulu Masen North 6 3 Kerinci Seblat NP West 27 6 Batang Hari production forest (PT AMT) West 81 9 Riau (Tesso Nilo, Kerumutan, Bukit East 10 2 Tigapuluh NP) Bukit Tigapuluh NP Central 3 1 Berbak NP East 3 1 Sembilang NP East 1 0 Bukit Barisan Selatan NP Southwest 0 - Way Kambas NP Southeast 7 5 Taman Safari Indonesia (zoo)* Mixed 10 10 Total 148 37

*Includes tigers from Riau, Jambi, Aceh, and Medan

- 111 - Figure 4.3 A total of 148 faecal samples were collected from 6 Tiger Conservation Landscapes, 3 protected areas and other human-tiger conflict zones on Sumatra.

4.5.2 PCR success The Qiagen stool mini kit was straightforward to use and I was able to extract DNA from all the captive and field samples without difficulty. DNA concentration per extract ranged from ~6-192ng/µl measured by NanoDrop (Thermo Scientific) but this

- 112 - did not appear to correlate with PCR success. Positive samples were relatively easy to score using the Tig 117F/231R primer which showed good amplification in all positive samples. Of the 148 samples tested 37 were positive, i.e. produced a single clear band in 2 positive PCRs (Table 2.3). Negative samples produced primer-dimer (small <100bp PCR product) or other non-specific products. With the mtDNA primers available it was not possible to determine the species origin of the negative samples without sequencing, and the cost and time involved precluded testing with other felid primers. The sex identification primers worked well with the captive samples but gave mixed success with those from the field. For the samples that were successful, the zinc finger markers produced fragments of 161bp and 165bp for males, and

165bp for females when binned using TANDEM. The zinc finger/DBY markers gave the expected product sizes of 156bp and 205bp as judged with a 100bp ladder. Samples were scored as male if they produced the 156bp alone or with the 205bp band. Samples were scored as female if they produced the 205bp band alone. The field samples produced varying results with a mix of good peaks, stutter peaks, artefacts, or peaks not of the expected size. Overall, I was able to sex 17 out of the 37 samples with a sex ratio of 12 males and 5 females.

For some of the duplex microsatellite primer groups one primer tended to amplify consistently well while the other did not. Altering the ratio of primers from 1:1 to 1:4 in favour of the poorly amplifying primer (Bercovich et al. 1999) appeared to have little effect. Pre-amplification protocols have also been suggested to improve PCR success (Piggott et al. 2004; Hedmark & Ellegren 2006; Lampa et al. 2008), but again this did not appear to provide any significant improvement. Loci Fca-91 and Fca-242 had to be removed from the microsatellite panel due to consistent null amplification. Loci Fca-5, Fca-126 and Fca-212 were also removed as PCRs could not be performed on all samples. I discarded 12 samples with the lowest PCR success (<10%) from further analysis. The remaining 25 individuals were thus genotyped at 24 loci. The mean number of positive PCRs estimated with GIMLET was 0.57 (range 0.25 - 0.88) across loci and 0.57 (range 0.36 - 0.77) across samples.

- 113 - The proportion of missing data per locus ranged from 12-72%. Overall, amplification with the microsatellite loci was not as successful as the initial cytochrome b test with Tig 117F/231R. This may be because the mitochondrial primer was designed to be tiger specific, but is more likely because mitochondrial DNA is present in relatively much larger quantities in each cell and microsatellite repeat arrays are more prone to template damage. The 7 PCRs under the multi-tubes approach was sufficient to genotype the majority of samples, although a small proportion required double this amount due to high rates of allelic dropout and some false alleles. Some microsatellite loci were also more prone to stutter peaks than others. TANDEM appeared to be consistent in assigning alleles to bins despite the variation in allele sizes between PCR repeats, and the transformed bins appeared to match an alternative method implemented in FlexiBin (Amos et al. 2007).

4.5.3 Error Although 37 samples clearly showed the presence of tiger DNA in the species identification test, only 25 produced adequate genotypes to allow further analysis. Modification of the PCR protocol and repeated extractions did not improve PCR success. This suggests that there may have been high levels of PCR inhibition. Faecal samples are known to contain many inhibitors such as humic acid or polysaccharides from the diet or environment in which they were collected (Bessetti 2007). I used the Cleanup of DNA protocol from the QIAamp DNA Investigator Handbook (Qiagen) which is designed to remove inhibitors by washing with Qiagen buffers and purification with the spin column system. Although this type of repeated extraction has been reported to give good results (Kemp et al. 2006; Fitzpatrick et al. 2010), it did not appear to greatly improve PCR success.

-8 The 24 microsatellite loci gave a PISIB value of 1.57 x 10 , and the most informative marker was Fca-94 as estimated by GIMLET. There was no evidence of null alleles on testing with MICRO-CHECKER. Locus Fca-220 was reported to have a homozygote excess which may be due to allelic dropout. Overall, GIMLET estimated a per-

- 114 - genotype allelic dropout rate of 0.40 and false allele rate of 0.05 across loci. This result was largely in agreement with PEDANT which estimated per-allele error rates of 0.39 and 0.09 respectively. These results are similar to other studies in which allelic dropout occurs much more frequently than the incidence of false alleles (Broquet & Petit 2004). For carnivores, allelic dropout rates vary from below 1% to ~27% and false allele rates are typically in the region of 3-8% when noninvasive samples are used (e.g. Frantz et al. 2003; Goossens et al. 2000; Lucchini et al. 2002; Piggott et al. 2006; Scandura et al. 2006). SHAZA detected one possible match between two northern samples in Ulu Masen. Visual inspection of the genotypes confirmed that they could be a match, and they were located close enough geographically to have originated from the same individual.

4.6 Discussion 4.6.1 Sample collection This study represents the first genetic survey of the wild Sumatran tiger population to include a large number of TCLs and protected areas. I was able to make use of freshly collected samples and those from NGOs that had collected scats as part of regular camera trap or occupancy surveys. Thus, the distribution of sampling sites provides a good representation of the tiger population on Sumatra. It was not possible to survey more sites in the time available for field work due to the time it took for survey and transport permits to be issued by the Department for Forest Protection and Nature Conservation Forestry. The application process took ~3-4 months to be approved, but each permit was then only valid for 3 months before it needed to be renewed. Genetic surveys such as this therefore rely on a huge amount of time and effort to be able to cover a large enough area and to generate a reasonable number of samples. Sampling over 15 months generated a large number of scats (n = 148) but only 37 tiger samples representing ~7% of the total Sumatran tiger population. This is in part due to the low population density of tigers with only 1-2 individuals/100km2 (Linkie et al. 2008b; Wibisono et al. 2011). It also serves to

- 115 - highlight that faecal studies of carnivores need to be conducted over at least 1 year to be effective.

It was difficult to find suitable candidates in-country for detection dog training. Due to the limitations of cost and import permits Rudi was bought from a breeder in Indonesia. Dog rescues are not common and hunting dogs are not easily trained for this type of work. Typical scent breeds such as the German Shepherd are available but because they are primarily used for police and customs work they are rarely sold privately. Rudi was a good choice because of his breed, temperament and willingness to learn. Integrating him into the field team required some adjustments to the standard survey procedures. Preparations for the surveys needed to include the provision of dog food and a cage to transport and hold Rudi overnight. Indonesians do not readily handle dogs or work closely with them as they often believe that they are dangerous and that their saliva is unclean. This provided the main challenge for obtaining permits to conduct surveys within national parks. However, following an explanation of the detection dog principle and demonstrations of Rudi's ability, I was able to obtain permits from the Department of Forestry for surveys in 3 protected areas. Rudi's presence did not appear to cause any significant problems for the other members of the field team, although the novelty of working with a dog and his friendly nature meant that members often tried to attract his attention which removed his focus from searching for scat.

The use of scent dogs has been recommended to increase the efficiency of sampling, but it is important that the dog is well trained and suited to the sampling environment. Even dogs that have shown good promise in the initial 4-6 week training period will not necessarily be suited to field work (Hurt et al. 2000; Smith et al. 2003; Kerley & Borisenko 2007). Although Rudi was able to find and indicate tiger scat along trails and open searches during training, once in the field it was clear that 3-4 weeks training was not sufficient for full length surveys. The high temperatures and humidity meant that regular breaks were necessary, and it took time to adjust to

- 116 - each new site. Motivation and focus also seemed to be a challenge. Tigers occur at very low density so scats will not commonly be encountered during a transect. This makes it difficult to maintain motivation for a detection dog when there is very little positive reward for consistent effort. Ongoing training and more survey experience may have helped, but ultimately the field teams were more confident conducting surveys on their own without the responsibilities of dog handling and training.

4.6.2 Sample preservation and PCR success Sample preservation is often challenging for field samples as it depends on the availability of reagents, storage equipment or space. It proved much easier to supply silica beads and zip loc bags to field teams, particularly if the samples were to be posted to the laboratory. This reduced the weight and amount of equipment to be carried and the risk of leakage while in transit. The majority of samples were then transferred to ethanol once they reached the laboratory. Many studies report higher DNA yield, PCR success and post-collection longevity with ethanol (Wasser et al. 1997; Murphy et al. 2002; Frantz et al. 2003; Nsubuga et al. 2004; Roeder et al. 2004), although there are also studies that report higher success with silica-dried samples (Wasser et al. 1997; Smith et al. 2006) or DET buffer (Seutin et al. 1991; Frantzen et al. 1998). The main criterion for any faecal preservation protocol is that the samples are dried sufficiently to prevent further DNA degradation from mould growth or the action of nuclease enzymes. DNA preservation may also be aided by freezing samples at -20°C or -70°C, although this requires the provision of large freezers that must be well maintained with a regular electricity supply. Unfortunately, there was not enough freezer space available at the Eijkman laboratory so all the samples remained in either silica or ethanol at room temperature.

The Qiagen stool mini kit has been used in many studies with good success. It is straightforward to use and has been designed to concentrate DNA and remove PCR inhibitors. Despite this, and despite the addition of the Cleanup of DNA protocol, there was still evidence of PCR inhibition in later reactions. Many different methods

- 117 - have been proposed to reduce the level of inhibition or to remove inhibitors from the DNA extract. Methods include serial dilution, increasing the amount of Taq polymerase per reaction, addition of BSA, sequential extraction, or isopropanol DNA precipitation (Kemp et al. 2006; Nechvatal et al. 2008; King et al. 2009; Fitzpatrick et al. 2010). Alternatively, some authors suggest homogenising the sample and extracting DNA with the original GuSCN method is more effective (Taberlet et al. 1996; Frantz et al. 2003; Rohland & Hofreiter 2007; Ariefdjohan et al. 2010). BSA had already been included in all the PCR reactions and a further modification would have been to dilute each extract (1:50 or 1:100) to reduce the effectiveness of any inhibitors (King et al. 2009). Initially I had decided not to dilute poor samples as they may have contained very low copy numbers of the target template. Positive control PCR can also be used as a screening test for inhibition. A small aliquot of the test DNA sample is added to a PCR containing a positive control template. In the absence of inhibition, amplification will occur and a PCR product of the expected size will be detected. However, if inhibitors are present in the test DNA sample, amplification will not occur, or will occur at a much lower rate (King et al. 2009; Fitzpatrick et al. 2010).

4.7 Conclusions Blood and tissue are the gold standard for genetic studies because they represent the best source of good quality DNA. Hair samples are a good alternative and have been used where hair traps have been shown to work (McKelvey et al. 2006; Schmidt & Kowalczyk 2006; Ruell & Crooks 2007). However, for many wildlife projects the easiest samples to obtain are faeces because they do not require direct handling or observation of the target species, and they are currently exempt from CITES control (Kohn et al. 1999; Farrell et al. 2000). This poses several challenges for conservation genetics studies including the sampling time and effort required to generate samples, and the variability in DNA quality and quantity. Sampling can be conducted alongside regular camera trap or detection/non-detection surveys and can be limited to a single protected area or extended across the wider species range.

- 118 - Laboratory preparation should include a pilot study to validate all mtDNA, microsatellite and sex identification markers. This was conducted for this study using faecal samples from captive individuals, but lab protocols still had to be modified to accommodate the use of different equipment in-country and the poorer quality of field-collected samples. Whilst PCR inhibition may have affected a proportion of reactions, relying on cross-species reactivity of domestic cat markers may also have played a part (Jane&ka et al. 2008). While the markers used in this study have been used successfully in other endangered felid work, a comparison with newer tiger- specific markers would have helped to determine the most effective marker panel for field samples (Williamson et al. 2002; Bhagavatula & Singh 2006; Zhang et al. 2006; Sharma et al. 2008; Wu et al. 2009; McEwing et al. 2011). Genotyping was not possible for all samples using microsatellite fragment analysis, but it may still have been possible to sequence the PCR product to obtain allele size data. Sequencing also plays a role in new developments in the field of conservation genomics. SNP panels are being developed for many non-model species (Seddon et al. 2006; Sacks & Louie 2008; Ogden 2011; Olsen et al. 2011), and the sensitivity and coverage of next generation sequencing techniques mean that it is becoming easier to amplify very small fragments of DNA from faeces (Valentini et al. 2009; Perry et al. 2010). This will also make it possible to foster better collaboration between projects as a single sample is then capable of generating enough data for many different purposes, e.g. landscape genetics, adaptive variation, prey selection, etc.

- 119 - Chapter 5

Population Genetics of the wild Sumatran tiger (Panthera tigris sumatrae)

- 120 - 5.1 Abstract Tiger numbers are declining and conservation managers are looking for new ways to safeguard the remaining populations. Of particular concern is the fragmentation that is affecting the remaining tiger habitat and the potential isolation of the populations they contain. These populations can no longer be managed individually and an overall picture of tiger distribution and how individuals are moving between regions is crucial to understanding how best to save them. Genetic tools offer an effective method for identifying important source populations, patterns of connectivity between them and isolated sink populations. I collected faecal samples from 9 different field sites covering 6 Tiger Conservation Landscapes on Sumatra. Using data from 24 microsatellite loci, I show that Sumatran tigers retain high levels of genetic variation

(He = 0.61) and that gene flow is continuing over much of the island. Overall, effective population size is small (Ne = 20-131) and tigers in a south Sumatra national park are partially isolated from the rest of the island (FST = 0.15). These results show that there is still a lot of genetic potential in the current Sumatran tiger population. They also show how deforestation and habitat loss can affect the population structure of a wide-ranging carnivore.

5.2 Introduction Tigers are critically endangered across their range and they are particularly threatened by deforestation and habitat fragmentation. Current estimates suggest that they are now restricted to less than 7% of their former distribution, with a 40% loss since 1995 (Sanderson et al. 2006). Sumatra is one of the top three countries (with India and Russia) that contain ~80% of remaining tiger habitat. The Sumatran tiger is of particular conservation concern because it is both morphologically and genetically distinct from mainland tigers (Cracraft et al. 1998; Hendrickson et al. 2000; Kitchener & Yamaguchi 2010). It is also the last remaining subspecies in Indonesia since the Java and Bali tigers became extinct in the 1980s and 1940s respectively (Tilson et al. 1993; Whitten et al. 2000). Despite this, it has been

- 121 - comparatively poorly studied and little is known about its ecology or genetics. The last PHVA published in 1993 put the estimated population at 400 individuals within protected areas and ~100 in unprotected land (Tilson et al. 1993). A recent collaborative island-wide occupancy survey has provided more information on the population status of the Sumatran tiger (Wibisono et al. 2011). Population genetics can complement this data by showing how the metapopulation is structured and how tigers disperse across the landscape. Measures of genetic variation can also give some indication of extinction risk. Small, island populations are considered to be more inbred and have been shown to have a greater risk of extinction (Brook et al. 2002).

In general, tigers are a challenging species to manage. They often occur at low densities and are wide-ranging, which means that large areas need to be conserved in order to protect demographically viable populations. However, available tiger habitat on Sumatra has been steadily decreasing since land conversion began in earnest in the 1970s (Imbernon 1999; FWI/GFW 2002). During this time the Indonesian government encouraged resettlement from overpopulated Java to Sumatra, and the rate of forest clearance has continued to escalate since then. Vast areas of fertile, lowland forest in Lampung province in southern Sumatra and now Riau province in eastern Sumatra have been cleared. The main drivers are land conversion for agriculture and plantations, although logging and human settlements are also prominent causes (Uryu et al. 2008). Due to the alarming rate at which primary forest was being lost, ecologists first proposed Tiger Conservation Units (TCUs) in the late 1990s and then Tiger Conservation Landscapes (TCLs) in 2006 to safeguard remaining tiger habitat (Wikramanayake et al. 1998; Sanderson et al. 2006). These areas were considered to contain sufficient habitat to safeguard healthy populations and to have the capacity to persist into the future. TCLs encompassed both protected areas and intervening forest that could support viable populations for up to 10 years. More recently, the concept of source sites was introduced (Walston et al. 2010). This concept proposed that conservation efforts

- 122 - should initially focus on those core areas containing >25 breeding females and that are embedded within a total landscape capable of supporting >50 breeding females.

In reality, the boundaries of many Sumatran protected areas are poorly enforced or respected, and encroachment is slowly eroding their edges (Gaveau et al. 2009a). Access roads are quickly colonised and provide an effective route for further encroachment and clearing, which in turn fragments existing areas and widens the gaps between them (Linkie et al. 2004; Ovando 2008). Despite these problems, the Indonesian government is still under pressure to build new roads through both Kerinci Seblat NP and Gunung Leuser NP, two of the oldest and largest protected areas on Sumatra (Harimaukita 2011). So what can conservationists do to manage tiger populations within this context? It is vital that we know which are the key populations to invest in. Population genetic tools are useful as a complement to traditional camera trap and occupancy surveys to characterise the existing metapopulation and to estimate the rates of movement (gene flow) between occupied areas. The type of management intervention will then vary depending on the character of the population in question. Large protected areas with high genetic variation and high levels of gene exchange with surrounding areas are assumed to be capable of supporting self-sustaining populations. They also have the potential to provide enough dispersing individuals to repopulate surrounding habitat. They could therefore be managed alone with intervention targeted towards policing their boundaries and reducing incursions into the interior forest (Ranganathan et al. 2008). In contrast, smaller protected areas are expected to have small population sizes and lower levels of genetic variation. In cases where rates of migration and gene exchange are low, they will also be differentiated as genes are lost through genetic drift. Intervention in these cases must therefore be geared towards management of the surrounding habitat matrix as connectivity to neighbouring source populations is what ensures persistence of these sinks.

- 123 - In order to identify potential source populations and the rates of gene flow within Sumatra I measured (i) genetic variation, (ii) population differentiation, and (iii) effective population size for 4 different regions. I defined the four regions to coincide with the current designation of TCLs. The northern group corresponds to samples collected in the Ulu Masen/Gunung Leuser ecosystem, the western group to samples collected from Kerinci Seblat NP and Batang Hari protection forest, the eastern group contains samples collected in Riau (namely Tesso Nilo NP, Kerumutan wildlife reserve and Berbak NP), and the southern group refers to samples taken from Way Kambas NP. Higher genetic variation is expected in larger, source populations, and where gene flow and dispersal rates are high I would expect low levels of differentiation. In contrast, I would expect greater differentiation of groups that have low rates of gene flow and thus higher rates of genetic drift. There are then two possible management strategies for Sumatra: (i) To target resources to a few healthy, well connected regions, or (ii) To reduce the extinction risk of small, isolated subpopulations by improving connectivity.

5.3 Methods 5.3.1 Sample collection and genotyping With the aid of field teams and a detection dog, I collected 148 faecal samples for DNA analysis during surveys conducted from July 2009 - October 2010. The samples were collected from 9 national parks and protected areas on Sumatra, and wild-caught individuals held in captivity. I used a modification of the Qiagen QIAamp stool mini kit protocol for DNA extraction and confirmed tiger samples using a Tig 117F/231R forensic primer (Wetton et al. 2004). I then derived consensus genotypes for all the tiger samples using a multi-tubes approach with 24 microsatellite loci: Fca-8, Fca-32, Fca-44, Fca-69, Fca-77, Fca-90, Fca-94, Fca-96, Fca-105, Fca-123, Fca-129, Fca-139, Fca-161, Fca-176, Fca-201, Fca-211, Fca-220, Fca-229, Fca-290, Fca-293, Fca-304, Fca-310, Fca-391, Fca-441 (Taberlet et al. 1996; Menotti-Raymond et al. 1999; Frantz et al. 2003). Full PCR protocols and conditions are described in Chapter 4.

- 124 - 5.3.2 Estimating genetic diversity

For the total data set I used the program GENEPOP v4.0 to test for Hardy-Weinberg equilibrium, heterozygote excess/deficiency, and linkage equilibrium (Guo &

Thompson 1992; Raymond & Rousset 1995; Rousset 2007). GENEPOP uses exact tests to estimate a p-value associated with the null hypothesis of Hardy-Weinberg and linkage equilibrium. I also used Wright's FIS to test for deviations from equilibrium. FIS measures the deviation of observed heterozygosity from expected values under Hardy-Weinberg equilibrium (Wright 1943; Hedrick 2005b). Under equilibrium FIS = 0; a negative value indicates an excess of heterozygotes and a positive value an excess of homozygotes relative to that expected at Hardy- Weinberg equilibrium.

I estimated observed and unbiased expected heterozygosity with GenAlEx v6.4 (Peakall & Smouse 2006) using genotype frequencies derived from the empirical data. Unbiased expected heterozygosity was used to account for small sample sizes at each locus (Nei 1978). As a comparative measure of genetic variation I also computed allelic richness (ag) to estimate the number of alleles found within the current sample:

I ˆ (5.1) ag = ! Pijg i=1 where Pijg is the probability of finding at least one copy of allele type i in a sample of g alleles from population j (Kalinowski 2004).

To assess regional differentiation I also used the mean number of private alleles per locus. This gives a measure of the number of alleles found in a population that are not found anywhere else:

- 125 - J * $ J ' - ! = ,P & Q ) / (5.2) g 0 , ijg # ij 'g / i=1 & j '=1 ) +, % j '" j ( ./

J is the total number of populations and Qij'g is the probability of finding no copies of allele i in a sample of g alleles from population j (Kalinowski 2004). To account for the different sample sizes between regions I used a rarefaction method implemented in ADZE v1.0 (Szpiech et al. 2008) to calculate allelic richness and private allelic richness as a function of a standardised sample size (g = 6) for a given set of loci (L = 7), and with the threshold for missing data set to 40%. I tested for significant differences in the values of genetic diversity between regions using a locus-by-locus two-way ANOVA in R (R Development Core Team 2011).

5.3.3 Is there evidence for population differentiation?

The most common measure of population differentiation is FST which can be thought of as a measure of allele frequency divergence among subpopulations. Values of FST range from 0 (no differentiation) to 1 (all subpopulations fixed for different alleles). The amount of divergence will depend on the balance between the homogenising effects of gene flow between populations, the divergent effects of genetic drift within isolated populations and the number of generations that populations have been separated for (Meirmans & Hedrick 2011; Whitlock 2011). In this study I used an unbiased weighted measure of FST ($w) that estimates the proportion of allele frequency variation that is between populations (Weir & Cockerham 1984). This estimator does not make assumptions about the number of populations, sample sizes or heterozygote frequencies and so is suited to small data sets. I used the program SPAGeDi v1.3 (Frantz et al. 2009) to estimate a global value for $w, and the program GENEPOP v4.0 to compute pairwise values of $w for all population pairs. I also used Fisher's exact test implemented in CHIFISH (Ryman 2006) to compute an unbiased estimate of the probability of differentiation, i.e. to test the null hypothesis that all populations share the same allele frequency distribution. Analysis of

- 126 - molecular variance (AMOVA) can also be used to find the grouping of populations that maximises differentiation (Excoffier et al. 1992; Michalakis & Excoffier 1996).

2 For a single group of populations, % a is the variance component due to differences

2 between groups, and % b is the variance due to differences in populations within a

2 group. The significance of the variation between groups (% a) and FST can be tested by permuting individual genotypes among groups. The structure of a population can

2 thus be determined by the grouping which has the highest significant value of % a. I tested for population structure using a locus-by-locus AMOVA in ARLEQUIN v3.1 (Excoffier et al. 2005) with 19 microsatellite loci and 16,000 permutations.

5.3.4 What is the effective population size?

The effective population size (Ne) is the size of an idealised population that would produce the same genetic properties as that observed for the real population under consideration, i.e. the number of individuals within a population that contribute gametes to the next generation (Cabellero 1994; Falconer & MacKay 1996; Hedrick 2000; Wang 2005). It is a particularly important measure for isolated or subdivided populations as it determines the magnitude of the effects of genetic drift and the loss of heterozygosity (Sugg et al. 1996; Hedrick 2005b; Hamilton 2009):

t " 1 % (5.3) Ht = $1! ' H 0 # 2Ne & where Ht is the current heterozygosity in generation t, Ne the effective population size and H0 initial heterozygosity in generation 0.

Effective population size can be estimated using single sample or temporal methods (Leberg 2005). Although temporal methods reportedly give estimates with higher precision, single sample estimators are more appropriate for species with long generation intervals as they do not require changes in allele frequencies between generations (Wang 2005). Because of the lack of temporal samples (tigers have a generation time of 7 years) I used two single sample methods to estimate Ne. The

- 127 - first is a sibship assignment method implemented in COLONY v2.0.1.1 that uses the pairwise probability of individuals being half sibs or full sibs (Jones & Wang 2009). For situations in which there is no difference between male and female offspring survival:

1 1+ 3! ! # 1 1 & (5.4) = (Q1 + Q2 + 2Q3 ) " % + ( Ne 4 2 $ Nm N f '

where ( is the deviation from Hardy-Weinberg proportions () FIS), Q1 is the probability of a pair of individuals being paternal half sibs, Q2 of them being maternal half sibs and Q3 of them being full sibs, Nm the number of males and Nf the number of females. I used the pairwise-likelihood approach accounting for missing data, with both male and female polygamy, allowing for inbreeding, and with sibship size prior

= 0 to give a lower bound estimate of Ne and sibship size prior = 1 to give an upper bound. Although this method relies on sibship relationships in a random sample, it is able to give an estimate of Ne if the total population is considered to have low genetic diversity or if there are duplicate genotypes in the data set. The second method implemented in LDNe v1.31 uses linkage disequilibrium to estimate Ne (Waples 2006; Waples & Do 2008). It also incorporates a bias correction to account for disequilibrium due to sampling error, particularly for small sample sizes (n < 30). I ran the program with the random mating model and allele frequencies set to >0.01. It was not possible to estimate Ne using the more recent approximate Bayesian computation method (ONESAMP; (Tallmon et al. 2008)) as it was not able to cope with the level of missing data within this data set.

I tested for the presence of a population bottleneck using the package MRATIO

(Garza & Williamson 2001). MRATIO uses the relationship between the total number of alleles (k) and the overall range in allele size (r) to determine if a reduction in population size has occurred in the recent past. A population that has reduced in size is assumed to have been affected by genetic drift and hence to have lost alleles compared to a population at equilibrium. The reduction in k is assumed to occur at a

- 128 - much faster rate than the reduction in r, thus the M ratio, given by k/r, is expected to be smaller in populations that have suffered a bottleneck. I used the program

CRITICAL_M to calculate Mc, the critical value below which population size reduction is assumed to have occurred with 95% certainty (Garza & Williamson 2001). I calculated both the M ratio and critical value Mc using 16 loci under Hardy-Weinberg equilibrium with the following parameter values: $ = 10, mutation rate = 5 x 10-4, the proportion of one-step mutations (ps) = 0.9, and the average size of non one-step mutations (!g) = 3.5.

I also used BOTTLENECK v1.2.02 (Cornuet & Luikart 1996) and MSVAR (Beaumont

1999) to look for evidence of a recent reduction (<4Ne generations in the past) in effective population size. Populations that have undergone a dramatic reduction in population size are expected to have lost a large number of alleles, but there will be little effect on heterozygosity. A useful test for a bottleneck is thus to compare the observed heterozygosity with that expected given the observed number of alleles. If a heterozygosity excess exists (observed heterozygosity > expected heterozygosity), then it is likely that a bottleneck has occurred in the recent past. Non-bottlenecked populations are assumed to be at mutation-drift equilibrium and so each locus will have an equal probability of heterozygosity excess or deficiency. BOTTLENECK implements 2 main methods: a sign test that calculates observed heterozygosity - expected heterozygosity, and a standardised differences test that compares the difference between observed and expected heterozygosity with a random distribution (Cornuet & Luikart 1996; Luikart et al. 1998a). There is also a graphical measure called the mode shift which looks for a change in allele frequency distribution. Low frequency alleles are assumed to be lost in a bottlenecked population such that intermediate and high frequency alleles become relatively more abundant (Luikart et al. 1998b). Two main mutation models are offered - IAM and SMM. Microsatellites are generally considered to follow the SMM model of mutation and this also provides a more conservative assessment. The IAM model tends to predict a lower equilibrium value for heterozygosity and is more likely to indicate that a significant

- 129 - heterozygosity excess exists (Luikart et al. 1998a). I compared results for both the

IAM and SMM models with 10,000 iterations. In contrast, MSVAR uses coalescent theory and MCMC analysis to determine if a population has undergone a size contraction in the recent past. Looking back in time the current population (N0) is assumed to have come from an ancestral population of size Nt, up to t generations ago. Both linear and exponential population growth models can be used and population sizes are scaled to mutation rate µ. Populations that have undergone a population bottleneck are expected to have a ratio r < 1, where r = N0/Nt. Based on published demographic information for tigers, I estimated N0 = 300 - 500, Nt = 7000-20,000, t = 7 - 75,000 years (equivalent to 1 - 1100 tiger generations) and µ = 10-2 - 10-5 (Foose 1987; Tilson & Seal 1987; Whitten et al. 2000; Luo et al. 2004;

Chundawat et al. 2011). This gave the following settings for MSVAR: $ (-5, 2), r (-4, 0), t (0, 5), maximum number of thinned update steps = 200,000, and degree of thinning = 25,000. All other parameters were set to default values.

5.3.5 Measuring rates of gene ﬂow

I used a Bayesian method implemented in BAYESASS v1.2 to estimate recent rates of gene flow between four defined regions: North, East, West and South (Fig. 5.1). The method used relaxes the key assumption that populations are in Hardy-Weinberg equilibrium by incorporating an inbreeding coefficient (Wilson & Rannala 2003). The joint posterior probability distribution of the model parameters m (migration rate), M (source of migrant ancestry for each individual), t (generation at which a migrant ancestor arrived), F (the inbreeding coefficient for each population), and p (the population frequencies of marker alleles) are estimated using MCMC methods given X (the multi-locus genotypes) and S (the population source for each sampled individual). I used the default settings of seed = 10, 3 x 106 iterations, 1 x 106 burn- in, and samples collected every 2000 iterations. To achieve the recommended acceptance rate of allele frequency, inbreeding and migration rate changes, I set the delta values to deltap = 0.35, deltam = 0.15 and deltaF = 0.35.

- 130 - Figure 5.1 A map showing the regional subdivision of sample sites: The North - Ulu Masen ecosystem, Kutacane and Medan. The West - Kerinci Seblat NP and Batang Hari protection forest. The East - Tesso Nilo NP, Kerumutan wildlife reserve, Bukit Tigapuluh NP and Berbak NP. The South - Way Kambas NP.

5.4 Results A total of 37 samples were positive for tiger DNA out of the 148 scats tested. Of these, 25 had sufficient rates of PCR success for subsequent population genetics analysis (Table 5.1). This represents a reasonable proportion (5%) of the total Sumatran tiger population.

- 131 - Table 5.1 The number of positive samples collected in each regional group across Sumatra.

Sampled area Regional group No. of tiger samples Ulu Masen North 4 Kutacane, Medan Kerinci Seblat NP West 5 Batang Hari Riau East 11 Berbak NP Jambi Way Kambas NP South 5 Total 25

5.4.1 Higher genetic variation than expected The amount of genetic variation over the whole data set was much higher than expected. To validate this result and to obtain a conservative measure of diversity, rare alleles with a frequency of <0.05 were removed from the data set (Table 5.2). Genotyping errors such as false alleles are likely to be observed as rare alleles and thus, they may be less reliable than more common alleles. Despite this adjustment, the values of observed and expected heterozygosity were greater than previously published values for Sumatran tigers (Luo et al. 2004). This may in part be due to sampling as I was able to sample over a wider area and include more individuals in the present study. Overall, the sampled population was not in Hardy-Weinberg equilibrium (exact test p = 0.0012), and across the 24 microsatellite loci used 7 pairs appeared to be in linkage disequilibrium. These loci are not physically linked as they have been mapped to different chromosomes in the domestic cat (Menotti-Raymond et al. 1999). There was a significant heterozygote deficiency (U test p = 0.0016) and a positive FIS value of 0.037. These results taken in combination suggest that there may be non-random mating, inbreeding or population subdivision within the sampled population. Loci may also be in linkage disequilibrium if they are under selection or if there is mixing of alleles from diverged groups (Hedrick 2005b).

- 132 - Table 5.2 A comparison of genetic diversity measures in subspecies of Panthera tigris.

Subspecies No. of No. of loci Observed Expected Reference individuals heterozygosity heterozygosity P. t. sumatrae 25 24 0.61 ± 0.08 0.61 ± 0.07 This study P. t. sumatrae 16 30 0.47 ± 0.02 0.49 ± 0.04 Luo et. al. (2004) P. t. altaica 34 30 0.47 ± 0.02 0.46 ± 0.04 Luo et. al. (2004) P. t. altaica 95 8 0.26 (0.11) - Henry et. al. (2009) P. t. corbetti 33 30 0.64 ± 0.02 0.67 ± 0.03 Luo et. al. (2004) P. t. jacksoni 22 30 0.56 ± 0.02 0.57 ± 0.03 Luo et. al. (2004) P. t. tigris 6 30 0.52 ± 0.04 0.57 ± 0.04 Luo et. al. (2004) P. t. tigris 73 5 0.70 (0.16) - Mondol et. al. (2009)

5.4.2 Heterozygosity and allelic diversity There were similar levels of genetic variation in all regions, with some of the lowest values in the southern group (Table 5.3). However, a two-way ANOVA found that there was no significant difference in unbiased heterozygosity between regions (p >

0.05). The northern grouping appeared to have a significant heterozygote excess (FIS = -0.216, p = 0.0103), and the eastern grouping a significant heterozygote deficiency (FIS = 0.046, p = 0.025). This indicates non-random mating in both these regions; e.g. due to different allele frequencies in the male and female parent populations or gene flow in the north, and inbreeding or population subdivision in the east (Chesser 1991; Wang 2005). Measures of allelic richness also appear to confirm that all regions have similar levels of allelic diversity, although the western group has a much lower number of private alleles (p < 0.05).

- 133 - Table 5.3 Regional values of observed heterozygosity (Ho), expected heterozygosity (He), unbiased expected heterozygosity (UHe), allelic richness (ag) and private alleles (#g).

** Region Mean HO HE UHE Mean ag Mean !g sample size/locus North 2.6 0.65 ± 0.15 0.43 ± 0.10 0.54 ± 0.12 2.48 ± 0.43 0.47 ± 0.63 East 7.2 0.59 ± 0.10 0.57 ± 0.07 0.60 ± 0.07 2.86 ± 0.45 0.53 ± 0.45 West 3.3 0.58 ± 0.14 0.47 ± 0.08 0.55 ± 0.10 2.34 ± 0.39 0.02 ± 0.04 South 2.0 0.45 ± 0.17 0.32 ± 0.11 0.43 ± 0.15 2.58 ± 0.98 0.39 ± 0.67

* Ho and UHe were measured using 24 microsatellite loci.

** ag and !g were measured using rarefaction down to 6 loci.

5.4.3 Genetic differentiation and restricted gene ﬂow in the south

Overall, the level of genetic differentiation between regions was low (FST = 0.08). However, the southern group appeared to be highly differentiated from the rest of the island with an FST value of up to 0.15 (Table 5.4). These results were supported by the AMOVA analysis which suggested that the greatest variance between groups could be achieved with two clusters: 1. North-West-East, and 2. South (Appendix 4). An appreciable amount (~14.8%) of the variance between groups could be attributed to differences between the south and the remainder of the island with an FST value of 0.15 (95% CI 0.05-0.18; p < 0.0001).

Table 5.4 Regional differentiation computed using pairwise FST (significant values are indicated in bold). North East West East 0.07 West 0.06 0.03 South 0.15 0.15 0.13

- 134 - Analysis with BAYESASS suggested that the majority of gene flow occurred from the west to the north, and from the west to the east (m = 0.12 and m = 0.20 respectively). In contrast there was little migration into or out of the southern region

(m * 0.05). This result in combination with the values of pairwise FST supports the suggestion that the south is partially isolated from the rest of Sumatra and that few individuals are able to disperse from the south to other regions and vice versa.

5.4.4 Small effective population size

Estimates of Ne using LDNe resulted in negative values with 95% confidence intervals including infinity. A negative estimate of Ne suggests that sampling error accounts for more of the disequilibrium noted in the sample than genetic drift (England et al. 2006; Waples & Do 2008). The sibship assignment method gave an estimate for Ne of 20 (95% CI 10 - 42). This is similar to an estimate of effective population size in Bengal tigers using variance in reproductive rate (Smith & McDougal 1991). Given that this result may have been affected by the presence of population structure in the Sumatran population (Chikhi et al. 2010), I repeated the analysis with Way Kambas samples excluded. This gave a similar result of 18 (95% CI 9 - 40) indicating that Sumatran tigers have a small effective population size. The result of analysis with MRATIO suggests that these low values are not due to a recent reduction in population size. Across the whole data set the M ratio was higher than the critical value Mc calculated given the data set parameters, and thus no population bottleneck was inferred. In contrast, BOTTLENECK found a significant heterozygosity excess under the IAM model for the entire data set and with Way Kambas samples removed. This result did not change with the removal of loci that may have been under selection or not in Hardy-Weinberg equilibrium. However, the SMM model found no evidence for a bottleneck in either the sign or standardised differences tests. Under both mutation models, the allele frequency distribution was normal.

Finally, analysis with MSVAR gave negative values for r suggesting a trend of population decline. However, point estimates for the ancestral population size could not be obtained due to lack of convergence, even with very long chains of 1.5 x 1010

- 135 - 5.5 Discussion 5.5.1 High genetic variation Heterozygosity provides one of the basic measures of genetic variation in population genetics, and low levels of heterozygosity correlate with a high risk of extinction for many species (Frankham 1998; Brook et al. 2002; Evans & Sheldon 2008). The Sumatran tiger appears to show high heterozygosity for an endangered island subspecies, and there does not appear to be any significant difference in genetic diversity between regions. Overall, the levels of expected and unbiased heterozygosity in this study (0.61 and 0.63 respectively) are comparable to mainland tiger subspecies (Luo et al. 2004; Mondol et al. 2009b), and other wild felids such as the jaguar and leopard (Eizirik et al. 2001; Uphyrkina et al. 2002). On average endangered species are thought to have ~60-65% of the microsatellite heterozygosity of related non-endangered species, and that island species in particular will have lower levels of variation compared to similar or related mainland species (Avise 1994; Frankham 1998; Frankham et al. 2002). This does not seem to be the case with the Sumatran tiger. This may mean that there has not been any great loss of genetic variation on Sumatra despite its apparent isolation from the mainland. This suggests that either Sumatra is not totally isolated and that there is still some gene flow with the mainland, or more likely that there has been insufficient time for genetic drift and loss of alleles. Given that heterozygosity is expected to be lost at a rate of 1/2Ne per generation (Hedrick 2005b; Hamilton 2009), the values of

Ne obtained in this study suggest that Sumatran tigers are only expected to lose up to 1-5% of their genetic variation every 7 years due to genetic drift. In the absence of other threats to the population, Sumatran tigers may therefore be able to avoid severe inbreeding effects due to their long generation time. The wild population may also have retained high diversity as it was historically part of a much larger population encompassing the Sunda plate, which consisted of peninsular Malaysia, Sumatra, Borneo and Java (Kinnaird et al. 2003; Kitchener & Yamaguchi 2010). Morphological studies also suggest that the Sumatran subspecies may represent a

- 136 - hybrid of mainland Southeast Asia tigers and Sunda island tigers that became isolated when sea levels rose following the late Pleistocene (Kitchener 1999; Mazak 2008; Kitchener & Yamaguchi 2010). This may have given them a much higher level of diversity due to the mixing of different populations and the addition of more (rare) alleles to the Sumatran population.

5.5.2 Small effective population size Extinction risk is thought to scale with population size. Small populations are more vulnerable to demographic, environmental and/or genetic stochasticity (genetic drift) and so have a higher risk of extinction (Falconer & MacKay 1996; Gaggiotti & Hanski

2004). The estimated effective population size in this study is low (Ne = 20-131) and it gives an Ne:Nc ratio of 0.04-0.26 where Nc represents the total population size of 500 tigers (Tilson et al. 1993; Faust & Tilson 1994). However, this is in the expected range given that the mean ratio across wildlife populations is 0.10-0.11 (Frankham

1995). Values of Ne = ~26 and 35 have been obtained in studies of Panthera tigris (Smith & McDougal 1991) and Panthera tigris altaica (Henry et al. 2009) respectively, with ratios of 0.41 and 0.05 - 0.07 (Smith & McDougal 1991; Henry et al. 2009). Among other cats species ratios of 0.25-0.64 have been recorded for the leopard, cheetah, ocelot, Florida panther and puma (Nowell & Jackson 1996; Spong et al. 2000; Kelly 2001).

The effective size of the population is expected to be much smaller than the total population size due to various factors such as unequal sex ratio, variation in family size, overlapping generations, and population bottlenecks (Falconer & MacKay 1996; Wang & Caballero 1999; Hedrick 2000; Frankham et al. 2002). The sex ratio in tigers is often in the range of 1:1 to 1:2 so it is unlikely that this is responsible for the reduction in Ne relative to Nc (Franklin 2002). The results from MRATIO and

BOTTLENECK under the IAM model suggest that there is no evidence that the Sumatran tiger population has suffered a severe bottleneck in the past 560-3500 years (equivalent to 4Ne generations). However, MSVar, which has been shown to

- 137 - outperform MRATIO and BOTTLENECK in simulation studies (Girod et al. 2011) suggests that the population has indeed contracted from its ancestral size. This seems more likely given the high rate of agricultural expansion and habitat loss that have occurred in Sumatra's past (Imbernon 1999). The government's transmigration programme, as well as ongoing deforestation and forest fragmentation, may have lead to a decrease in overall population size as a result of reduced habitat and prey availability, and higher rates of poaching (Dinerstein et al. 2006). Another contributing factor to the small value of Ne may be the tiger's mating system. Studies in the domestic cat and wild felids suggest that a polygynous mating system and variance in family size can greatly reduce Ne relative to total population size (Smith & McDougal 1991; Nunney 1993; Cabellero 1994; Kelly 2001). Polygynous groups in which dominant males mate with the most females have up to 20% reduction in Ne/

Nc compared to promiscuous groups in which all males mate with equal probability (Kaeuffer et al. 2004). Less dominant males are also expected to have reduced mating success because they have lower survival probabilities due to territorial disputes, etc.

5.5.3 Population differentiation and gene ﬂow In population genetics studies, population isolation may be measured via estimates of gene flow and differentiation. In the absence of gene flow, populations lose alleles under the influence of genetic drift and become increasingly differentiated. It is surprising that there was not more population differentiation on Sumatra despite the presence of geographic features such as Lake Toba and the Bukit Barisan mountain range. These landscape barriers appear to split the distribution of other large mammals such as the tapir, orangutan and Sumatran rhino (Morales et al. 1997; Whitten et al. 2000; Wikramanayake et al. 2002) but have not caused any obvious genetic discontinuity in the Sumatran tiger. The high concentration of roads, settlements and plantations across parts of Sumatra are thought to act as barriers to tiger dispersal and so could also cause genetic differentiation between regions (Smith 1993; Smith et al. 1998; Kerley et al. 2002; Linkie et al. 2006). However, the

- 138 - low differentiation found in this study perhaps reflects the tiger's dispersal ability (Karanth & Stith 1999; Kerley et al. 2002; O'Brien et al. 2003). In tropical habitats males disperse up to 65km and females up to 33km, with home ranges of up to 243km2 and 78km2 respectively (Goodrich et al. 2010). They can also make use of a wide variety of habitat types including lowland, montane and peat swamp forest (Seidensticker et al. 1999). This means that they may be capable of maintaining a fairly continuous distribution across a varied landscape.

The results of the migration analysis and the number of private alleles suggest that there is a net movement of individuals out of the west to the north and east. The western group broadly corresponds to the Batang Hari/Kerinci Seblat NP complex, and this region may be acting as a source of tigers for other regions. This is an important result for future conservation as NGOs are beginning to discuss the feasibility of creating corridors between Riau in the east and west Sumatra. There has also been some discussion about connecting all of the protected areas along the Bukit Barisan mountain range which runs along the western spine of Sumatra. If the forest within Kerinci Seblat NP is indeed acting as a source for tigers, this provides compelling support for securing its boundaries and for protecting the linking habitat between it and neighbouring protected areas.

In contrast, there appears to be very little gene flow into or out of Way Kambas NP in the southern tip of Sumatra. This national park showed the highest pairwise FST and the lowest migration rates. Marked deforestation and land conversion surrounds the park and this may have caused enough isolation for population differentiation to occur (Nyhus & Tilson 2004a). These high FST values represent a separation from the main TCLs in Kerinci and Riau, although there could be some exchange of individuals with forest patches closer to Way Kambas. This could easily be tested by obtaining samples from tiger populations in Sembilang NP and Bukit Balai Renjang NP. Sadly, the prospects for increasing connectivity in this region are bleak. Way Kambas NP does not have a significant buffer zone or forest patches suitable for

- 139 - wildlife adjacent to the park (Nyhus & Tilson 2004a), and although tigers have broad habitat requirements, they will tend to avoid areas of high human population density (Smith 1993; O'Brien et al. 2003). This region has suffered some of the fastest human population growth rates and has some of the highest human population densities on Sumatra. It was one of the first areas targeted by the Indonesian government’s transmigration policy and it has been colonised by many agricultural plantations (Imbernon 1999; Nyhus & Tilson 2004a). The protected areas in Riau may eventually suffer the same fate as this province is also subject to high rates of encroachment and land conversion (Uryu et al. 2008; Broich et al. 2011).

5.5.4 Conservation management The optimal long term management strategy for the tiger population on Sumatra may be to target those areas that are most under threat from habitat loss and thus have the highest risk of extinction (Chauvenet et al. 2010). Taking habitat quality, genetic variation and gene flow rates into consideration suggests that conservation managers could direct proportionally more resources to Way Kambas NP (South), then Riau (East), and lastly the Ulu Masen/Gunung Leuser ecosystem (North) and Kerinci Seblat NP/Batang Hari complex (West) to achieve this objective. For Way Kambas and Riau in particular, conservation intervention could involve law enforcement and the reduction of threats at their borders, e.g. human-tiger conflict and land conversion, as this has a significant effect on population survival (Woodroffe & Ginsberg 1998; Nyhus & Tilson 2004b; Ranganathan et al. 2008; Uryu et al. 2008). Tigers could also be translocated from the nearest national parks to both add to the current population size and to add genetic variation in order to prevent inbreeding. Such methods have already been used successfully for tiger populations in India and for problem tigers in Sumatra (Priatna et al. 2012).

An alternative, and perhaps more controversial, management scenario would be to concentrate primarily on those regions that have the highest chance of persistence (McDonald-Madden et al. 2008). Deforestation, land conversion, and population

- 140 - growth is occurring at such a high rate in lowland areas that in reality it will be difficult to secure all the tiger populations here. The Ulu Masen/Gunung Leuser ecosystem and the Kerinci Seblat NP/Batang Hari complex contain mostly submontane/ montane forest so they are less productive and surrounded by lower human population densities than Riau and south Sumatra. Also, under the source site criteria Kerinci Seblat in particular represents a source population and so has the potential to provide individuals to surrounding areas. For larger protected areas such as these it is preferable to prioritise interventions within their borders. It is particularly important to resist the pressure to build roads across Kerinci so that the population dynamics in this region are not disrupted. Splitting the park into subgroups may result in reduced rates of gene flow and differentiation of the smaller core areas contained with the park (Linkie et al. 2006).

5.6 Conclusions Overall, the results from this study suggest that the current Sumatran tiger population contains a high level of diversity. This means that there is a low risk of inbreeding depression and thus a low risk of extinction due to genetic effects now and in the immediate future. There is no evidence of a recent population bottleneck, and the effective population size is low but is expected given the tiger's mating system. The combination of landscape change, low rates of gene flow and low population size (N = 20-30 tigers) mean that Way Kambas NP may be acting as a sink with a poor prognosis for future persistence. The park is still an important component of the Sumatra population, and so efforts should be targeted to reducing human-tiger conflict at the park borders. In contrast, Kerinci Seblat NP should be protected as an important source population. It has already been recognised as a globally important region for tiger conservation, and the results of this study show that it is an important part of the local Sumatran population as it has the highest rates of gene flow and one of the largest tiger populations (N = 145 tigers) on Sumatra.

- 141 - Chapter 6

Population Structure of the Sumatran tiger

- 142 - 6.1 Abstract Landscape-scale management is an emerging field for the conservation of wide- ranging, low density carnivores such as the Sumatran tiger. Effective management depends on knowledge of habitat preference, current range/distribution and the level of connectivity between subpopulations. It is possible to determine habitat preference and distribution from camera trap surveys and satellite imagery, but very little empirical data is available for tiger movement in tropical habitats. Tiger Conservation Landscapes were proposed to encompass blocks of connected tiger habitat, however, there have not been any studies of matrix connectivity between Sumatran TCLs. Here I used genetic techniques to determine the pattern of gene flow on Sumatra and thus provide an indirect measure of connectivity. I tested for isolation-by-distance and used Bayesian clustering programs (STRUCTURE, TESS, BAPS and GENELAND) to look for genetic clusters. I also used CIRCUITSCAPE to map potential dispersal routes across Sumatra. My results suggest that there is very little genetic structure in the current population. There is significant isolation-by-distance but no obvious evidence for barriers to dispersal or separate genetic clusters. This means that gene flow on Sumatra is primarily determined by dispersal ability and geographic distance, and the high levels of human activity and land cover conversion in Lampung and Riau province have not yet introduced genetic structure. Tigers are able to make use of a wide range of habitats and there may be sufficient degraded forest or scrub to provide connectivity between TCLs. However, with continued landscape change it is likely that the tiger's range will continue to contract and connectivity will be limited to areas of low human population density.

6.2 Introduction Conservationists are increasingly looking to landscape-scale management to secure populations of wide-ranging species such as the tiger (Sanderson et al. 2002b; Redford et al. 2011; Wikramanayake et al. 2011). Like many large carnivores, tigers exist at very low densities, have large home ranges and thus require vast areas for

- 143 - hunting and breeding. To reflect this need for space researchers first proposed Tiger Conservation Units (TCUs) and then Tiger Conservation Landscapes (TCLs) to encompass the remaining fragments of tiger habitat across Asia (Wikramanayake et al. 1998; Sanderson et al. 2006). Both TCUs and TCLs were delineated to represent blocks of habitat that could support viable tiger populations for up to 10 years with adequate levels of protection and low levels of threats. Unfortunately, the TCL designations do not translate directly to protection on the ground. Only ~12.5% of the total TCL area is under protection (IUCN categories I-IV), and ~40% of available tiger habitat was lost in the period between 1995 and 2005 (Sanderson et al. 2006). Sumatra holds much of southeast Asia's remaining tropical forest but it is also subject to some of the highest rates of deforestation in this region (Miettinen et al. 2011a). In the early 1980s land clearance was largely targeted to the lowland forests of southern Sumatra, and since 2000 the focus has changed to peat swamp forests in the east (Koh et al. 2011; Miettinen et al. 2011b). These peat habitats are the last remnants of forest at low altitudes and represent an important source of carbon stocks under the new REDD schemes (Uryu et al. 2008). If this high rate of land conversion continues Sumatran tigers may be restricted to small patches of degraded forest or montane habitat in the north and west of the island.

The interdigitation of forest and other land uses poses another problem for the tiger. Most authors agree that connectivity is crucial to species persistence within a fragmented landscape (With 2004), and the same is expected to be true for the Sumatran tiger. Individuals are assumed to occupy patches of suitable habitat and to use landscape corridors to move between them for hunting and breeding. However, it is difficult to define the elements of an effective corridor as connectivity is determined by many landscape features such as habitat type, degree of fragmentation, species' habitat preference, dispersal ability and the level of human activity. For many species there is also a blurred distinction between 'habitat' and 'non-habitat' (Fahrig 2003). Instead there may be a mosaic of different habitat types or levels of disturbance which will affect dispersal rates in different ways (Cushman

- 144 - et al. 2010b). Habitat generalists that can make use with a wider variety of habitat types are expected to cope with more habitat disturbance and loss of preferred habitat before local populations become isolated (With & Crist 1995; Crooks 2002; D'Eon et al. 2002; With 2004; de Angelo et al. 2011). The boundaries of the current TCLs were delineated based on the assumption that tigers are less likely to cross non-habitat areas of >4km (Sanderson et al. 2006). Thus, there should be much higher rates of dispersal within TCL boundaries than between them. However, many studies have also shown that including matrix quality explains far more variation in dispersal rates than patch size and isolation alone (Ricketts 2001; Tischendorf et al. 2003; Bender & Fahrig 2005; Prugh et al. 2008). Tigers have been found in a gradient of different land cover types from pristine primary forest, through degraded secondary forest (selectively logged forest) to scrub, although they occur in very low numbers near plantations such as oil palm (Maddox et al. 2007; Linkie et al. 2008b; Sunarto et al. 2012). Tiger abundance has also been linked to human activity such that tigers are more likely to be found in areas of low human population density or away from major roads (Kerley et al. 2002; Kinnaird et al. 2003; Uphyrkina & O’Brien 2003; Carroll & Miquelle 2006; Linkie et al. 2006). Landscape-scale management may therefore have to consider not only the protection of suitable habitat patches but the factors affecting dispersal in the matrix between them (Sanderson et al. 2002b; Wikramanayake et al. 2004).

In the absence of empirical movement data, estimates of tiger distribution through surveys and satellite imagery can be used to determine habitat availability and connectivity. An alternative method is to use genetic techniques to determine spatial genetic structure and gene flow (Cushman et al. 2006; Balkenhol et al. 2009). The TCL boundaries already account for available tiger habitat, areas of low human influence and connectivity. They should therefore provide a good reflection of dispersal probabilities across Sumatra and the network of TCLs should mirror the tiger's genetic structure. Once the genetic structure is known this can be related to landscape variables to try to determine the relationship between land cover type or

- 145 - human influence and gene flow. Each land cover type is expected to provide a different level of resistance to movement and the cumulative rates of gene flow between occupied areas should reflect this. Here it is important to distinguish between dispersal and gene flow as although these terms are often used interchangeably they refer to different processes in a genetic framework. I define dispersal as the movement of individuals from their natal range, and gene flow as the movement of genes between areas, i.e. the combination of dispersal and successful mating. In the absence of a clear relationship between landscape structure and gene flow, genetic differences will simply be a consequence of the physical distance between individuals, i.e. how likely they are to meet and mate (Hardy & Vekemans 1999). However, if there is some influence of habitat type, genetic differences will be due to both distance and dispersal success. Groups either side of a dispersal barrier are therefore expected to become differentiated over time and there will be a strong signal of population structure. The barriers themselves may not be obvious, but can be determined by visualising the genetic groups on a GIS map to try to identify the relevant landscape features, e.g. water bodies, human developments or roads (Goossens et al. 2005; Riley et al. 2006; Millions & Swanson 2007; Coulon et al. 2008; Latch et al. 2008).

In order to identify the spatial genetic pattern for tigers on Sumatra I measured the correlation between individual genetic differentiation and geographic distance

(isolation-by-distance), and performed Bayesian clustering using STRUCTURE, TESS,

GENELAND and BAPS. I also used the most recent tiger distribution data for Sumatra and landscape resistance analysis to determine the potential for connectivity. Overall, a strong relationship is expected between genetic distance and geographic distance because of the dispersal capability of tigers. However, the presence of potential barriers such as large areas of deforestation, plantations or high human density are expected to introduce spatial structure. Satellite maps suggest that the biggest potential for gene flow will be along the western edge of Sumatra (Bukit

- 146 - Barisan mountain range) and the stepping stone habitats between west and east Sumatra.

6.3 Methods 6.3.1 Sample collection and genotyping The methods for sample collection, DNA extraction and construction of multilocus genotypes were as described in Chapter 5. All analysis was performed on the 25 positive samples considered to have sufficient rates of PCR success.

6.3.2 Neighbourhood size Under Wright’s model of isolation-by-distance (IBD) there are no sharp boundaries separating groups of individuals into discrete groups. Instead, there is a gradual change in gene frequency between regions and individuals are subdivided into neighbourhoods. The size of a neighbourhood is determined by dispersal distance and an individual is more likely to mate with others within a neighbourhood than with those outside it (Wright 1943; Falconer & MacKay 1996). In this way neighbourhood size (Nb) determines the genetic properties of the population in the same way as effective population size (Wright 1946). In the more general lattice model and in a 2- dimensional habitat, there is a linear relationship between genetic differentiation a(r) and the logarithm of geographic distance ln(r):

ln(r) a(r) ! + constant (6.1) 4" D# 2 The inverse of the slope of the linear regression of a(r) versus ln(r) can then be used to estimate neighbourhood size (4#D%2), where D is population density and "2 is the variance of parent-offspring axial dispersal distance (Wright 1946; Crawford 1984; Rousset 2000; Sumner et al. 2001; Watts et al. 2007). The spatial dimensions of the genetic neighbourhood are thus a function of "2 which is given by:

- 147 - x2 + y2 2 (" " ) ! = (6.2) axial 2n where x is the dispersal distance measured on the x-axis and y the dispersal distance measured on the y-axis for n observations (Crawford 1984).

I used the Mantel test in GenAlEx v6.4 (Peakall & Smouse 2006) to test for a correlation between geographic distance and genetic differentiation. GenAlEx uses a pairwise, individual x individual (N x N) genotypic distance matrix to compute genetic distance (Smouse & Peakall 1999) and a modification of the haversine formula developed by Sinnott (1984) to compute geographic distance between individuals (Peakall & Smouse 2005). I also used a spatial autocorrelation test in GenAlEx to look for a local, fine-scale correlation between genetic distance and geographic distance with distance classes 50-1500km (Peakall et al. 2003; Watts et al. 2004; Vignieri 2005). Under locally-restricted gene flow, populations exhibit positive spatial- genetic autocorrelation at short distance classes, subsequently declining through zero and becoming negative. I used the program SPAGeDi v1.3 to test for a significant linear relationship between Rousset’s estimator of genetic differentiation (â) and the logarithm of geographic distance between individuals (Frantz et al. 2009). I then used the inverse of the jackknife mean slope to estimate neighbourhood size.

6.3.3 Barriers to gene ﬂow Where there are barriers to dispersal, a population may be structured into subpopulations and gene frequencies may change abruptly across a landscape. Bayesian clustering algorithms use genetic distance and Markov chain Monte Carlo (MCMC) statistics to derive a posterior probability distribution of k (the number of subpopulations or clusters). They search for the most likely value of k which minimises deviations from Hardy-Weinberg and linkage disequilibrium for each subpopulation. For a given value of k, individuals are assigned to each cluster where the posterior probability of the cluster being the source population is highest.

Bayesian algorithms such as STRUCTURE and TESS allow individuals to have mixed

- 148 - ancestry, i.e. parents from two or more populations, and will proportionally assign individuals to the relevant clusters (Pritchard et al. 2000; François et al. 2006; Latch et al. 2006). Spatial methods, e.g. GENELAND and BAPS, are especially useful as they can be used to identify potential geographic boundaries. The use of spatial data is particularly recommended for data sets where the signal of population structure may be relatively weak (Guillot et al. 2005; Hubisz et al. 2009). I used four spatial Bayesian clustering methods to infer the number of subpopulations in the sampled population (Appendix 5): STRUCTURE v2.3.3 (Pritchard et al. 2000), TESS v2.3.1

(François et al. 2006; Durand et al. 2009), BAPS5 v5.4 (Corander et al. 2008) and

GENELAND (Guillot et al. 2005). I also used the program CLUMPP v1.1.2 (Jakobsson & Rosenberg 2007) to determine individual membership assignments for each cluster. I assigned individuals with a membership coefficient of q + 0.7 to a single cluster. Those with membership coefficients of 0.25 ' q ' 0.7 I assigned to more than one cluster, i.e. I considered them to have shared membership (Coulon et al. 2008).

6.3.4 Connectivity The most probable routes for dispersal are more commonly identified using least cost path methods (e.g. Wikramanayake et al. 2004; Sawyer et al. 2011). These assign costs to a landscape surface that reflect the relative difficulty of movement through different habitat types. Potential corridors can then be identified by plotting optimal routes based on the path of least resistance (e.g. Spear et al. 2005; Vignieri 2005; Michels et al. 2001; Adriaensen et al. 2003). Circuit theory models extend this analysis by using algorithms that consider all possible pathways connecting population pairs. Multiple or wider habitat swaths connecting populations are considered to allow greater gene flow than narrow or single swaths (McRae 2006; McRae et al. 2008). Traditional least cost paths or low resistance paths under circuit theory usually explain more variation in gene flow than the straight-line Euclidean distance used in traditional IBD studies (Arnaud 2003; Broquet et al. 2006; Cushman et al. 2006; McRae & Beier 2007). The expectation is that populations connected by good quality habitat are likely to be subject to more dispersal and gene flow, and

- 149 - therefore to be less differentiated than populations with an intervening matrix of poor quality habitat.

I used the PATHMATRIX v1.1 extension in ArcView 3.1 (Ray 2005) to measure least cost distances between samples using the human footprint index from the Last of the Wild v2 (Society & Center for International Earth Science Information Network 2005) and the FAO GeoNetwork land cover map of Indonesia (Latham 2005) as friction maps. The human footprint map represents the sum of the ecological footprints of the human population in year 2000. It uses human influence scores from 0 to 100 (human population density, land transformation, human access and power infrastructure) normalised to each regional biome (Sanderson et al. 2002a). The FAO land cover map uses a generalised land cover classification system from 2005 and twelve LCCS classes are present in Sumatra. I ranked friction values based on known habitat preference for tigers from 1 - most favourable, least resistance, to 5 - least favourable, most resistance (Wikramanayake et al. 2004). I then regressed least cost distances against genetic distance in a Mantel test (GenAlEx v6.4) to test for a significant correlation. I also implemented the circuit theory method in

CIRCUITSCAPE v3.5.4 (Shah & McRae 2008) to compute relative rates of gene flow between occupied tiger habitat on Sumatra. I used the human footprint index as a baseline raster grid to define resistance to movement for tigers. I used the human footprint score as a direct cost value, i.e. I considered a high human footprint score to equate to a higher cost of movement. I assumed that where human influence was highest the landscape would be most modified and tigers would be under the most pressure from human activity (Carroll et al. 2001). I used a tiger presence map (Wibisono & Pusparini 2010) and field sample GPS coordinates as source points for the gene flow analysis. All maps were prepared in ArcMap v9.2 (Esri, USA) and converted into ASCII files for import into CIRCUITSCAPE using the Export to

Circuitscape Tool v1.0.87 (Jenness 2010). I used the default settings in CIRCUITSCAPE to produce cumulative current output maps.

- 150 - 6.4 Results 6.4.1 Small neighbourhood size Overall, there was a significant pattern of isolation by distance using the genotypic distance measure in GenAlEx (Fig. 6.1) and Rousset's estimator â. The slope of the linear regression between Rousset's â and the logarithm of geographic distance gave a neighbourhood estimate of 35.7 individuals (95% CI 18.1 - 1000). This is comparable to the estimate for effective population size derived from the sibship assignment method in Chapter 5. The result of the spatial autocorrelation test in GenAlEx suggests that this spatial genetic pattern is only detectable over distance classes of up to 800km. This is roughly equivalent to half the long axis of Sumatra from north to south.

Figure 6.1 A test for isolation-by-distance using linear regression of genetic distance and the logarithm of geographic distance (Rxy = 0.371, p-value 0.000).

- 151 - 6.4.2 Two genetic clusters

Initial results from TESS, BAPS and GENELAND found multiple clusters with the majority of individuals assigned to 2 subpopulations. Broadly speaking, the samples from Way Kambas NP were assigned to a single separate cluster. After cluster assignment the outputs from all the Bayesian programs suggested k = 2 (Table. 6.1).

However, only TESS and GENELAND maintained the separation of the southern individuals into a separate cluster. The individual assignments from STRUCTURE and

BAPS did not appear to follow any geographic pattern, although STRUCTURE grouped the eastern samples into one group. This may reflect true structure or the inequality in sample size between regions (Kalinowski 2011). Both GENELAND and TESS have been found to be more powerful at detecting population clusters when either permeable barriers or isolation by distance is present (Coulon et al. 2006; Safner et al. 2011; Blair et al. 2012). Thus, although there is weak power to detect population structure due to the small sample size, these results taken together provide some evidence for separation of the southern Way Kambas NP samples into a separate cluster.

Table 6.1 The number of genetic clusters estimated by STRUCTURE, TESS, BAPS and

GENELAND. The inferred value of k refers to the initial output from each program. The corrected value of k refers to the value after individuals were assigned to clusters.

Method Inferred value of k Corrected value of k

STRUCTURE 2 2

TESS 4 2

BAPS 5 2

GENELAND 3 2

- 152 - 6.4.3 Multiple routes for dispersal Least cost distances derived from the human footprint and land cover classification maps both showed a significant positive relationship with genetic distance. Regression of least cost distances with Rousset’s estimator â gave a jackknife mean slope of ~0.034 (s.e. 0.013, p-value 0.023) and a mean estimate of Nb = 29 individuals (95% CI 16 - 115). Similar results were obtained with the FAO land cover data. The results of the circuit theory analysis suggest that there is still much potential for gene flow on Sumatra. The lowest resistance paths appear to occur across the central portion of Sumatra from east to west encompassing protected areas from Kerinci Seblat NP in the west to Berbak NP in the east (Fig. 6.2). It also appears that the highest resistance paths occur between Way Kambas NP and other tiger habitat (Table 6.2).

Figure 6.2 CIRCUITSCAPE output showing landscape resistance for tigers on Sumatra. Habitat patches >200km2 are indicated in green. Areas of low resistance are highlighted in red/purple. Areas of high resistance are highlighted in yellow/orange.

- 153 - Table 6.2 Cumulative pairwise resistance values between the sampled patches of tiger habitat on Sumatra as estimated by CIRCUITSCAPE. The human footprint index was used as the baseline landscape map.

Gunung Ulu Masen Kerinci Tesso Nilo Bukit Giam Siak Kampar Kerumutan Berbak/ Bukit leuser Seblat/ Tigapuluh Kecil Sembilang Barisan Batang hari Selatan Ulu Masen 1.471

Kerinci Seblat/ 40.018 41.489 Batang Hari Tesso Nilo 46.720 48.190 12.880

Bukit 45.123 46.594 8.643 12.516 Tigapuluh Giam Siak Kecil 53.300 54.771 25.461 24.794 25.155

Kampar 49.119 50.590 15.722 13.281 12.139 20.171

Kerumutan 48.289 49.759 14.581 12.137 10.775 20.583 1.802

Berbak/ 63.958 65.429 25.597 34.297 27.137 46.676 35.228 33.969 Sembilang Bukit Barisan 67.064 68.535 27.502 39.229 34.154 51.762 41.557 40.385 41.423 Selatan Way Kambas 107.090 108.561 67.742 78.931 73.475 91.439 81.016 79.829 75.499 53.357

- 154 - 6.5 Discussion 6.5.1 Isolation by distance The first significant result from this study is that there is a significant relationship between genetic differentiation and geographic distance giving an estimated neighbourhood size of 29 - 35 individuals using least-cost path and haversine geographic distances respectively. Correlation between genetic differentiation and the LCP distance was slightly greater suggesting that the human footprint index provides a more realistic representation of movement through the landscape compared to straight-line distance. While the estimate of Nb is similar to that of effective population size derived using a sibship assignment method in Chapter 5, care must be taken when relating the two parameters. Effective population size gives some indication of the strength of genetic drift acting on the population (Falconer & MacKay 1996). In contrast, neighbourhood size explains the degree of local differentiation that occurs within a continuously distributed population as a result of restricted dispersal (Wright 1943; Wright 1946). Where no dispersal barriers exist, differentiation occurs because individuals are more likely to mate with close neighbours. This is particularly true where there are large areas of homogeneous habitat or for habitat generalists that are less affected by landscape structure. Historically tigers would have been widely distributed across much of mainland Asia and the Sunda islands mirroring the continuous distribution of tropical forest and prey species there (Seidensticker et al. 1999). In genetic terms, it appears that the current Sumatran population still shows some evidence of this continuous distribution at a landscape scale.

Whilst many studies look for evidence of IBD very few use the relationship to estimate neighbourhood size, preferring to estimate effective population size instead. Most populations will not meet all the assumptions of Wright's IBD model, such as normal distribution of dispersal distances and equal dispersal between sexes, and Ne may provide a more accurate reflection of a population's genetic properties.

- 155 - However, Nb still gives some indication of the level of differentiation to be expected in a given population. Small neighbourhoods of ~20 individuals are likely to produce far more local differentiation than larger groups (Wright 1946; Wright 1969). Intuitively, tigers would be expected to have a large neighbourhood size because they are capable of large dispersal distances. However, they also occur at low population density with large home ranges so in reality a large area may contain very few individuals. Surprisingly, the estimate of Nb in this study is comparable to similar studies of the prickly forest skink, European mink, Myobatrachid frogs, and hazel grouse, which have all generated similar estimates of 2-185 individuals (Driscoll 1999; Sumner et al. 2001; Lodé & Peltier 2005; Sahlsten et al. 2008). Therefore, tigers may breed with far fewer individuals than expected from their dispersal abilities. Sex-biased dispersal may also act to reduce neighbourhood size relative to that expected when both sexes disperse equally. Male tigers, as with most wild felids, disperse over greater distances compared to females who tend to have smaller and overlapping home ranges close to their natal site (Smith et al. 1987; Smith 1993; Goodrich et al. 2010; Barlow et al. 2011). Dispersal also influences the size of the neighbourhood area (Wright 1969). Individuals within a neighbourhood area form a randomly mating group, and the size of this area determines the scale over which differentiation occurs (Rousset 2004). In this study it appears that although neighbourhood size is small, neighbourhood area is relatively large with differentiation occurring over a distance of up to 850km. It is unlikely that tigers in a tropical habitat would disperse over such a large distance in one journey. Subadult males often have to move from place to place until they find a vacant region in which to create their own territory. There are sufficient patches of forest within much of Sumatra, but particularly in the central and western parts of the island, for this stepping-stone migration to occur (Wibisono & Pusparini 2010).

6.5.2 Spatial genetic structure It is possible for natural populations to show evidence of both IBD and population structure (Wright 1940). There appears to be weak support for two genetic clusters

- 156 - on Sumatra, although only two of the four Bayesian clustering programs produced clusters that were biologically meaningful. I used non-spatial models and randomised locations to test the weight given to geographic location in the cluster assignment. For all the Bayesian methods, removing the location data resulted in no clear population structure. This suggests that in the presence of weak genetic structure, spatial Bayesian models may be biased by geographic information in the data set. Using separate Mantel tests, there was also no clear indication that the degree of genetic differentiation between Way Kambas and the rest of Sumatra is any greater than that between individuals overall. The slope of IBD plots were the same for both central Sumatra and for the whole of the island. If an impermeable dispersal barrier did exist around Way Kambas, individuals on the same side of the barrier would be expected to have greater relatedness than individuals on either side of the barrier (Frantz et al. 2010). There are three possible reasons for the lack of population structure noted in this study. The first is that population structure does not exist, and that tigers are still able to disperse effectively through the matrix of different land uses on Sumatra. This could occur if there was sufficient vegetation cover and prey availability within each individual's home range. The second relates to the time lag between changes to the landscape and a resultant change in gene flow. It is possible that given the tigers long generation time of 7 years, it could take up to 105 years (15 generations) for a clear signal of a landscape barrier to be detectable (Holzhauer et al. 2006; Clark et al. 2010; Landguth et al. 2010). Thus, although land cover conversion has been severe, there may be sufficient patches of remnant forest to allow tigers to permeate through the landscape. Finally, studies of the effect of sample size on STRUCTURE analysis suggest that unequal sampling or sample sizes below 10 individuals may bias results (Chapter 2). STRUCTURE may either fail to find genetic clusters where they exist, or may preferentially group together individuals from the sample with the biggest size (Kalinowski 2011).

The question still remains of how tigers will be affected by ongoing landscape change. Field abundance studies and individual-based models conclude that tiger

- 157 - abundance is lowest in areas of high human population density or human disturbance (O'Brien et al. 2003; Carroll & Miquelle 2006; Johnson et al. 2006; Linkie et al. 2006). Tropical forest is the most favoured habitat, followed by logging concessions and agroforestry, then acacia, settlements and oil palm (Imron et al. 2011). This suggests that although there is no evidence for population structure now, this is likely to change in the future. The lowland forests of Lampung and Riau province have been severely affected by deforestation in the past 20 years. Satellite maps and ground surveys suggest that the two national parks, Way Kambas and Bukit Barisan Selatan NP, are the only viable tiger habitat in Lampung (Nyhus et al. 2000). Way Kambas in particular is now surrounded by a zone of urbanisation that could be sufficient to isolate the tigers within its borders (Imbernon 1999; Nyhus & Tilson 2004a; Gaveau et al. 2007). Lowland forest in this area has been replaced by more open habitat consisting of land clearance, agricultural crops or small plantations, and remnant patches of forest (Miettinen et al. 2008; Miettinen & Liew 2010). Lowland peat swamp forest in Riau is also being cleared for agriculture, paper and pulp or oil palm plantations (Uryu et al. 2008; Hansen et al. 2009; Broich et al. 2011; Koh et al. 2011; Miettinen et al. 2011b). The tiger population here is certain to be affected given that tigers avoid areas of high human activity and that the probability of persistence is also much lower for these types of habitat (Imron et al. 2011). In reality, tigers in human-modified landscapes may be able to cope with more land cover change than we expect. They are still able to make use of secondary or degraded forest, and can hunt as long as sufficient ground cover is available (Smith 2009; Sunarto et al. 2012). Perhaps then the continued loss of primary forest will not result in any hard edges between habitat types, but will create a reorganisation of the population with contraction towards the montane forest in the north and west. Tigers would then persist within a series of inviolate core areas with individuals dispersing out into the wider landscape through any remaining forest fragments.

- 158 - 6.5.3 Potential for connectivity The human footprint index appears to be an adequate surrogate for a gene flow resistance map for tigers as the CIRCUITSCAPE output was very similar to that obtained using the FAO land cover data. Given the relative values in the human footprint index, there seems to be much potential for connectivity on Sumatra. The highest flow rates appear in central Sumatra incorporating the forest patches between the large TCLs in Kerinci Seblat NP and Riau. Another major route follows the Bukit Barisan mountain range along the western edge of the island. The circuit map also highlights Kerinci Seblat NP as a key site central to both north-south and east-west connectivity. The extreme ends of the island are subject to the lowest rates of flow, but this is expected given that there are fewer options for linkages to other regions. In its current form the circuit map can serve only as a guide to potential routes for gene flow on Sumatra. CIRCUITSCAPE was not able to fully resolve resistance values between all individuals in the data set and so it was not possible to determine if genetic distance was correlated with landscape resistance. It is difficult to accurately parameterise a resistance model for tropical tigers as there is very little data relating to rates of movement in different habitat types, and there have not been any genetic distance studies prior to this one. Species occurrence studies have largely shown a preference for forest (both primary and secondary or disturbed forest) and the avoidance of roads and people (Carroll & Miquelle 2006; Maddox et al. 2007; Linkie et al. 2008b). A recent study suggests a link to high altitude and permanent water sources, but this could be due to the correlation with human population density (Sunarto et al. 2012). Least-cost models for tigers can therefore only incorporate crude classifications of land cover type and human presence to specify resistance to movement. This appears to be typical of the analysis for habitat generalists (Beier et al. 2008; Beier et al. 2009). However, such analysis can still highlight how gene flow patterns could change with the loss of forest fragments between TCLs (Koen et al.

2012). For example, CIRCUITSCAPE analysis with parts of the Riau fragments removed show that gene flow would be largely limited to connections between Gunung Leuser, Kerinci Seblat NP and Bukit Barisan Selatan NP. The reduced productivity in

- 159 - these montane and submontane regions could mean reduced capacity to support a large population of tigers.

6.6 Conclusions For the current Sumatran tiger population it appears that dispersal ability may be the major determinant of spatial genetic structure. Whilst the TCLs provide a convenient measure of habitat availability for tigers, they do not appear to correspond to gene flow patterns on the ground. It is likely that the TCLs act as refugia in a matrix of disturbance with individuals dispersing into the wider landscape by using smaller patches of degraded forest and scrub within dispersal distance. In the absence of empirical movement data, genetic distance is a useful tool to determine patterns of gene flow. However, it must be combined with traditional occurrence studies to provide a more complete picture of habitat preference and tiger distribution. This multifaceted approach will then allow conservationists to identify subpopulations at risk and the regions that provide important linkages in the landscape.

- 160 - Chapter 7

General discussion

- 161 - 7.1 Key ﬁndings The main aim of this work was to address the question of species persistence in fragmented landscapes by combining the fields of population genetics and landscape ecology. Habitat fragmentation typically results in population subdivision and this prompts conservation managers to consider how this might affect extinction risk. In this study I used the tiger populations of Sumatra (Panthera tigris sumatrae) as an example of a wide-ranging carnivore that has been subject to extensive habitat fragmentation over the last 20-30 years. This study is the first to gather data on the genetic status of the overall population and as such provides baseline measures of key genetic parameters: genetic variation, effective population size, neighbourhood size, population structure and connectivity. Confronted by the small sample sizes often associated with studies of rare, elusive species, I conducted a comprehensive series of computer simulations to show that as few as 5-10 individuals can be used to obtain unbiased estimates of genetic variation and population structure. These estimates are also robust to reasonably high levels of genotyping error. Overall, this body of work lends support to the increasing popularity of noninvasive genetics techniques as a means of assessing and monitoring the status of wildlife populations at landscape scales.

High genetic variation Despite being a relatively small island population, tigers on Sumatra appear to have retained high levels of genetic diversity. Previous studies have indicated that the Sumatran subspecies may be genetically impoverished compared to those on the mainland (Luo et al. 2004; Mondol et al. 2009b), but the more extensive sampling presented here suggests that this is not the case. The high levels of diversity observed on Sumatra are possible for two reasons. First, the island once formed part of a larger population within the Sunda plate (Whitten et al. 2000; Kinnaird et al. 2003) and second, the relatively long generation time of tigers, typically 7 years (Tilson & Seal 1987), means that a decline in heterozygosity caused by genetic drift

- 162 - will be slow to take effect (Hamilton 2009). This means that habitat fragmentation on Sumatra has occurred relatively recently, within 3-5 tiger generations, and there has been little time for large genetic changes to occur. Therefore, the risk of inbreeding effects appears to be low on Sumatra and a loss of genetic diversity is unlikely to be the immediate cause of a population decline.

Genetic neighbourhoods Analysis of both effective population size and neighbourhood size suggests that the tigers in this system are largely structured by dispersal distances and the mating system. Spatial genetic structure occurred over an approximate distance of 850km and panmictic (randomly mating) groups consisted of up to 30 individuals. This is in line with previous work in Nepal, which estimated effective population size from data on reproductive output (Smith & McDougal 1991). In the absence of habitat fragmentation tigers may preferentially distribute themselves evenly across a landscape and breed within a limited area. Tigers are polygynous and each resident male has a large territory that overlaps with several females (Sunquist 1981). Thus, breeding mostly occurs in units delineated by each male's home range.

Connectivity An additional and key result for Sumatran tigers is that there was low regional differentiation and no genetic evidence for dispersal barriers or strong population structure. This could be due to sustained connectivity between regions or insufficient time for fragmentation to take effect on the population. However, the results of this study suggest the primary reason may be connectivity. Analysis using resistance maps derived from land cover types and a composite measure of human disturbance (the human footprint) indicated there are sufficient forest patches to facilitate connectivity throughout much of Sumatra. The highest rates of gene flow were found between the Batang Hari/Kerinci Seblat ecosystem (west Sumatra) and Ulu Masen/ Gunung Leuser (north Sumatra) and Riau province (east Sumatra). Lowest rates were associated with Way Kambas NP in the south. The Tiger Conservation

- 163 - Landscapes in the west and north represent the largest contiguous swaths of forest on Sumatra and support up to 150 tigers each (Linkie et al. 2006; Sanderson et al. 2006). They are therefore large enough to serve as source populations for the surrounding regions. In contrast, Way Kambas is home to 20-30 individuals and lies in an isolated position (Nyhus & Tilson 2004a). This suggests a different trajectory in each region in response to continued landscape change. The north and west have a relatively low risk of population extinction, whilst continued isolation could increase extinction risk in the south.

7.2 Implications for Sumatra Taken together these results suggest that whilst there has been severe deforestation on Sumatra, it has not yet been reflected in the genetic structure or patterns of gene flow in the current population. Effective population size is within expected limits, and both genetic variation and connectivity are of sufficient levels to maintain population persistence in the current landscape. However, there are still 2 main problems that conservation managers face. The first is the isolation and differentiation of the smaller subpopulations on Sumatra. Way Kambas NP showed the highest levels of differentiation of all the sampled regions, and it is likely that this is due to the profound land clearance that has occurred in Lampung Province (Imbernon 1999; Nyhus et al. 2000). Similar habitat loss is now occurring in the lowland peat swamp forests of east Sumatra (Broich et al. 2011; Miettinen et al. 2011b), and the concern is that this could introduce similar levels of isolation. Whilst there is evidence of connectivity in central Sumatra, this is largely dependent on connections to the populations in the west of the island. Severing these links would reduce the potential for migration. The second issue is the potential vulnerability of the larger subpopulations. Intact, they represent valuable source populations that are likely to persist into the future. However, development pressures mean that they are constantly threatened by the construction of large access roads and subsequent human settlement (Harimaukita 2011). Bisecting the populations with one or two roads may seem benign, but increasing access to previously untouched forest often

- 164 - results in human colonisation and exploitation of forest resources (Linkie et al. 2004). Subdivision of the habitats supporting these tiger populations poses a significant threat to the long-term viability of tigers in this region.

7.3 Application to landscape species In this thesis I have shown that genetic monitoring is a valuable tool for the assessment of species persistence in fragmented landscapes. For species that occur at low density and that are distributed over very large areas, noninvasive monitoring offers an effective way to sample the population. Since the most useful metrics of extinction risk are population size and connectivity, noninvasive genetic sampling offers the opportunity to address both of these issues from a single round of sampling. Total population size can be estimated using capture-recapture statistics or alternatively effective population size and neighbourhood size can be derived from measures of genetic distance between individuals. For many regions, sample collection could be incorporated into yearly camera-trap or presence/ absence surveys that are currently employed to monitor population trends. A comparison between camera trap and genetic surveys provides an effective method to validate any population size estimates. Connectivity can be determined in several ways including estimation of differentiation (e.g. FST), direct estimates of gene flow, and least-cost path or circuit theory methods that take into account the species' perception of landscape hostility. This information could also be used to identify the landscape features that impede connectivity. Despite the challenges involved in collecting and processing noninvasive samples, neither sample size nor genotyping errors limited the quality of the information obtained from the population described here. However, sampling should still be extensive and should cover as many regions as possible to ensure an accurate representation of the total population. In future studies on Sumatra, it would be best to target additional populations in east Sumatra (e.g. Berbak NP, Sembilang NP) and south Sumatra (e.g. Bukit Bali Renjang NP, Bukit Barisan Selatan NP) as this would give more power to determine population structure and the influence of geographic distance on population differentiation.

- 165 - There are many advantages to a purely genetic approach but also a few limitations in this research context. Genetic data alone cannot be used to address the distribution of habitat, species distribution or vital rates (reproduction and mortality) in each habitat type. However, these issues can be largely overcome by incorporating aspects of landscape ecology. This combined approach is best described by the field of landscape genetics, which aims to relate genetic information on gene flow and population structure to factors such as habitat quality and landscape configuration (Manel et al. 2003; Storfer et al. 2007; Holderegger & Wagner 2008). This provides a more complete picture of a species' interaction with its environment and the important factors affecting population persistence in different habitat types. Regular assessment of landscape structure would also account for the potential time lag in genetic studies (Holzhauer et al. 2006; Clark et al. 2010; Landguth et al. 2010). In many cases habitat loss is not expected to affect species distribution until some critical threshold is reached (the extinction threshold), and thus there may not be any changes in estimates of gene flow until a large proportion of available habitat has been lost (Hanski 1998; Fahrig 2003). Similarly, there may be a delay in the response of a population to habitat disturbance. This constitutes an extinction debt and it is expected to be larger for low density species (Tilman et al. 1994; With 2004). For large carnivores there is also the added effect of poaching and conflict with humans. For regions in which the amount of available habitat lies above the extinction threshold, increased mortality due to the interaction with people could be the primary factor influencing extinction risk (Linkie et al. 2006; Chapron et al. 2008; Packer et al. 2009; Goodrich et al. 2010). This has not been widely quantified for tigers, but resistance maps could reflect the combined effect of both human population density and active persecution of tigers.

- 166 - 7.4 Conclusions The research presented here combined with ecology studies on Sumatra clearly demonstrate that while the Sumatran tiger faces significant threats, large well connected populations still remain on the island (Linkie et al. 2006; Wibisono & Pusparini 2010; Wibisono et al. 2011). This is in contrast to studies across the tiger's geographic range, which have found structured populations in response to landscape change (Henry et al. 2009; Joshi 2010; Sharma et al. 2010; Reddy et al. 2012). The techniques described here have provided a valuable snapshot of tigers on Sumatra and suggest that with adequate protection from persecution and the loss of their habitats they could persist well into the future.

- 167 - Bibliography

Adams, J.R. & Waits, L.P. (2007) An efficient method for screening faecal DNA genotypes and detecting new individuals and hybrids in the red wolf (Canis rufus) experimental population area. Conservation Genetics, 8, 123-131.

Adriaensen, F., Chardon, J.P., De Blust, G., Swinnen, E., Villalba, S., Gulinck, H. & Matthysen, E. (2003) The application of ‘least-cost’ modelling as a function of a landscape model. Landscape and Urban Planning, 64, 233-247.

Ahearn, S.C., Smith, J.L.D., Joshi, A. & Ding, J. (2001) TIGMOD: an individual-based spatially explicit moel for simulating tiger/human interaction in multiple use forests. Ecological Modelling, 140, 81-97.

Akçakaya, H.R. & Atwood, J.L. (1997) A habitat-based metapopulation model of the California gnatcatcher. Conservation Biology, 11, 422-434.

Akçakaya, H.R., Mills, G. & Doncaster, C.P. (2006) The role of metapopulations in conservation. In Key Topics in Conservation Biology., Wiley-Blackwell, pp. 64.

Akçakaya, H.R., Radeloff, V.C., Mladenoff, D.J. & He, H.S. (2004) Integrating landscape and metapopulation modeling approaches: Viability of the sharp-tailed grouse in a dynamic landscape. Conservation Biology, 18, 526-537.

Al-soud, W.A. & Rådström, P. (2000) Effects of amplification facilitators on diagnositic PCR in the presence of blood, feces, and meat. Journal of Clinical Microbiology, 38, 4463-4470.

Alassad, S., Soriguer, R.C., Chelomina, G., Sushitsky, Y.P. & Fickel, J. (2011) Siberian tiger’s recent population bottleneck in the Russian Far East revealed by microsatellite markers. doi: 10.1016/j.mambio.2011.01.011

Alberto, F. (2009) MsatAllele_1.0: An R package to visualize the binning of microsatellite alleles. Journal of Heredity, 100, 394-397.

Allendorf, F.W. (1986) Genetic drift and the loss of alleles versus heterozygosity. Zoo Biology, 5, 181-190.

Allendorf, F.W. & Luikart, G. (2007) Conservation and the genetics of populations. Blackwell Publishing,

Amarasekare, P. (1998) Allee effects in metapopulation dynamics. American Naturalist, 152, 298-302.

Amos, W. (2006) The hidden value of missing genotypes. Molecular Biology and Evolution, 23, 1995-1996.

Amos, W., Hoffman, J.I., Frodsham, A., Zhang, L., Best, S. & Hill, A.V.S. (2007) Automated binning of microsatellite alleles: problems and solutions. Molecular Ecology Notes, 7, 10-14.

- 168 - Angers, B., Magnan, P., Plante, M. & Bernatchez, L. (1999) Canonical correspondence analysis for estimating spatial and environmental effects on microsatellite gene diversity in brook char (Salvelinus fontinalis). Molecular Ecology, 8, 1043-1053.

Arandjelovic, M., Guschanski, K., Schubert, G., Harris, T.R., Thalmann, O., Siedel, H. & Vigilant, L. (2009) Two-step multiplex polymerase chain reaction improves the speed and accuracy of genotyping using DNA from noninvasive and museum samples. Molecular Ecology Resources, 9, 28-36.

Archie, J.W. (1985) Statistical analysis of heterozygosity data: independent sample comparisons. Evolution, 39, 623-637.

Ariefdjohan, M.W., Savalano, D.A. & Nakatsu, C.H. (2010) Comparison of DNA extraction kits for PCR-DGGE analysis of human intestinal microbial communities from fecal specimens. Nutrition Journal, 9, 23.

Arnaud, J. (2003) Metapopulation genetic structure and migration pathways in the land snail Helix aspersa: influence of landscape heterogeneity. Landscape Ecology, 18, 333-346.

Aspi, J.,Roininen E.,Ruokonen M.,Kojola I., C., V. (2006) Genetic diversity, population structure, effective population size and demographic history of the Finnish wolf population. Molecular Ecology, 15, 1561-1576.

Avise, J.C. (1994) Molecular markers, natural history and evolution. Chapman & Hall Inc., New York.

Baeza, A. & Estades, C.F. (2010) Effect of the landscape context on the density and persistence of a predator population in a protected area subject to environmental variability. Biological Conservation, 143, 94-101.

Balding, D.J., Bishop, M. & Cannings, C. (2007) Handbook of statistical genetics. Wiley & Sons Ltd, Chicester, West Sussex.

Balkenhol, N., Waits, L.P. & Dezzani, R.J. (2009) Statistical approaches in landscape genetics: an evaluation of methods for linking landscape and genetic data. Ecography, 32, 818-830.

Ballou, J.D. & Seidensticker, J. (1987) The genetic and demographic characteristics of the 1983 captive population of Sumatran tigers (Panthera tigris sumatrae). In Tigers of the world: The biology, biopolitics, management and conservation of an endangered species, (Eds, Tilson, R. & Seal, U.S.) Noyes Publications, New Jersey, USA, pp. 329-347.

Balloux, F. (2001) EASYPOP (Version 1.7): a computer program for population genetics simulations. Journal of Heredity, 92, 301-302.

Balloux, F. & Goudet, J. (2002) Statistical properties of population differentiation estimators under stepwise mutation in a finite island model. Molecular Ecology, 11, 771-783.

Barlow, A.C.D., Smith, J.L.D., Ahmad, I.U., Hossain, A.N.M., Rahman, M. & Howlader, A. (2011) Female tiger Panthera tigris home range size in the Bangladesh Sundarbans: the

- 169 - value of this mangrove ecosystem for the species’ conservation. Oryx, 45, 125-128.

Barone, M.A., Roelke, M.E., Howard, J.G., Brown, J.L., Anderson, A.E. & Wildt, D.E. (1994) Reproductive characteristics of male Florida panthers: Comparative studies from Florida, Texas, Colorado, Latin America, and North American zoos. Journal of Mammalogy, 75, 150-162.

Beaumont, M.A. (1999) Detecting population expansion and decline using microsatellites. Genetics, 153, 2013-2029.

Beaumont, M.A. & Nichols, R.A. (1996) Evaluating loci for use in the genetic analysis of population structure. Proceedings of the Royal Society of London - B. Biological Sciences, 263, 1619-1626.

Beebee, T. & Rowe, G. (2004) Molecular identification: species, individuals and sex. In An Introduction to Molecular Ecology, (Eds, Beebee, T. & Rowe, G.) Oxford University Press, Oxford, pp. 61.

Beerli, P. & Felsenstein, J. (1999) Maximum-likelihood estimatin of migration rates and effective population numbers in two populations using a coalescent approach. Genetics, 152, 763-773.

Beier, P., Majka, D.R. & Newell, S.L. (2009) Uncertainty analysis of least-cost modeling for designing wildlife linkages. Ecological Applications, 19, 2067-2077.

Beier, P., Majka, D.R. & Spencer, W.D. (2008) Forks in the road: choices in procedures for designing wildland linkages. Conservation Biology, 22, 836-851.

Beier, P. & Noss, R.F. (1998) Do habitat corridors provide connectivity? Conservation Biology, 12, 1241-1252.

Belant, J.L.,Van Stappen J. F., D., P. (2005) American black bear population size and genetic diversity at Apostle Islands National Lakeshore. Ursus, 16, 85-92.

Bellemain, E., Swenson, J.E., Tallmon, D., Brunberg, S. & Taberlet, P. (2005) Estimating population size of elusive animals with DNA from hunter-collected feces: four methods for brown bears. Conservation Biology, 19, 150-161.

Bellemain, E. & Taberlet, P. (2004) Improved noninvasive genotyping method: application to brown bear (Ursus arctos) faeces. Molecular Ecology Notes, 4, 519-522.

Bender, D.J. & Fahrig, L. (2005) Matrix structure obscures the relationship between interpatch movement and patch size and isolation. Ecology, 86, 1023-1033.

Bercovich, D., Regev, Z., Ratz, T., Luder, A., Plotsky, Y. & Gruenbaum, Y. (1999) Quantitative ratio of primer pairs and annealing temperature affecting PCR products in duplex amplification. BioTechniques, 27, 762-770.

Bessetti, J. (2007) An introduction to PCR inhibitors. Promega Corporation,

- 170 - Bhagavatula, J. & Singh, L. (2006) Genotyping faecal samples of Bengal tiger Panthera tigris tigris for population estimation: A pilot study. BMC Genetics, 7, 48.

Bijlsma, R., Bundgaard, J. & Boerema, A.C. (2000) Does inbreeding affect the extinction risk of small populations?: predictions from Drosophila. Journal of Evolutionary Biology, 13, 502-514.

Biswas, S. & Sankar, K. (2002) Prey abundance and food habit of tigers (Panthera tigris tigris) in Pench National Park, Madhya Pradesh, India. Journal of Zoology (London), 256, 411-420.

Blair, C., Weigel, D.E., Balazik, M., Keeley, A.T.H., Walker, F.M., Landguth, E., Cushman, S., Murphy, M., Waits, L. & Balkenhol, N. (2012) A simulation-based evalutation of methods for inferring linear barriers to gene flow. doi: 10.1111/j.1755-0998.2012.03151.x

Boecklen, W.J. & Howard, D.J. (1997) Genetic analysis of hybrid zones: numbers of markers and power of resolution. Ecology, 78, 2611-2616.

Bonin, A., Bellemain, E., Bronken Eidensen, P., Pompanon, F., Brochmann, C. & Taberlet, P. (2004) How to track and assess genotyping errors in population genetics studies. Molecular Ecology, 13, 3261-3273.

Boom, R., Sol, C.J.A., Salimans, M.M.M., Jansen, C.L., Wertheim van Dillen, P.M.E. & van der Noordaa, J. (1990) Rapid and simple method for purification of nucleic acids. Journal of Clinical Microbiology, 1990, 495-503.

Borthakur, U., Barman, R.D., Das, C., Basumatary, A., Talukdar, A., Ahmed, M.F., Talukdar, B.K. & Bharali, R. (2010) Noninvasive genetic monitoring of tiger (Panthera tigris tigris) population of Orang National Park in the Brahmaputra floodplain, Assam, India. doi: 10.1007/s10344-010-0471-0

Bovo, D., Rugge, M. & Shiao, Y.-H. (1999) Origin of spurious multiple bands in the amplification of microsatellite sequences. Molecular Pathology, 52, 50-51.

Bradley, B.J. & Vigilant, L. (2002) False alleles derived from microbial DNA pose a potential source of error in microsatellite genotyping of DNA from faeces. Molecular Ecology Notes, 2, 602-605.

Breazel, E.H. (2008) Effects of common errors in microsatellite data on estimates of population differentiation, and Inferring genotypic structure of complex disease loci using genome-wide expression data. PhD thesis, The University of Georgia.

Broich, M., Hansen, M., Stolle, F., Potapov, P., Margono, B.A. & Adusei, B. (2011) Remotely sensed forest cover loss shows high spatial and temporal variation across Sumatera and Kalimantan, Indonesia 2000-2008. Environmental Research Letters, 6, 014010.

Brook, B.W., Tonkyn, D.W., O’Grady, J.J. & Frankham, R. (2002) Contribution of inbreeding to extinction risk in threatened species. Conservation Ecology, 6, 16.

Broquet, T., Ménard, N. & Petit, E. (2007) Noninvasive population genetics: a review of

- 171 - sample source, diet, fragment length and microsatellite motif effects on amplification success and genotyping error rates. Conservation Genetics, 8, 249-260.

Broquet, T. & Petit, E. (2004) Quantifying genotyping errors in noninvasive population genetics. Molecular Ecology, 13, 3601-3608.

Broquet, T., Ray, N., Petit, E., Fryxell, J.M. & Burel, F. (2006) Genetic isolation by distance and landscape connectivity in the American marten (Martes americana). Landscape Ecology, 21, 877-889.

Brownstein, M.J., Carpten, J.D. & Smith, J.R. (1996) Modulation of non-templated nucleotide addition by Taq DNA polymerase: primer modifications that facilitate genotyping. BioTechniques, 20, 1004-1010.

Buchan, J.C., Archie, E.A., van Horn, R.C., Moss, C.J. & Alberts, S.C. (2005) Locus effects and sources of error in noninvasive genotyping. Molecular Ecology Notes, 5, 680-683.

Buckley-Beason, V.A., Johnson, W.E., Nash, W.G., Stanyon, R., Menninger, J.C., Driscoll, C.A., Howard, J., Bush, M., Page, J.E., Roelke, M.E., Stone, G., Martelli, P.P., Wen, C., Ling, L., Duraisingam, R.K., Lam, P.V. & O’Brien, S.J. (2006) Molecular Evidence for Species-Level Distinctions in Clouded Leopards. Current Biology, 16, 2371-2376.

Cabellero, A. (1994) Developments in the prediction of effective population size. Heredity, 73, 657-679.

Callen, D.F., Thompson, A.D., Shen, Y., Phillips, H.A., Richards, R.I., Mulley, J.C. & Sutherland, G.R. (1993) Incidence and origin of “null” alleles in the (AC)n microsatellite markers. American Journal of Human Genetics, 52, 922-927.

Cardillo, M., Purvis, A., Bielby, J., Mace, G.M., Sechrest, W. & Gittleman, J.L. (2004) Human population density and extinction risk in the world’s carnivores. PLoS Biology, 2, 0909-0914.

Carr, L.W. & Fahrig, L. (2001) Effect of road traffic on two amphibian species of differing vagility. Conservation Biology, 15, 1071-1078.

Carroll, C. & Miquelle, D.G. (2006) Spatial viability analysis of Amur tiger Panthera tigris altaica in the Russian Far East: The role of protected areas and landscape matrix in population persistence. Journal of Applied Ecology, 43, 1056-1068.

Carroll, C., Noss, R.F. & Paquet, P.C. (2001) Carnivores as focal species for conservation planning in the Rocky Mountain region. Ecological Applications, 11, 961-980.

Carroll, C., Noss, R.F., Paquet, P.C. & Schumaker, N.H. (2003a) Use of population viability analysis and reserve selection algorithms in regional conservation plans. Ecological Applications, 13, 1773-1789.

Carroll, C., Noss, R.F., Paquet, P.C. & Schumaker, N.H. (2004) Extinction debt of protected areas in developing landscapes. Conservation Biology, 18, 1110-1120.

Carroll, C., Phillips, M.K., Schumaker, N.H. & Smith, D.W. (2003b) Impacts of landscape

- 172 - change on wolf restoration success: planning a reintroduction program based on static and dynamic spatial models. Conservation Biology, 17, 536-548.

Castilho, C.S., Marins-Sá, L.G., Benedet, R.C. & Freitas, T.O. (2010) Landscape genetics of mountain lions (Puma concolor) in southern Brazil. doi: 10.1016/j.mambio.2010.08.002

Cavers, S., Degen, B., Caron, H., Lemes, M.R., Margis, R., Salgueiro, F. & Lowe, A.J. (2005) Optimal sampling strategy for estimation of spatial genetic structure in tree populations. Heredity, 95, 281-289.

Cegelski, C.C.,Waits L. P.,Anderson N. J.,Flagstad O.,Strobeck C., J., K.C. (2006) Genetic diversity and population structure of wolverine (Gulo gulo) populations at the southern edge of their current distribution in North America with implications for genetic viability. Conservation Genetics, 7, 197-211.

Cegelski, C.C., Waits, L.P. & Anderson, N.J. (2003) Assessing population structure and gene flow in Montana wolverines (Gulo gulo) using assignment-based approaches. Molecular Ecology, 12, 2907-2918.

Chapron, G., Miquelle, D.G., Lambert, A., Goodrich, J.M., Legendre, S. & Clobert, J. (2008) The impact on tigers of poaching versus prey depletion. Journal of Applied Ecology, 45, 1667-1674.

Chapuis, M. & Estoup, A. (2007) Microsatellite null alleles and estimation of population differentiation. Molecular Biology and Evolution, 24, 621-631.

Charruau, P., Fernandes, C., Orozco-terWengel, P., Peters, J., Hunter, L., Ziaie, H., Jourabchian, A., Jowkar, H., Schaller, G., Ostrowski, S., Vercammen, P., Grange, T., Schlötterer, C., Kotze, A., Geigl, E.M., Walzer, C. & Burger, P.A. (2011) Phylogeography, genetic structure and population divergence time of cheetahs in Africa and Asia: evidence for long-term geographic isolates. doi: 10.1111/j.1365-294X.2010.04986.x

Chauvenet, A.L.M., Baxter, P.W.J., McDonald-Madden, E. & Possingham, H.P. (2010) Optimal allocation of conservation effort among subpopulations of a threatened species: How important is patch quality? Ecological Applications, 20, 789-797.

Chen, C., Durand, E., Forbes, F. & Francois, O. (2007) Bayesian clustering algorithms ascertaining spatial population structure: a new computer program and a comparison study. Molecular Ecology Notes, 7, 747-756.

Chesser, R.K. (1991) Gene diversity and female philopatry. Genetics, 127, 437-447.

Chikhi, L., Sousa, V.C., Luisi, P., Goossens, B. & Beaumont, M.A. (2010) The confounding effects of population structure, genetic diversity and the sampling scheme on the detection and quantification of population size changes. Genetics, 186, 983-995.

Chundawat, R.S., Habib, B., Karanth, U., Kawanishi, K., Ahmad Khan, J., Lynam, T., Miquelle, D., Nyhus, P., Sunarto, S., Tilson, R. & Wang, S. (2011) Panthera tigris. IUCN 2011. www.iucnredlist.org/apps/redlist/details/15955/0

- 173 - Clark, R.W., Brown, W.S., Stechert, R. & Zamudio, K.R. (2010) Roads, interrupted dispersal, and genetic diversity in timber rattlesnakes. Conservation Biology, 24, 1059-1069.

Clarke, L.A., Rebelo, C.S., Gonçalves, J., Boavida, M.G. & Jordan, P. (2001) PCR amplification introduces errors into mononucleotide and dinucleotide repeat sequences. Molecular Pathology, 54, 351-353.

Coltman, D.W., Don Bowen, W. & Wright, J.M. (1998) Birth weight and neonatal survival of harbour seal pups are positively correlated with genetic variation measured by microsatellites. Proceedings of the Royal Society of London. Series B, 265, 803-809.

Corander, J., Sirén, J. & Arjas, E. (2008) Bayesian spatial modeling of genetic populations structure. Computational Statistics, 23, 111-129.

Corander, J., Waldmann, P. & Sillanpää, M.J. (2003) Bayesian analysis of genetic differentiation between populations. Genetics, 163, 367-374.

Cornuet, J.-M. & Luikart, G. (1996) Description and power analysis of two tests for detecting recent population bottlenecks from allele frequency data. Genetics, 144, 2001-2014.

Coulon, A., Fitzpatrick, J.W., Bowman, R., Stith, B.M., Makarewich, C.A., Stenzler, L.M. & Lovette, I.J. (2008) Congruent population structure inferred from dispersal behaviour and intensive genetic surveys of the threatened Florida scrub-jay (Aphelocoma cœrulescens). Molecular Ecology, 17, 1685-1701.

Coulon, A., Guillot, G., Cosson, J.-F., Angibault, J.M.A., Aulangnier, S., Cargnelutti, B., Galan, M. & Hewison, A.J.M. (2006) Genetic structure is influenced by landscape features: empirical evidence from a roe deer population. Molecular Ecology, 15, 1669-1679.

Cracraft, J., Feinstein, J., Vaughn, J. & Helm Bychowski, K. (1998) Sorting out tigers (Panthera tigris): Mitochondrial sequences, nuclear inserts, systematics, and conservation genetics. Animal Conservation, 1, 139-150.

Cramer, P.C. & Portier, K.M. (2001) Modeling Florida panther movements in response to human attributes of the landscape and ecological settings. Ecological Modelling, 140, 51-80.

Crawford, T.J. (1984) The estimation of neighbourhood parameters for plant populations. Heredity, 52, 273-283.

Creel, S., Spong, G., Sands, J.L., Rotella, J., Zeigle, J., Joe, L., Murphy, K.M. & Smith, D. (2003) Population size estimation in Yellowstone wolves with error-prone noninvasive microsatellite genotypes. Molecular Ecology, 12, 2003-2009.

Crnokrak, P. & Roff, D.A. (1999) Inbreeding depression in the wild. Heredity, 83, 260-270.

Crooks, K.R. (2002) Relative sensitivities of mammalian carnivores to habitat fragmentation. Conservation Biology, 16, 488-502.

Csiki, I., Lam, C., Key, A., Coulter, E., Clark, J.D., Pace III, R.M., Smith, K.G. & Rhoads, D.D. (2003) Genetic variation in black bears in Arkansas and Louisiana using microsatellite

- 174 - DNA markers. Journal of Mammalogy, 84, 691-701.

Culver, M., Menotti-Raymond, M.A. & O’Brien, S.J. (2001) Patterns of size homoplasy at 10 microsatellite loci in Pumas (Puma concolor). Molecular Biology and Evolution, 18, 1151-1156.

Currat, M., Ray, N. & Excoffier, L. (2004) SPLATCHE: a program to simulate genetic diversity taking into account environmental heterogeneity. Molecular Ecology Notes, 4, 139-142.

Cushman, S.A., Chase, M. & Curtice, G. (2010a) Mapping landscape resistance to identify corridors and barriers for elephant movement in southern Africa. In Spatial complexity, informatics, and wildlife conservation., (Eds, Cushman, S.A. & Huettmann, F.) Springer, New York, pp. 349-367.

Cushman, S.A. & Landguth, E.L. (2010) Spurious correlations and inference in landscape genetics. Molecular Ecology, 19, 3592-3602.

Cushman, S.A., McKelvey, K.S., Hayden, J. & Schwartz, M.K. (2006) Gene flow in complex landscapes: testing multiple hypotheses with causal modeling. The American Naturalist, 168, 486-499.

Cushman, S.A., Compton, B.W. & McGarigal, K. (2010b) Habitat fragmentation effects depend on complex interactions between population size and dispersal ability: Modeling influences of roads, agriculture and residential development across a range of life-history characteristics. In Spatial complexity, informatics, and wildlife conservation., (Eds, Cushman, S.A. & Huettmann, Falk) Springer, New York, pp. 369-385.

Cushman, S.A., Evans, J.S. & McGarigal, K. (2010c) Landscape ecology: Past, present, and future. In Spatial complexity, informatics, and wildlife conservation., (Eds, Cushman, S.A. & Huettmann, F.) Springer, New York, pp. 65-82.

Cushman, S.A., Gutzweiler, K., Evans, J.S. & McGarigal, K. (2010d) The Gradient Paradigm: A conceptual and analytical framework for landscape ecology. In Spatial complexity, informatics, and wildlife conservation., (Ed, Cushman, S.A.H., Falk) Springer, New York, pp. 83-108.

D’Eon, R.G., Glenn, S.M., Parfitt, I. & Fortin, M. (2002) Landscape connectivity as a function of scale and organism vagility in a real forested landscape. Conservation Ecology, 6, 10.

Dakin, E.E. & Avise, J.C. (2004) Microsatellite null alleles in parentage analysis. Heredity, 93, 504-509.

Dallas, J.F., Marshall, F., Piertney, S.B., Bacon, P.J. & Racey, P.A. (2002) Spatially restricted gene flow and reduced microsatellite polymorphism in the Eurasian otter Lutra lutra in Britain. Conservation Genetics, 3, 15-29.

Dalton, R. (2006) Doing conservation by numbers. Nature, 442, 12.

Dawson, K.J. & Belkhir, K. (2001) A bayesian approach to the identification of panmictic

- 175 - populations and the assignment of individuals. Genetic Research Cambridge, 78, 59-77. de Angelo, C., Paviolo, A. & di Bitetti, M. (2011) Differential impact of landscape transformation on pumas (Puma concolor) and jaguars (Panthera onca) in the Upper Paraná Atlantic Forest. Diversity and Distributions, 1-15.

Degen, B., Streiff, R. & Ziegenhagen, B. (1999) Comparative study of genetic variation and differentiation of two pedunculate oak (Quercus robur) stands using microsatellite and allozyme loci. Heredity, 83, 597-603.

Dinerstein, E., Loucks, C., Heydlauff, A., Wikramanayake, E., Bryja, G., Forrest, J., Ginsberg, J., Klenzendorf, S., Leimgruber, P., O’Brien, T., Sanderson, E., Seidensticker, J. & Songer, M. (2006) Setting Priorities for the Conservation and Recovery of Wild Tigers: 2005 - 2015. A User’s Guide. WWF, WCS, Smithsonian, and NFWF-STF,

Diniz Filho, J.A.F. & De Campos Telles, M.P. (2002) Spatial autocorrelation analysis and the identification of operational units for conservation in continuous populations. Conservation Biology, 16, 924-935.

Dixon, J.D., Oli, M.K., Wooten, M.C., Eason, T.H., McCown, J.W. & Paetkau, D. (2006) Effectiveness of a regional corridor in connecting two Florida black bear populations. Conservation Biology, 20, 155-162.

Donovan, T.M., Lamberson, R.H., Kimber, A., Thompson III, F.R. & Faaborg, J. (1995) Modeling the effects of habitat fragmentation on source and sink demography of neotropical migrant birds. Conservation Biology, 9, 1396-1407.

Driscoll, C., Yamaguchi, N., Bar-Gal, G.K., Roca, A.L., Luo, S.-J., Macdonald, D.W. & O’Brien, S.J. (2009) Mitochondrial phylogeography illuminates the origin of the extinct Caspian tiger and its relationship to the Amur tiger. PloS One, 4, e4125.

Driscoll, D.A. (1999) Genetic neighbourhood and effective population size for two endangered frogs. Biological Conservation, 88, 221-229.

Dunning, J.B., Danielson, B.J., Noon, B.R., Root, T.L., Lamberson, R.H. & Stevens, E.E. (1995) Spatially explicit population models: current forms and future uses. Ecological Applications, 5, 3-11.

Durand, E., Chen, C. & François, O. (2009) Comment on ‘On the inference of spatial structure from population genetics data’. Bioinformatics, 25, 1802-1804.

Dyer, R.J. & Sork, V.L. (2001) Pollen pool heterogeneity in shortleaf pine, Pinus echinata Mill. Molecular Ecology, 10, 859-866.

Earl, D.A. (2009) Structure Harvester v0.3. BME Department, UCSC. http:/ /taylor0.biology.ucla.edu/struct_harvest/

Earl, D.A. & von Holdt, B.M. (2011) STRUCTURE HARVESTER: a website and program for visualizing STRUCTURE output and implementing the Evanno method. doi: 10.1007/ s12686-011-9548-7

- 176 - Eggert, L.S., Eggert, J.A. & Woodruff, D.S. (2003) Estimating population sizes for elusive animals: the forst elephants of Kakum National Park, Ghana. Molecular Ecology, 12, 1389-1402.

Eizirik, E., Jae-Heup, K., Menotti Raymond, M., Crawshaw, P.G., O’Brien, S.J. & Johnson, W.E. (2001) Phylogeography, population history and conservation genetics of jaguar (Panthera onca, Mammalia, Felidae). Molecular Ecology, 10, 65-79.

El Mousadik, A. & Petit, R.J. (1996) High level of genetic differentiation for allelic richness among populations of the argan tree [Argania spinosa (L.) Skeels] endemic to Morocco. Theoretical and Applied Genetics, 92, 832-839.

England, P.R., Cornuet, J.-M., Berthier, P., Tallmon, D.A. & Luikart, G. (2006) Estimating effective population size from linkage disequilibrium: severe bias in small samples. Conservation Genetics, 7, 303-308.

Epperson, B.K. (1990) Spatial autocorrelation of genotypes under directional selection. Genetics, 124, 757-771.

Epperson, B.K. (2005) Estimating dispersal from short distance spatial autocorrelation. Heredity, 95, 7-15.

Epps, C.W., Wehausen, J.D., Bleich, V.C., Torres, S.G. & Brashares, J.S. (2007) Optimizing dispersal and corridor models using landscape genetics. Journal of Applied Ecology, 44, 714-724.

Ernest, H.B., Boyce, W.M., Bleich, V.C., May, B., Stiver, S.J. & Torres, S.G. (2003) Genetic structure of mountain lion (Puma concolor) populations in California. Conservation Genetics, 4, 353-366.

Ernest, H.B., Penedo, M.C.T., May, B.P., Syvanen, M. & Boyce, W.M. (2000) Molecular tracking of mountain lions in the Yosemite Valley region in California: genetic analysis using microsatellites and faecal DNA. Molecular Ecology, 9, 433-441.

Evanno, G., Regnaut, S. & Goudet, J. (2005) Detecting the number of clusters of individuals using the software STRUCTURE: a simulation study. Molecular Ecology, 14, 2611-2620.

Evans, S.R. & Sheldon, B.C. (2008) Interspecific patterns of genetic diversity in birds: correlations with extinction risk. Conservation Biology, 22, 1016-1025.

Excoffier, L. (2007) Analysis of population subdivision. In Handbook of Statistical Genetics, (Eds, Balding, D.J., Bishop, M. & Cannings, C.) John Wiley & Sons Ltd, Chicester, UK, pp. 980.

Excoffier, L., Laval, G. & Schneider, S. (2005) ARLEQUIN (version 3.0): An integrated software package for population genetics data analysis. Evolutionary Bioinformatics Online, 1, 47-50.

Excoffier, L., Smouse, P.E. & Quattro, J.M. (1992) Analysis of molecular variance inferred from metric distances among DNA haplotypes: Application to human mitochondrial DNA

- 177 - restriction data. Genetics, 131, 479-491.

Fahrig, L. (2003) Effects of habitat fragmentation on biodiversity. Annual Review of Ecology and Systematics, 34, 487-515.

Fahrig, L. (2007) Landscape heterogeneity and metapopulation dynamics. In Key Topics and Perspectives in Landscape Ecology., Cambridge University Press, Cambridge, pp. 78-89.

Fahrig, L. & Merriam, G. (1985) Habitat patch connectivity and population survival. Ecology, 66, 1762-1768.

Fahrig, L. & Merriam, G. (1994) Conservation of fragmented populations. Conservation Biology, 8, 50-59.

Falconer, D.S. & MacKay, T.F.C. (1996) Introduction to quantitative genetics. Longman Group Ltd,

Farrell, L.E., Roman, J. & Sunquist, M. (2000) Dietary separation of sympatric carnivores identified by molecular analysis of scats. Molecular Ecology, 9, 1583-1590.

Faubet, P. & Gaggiotti, O.E. (2008) A new Bayesian method to identify environmental factors that influence recent migration. Genetics, 178, 1491-1504.

Faust, T. & Tilson, R. (1994) Estimating how many tigers are in Sumatra: A beginning. In Sumtran tiger population and habitat viability analysis report, (Eds, Tilson, R., Soemarna, K., Ramono, W., Lusli, S., Traylor-Holzer, K. & Seal, U.) IUCN/SSC Conservation Breeding Specialist Group, pp. 11-38.

Fernando, P., Vidya, T.N.C., Rajapakse, C., Dangolla, A. & Melnick, D.J. (2003) Reliable noninvasive genotyping: fantasy or reality? Journal of Heredity, 94, 115-123.

Fischer, J. & Lindenmayer, D.B. (2007) Landscape modificiation and habitat fragmentation: a synthesis. Global Ecology and Biogeography, 16, 265-280.

Fitzpatrick, B.M. (2009) Power and sample size for nested analysis of molecular variance. Molecular Ecology, 18, 3961-3955.

Fitzpatrick, K.A., Kersh, G.J. & Massung, R.F. (2010) Practical method for extraction of PCR-quality DNA from environmental soil samples. Applied and Environmental Microbiology, 76, 4571-4573.

Fogelqvist, J., Niittyvuopio, A., Ågren, J., Savolainen, O. & Lascoux, M. (2010) Cryptic population genetic structure: the number of inferred clusters depends on sample size. Molecular Ecology Resources, 10, 314-323.

Foose, T.J. (1987) Species Survival Plans and Overall Management Strategies. In Tigers of the World: The Biology, Biopolitics, Management, and Conservation of an Endangered Species., (Eds, Tilson, R.L. & Seal, U.S.) Noyes Publications, Park Ridge, New Jersey, USA, pp. 304-316.

- 178 - Foppen, R.P.B., Chardon, J.P. & Liefveld, W. (2000) Understanding the role of sink patches in source-sink metapopulations: Reed Warbler in an agricultural landscape. Conservation Biology, 14,

Forbes, B.A. & Hicks, K.E. (1996) Substances interfering with direct detection of Mycobacterium tuberculosis in clinical specimens by PCR: effects of bovine serum albumin. Journal of Clinical Microbiology, 34, 2125-2128.

Foucault, F., Praz, F., Jaulin, C. & Amor-Gueret, M. (1996) Experimental limits of PCR analysis of (CA)a repeat alterations. Trends in Genetics, 12, 450-452.

Foulley, J. & Ollivier, L. (2006) Estimating allelic richness and its diversity. Livestock Science, 101, 150-158.

François, O., Ancelet, S. & Guillot, G. (2006) Bayesian clustering using hidden Markov random fields in spatial population genetics. Genetics, 174, 805-816.

Frankham, R. (1995) Effective population size/adult population ratios in wildlife: a review. Genetics Research Cambridge, 66, 95-107.

Frankham, R. (1998) Inbreeding and extinction: Island populations. Conservation Biology, 12, 665-675.

Frankham, R., Ballou, J.D. & Briscoe, D.A. (2002) Introduction to Conservation Genetics. Cambridge University Press, Cambridge.

Frankham, R. & Ralls, K. (1998) Inbreeding leads to extinction. Nature, 392, 441-442.

Franklin, N. (2002) Conservation biology of the Sumatran tiger in Way Kambas National Park, Sumatra, Indonesia. PhD thesis, University of York.

Frantz, A.C., Bertouille, S., Eloy, M.C., Licoppe, A., Chaumont, F. & Flamand, M.C. (2012) Comparative landscape genetic analyses show a Belgian motorway to be a gene flow barrier for red deer (Cervus elaphus), but not wild boars (Sus scrofa). doi: 10.1111/ j.1365-294X.2012.05623.x

Frantz, A.C., Pope, L.C., Etherington, T.R., Wilson, G.J. & Burke, T. (2010) Using isolation- by-distance-based approaches to assess the barrier effect of linear landscape elements on badger (Meles meles) dispersal. Molecular Ecology, 19, 1663-1674.

Frantz, A.C. & Roper, T.J. (2006) Simulations to assess the performance of different rarefaction methods in estimating population size using small datasets. Conservation Genetics, 7, 315-318.

Frantz, A.C., Cellina, S., Krier, A., Schley, L. & Burke, T. (2009) Using spatial Bayesian methods to determine the genetic structure of a continuously distributed population: clusters or isolation by distance? Journal of Applied Ecology, 46, 493-505.

Frantz, A.C., Pope, L.C., Carpenter, P.J., Roper, T.J., Wilson, G.J., Delahay, R.J. & Burke, T. (2003) Reliable microsatellite genotyping of the Eurasian badger (Meles meles) using

- 179 - faecal DNA. Molecular Ecology, 12, 1649-1661.

Frantzen, M.A.J., Silk, J.B., Ferguson, J.W.H., Wayne, R.K. & Kohn, M.H. (1998) Empirical evaluation of preservation methods for faecal DNA. Molecular Ecology, 7, 1423-1428.

Futuyma, D. (1998) Evolutionary biology. Sinauer Associates Inc., Sunderland, Massachusetts.

FWI/GFW (2002) The State of the Forest: Indonesia. Forest Watch Indonesia & Global Forest Watch,

Gaggiotti, O. (2004) Multilocus genotypes methods for the study of metapopulation processes. In Ecology, Genetics, and Evolution of Metapopulations., (Eds, Hanski, I. & Gaggiotti, O.E.) Elsevier Academic Press, pp. 367-386.

Gaggiotti, O.E. & Hanski, I. (2004) Mechanisms of population extinction. In Ecology, Genetics, and Evolution of Metapopulations., (Eds, Hanski, I. & Gaggiotti, O.E.) Elsevier Academic Press, pp. 337-366.

Gaggiotti, O.E., Lange, O., Rassmann, K. & Gliddons, C. (1999) A comparison of two indirect methods for estimating average levels of gene flow using microsatellite data. Molecular Ecology, 8, 1513-1520.

Gagneux, P., Boesch, C. & Woodruff, D.S. (1997) Microsatellite scoring errors associated with noninvasive genotyping based on nuclear DNA amplified from shed hair. Molecular Ecology, 6, 861-868.

Galpern, P., Manseau, M., Hettinga, P., Smith, K. & Wilson, P. (2012) ALLELEMATCH: an R package for identifying unique multilocus genotypes where genotyping error and missing data may be present. Molecular Ecology Resources, 12, 771-778.

Garza, J.C. & Williamson, E.G. (2001) Detection of reduction in population size using data from microsatellite loci. Molecular Ecology, 10, 305-318.

Gaveau, D.L.A., Epting, J., Adnan, B., Kumara, I. & Suyikno, B. (2007) Deforestation map (1990-2006) of northern Sumatra at 150,000 scale.

Gaveau, D.L.A., Epting, J., Lyne, O., Linkie, M., Kumara, I., Kanninen, M. & Leader Williams, N. (2009a) Evaluating whether protected areas reduce tropical deforestation in Sumatra. Journal of Biogeography, 36, 2165-2175.

Gaveau, D.L.A., Linkie, M., Suyadi, Levang, P. & Leader-Williams, N. (2009b) Three decades of deforestation in southwest Sumatra: Effects of coffee prices, law enforcement and rural poverty. Biological Conservation, 142, 597-605.

Geffen, E., Anderson, M.J. & Wayne, R.K. (2004) Climate and habitat barriers to dispersal in highly mobile grey wolf. Molecular Ecology, 13, 2481-2490.

Geist, H.J. & Lambin, E.F. (2002) Proximate causes and underlying driving forces of tropical deforestation. BioScience, 52, 143-150.

- 180 - Ginot, F., Bordelais, I., Nguyen, S. & Gyapay, G. (1996) Correction of some genotyping errors in automated fluorescent microsatellite analysis by enzymatic removal of one base overhangs. Nucleic Acids Research, 24, 540-541.

Girod, C., Vitalis, R., Leblois, R. & Fréville, H. (2011) Inferring population decline and expansion from microsatellite data: a simulation-based evaluation of the MSVAR method. Genetics, 188, 165-179.

Glaubitz, J.C. (2004) CONVERT: A user-friendly program to reformat diploid genotypic data for commonly used population genetic software packages. Molecular Ecology Notes, 4, 309-310.

Glaubitz, J.C., Murrell, J.C. & Moran, G.F. (2003) Effects of native forest regeneration practices on genetic diversity in Eucalyptus consideniana. Theoretical and Applied Genetics, 107, 422-431.

Goodman, S.J. (1997) Rst Calc: a collection of computer programs for calculating estimates of genetic differentiation from microsatellite data and determining their significance. Molecular Ecology, 6, 881-885.

Goodrich, J.M., Miquelle, D.G., Smirnov, E.N., Kerley, L.L., Quigley, H.B. & Hornocker, M.G. (2010) Spatial structure of Amur (Siberian) tigers (Panthera tigris altaica) on Sikhote-Alin Biosphere Zapovednik, Russia. Journal of Mammalogy, 91, 737-748.

Goossens, B., Chikhi, L., Jalil, M.F., Ancrenaz, M., Lackman-Ancrenaz, Mohamed, M., Andau, P. & Bruford, M.W. (2005) Patterns of genetic diversity and migration in increasingly fragmented and declining orang-utan (Pongo pygmaeus) populations from Sabah, Malaysia. Molecular Ecology, 14, 441-456.

Goossens, B., Chikhi, L., Utami, S.S., de Ruiter, J. & Bruford, M.W. (2000) A multi-samples, multi-extracts approach for microsatellite analysis of faecal samples in an arboreal ape. Conservation Genetics, 1, 157-162.

GTRP (2011) Global Tiger Recovery Program 2010-2022. Global Tiger Initiative Secretariat,

Guillot, G., Mortier, F. & Estoup, A. (2005) GENELAND: a computer package for landscape genetics. Molecular Ecology Notes, 5, 712-715.

Guo, S.W. & Thompson, E.A. (1992) Performing the exact test of Hardy-Weinberg proportions for multiple alleles. Biometrics, 48, 361-372.

Haag, T., Santos, A.S., Sana, D.A., Morato, R.G., Cullen Jr, L., Crawshaw Jr, P.G., de Angelo, C., di Bitetti, M.S., Salzano, F.M. & Eizirik, E. (2010) The effect of habitat fragmentation on the genetic structure of a top predator: loss of diversity and high differentiation among remannt populations of Atlantic Forest jaguars (Panthera onca). doi: 10.1111/j.1365-294X.2010.04856.x

Haddad, N.M. (1999) Corridor use predicted from behaviors at habitat boundaries. The American Naturalist, 153, 215-227.

- 181 - Hájková, P., Zemanová, B., Bryja, J., Hájek, B., Roche, K., Tkadlec, E. & Zima, J. (2006) Factors affecting success of PCR amplification of microsatellite loci from otter faeces. Molecular Ecology Notes, 6, 559-562.

Hale, K.A. & Briskie, J.V. (2007) Decreased immunocompetence in a severely bottlenecked population of an endemic New Zealand bird. Animal Conservation, 10, 2-10.

Hamilton, M. (2009) Population Genetics. Wiley-Blackwell,

Han, K., Kim, K. & Park, T. (2010) Unbalanced sample size effect on the genome-wide population differentiation studies. 2010 IEEE International Conference on Bioinformatics and Biomedicine Workshops, 347-352.

Hansen, H., Ben-David, M. & McDonald, D.B. (2008) Effects of genotyping protocols on success and errors in identifying individual river otters (Lontra canadensis) from their faeces. Molecular Ecology Resources, 8, 282-289.

Hansen, M.C., Stehman, S.V., Potapov, P.V., Arunarwati, B., Stolle, F. & Pittman, K. (2009) Quantifying changes in the rates of forest clearing in Indonesia from 1990 to 2005 using remotely sensed data sets. Environmental Research Letters, 4, 034001.

Hansen, M.M. & Jacobsen, L. (1999) Identification of mustelid species: otter (Lutra lutra), American mink (Mustela vison) and polecat (Mustela putorius), by analysis of DNA from faecal samples. Journal of Zoology (London), 247, 177-181.

Hanski, I. (1991) Single-species metapopulation dynamics: concepts, models and observations. Biological Journal of the Linnean Society, 42, 17-38.

Hanski, I. (1994) A practical model of metapopulation dynamics. Journal of Animal Ecology, 63, 151-162.

Hanski, I. (1998) Metapopulation dynamics. Nature, 396, 41-49.

Hanski, I. & Ovaskainen, O. (2003) Metapopulation theory for fragmented landscapes. Theoretical Population Biology, 64, 119-127.

Hardy, O.J., Charbonnel, N., Fréville, H. & Heuertz, M. (2003) Microsatellite allele sizes: a simple test to assess their significance on genetic differentiation. Genetics, 163, 1467-1482.

Hardy, O.J. & Vekemans, X. (1999) Isolation by distance in a continuous population: reconciliation between spatial autocorrelation analysis and population genetic models. Heredity, 83, 145-154.

Hardy, O.J. & Vekemans, X. (2002) SPAGeDi: a versatile computer program to analyse spatial genetic structure at the individual or population levels. Molecular Ecology Notes, 2002, 618-620.

Harimaukita (2011) Harimaukita minta Menhut tolak jalan tembus TN Kerinci Seblat. Forum Harimaukita.

- 182 - Wibisono, H.T.,Matthew Linkie,Gurutzeta Guillera-Arroita,Joseph A. Smith,Sunarto, Pusparini, W., Asriadi, Baroto, P., Brickle, N., Dinata, Y., Gemita, E.,Donny, Gunaryadi, Haidir, I.A., Herwansyah, Karina, I., Kiswayadi, D.,Decki, Kristiantono, Kurniawan, H., Lahoz-Monfort, J.J., Leader-Williams, N., Maddox, T., Martyr, D.J., Maryati, Nugroho, A., Parakkasi, K., Priatna, D.,Eka, Ramadiyanta, Ramono, W.S., Reddy, G.V., Rood, E.J.J., Saputra, D.Y., Sarimudi, A., Salampessy, A., Septayuda, E., Suhartono, T., Ade Sumantri1, Susilo, Tanjung, I., Tarmizi, Yulianto, K., Yunus, M. & Zulfahmi (2011) Population status of a cryptic top predator: an island-wide assessment of tigers in Sumatran rainforests. PloS One, 6, e25931.

Hauge, X.Y. & Litt, M. (1993) A study of the origin of ‘shadow bands’ seen why typing dinucleotide repeat polymorphisms by the PCR. Human Molecular Genetics, 2, 411-415.

Hedmark, E. & Ellegren, H. (2006) A test of the multiplex pre-amplification approach in microsatellite genotyping of wolverine faecal DNA. Conservation Genetics, 7, 289-293.

Hedrick, P.W. (2005a) A standardized genetic differentiation measure. Evolution, 59, 1633-1638.

Hedrick, P.W. (2000) Genetics of populations. Jones & Bartlett Publishers,

Hedrick, P.W. (2005b) Genetics of populations. Jones & Bartlett Publishers,

Hendrickson, S.L., Mayer, G.C., Wallen, E.P. & Quigley, K. (2000) Genetic variability and geographic structure of three subspecies of tigers (Panthera tigris) based on MCH class I variation. Animal Conservation, 3, 135-143.

Henry, P., Miquelle, D., Sugimoto, T., McCullough, D.R., Caccone, A. & Russello, M.A. (2009) In situ population structure and ex situ representation of the endangered Amur tiger. doi: 10.1111/j.1365-294X.2009.04266.x

Hoehn, M., Sarre, S.D. & Henle, K. (2007) The tales of two geckos: does dispersal prevent extinction in recently fragmented populations? Molecular Ecology, 16, 3299-3312.

Hoffman, J.I. & Amos, W. (2005) Microsatellite genotyping errors: detection approaches, common sources and consequences for paternal exclusion. Molecular Ecology, 14, 599-612.

Holderegger, R. & Wagner, H.H. (2008) Landscape genetics. BioScience, 58, 199-207.

Holleley, C.E. & Geerts, P.G. (2009) Multiplex Manager 1.0: a cross-platform computer program that plans and optimizes multiplex PCR. BioTechniques, 46, 511-517.

Holm, L., Loeschcke, V. & Bendixen, C. (2001) Elucidation of the molecular basis of a null allele in a rainbow trout microsatellite. Marine Biotechnology, 3, 555-560.

Holzhauer, S.I.J., Ekschmitt, K., Sander, A., Dauber, J. & Wolters, V. (2006) Effect of historic landscape change on the genetic structure of the bush-cricket Metrioptera roeseli. Landscape Ecology, 21, 891-899.

Höss, M. & Pääbo, S. (1993) DNA extraction from Pleistocene bones by a silica-based

- 183 - purification method. Nucleic Acids Research, 21, 3913-3914.

Hubisz, M.J., Falush, D., Stephens, M. & Pritchard, J.K. (2009) Inferring weak population structure with the assistance of sample group information. Molecular Ecology Resources, 9, 1322-1332.

Hurt, A., Davenport, B. & Greene, E. (2000) Training dogs to distinguish between black bear (Ursus americanus) and grizzly bear (Ursus arctos) faeces. UM Biology Undergraduate Journal, 1,

Idury, R.M. & Cardon, L.R. (1997) A simple method for automated allele binning in microsatellite markers. Genome Research, 7, 1104-1109.

Imbernon, J. (1999) Changes in agricultural practice and landscape over a 60-year period in North Lampung, Sumatra. Agriculture, Ecosystems and Environment, 76, 61-66.

Imron, M.A., Herzog, S. & Berger, U. (2011) The influence of agroforestry and other land- use types on the persistence of a Sumatran tiger (Panthera tigris sumatrae) population: an individual-based model approach. Environmental Management, 48, 276-288.

Inskip, C. & Zimmermann, A. (2009) Human-felid conflict: a review of patterns and priorities worldwide. Oryx, 43, 18-34.

Iyengar, A., Babu, V.N., Hedge, S., Venkataraman, A.B., Maclean, N. & Morin, P.A. (2005) Phylogeography, genetic structure, and diversity in the dhole (Cuon alpinus). Molecular Ecology, 14, 2281-2297.

Jakobsson, M. & Rosenberg, N.A. (2007) CLUMPP: a cluster matching and permutation program for dealing with label switching and multimodality in analysis of population structure. Bioinformatics, 23, 1801-1806.

Jane&ka, J.E., Jackson, R., Yuquang, Z., Diqiang, L., Munkhtsog, B., Buckley-Beason, V. & Murphy, W.J. (2008) Population monitoring of snow leopards using noninvasive collection of scat samples: a pilot study. Animal Conservation, 11, 401-411.

Jenness, J. (2010) Export to Circuitscape Tool for ArcGIS 9.x (v.1.0.87). http:/ /www.jennessent.com/arcgis/Circuitscape_Exp.htm

Johansson, Å., Karlsson, P. & Gyllensten, U. (2003) A novel method for automatic genotyping of microsatellite markers based on parametric pattern recognition. Human Genetics, 113, 316-324.

Johnson, A., Vongkhamheng, C., Hedemark, M. & Saithongdam, T. (2006) Effects of human-carnivore conflict on tiger (Pantera tigris) and prey populations in Lao PDR. Animal Conservation, 9, 421-430.

Johnson, P.C.D. & Haydon, D.T. (2007a) Maximum-likelihood estimation of allelic dropout an false allele error rates from microsatellite genotypes in the absence of reference data. Genetics, 175, 827-842.

- 184 - Johnson, P.C.D. & Haydon, D.T. (2007b) Software for quantifying and simulating microsatellite genotyping error. Bioinformatics & Biology Insights, 1, 71-75.

Jones, O. & Wang, J. (2009) COLONY: a program for parentage and sibship inference from multilocus genotype data. Molecular Ecology Resources, 10, 551-555.

Joshi, A. (2010) Genetic structure and connectivity of tiger (Panthera tigris tigris) populations in central Indian forests. Master of Science thesis, Tata Institute of Fundamental Research.

Jost, L. (2008) Gst and its relatives do not measure differentiation. Molecular Ecology, 17, 4015-4026.

Kaeuffer, R., Pontier, D., Devillard, S. & Perrin, N. (2004) Effective size of two feral domestic cat populations (Felis catus L.): effect of the mating system. Molecular Ecology, 13, 483-490.

Kalinowski, S.T. (2002) How many alleles per locus should be used to estimate genetic distances? Heredity, 88, 62-65.

Kalinowski, S.T. (2004) Counting alleles with rarefaction: private alleles and hierarchical sampling designs. Conservation Genetics, 5, 539-543.

Kalinowski, S.T. (2005) Do polymorphic loci require large sample sizes to estimate genetic distances? Heredity, 94, 33-36.

Kalinowski, S.T. (2011) The computer program STRUCTURE does not reliably identify the main genetic clusters within species: simulations and implications for human population structure. Heredity, 106, 625-632.

Kalinowski, S.T., Taper, M.L. & Creel, S. (2006) Using DNA from non-invasive samples to identify individuals and census populations: an evidential approach tolerant of genotyping errors. Conservation Genetics, 7, 319-329.

Kang, S.J., Gordon, D., Brown, A.M., Ott, J. & Finch, S.J. (2004) Tradeoff between no-call reduction in genotyping error rate and loss of sample size for genetic case/control association studies. Paciﬁc Symposium on Biocomputing, 9, 116-127.

Karanth, K.U. & Stith, B.M. (1999) Prey depletion as a critical determinant of tiger population viability. In Riding the Tiger: Tiger conservation in human-dominated landscapes., (Eds, Seidensticker, J., Christie, S. & Jackson, P.) Cambridge University Press, Cambridge, pp. 100-113.

Karanth, U.K., Nichols, J.D., Kumar, N.S., Link, W.A. & Hines, J.E. (2004) Tigers and their prey: Predicting carnivore densities from prey abundance. Proceedings of the National Academy of Sciences of the United States of America, 101, 4854-4858.

Kautz, R.,Kawula R.,Hoctor T.,Comiskey J.,Jansen D.,Jennings D.,Kasbohm J.,Mazzotti F.,McBride R.,Richardson L., K., R. (2006) How much is enough? Landscape-scale conservation for the Florida panther. Biological Conservation, 130, 118-133.

Keller, L.F. & Waller, D.M. (2002) Inbreeding effects in wild populations. Trends in Ecology

- 185 - and Evolution, 17, 230-241.

Kelly, A.C., Mateus-Pinilla, N.E., Douglas, M., Shelton, P. & Novakofski, J. (2011) Microsatellites behaving badly: empirical evaluation of genotyping errors and subsequent impacts on population studies. Genetics and Molecular Research, 10, 2534-2553.

Kelly, M.A. (2001) Lineage loss in Serengeti cheetahs: consequences of high reproductive variance and heritability of fitness on effective population size. Conservation Biology, 15, 137-147.

Kemp, B.M., Monroe, C. & Smith, D.G. (2006) Repeat silica extraction: a simple technique for the removal of PCR inhibitors from DNA extracts. Journal of Archaeological Science, 33, 1680-1689.

Kenney, J.S., Smith, J.L.D., Starfield, A.M. & McDougal, C.W. (1995) The long-term effects of tiger poaching on population viability. Conservation Biology, 9, 1127-1133.

Kerley, L. (2002) A final report to Save the Tiger Fund: Scent dog monitoring of Amur tigers- II (2002-0301-016) March 1, 2002 - March 1, 2003.

Kerley, L. & Borisenko, M.I. (2007) A final report to Wildlife Conservation Society, June 2007: Using scat detection dogs to collect Amur leopard and tiger scats for comparative analysis.

Kerley, L.L., Goodrich, J.M., Miquelle, D.G., Smirnov, E.N., Quigley, H.B. & Hornocker, M.G. (2002) Effects of roads and human disturbance on Amur tigers. Conservation Biology, 16, 97-108.

Keyghobadi, N., Roland, J. & Strobeck, C. (1999) Influence of landscape on the population genetic structure of the alpine butterfly Parnassius smintheus (Papilionidae). Molecular Ecology, 8, 1481-1495.

King, C.E., Debruyne, R., Kuch, M., Schwarz, C. & Poinar, H.N. (2009) A quantitative approach to detect and overcome PCR inhibition in ancient DNA extracts. BioTechniques, 47, 941-949.

Kinnaird, M.F., O’Brien, T.G., Wibisono, H.T., Sanderson, E.W. & Woolmer, G. (2003) Deforestation trends in a tropical landscape and implications for endangered large mammals. Conservation Biology, 17, 245-257.

Kishore, R., Hardy, W.R., Anderson, V.J., Sanchez, N.A. & Buoncristiani, M.R. (2006) Optimization of DNA extraction from low-yield and degraded samples using the BioRobot® EZ1 and BioRobot® M48. Journal of Forensic Science, 51, 1-7.

Kitchener, A.C. (1999) Tiger distribution, phenotypic variation and conservation issues. In Riding the Tiger: Tiger conservation in human-dominated landscapes., Cambridge University Press, Cambridge, pp. 19-39.

Kitchener, A.C. & Dugmore, A.J. (2000) Biogeographical change in the tiger, Panthera tigris. Animal Conservation, 3, 113-124.

- 186 - Kitchener, A.C. & Yamaguchi, N. (2010) What is a tiger? Biogeography, morphology, and taxonomy. In Tigers of the world: The science, politics, and conservation of Panthera tigris, (Eds, Tilson, R. & Nyhus, P.J.) Academic Press, London, pp. 53-84.

Knapp, S.M., Craig, B.A. & Waits, L.P. (2009) Incorporating genotyping error into noninvasive DNA-based mark-recapture population estimates. Journal of Wildlife Management, 73, 598-604.

Koen, E.L., Bowman, J. & Walpole, A.A. (2012) The effect of cost surface parameterization on landscape resistance estimates.

Koh, L.P., Miettinen, J., Liew, S.C. & Ghazoul, J. (2011) Remotely sensed evidence of tropical peatland conversion to oil palm. Proceedings of the National Academy of Sciences USA, 108, 5127-5132.

Kohn, M.H., Murphy, W.J., Ostrander, E.A. & Wayne, R.K. (2006) Genomics and conservation genetics. TRENDS in Ecology and Evolution, 21, 629-637.

Kohn, M.H., York, E.C., Kamradt, D.A., Haught, G., Sauvajot, R.M. & Wayne, R.K. (1999) Estimating population size by genotyping faeces. Proceedings of the Royal Society of London. Series B, 266, 657-663.

Kreader, C.A. (1996) Relief of Amplification Inhibition in PCR with Bovine Serum Albumin or T4 Gene 32 Protein. Applied and Environmental Microbiology, 62, 1102-1106.

Kurose, N., Masuda, R. & Tatara, M. (2005) Fecal DNA analysis for identifying species and sex of sympatric carnivores: A noninvasive method for conservation on the Tsushima Islands, Japan. Journal of Heredity, 96, 688-697.

Lampa, S., Gruber, B., Henle, K. & Hoehn, M. (2008) An optimisation approach to increase DNA amplification success of otter faeces. Conservation Genetics, 9, 201-210.

Landguth, E.L. & Cushman, S.A. (2010) CDPOP: a spatially explicit cost distance population genetics program. Molecular Ecology Resources, 10, 156-161.

Landguth, E.L., Cushman, S.A., Schwartz, M.K., McKelvey, K.S., Murphy, M. & Luikart, G. (2010) Quantifying the lag time to detect barriers in landscape genetics. Molecular Ecology, 19, 4179-4191.

Latch, E.K., Dharmarajan, G., Glaubitz, J.C. & Rhodes Jr, O.E. (2006) Relative performance of Bayesian clustering software for inferring population substructure and individual assignment at low levels of population differentiation. Conservation Genetics, 7, 295-302.

Latch, E.K., Scognamillo, D.G., Fike, J.A., Chamberlain, M.J. & Rhodes Jr, O.E. (2008) Deciphering Ecological Barriers to North American River Otter (Lontra canadensis) Gene Flow in the Louisiana Landscape. Journal of Heredity, 99, 265-274.

Latham, J. (2005) Landcover of Indonesia. FAO GeoNetwork. http://www.fao.org/ geonetwork/srv/en/metadata.show?id=37156&currTab=simple

- 187 - Leberg, P.L. (1992) Effects of population bottlenecks on genetic diversity as measured by allozyme electrophoresis. Evolution, 46, 477-494.

Leberg, P.L. (2002) Estimating allelic richness: Effects of sample size and bottlenecks. Molecular Ecology, 11, 2445-2449.

Leberg, P. (2005) Genetic approaches for estimating the effective size of populations. Journal of Wildlife Management, 69, 1385-1399.

Legendre, P. & Fortin, M. (2010) Comparison of the Mantel test and alternative approaches for detecting complex multivariate relationships in the spatial analysis of genetic data. Molecular Ecology Resources, 10, 831-844.

Legendre, P. & Troussellier, M. (1988) Aquatic heterotropic bacteria: Modeling in the presence of spatial autocorrelation. Limnology & Oceanography, 33, 1055-1067.

León-Cortés, J.L., Lennon, J.J. & Thomas, C.D. (2003) Ecological dynamics of extinct species in empty habitat networks. 1. The role of habitat pattern and quantity, stochasticity and dispersal. Oikos, 102, 449-464.

Linkie, M., Borysiewicz, R.S., Dinata, Y., Nugroho, A., Achmad Haidir, I., Ridout, M.S., Morgan, B.J.T. & Leader-Williams, N. (2008a) Predicting the spatio-temporal patterns of tiger and their prey across a primary-distrubed forest landscape in Sumatra.

Linkie, M., Haidir, I.A., Nugroho, A. & Dinata, Y. (2008b) Conserving tigers Panthera tigris in selectively logged Sumatran forests. Biological Conservation, 141, 2410-2415.

Linkie, M., Rood, E. & Smith, R.J. (2010) Modelling the effectiveness of enforcement strategies for avoiding tropical deforestation in Kerinci Seblat National Park, Sumatra. Biodiversity and Conservation, 19, 973-984.

Linkie, M., Smith, R.J. & Leader-Williams, N. (2004) Mapping and predicting deforestation patterns in the lowlands of Sumatra. Biodiversity and Conservation, 13, 1809-1818.

Linkie, M., Wibisono, H.T., Martyr, D.J. & Sunarto, S. (2008c) Panthera tigris ssp. sumatrae. www.iucnredlist.org/apps/redlist/details/15966/0

Linkie, M., Chapron, G., Martyr, D.J., Holden, J. & Leader Williams, N. (2006) Assessing the viability of tiger subpopulations in a fragmented landscape. Journal of Applied Ecology, 43, 576-586.

Linkie, M., Martyr, D.J., Holden, J., Yanuar, A., Hartana, A.T., Sugardjito, J. & Leader Williams, N. (2003) Habitat destruction and poaching threaten the Sumatran tiger in Kerinci Seblat National Park, Sumatra. Oryx, 37, 41-48.

Lodé, T. & Peltier, D. (2005) Genetic neighbourhood and effective population size in the endangered European mink Mustela lutreola. Biodiversity and Conservation, 14, 251-259.

Lu, Z., Johnson, W.E., Menotti-Raymond, M., Yuhki, N., Martenson, J.S., Mainka, S., Shi- Qiang, H., Zhihe, Z., Li, G., Pan, W., Mao, X. & O’Brien, S.J. (2001) Patterns of genetic

- 188 - diversity in remaining giant panda populations. Conservation Biology, 15, 1596-1607.

Lucchini, V., Fabbri, E., Marucco, F., Ricci, S., Boitani, L. & Randi, E. (2002) Noninvasive molecular tracking of colonizing wolf (Canis lupus) packs in the western Italian Alps. Molecular Ecology, 11, 857-868.

Luikart, G., Allendorf, F.W., Cornuet, J.-M. & Sherwin, W.B. (1998a) Distortion of allele frequency distributions provides a test for recent population bottlenecks. Journal of Heredity, 89, 238-247.

Luikart, G., Sherwin, W.B., Steele, B.M. & Allendorf, F.W. (1998b) Usefulness of molecular markers for detecting population bottlenecks via monitoring genetic change. Molecular Ecology, 7, 963-974.

Lukacs, P.M. & Burnham, K.P. (2005) Review of capture-recapture methods applicable to noninvasive genetic sampling. Molecular Ecology, 14, 3909-3919.

Luo, S.-J., Kim, J.-H., Johnson, W.E., van der Walt, J., Martenson, J., Yuhki, N., Miquelle, D.G., Uphyrkina, O., Goodrich, J.M., Quigley, H.B., Tilson, R., Brady, G., Martelli, P., Subramaniam, V., McDougal, C., Hean, S., Huang, S.-Q., Pan, W., Karanth, U.K., Sunquist, M., Smith, J.L.D. & O’Brien, S.J. (2004) Phylogeography and genetic ancestry of tigers (Panthera tigris). PLoS Biology, 2, e442.

Macbeth, G.M., Broderick, D., Ovenden, J.R. & Buckworth, R.C. (2010) SHAdow Zone Analysis software (SHAZA). Molecular Fisheries Laboratory.

Maddox, T., Priatna, D., Gemita, E. & Salampessy, A. (2007) The conservation of tigers and other wildlife in oil palm plantations. Jambi Province, Sumatra, Indonesia (October 2007). ZSL Conservation Report No. 7. The Zoological Society of London,

Magnuson, V.L., Ally, D.S., Nylund, S.J., Karanjawala, Z.E., Rayman, J.B., Knapp, J.I., Lowe, A.L., Ghosh, S. & Collins, F.S. (1996) Substrate nucleotide-determined non-templated addition of adenine by Taq DNA polymerase: implications for PCR-based genotyping and cloning. BioTechniques, 21, 700-709.

Maheshwari, A. (2006) Food habits and prey abundance of leopard (Panthera pardus fusca) in Gir National Park and Wildlife Sanctuary. MSc thesis, Aligarh Muslim University.

Manel, S., Bellemain, E., Swenson, J.E. & Francois, O. (2004) Assumed and inferred spatial structure of populations: The Scandinavian brown bears revisited. Molecular Ecology, 13, 1327-1331.

Manel, S., Schwartz, M.K., Luikart, G. & Taberlet, P. (2003) Landscape genetics: combining landscape ecology and population genetics. Trends in Ecology and Evolution, 18, 189-197.

Manoukis, N.C. (2007) FORMATOMATIC: a program for converting diploid allelic data between common formats for population genetic analysis. Molecular Ecology Notes, 7, 592-593.

Matschiner, M. & Salzburger, W. (2009) TANDEM: integrating automated allele binning into

- 189 - genetics and genomics workflows. Bioinformatics, 25, 1982-1983.

Matsumoto, T., Yukawa, W., Nozaki, Y., Nakashige, R., Shinya, M., Makino, S., Yagura, M., Ikuta, T., Imanishi, T., Inoko, H., Tamiya, G. & Gojobori, T. (2004) Novel algorithm for automated genotyping of microsatellites. Nucleic Acids Research, 32, 6069-6077.

Mazak, J.H. (2008) Craniometric variation in the tiger (Panthera tigris): implications for patterns of diversity, taxonomy and conservation. doi: 10.1016/j.mambio.2008.06.003

McDonald-Madden, E., Baxter, P.W.J. & Possingham, H.P. (2008) Making robust decisions for conservation with restricted money and knowledge. Journal of Applied Ecology, 45, 1630-1638.

McEwing, R., Ouitavon, K., Rovie-Ryan, J.J, Wulansari, Sitam, F.T. & Ogden, R. (2011) Molecular sexing of tigers, Panthera tigris. doi: 10.1007/s12686-011-9529-x

McKelvey, K.S. & Schwartz, M.K. (2004) Genetic errors associated with population estimation using non-invasive molecular tagging: problems and new solutions. Journal of Wildlife Management, 68, 439-448.

McKelvey, K.S. & Schwartz, M.K. (2005) DROPOUT: a program to identiy problem loci and samples for noninvasive genetic samples in a capture-mark-recapture framework. Molecular Ecology Notes, 5, 716-718.

McKelvey, K.S., von Kienast, J., Aubry, K.B., Koehler, G.M., Maletzke, B.T., Squires, J.R., Lindquist, E.L., Loch, S. & Schwartz, M.K. (2006) DNA analysis of hair and scat collected along snow tracks to document the presence of Canada lynx. Wildlife Society Bulletin, 34, 451-455.

McRae, B.H., Huynh, L.Y., Keim, P., Beier, P. & Dewald, L.E. (2005) Habitat barriers limit gene flow and illuminate historical events in a wide-ranging carnivore, the American puma. Molecular Ecology, 14, 1965-1977.

McRae, B.H. (2006) Isolation by resistance. Evolution, 60, 1551-1561.

McRae, B.H. & Beier, P. (2007) Circuit theory predicts gene flow in plant and animal populations. Proceedings of the National Academy of Sciences of the United States of America, 104, 19885-19890.

McRae, B.H., Dickson, B.G., Keitt, T.H. & Shah, V.B. (2008) Using circuit theory to model connectivity in ecology, evolution, and conservation. Ecology, 89, 2712-2724.

Meadows, R. (2002) Scat sniffing dogs. ZooGoer, 31,

Meirmans, P.G. & Hedrick, P.W. (2011) Assessing population structure: Fst and related measures. Molecular Ecology Resources, 11, 5-18.

Menotti-Raymond, M., David, V.A., Lyons, L., A, Schaffer, A.A., Tomlin, J.F., Hutton, M.K. & O’Brien, S.J. (1999) A genetic linkage map of microsatellites in the domestic cat (Felis catus). Genomics, 57, 9-23.

- 190 - Menotti-Raymond, M. & O’Brien, S.J. (1995) Evolutionary conservation of ten microsatellite loci in four species of Felidae. Journal of Heredity, 84(4), 319-322.

Menotti-Raymond, M., David, V.A., Chen, Z.Q., Menotti, K.A., Sun, S., Schaffer, A.A., Agarwala, R., Tomlin, J.F., O’Brien, S.J. & Murphy, W.J. (2003) Second-generation integrated genetic linkage/radiation hybrid maps of the domestic cat (Felis catus). Journal of Heredity, 94, 95-106.

Michalakis, Y. & Excoffier, L. (1996) A generic estimation of population subdivision using distances between alleles with special reference for microsatellite loci. Genetics, 142, 1061-1064.

Michalski, F.,Boulhosa R. L. P.,Faria A., A., P.C. (2006) Human-wildlife conflicts in a fragmented Amazonian forest landscape: Determinants of large felid depredation on livestock. Animal Conservation, 9, 179-188.

Michels, E., Cottenie, K., Neys, L., de Gelas, K., Coppin, P. & de Meester, L. (2001) Geographical and genetic distances among zooplankton populations in a set of interconnected ponds: a plea for using GIS modelling of the effective geographical distance. Molecular Ecology, 10, 1929-1938.

Miettinen, J. & Liew, S.C. (2010) Degradation and development of peatlands in Peninsular Malaysia and in the islands of Sumatra and Borneo since 1990. Land Degradation and Development, 21, 285-296.

Miettinen, J., Shi, C. & Liew, S.C. (2011a) Deforestation rates in insular Southeast Asia between 2000 and 2010. Global Change Biology, 17, 2261-2270.

Miettinen, J., Shi, C. & Liew, S.C. (2011b) Two decades of destruction in Southeast Asia’s peat swamp forests. doi: 10.1890/100236

Miettinen, J., Wong, C.M. & Liew, S.C. (2008) New 500m spatial resolution land cover map of the western insular Southeast Asia region. International Journal of Remote Sensing, 29, 6075-6081.

Miller, C.R. & Waits, L.P. (2003) The history of effective population size and genetic diversity in the Yellowstone grizzly (Ursus arctos): Implications for conservation. Proceedings of the National Academy of Sciences of the United States of America, 100, 4334-4339.

Miller, C.R., Joyce, P. & Waits, L.P. (2002) Assessing allelic dropout and genotype reliability using maximum likelihood. Genetics, 160, 357-366.

Millions, D.G. & Swanson, B.J. (2007) Impact of natural and artificial barriers to dispersal on the population structure of Bobcats. Journal of Wildlife Management, 71, 96-102.

Mills, L.S., Citta, J.J., Lair, K.P., Schwartz, M.K. & Tallmon, D.A. (2000) Estimating animal abundance using noninvasive DNA sampling: Promise and pitfalls. Ecological Applications, 10, 283-294.

Mills, L.S. & Allendorf, F.W. (1996) The one-migrant-per-generation rule in conservation and

- 191 - management. Conservation Biology, 10, 1509-1518.

Miquel, C., Bellemain, E., Poillot, C., Bessiére, J., Durand, A. & Taberlet, P. (2006) Quality indexes to assess the reliability of genotypes in studies using noninvasive sampling and multi-tube approach. Molecular Ecology Notes, 6, 985-988.

Miquelle, D.G., Smirnov, E.N., Merrill, T.W., Myslenkov, A.E., Quigley, H.B., Hornocker, M.G. & Schleyer, B. (1999) Hierarchical spatial analysis of Amur tiger relationships to habitat and prey. In Riding the tiger: Tiger conservation in human-dominated landscapes, (Eds, Seidensticker, J., Christie, S. & Jackson, P.) Cambridge University Press, Cambridge, pp. 71-99.

Miththapala, S., Seidensticker, J. & O’Brien, S.J. (1996) Phylogeographic subspecies recognition in leopards (Panthera pardus): Molecular genetic variation. Conservation Biology, 10, 1115-1132.

Moilanen, A. (2004) SPOMSIM: software for stochastic patch occupancy models of metapopulation dynamics. Ecological Modelling, 179, 533-550.

Moilanen, A. & Nieminen, M. (2002) Simple connectivity measures in spatial ecology. Ecology, 83, 1131-1145.

Mondol, S., Karanth, K.U., Kumar, N.S., Gopalaswamy, A.M., Andheria, A. & Ramakrishnan, U. (2009a) Evaluation of non-invasive genetic sampling methods for estimating tiger population size. doi: 10.1016/j.biocon.2009.05.014

Mondol, S., Karanth, K.U. & Ramakrishnan, U. (2009b) Why the Indian Subcontinent Holds the Key to Global Tiger Recovery. PLoS Genetics, 5, e1000585.

Monteiro, L., Bonnemaison, D., Vekris, A., Petry, K.G., Bonnet, J., Vidal, R., Cabrita, J. & Mégraud, F. (1997) Complex polysaccharides as PCR inhibitors in feces: Helicobacter pylori model. Journal of Clinical Microbiology, 35, 995-998.

Morales, J.C., Andau, P.M., Supriatna, J., Zainuddin, Z.-Z. & Melnick, D.J. (1997) Mitochondrial DNA variability and conservation genetics of the Sumatran rhinoceros. Conservation Biology, 11, 539-543.

Morin, P.A., Chambers, K.E., Boesch, C. & Vigilant, L. (2001) Quantitative polymerase chain reaction analysis of DNA from noninvasive samples for accurate microsatellite genotyping of wild chimpanzees (Pan troglodytes verus). Molecular Ecology, 10, 1835-1844.

Morin, P.A., Manaster, C., Mesnick, S.L. & Holland, R. (2009) Normalization and binning of historical and multi-source microsatellite data: overcoming the problems of allele size shift with ALLELOGRAM. Molecular Ecology Resources, 9, 1451-1455.

Morin, P.A., Luikart, G., Wayne, R.K. & group, S.N.P.w. (2004) SNPs in ecology, evolution and conservation. Trends in Ecology and Evolution, 19, 208-216.

Mowat, G. & Paetkau, D. (2002) Estimating marten Martes americana population size using hair capture and genetic tagging. Wildlife Biology, 8, 201-209.

- 192 - Munson, L., Brown, J.L., Bush, M., Packer, C., Janssen, D., Reiziss, S.M. & Wildt, D.E. (1996) Genetic diversity affects testicular morphology in free-ranging lions (Panthera leo) of the Serengeti Plains and Ngorongoro Crater. Journal of Reproduction and Fertility, 108, 11-15.

Murphy, M.A., Waits, L.P., Kendall, K.C., Wasser, S.K., Higbee, J.A. & Bogden, R. (2002) An evaluation of long-term preservation methods for brown bear (Ursus arctos) faecal DNA samples. Conservation Genetics, 3, 435-440.

Nauta, M.J. & Weissing, F.J. (1996) Constraints on allele size at microsatellite loci: implications for genetic differentiation. Genetics, 143, 1021-1032.

Navidi, W., Arnheim, N. & Waterman, M.S. (1992) A multiple-tubes approach for accurate genotyping of very small DNA samples using PCR: statistical considerations. American Journal of Human Genetics, 50, 347-359.

Nechvatal, J.M., Ram, J.L., Basson, M.D., Namprachan, P., Niec, S.R., Badsha, K.Z., Matherly, L.H., Majumdar, A.P.N. & Kato, I. (2008) Fecal collection, ambient preservation, and DNA extraction for PCR amplification of bacterial and human markers from human feces. Journal of Microbiology Methods, 72, 124-132.

Neel, M.C. & Cummings, M.P. (2003) Effectiveness of conservation targets in capturing genetic diversity. Conservation Biology, 17, 219-229.

Neff, B.D. & Gross, M.R. (2001) Microsatellite evolution in vertebrates: inference from AC dinucleotide repeats. Evolution, 55, 1717-1733.

Nei, M. (1973) Analysis of gene diversity in subdivided populations. Proceedings of the National Academy of Sciences of the United States of America, 70, 3321-3323.

Nei, M. (1978) Estimation of average heterozygosity and genetic distance from a small number of individuals. Genetics, 89, 583-590.

Nei, M., Maruyama, T. & Chakraborty, R. (1975) The bottleneck effect and genetic variability in populations. Evolution, 29, 1-10.

Nei, M. & Roychoudhury, A.K. (1974) Sampling variances of heterozygosity and genetic distance. Genetics, 76, 379-390.

Nowell, K. (2000) Far from a cure: The tiger trade revisited. TRAFFIC,

Nowell, K. & Jackson, P. (1996) Wild Cats: Status survey and conservation action plan. IUCN/SSC Cat Specialist Group,

Nsubuga, A.M., Robbins, M.M., Roder, A.D., Morin, P.A., Boesch, C. & Vigilant, L. (2004) Factors affecting the amount of genomic DNA extracted from ape faeces and the identification of an improved sample storage method. Molecular Ecology, 13, 2089-2094.

Nunney, L. (1993) The influence of mating system and overlapping generations on effective population size. Evolution, 47, 1329-1341.

- 193 - Nyhus, P. & Tilson, R. (2004a) Agroforestry, elephants, and tigers: balancing conservation theory and practice in human-dominated landscapes of Southeast Asia. Agriculture, Ecosystems and Environment, 104, 87-97.

Nyhus, P.J. & Tilson, R. (2004b) Characterising human-tiger conflict in Sumatra, Indonesia: implications for conservation. Oryx, 38, 68-74.

Nyhus, P.J., Tilson, R., Franklin, N., Bastoni, Sriyanto, Yunus, M. & Sumianto (2000) Tigers, cameras, and satellites: interdisciplinary efforts to map the habitat and threats of the last Sumtran tigers. Esri International User Conference,

O’Brien, T.G., Kinnaird, M.F. & Wibisono, H.T. (2003) Crouching tigers, hidden prey: Sumatran tiger and prey populations in a tropical forest landscape. Animal Conservation, 6, 131-139.

Ogden, R. (2011) Unlocking the potential of genomic technologies for wildlife forensics. Molecular Ecology Resources, 11, 109-116.

Ogden, R., Dawnay, N. & McEwing, R. (2009) Wildlife DNA forensics - bridging the gap between conservation genetics and law enforcement. doi: 10.3354/esr00144

Okello, J.B.A., Wittemyer, G., Rasmussen, H.B., Douglas-Hamilton, I., Nyakaana, S., Arctander, P. & Siegismund, H.R. (2005) Noninvasive genotyping and mendelian analysis of microsatellites in African savannah elephants. Journal of Heredity, 96, 679-687.

Olsen, M.T., Volny, V.H., Bérubé, M., Dietz, R., Lydersen, C., Kovacs, K.M., Dodd, R.S. & Palsbøll, P.J. (2011) A simple route to single-nucleotide polymorphisms in a nonmodel species: identification and characterization of SNPs in the Arctic ringed seal (Pusa hispida hispida). Molecular Ecology Resources, 11, 9-19.

Opdam, P. & Wascher, D. (2004) Climate chane meets habitat fragmentation: linking landscape and biogeographical scale levels in research and conservation. Biological Conservation, 117, 285-297.

Ovando, D.A.C. (2008) Road impact on deforestation and jaguar habitat loss in the Mayan forest. PhD thesis, Duke University.

Packer, C., Kosmala, M., Cooley, H.S., Brink, H., Pintea, L., Garshelis, D., Purchase, G., Strauss, M., Swanson, A., Balme, G., Hunter, L. & Nowell, K. (2009) Sport hunting, predator control and conservation of large carnivores. PloS One, 4, e5941.

Paetkau, D. (2003) An empirical exploration of data quality in DNA-based population inventories. Molecular Ecology, 12, 1375-1387.

Paetkau, D., Slade, R., Burden, M. & Estoup, A. (2004) Genetic assignment methods for the direct, real-time estimation of migration rate: a simulation-based exploration of accuracy and power. Molecular Ecology, 13, 55-65.

Paetkau, D., Strobeck, C., Waits, L.P., Ward, R., Clarkson, P.L., Craighead, L. & Vyse, E. (1998) Variation in genetic diversity across the range of North American brown bears.

- 194 - Conservation Biology, 12, 418-429.

Palsbøll, P.J., Allen, J., Bérubé, M., Clapham, P.J., Feddersen, T.P., Hammond, P.S., Hudson, R.R., Jørgensen, H., Katona, S., Larsen, A.H., Larsen, F., Lien, J., Mattila, D.K., Sigurjónsson, J., Sears, R., Smith, T., Sponer, R., Stevick, P. & Øien, N. (1997) Genetic tagging of humpback whales. Nature, 388, 767-769.

Palstra, F.P. & Ruzzante, D.E. (2008) Genetic estimates of contemporary effective population size: what can they tell us about the importance of genetic stochasticity for wild population persistence? Molecular Ecology, 17, 3428-3447.

Panthera Tiger Corridor Initiative. http://www.panthera.org/video/tiger-corridors

Paquette, S.R. (2011) Package “PopGenKit” v.1.0. The Comprehensive R Archive Network. http://cran.r-project.org/web/packages/PopGenKit/index.html

Patterson, N., Price, A.L. & Reich, D. (2006) Population structure and eigenanalysis. Plos Genetics, 2, e190.

Peakall, R., Ruiball, M. & Lindenmayer, D.B. (2003) Spatial autocorrelation analysis offers new insights into gene flow in the Australian bush rat, Rattus fuscipes. Evolution, 57, 1182-1195.

Peakall, R. & Smouse, P. (2005) Appendix 2 - Spatial autocorrelation in GenAlEx 6.

Peakall, R. & Smouse, P.E. (2006) GENALEX 6: genetic analysis in Excel. Population genetic software for teaching and research. Molecular Ecology Notes, 6, 288-295.

Perry, G.H., Marioni, J.C., Melsted, P. & Gilad, Y. (2010) Genomic-scale capture and sequencing of endogenous DNA from feces. Molecular Ecology, 19, 5332-5344.

Piertney, S.B., MacColl, A.D.C., Bacon, P.J. & Dallas, J.F. (1998) Local genetic structure in red grouse (Lagapus lagopus scoticus): evidence from microsatellite DNA markers. Molecular Ecology, 7, 1645-1654.

Piggott, M.P. (2004) Effect of sample age and season of collection on the reliability of microsatellite genotyping of faecal DNA. Wildlife Research, 31, 485-493.

Piggott, M.P., Banks, S.C., Stone, N. & Taylor, A.C. (2006) Estimating population size of endangered brush-tailed rock-wallaby (Petrogale penicillata) colonies using faecal DNA. Molecular Ecology, 15, 81-91.

Piggott, M.P., Bellemain, E., Taberlet, P. & Taylor, A.C. (2004) A multiplex pre-amplification method that significantly improves microsatellite amplification and error rates for faecal DNA in limiting conditions. Conservation Genetics, 5, 417-420.

Pilgrim, K.L., McKelvey, K.S., Riddle, A.E. & Schwartz, M.K. (2005) Felid sex identification based on noninvasive genetic samples. Molecular Ecology Notes, 5, 60-61.

Pilot, M., Jedrzejewski, W., Branicki, W., Sidorovich, V.E., Jedrzejewska, B., Stachura, K. &

- 195 - Funk, S.M. (2006) Ecological factors influence population genetic structure of European grey wolves. Molecular Ecology, 15, 4533-4553.

Pocock, R.I. (1929) Tigers. Journal of the Bombay Natural History Society, 33, 505-541.

Pompanon, F., Bonin, A., Bellemain, E. & Taberlet, P. (2005) Genotyping errors: causes, consequences and solutions. Nature Reviews Genetics, 6, 847-859.

Pope, L.C.,Domingo-Roura X.,Erven K., T., B. (2006) Isolation by distance and gene flow in the Eurasian badger (Meles meles) at both a local and broad scale. Molecular Ecology, 15, 371-386.

Priatna, D., Santosa, Y., Prasetyo, L.B. & Kartono, A.P. (2012) Home range and movements of male translocated problem tigers in Sumatra. Asian Journal of Conservation Biology, 1, 20-30.

Prigioni, C., Remonti, L., Balestrieri, A., Sgrosso, S., Priore, G., Mucci, N. & Randi, E. (2006) Estimation of European otter (Lutra lutra) population size by fecal DNA typing in southern Italy. Journal of Mammalogy, 87, 855-858.

Primack, R.B. (2002) Problems of small populations. In Essentials of Conservation Biology, (Ed, Primack, R.B.) Sinauer Associates Inc, Sunderland, MA, pp. 297-328.

Pritchard, J.K., Stephens, M. & Donnelly, P. (2000) Inference of population structure using multilocus genotype data. Genetics, 155, 945-959.

Pritchard, J.K., Wen, X. & Falush, D. (2010) Documentation for Structure Software: Version 2.3. http://pritch.bsd.uchicago.edu/structure.html

Pruett, C.L. & Winker, K. (2008) The effects of sample size on population genetic diversity estimates in song sparrows Melospiza melodia. Journal of Avian Biology, 39, 252-256.

Prugh, L.R., Hodges, K.E., Sinclair, A.R.E. & Brashares, J.S. (2008) Effect of habitat area and isolation on fragmented animal populations. Proceedings of the National Academy of Sciences USA, 105, 20770-20775.

Prugh, L.R., Ritland, C.E., Arthur, S.M. & Krebs, C.J. (2005) Monitoring coyote population dynamics by genotyping faeces. Molecular Ecology, 14, 1585-1596.

Pulliam, H.R., Dunning, J.B. & Liu, J. (1992) Population dynamics in complex landscapes: a case study. Ecological Applications, 2, 165-177.

R Development Core Team (2011) R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing. http://www.R-project.org

Rabinowitz, A. & Zeller, K.A. (2010) A range-wide model of landscape connectivity and conservation for the jaguar, Panthera onca. doi: 10.1016/j.biocon.2010.01.002

Ralls, K., Ballou, J. & Templeton, A. (1988) Estimates of lethal equivalents and the cost of inbreeding in mammals. Conservation Biology, 2, 185-193.

- 196 - Ranganathan, J., Chan, K.M.A., Karanth, K.U. & Smith, J.L.D. (2008) Where can tigers persist in the future? A landscape-scale, density-based population model for the Indian subcontinent. Biological Conservation, 141, 67-77.

Rannala, B. & Mountain, J.L. (1997) Detecting immigration by using multilocus genotypes. Proceedings of the National Academy of Sciences USA, 94, 9197-9201.

Rao, C.B. (2001) Sample size considerations in genetic polymorphism studies. Human Heredity, 52, 191-200.

Ray, N. (2005) PATHMATRIX: a geographical information system tool to compute effective distances among samples. Molecular Ecology Notes, 5, 177-180.

Raymond, M. & Rousset, F. (1995) GENEPOP (Version 1.2): Population genetics software for exact tests and ecumenicism. Journal of Heredity, 86, 248-249.

Reddy, H.S., Srinivasulu, C. & Thulsi Rao, K. (2004) Prey selection by the Indian tiger (Panthera tigris tigris) in Nagarjunasagar Srisailam Tiger Reserve, India. Mammalian Biology (Zeitschrift fur Saugetierkunde), 69, 384-391.

Reddy, P.A., Gour, D.S., Bhavanishankar, M., Jaggi, K., Hussain, S.M., Harika, K. & Shivaji, S. (2012) Genetic evidence of tiger population structure and migration within an isolated and fragmented landscape in Northwest India. PLoS One, 7, e29827.

Redford, K.H., Ray, J.C. & Boitani, L. (2011) Mapping and navigating mammalian conservation from analysis to action. Philosophical Transactions of the Royal Society of London: Series B Biological Sciences, 366, 2712-2721.

Reed, D.H. (2004) Extinction risk in fragmented habitats. Animal Conservation, 7, 181-191.

Reed, D.H. & Frankham, R. (2003) Correlation between Fitness and Genetic Diversity. Conservation Biology, 17, 230-237.

Reed, J.Z., Tollit, D.J., Thompson, P.M. & Amos, W. (1997) Molecular scatology: the use of molecular genetic analysis to assign species, sex and individual identity to seal faeces. Molecular Ecology, 6, 225-234.

Ricketts, T.H. (2001) The matrix matters: effective isolation in fragmented landscapes. The American Naturalist, 158, 87-99.

Ries, L., Fletcher, R.J., Battin, J. & Sisk, T.D. (2004) Ecological responses to habitat edges: mechanisms, models, and variability explained. Annual Review of Ecology, Evolution, and Systematics, 35, 491-522.

Riley, S.P.D., Pollinger, J.P., Sauvajot, R.M., York, E.C., Bromley, C., Fuller, T.K. & Wayne, R.K. (2006) A southern California freeway is a physical and social barrier to gene flow in carnivores. Molecular Ecology, 15, 1733-1741.

Roeder, A.D., Archer, F.I., Poinar, H.N. & Morin, P.A. (2004) A novel method for collection and preservation of faeces for genetic studies. Molecular Ecology Notes, 4, 761-764.

- 197 - Roelke, M.E., Martenson, J.S. & O’Brien, S.J. (1993) The consequences of demographic reduction and genetic depletion in the endangered Florida panther. Current Biology, 3, 340-350.

Rohland, N. & Hofreiter, M. (2007) Comparison and optimization of ancient DNA extraction. BioTechniques, 42, 343-352.

Römpler, H., Dear, P.H., Krause, J., Meyer, M., Rohland, N., Schöneberg, T., Spriggs, H., Stiller, M. & Hofreiter, M. (2006) Multiplex amplification of ancient DNA. Nature Protocols, 1, 720-728.

Roon, D.A., Waits, L.P. & Kendall, K.C. (2003) A quantitative evaluation of two methods for preserving hair samples. Molecular Ecology Notes, 3, 163-166.

Rosenberg, N.A., Burke, T., Elo, K., Feldmen, M.W., Freidlin, P.J., Groenen, M.A.M., Hillel, J., Mäki-Tanila, A., Tixier-Boichard, M., Vignal, A., Wimmers, K. & Weigend, S. (2001) Empirical evaluation of genetic clustering methods using multilocus genotypes from 20 chicken breeds. Genetics, 159, 699-713.

Rousset, F. (1996) Equilibrium values of measures of population subdivision for stepwise mutation processes. Genetics, 142, 1357-1362.

Rousset, F. (2000) Genetic differentiation between individuals. Journal of Evolutionary Biology, 13, 58-62.

Rousset, F. (2004) Spatially homogeneous dispersal: the island model and isolation by distance. In Genetic Structure and Selection in Subdivided Populations., (Eds, Levin, S.A. & Horn, H.S.) Princeton University Press, Princeton, USA, pp. 23-52.

Rousset, F. (2007) Neutral models of geographic variation. In Handbook of Statistical Genetics, (Eds, Balding, D.J., Bishop, M. & Cannings, C.) John Wiley & Sons Ltd, Chicester, UK, pp. 945.

Rousset, F. (1997) Genetic differentiation and estimation of gene flow from F-statistics under isolation by distance. Genetics, 145, 1219-1228.

Ruell, E.W. & Crooks, K.R. (2007) Evaluation of noninvasive genetic sampling methods for felid and canid populations. Journal of Wildlife Management, 71, 1690-1694.

Ruiz Garcia, M. (2003) Molecular population genetic analysis of the spectacled bear (Tremarctos ornatus) in the northern Andean area. Hereditas, 138, 81-93.

Ruzzante, D.E. (1998) A comparison of several measures of genetic distance and population structure with microsatellite data: bias and sampling variance. Canadian Journal of Fisheries and Aquatic Sciences, 55, 1-14.

Ryman, N. (2006) CHIFISH: a computer program testing for genetic heterogeneity at multiple loci using chi-square and Fisher’s exact test. Molecular Ecology Notes, 6, 285-287.

Saccheri, I.J., Kuussaari, M., Kankare, M., Vikman, P., Fortelius, W. & Hanski, I. (1998)

- 198 - Inbreeding and extinction in a butterfly metapopulation. Nature, 392, 491-494.

Sacks, B.N., Brown, S.K. & Ernes, H.B. (2004) Population structure of California coyotes corresponds to habitat-specific breaks and illuminates species history. Molecular Ecology, 13, 1265-1275.

Sacks, B.N. & Louie, S. (2008) Using the dog genome to find single nucleotide polymorphisms in red foxes and other distantly related members of the Canidae. Molecular Ecology Resources, 8, 35-49.

Safner, T., Miller, M.P., McRae, B.H., Fortin, M. & Manel, S. (2011) Comparison of Bayesian Clustering and Edge Detection Methods for Inferring Boundaries in Landscape Genetics. Internation Journal of Molecular Sciences, 12, 865-889.

Sahlsten, J., Thörngren, H. & Höglund, J. (2008) Inference of hazel grouse population structure using multilocus data: a landscape genetic approach. Heredity, 101, 475-482.

Sanderson, E., Forrest, J., Loucks, C., Ginsberg, J., Dinerstein, E., Seidensticker, J., Leimgruber, P., Songer, M., Heydlauff, A., O’Brien, T., Bryja, G., Klenzendorf, S. & Wikramanayake, E. (2006) Setting Priorities for the Conservation and Recovery of Wild Tigers: 2005 - 2015. The Technical Assessment. WCS, WWF, Smithsonian, and NFWF- STF,

Sanderson, E.W., Jaiteh, M., Levy, M.A., Redford, K.H., Wannebo, A.V. & Woolmer, G. (2002a) The human footprint and the last of the wild. BioScience, 52, 891-904.

Sanderson, E.W., Redford, K.H., Vedder, A., Coppolillo, P.B. & Ward, S.E. (2002b) A conceptual model for conservation planning based on landscape species requirements. Landscape and Urban Planning, 58, 41-56.

Santiapillai, C. & Ramono, W.S. (1987) Tiger numbers and habitat evaluation in Indonesia. In Tigers of the world: The biology, biopolitics, management and conservation of an endangered species, (Eds, Tilson, R. & Seal, U.S.) Noyes Publications, New Jersey, USA, pp. 85-91.

Santini, A., Lucchini, V., Fabbri, E. & Randi, E. (2007) Ageing and environmental factors affect PCR success in wolf (Canis lupus) excremental DNA samples. doi: 10.1111/ j.1471-8286.2007.01829.x

Sawyer, S.C., Epps, C.W. & Brashares, J.S. (2011) Placing linkages among fragmented habitats: do least-cost models reflect how animals use landscapes? Journal of Applied Ecology, 48, 668-678.

Scandura, M., Capitani, C., Iacolina, L. & Marco, A. (2006) An empirical approach for reliable microsatellite genotyping of wolf DNA from multiple noninvasive sources. Conservation Genetics, 7, 813-823.

Schadt, S., Revilla, E., Wiegand, T., Knauer, F., Kaczensky, P. & Trepl, L. (2002) Rule- based assessment of suitable habitat and patch connectivity for the Eurasian lynx. Ecological Applications, 12, 1469-1483.

- 199 - Schlüter, P.M. & Harris, S.A. (2006) Analysis of multilocus fingerprinting data sets containing missing data. Molecular Ecology Notes, 6, 569-572.

Schmidt, K. & Kowalczyk, R. (2006) Using scent-marking stations to collect hair samples to monitor Eurasian lynx populations. Wildlife Society Bulletin, 34, 462-466.

Schtickzelle, N. & Baguette, M. (2003) Behavioural responses to habitat patch boundaries restrict dispersal and generate emigration-patch area relationships in fragmented landscapes. Journal of Animal Ecology, 72, 533-545.

Schultz, C.B. & Crone, E.E. (2001) Edge-mediated dispersal behaviour in a prairie butterfly. Ecology, 82, 1879-1892.

Schumaker, N.H. (1998) A users guide to the PATCH model. US Environmental Protection Agency, Environmental Research Laboratory.

Schwartz, M.K., Copeland, J.P., Anderson, N.J., Squires, J.R., Inman, R.M., McKelvey, K.S., Pilgrim, K.L., Waits, L.P. & Cushman, S.A. (2009) Wolverine gene flow across a narrow climatic niche. Ecology, 90, 3222-3232.

Schwartz, M.K., Cushman, S.A., McKelvey, K.S., Hayden, J. & Engkjer, C. (2006) Detecting genotyping errors and describing American black bear movement in northern Idaho. Ursus, 17, 138-148.

Schwartz, M.K., Mills, L.S., Ortega, Y., Ruggiero, L.F. & Allendorf, F.W. (2003) Landscape location affects genetic variation of Canada lynx (Lynx canadensis). Molecular Ecology, 12, 1807-1816.

Schwartz, M.K., Pilgrim, K.L., Ralls, K., Fleischer, R.C., Williams, D.F. & Cypher, B.L. (2005) Gene flow among San Joaquin kit fox populations in a severely changed ecosystem. Conservation Genetics, 6, 25-37.

Seddon, J.M.,Sundqvist A. K.,Bjornerfeldt S., H., E. (2006) Genetic identification of immigrants to the Scandinavian wolf population. Conservation Genetics, 7, 225-230.

Segelbacher, G. (2002) Noninvasive genetic analysis in birds: testing reliability of feather samples. Molecular Ecology Notes, 2, 367-369.

Seidensticker, J. (1987) Bearing witness: Obeservations on the extinction of Panthera tigris balica and Panthera tigris sondaica. In Tigers of the world: The biology, biopolitics, management and conservation of an endangered species, (Eds, Tilson, R. & Seal, U.S.) Noyes Publications, New Jersey, USA, pp. 1-8.

Seidensticker, J., Christie, S. & Jackson, P. (1999) Riding the tiger: Tiger conservation in human-dominated landscapes. Cambridge University Press, Cambridge.

Sekhar, N.U. (1998) Crop and livestock depredation caused by wild animals in protected areas: the case of Sariska Tiger Reserve, Rajasthan, India. Environmental Conservation, 25, 160-171.

- 200 - Seutin, G., White, B.N. & Boag, P.T. (1991) Preservation of avian blood and tissue samples for DNA analyses. Canadian Journal of Zoology, 69, 82-90.

Shah, V.B. & McRae, B.H. (2008) Circuitscape: A tool for landscape ecology. 7th Python in Science Conference (SciPy 2008), 62-66.

Sharma, R., Stuckas, H., Bhaskar, R., Khan, I., Goyal, S.P. & Tiedemann, R. (2010) Genetically distinct population of Bengal tiger (Panthera tigris tigris) in Terai Arc Landscape (TAL) of India. doi: 10.1016/j.mambio.2010.10.005

Sharma, R., Stuckas, H., Bhaskar, R., Rajput, S., Khan, I., Goyal, S.P. & Tiedemann, R. (2009) mtDNA indicates profound population structure in Indian tiger (Panthera tigris tigris). Conservation Genetics, 10, 909-914.

Sharma, R., Stuckas, H., Moll, K., Khan, I., Bhaskar, R., Goyal, S.P. & Tiedemann, R. (2008) Fourteen new di- and tetranucleotide microsatellite loci for the critically endangered Indian tiger (Panthera tigris tigris). Molecular Ecology Resources, 8, 1480-1482.

Shaw, K.J., Thain, L., Docker, P.T., Dyer, C.E., Greenman, J., Greenway, G.M. & Haswell, S.J. (2009) The use of carrier RNA to enhance DNA extraction from microfluidic-based silica monoliths. Analytica Chimica Acta, 652, 231-233.

Shepherd, C.R. & Magnus, N. (2004) Nowhere to hide: The trade in Sumatran tiger. TRAFFIC Southeast Asia, 108.

Shriver, M.D., Smith, M.W., Jin, L., Marcini, A., Akey, J.M., Deka, R. & Ferrell, R.E. (1997) Ethnic-affiliation estimation by use of population-specific DNA markers. American Journal of Human Genetics, 60, 957-964.

Sinclair, E.A. & Hobbs, R.J. (2009) Sample size effects on estimates of population genetic structure: implications for ecological restoration. Restoration Ecology, 17, 837-844.

Sinnott, R.W. (1984) Virtues of the Haversine. Sky and Telescope, 68, 159.

Sirkkomaa, S. (1983) Calculations on the decrease of genetic variation due to the founder effect. Hereditas, 99, 11-20.

Sjögren, P. & Wyoni, P.-I. (1994) Conservation genetics and detection of rare alleles in finite populations. Conservation Biology, 8, 267-270.

Slatkin, M. (1995) A measure of population subdivision based on microsatellite allele frequencies. Genetics, 139, 457-462.

Sloane, M.A., Sunnucks, P., Alpers, D., Beheregaray, L.B. & Taylor, A.C. (2000) Highly reliable genetic identification of individual northern hairy-nosed wombats from single remotely collected hairs: a feasible censusing method. Molecular Ecology, 9, 1233-1240.

Smirnov, E.N. & Miquelle, D.G. (1999) Population dynamics of the Amur tiger in Sikhote-Alin Zapovednik, Russia. In Riding the Tiger: Tiger conservation in human-dominated landscapes., (Eds, Seidensticker, J., Christie, S. & Jackson, P.) Cambridge University Press,

- 201 - Cambridge, pp. 61-70.

Smith, D.A., Ralls, K., Hurt, A., Adams, B., Parker, M., Davenport, B., Smith, M.C. & Maldonado, J.E. (2003) Detection and accuracy rates of dogs trained to find scats of San Joaquin kit foxes (Vulpes macrotis mutica). Animal Conservation, 6, 339-346.

Smith, D.A., Ralls, K., Hurt, A., Adams, B., Parker, M. & Maldonado, J.E. (2006) Assessing reliability of microsatellite genotypes from kit fox faecal samples using genetic and GIS analyses. Molecular Ecology, 15, 387-406.

Smith, J.L.D., McDougal, C.W. & Sunquist, M.E. (1987) Female land tenure system in tigers. In Tigers of the world: The biology, biopolitics, management and conservation of an endangered species, (Eds, Tilson, R. & Seal, U.S.) Noyes Publications, New Jersey, USA, pp. 97-109.

Smith, J.L.D. & McDougal, C. (1991) The contribution of variance in lifetime reproduction to effective population size in tigers. Conservation Biology, 5, 484-490.

Smith, J.A. (2009) Mammalian diversity and distribution in human-altered tropical landscapes. PhD thesis, Imperial College London.

Smith, J.R., Carpten, J.D., Brownstein, M.J., Ghosh, S., Magnuson, V.L., Gilbert, D.A., Trent, J.M. & Collins, F.S. (1995) Approach to genotyping errors caused by nontemplated nucleotide addition by Taq DNA polymerase. Genome research, 5, 312-317.

Smith, J.L.D., Ahern, S.C. & McDougal, C. (1998) Landscape analysis of tiger distribution and habitat quality in Nepal. Conservation Biology, 12, 1338-1346.

Smith, J.L.D. (1993) The role of dispersal in structuring the Chitwan tiger population. Behaviour, 124, 165-195.

Smouse, P.E., Long, J.C. & Sokal, R.R. (1986) Multiple regression and correlation extensions of the Mantel test of matrix correspondence. Systematic Zoology, 35, 627-632.

Smouse, P.E. & Peakall, R. (1999) Spatial autocorrelation analysis of individual multiallele and multilocus genetic structure. Heredity, 82, 561-573.

Spear, S.F., Balkenhol, N., Fortin, M., McRae, B.H. & Scribner, K. (2010) Use of resistance surfaces for landscape genetic studies: considerations for parameterisation and analysis. Molecular Ecology, 19, 3576-3591.

Spear, S.F., Peterson, C.R., Matocq, M.D. & Storfer, A. (2005) Landscape genetics of the blotched tiger salamander (Ambystoma tigrinum melanostictum). Molecular Ecology, 14, 2553-2564.

Spencer, C.C., Neigel, J.E. & Leberg, P.L. (2000) Experimental evaluation of the usefulness of microsatellite DNA for detecting demographic bottlenecks. Molecular Ecology, 9, 1517-1528.

Spong, G., Johansson, M. & Björklund, M. (2000) High genetic variation in leopards

- 202 - indicates large and long-term stable effective population size. Molecular Ecology, 9, 1773-1782.

Stamps, J.A., Buechner, M. & Krishnan, V.V. (1987) The effects of edge permeability and habitat geometry on emigration from patches of habitat. The American Naturalist, 129, 533-552.

Stenglein, J.L., de Barba, M., Ausband, D.E. & Waits, L.P. (2010) Impacts of sampling location within a faeces on DNA quality in two carnivore species. Molecular Ecology Resources, 10, 109-114.

Stoen, O.G. & Wegge, P. (1996) Prey selection and prey removal by tiger (Panthera tigris) during the dry season in lowland Nepal. Mammalia, 60, 363-373.

Storfer, A., Murphy, M.A., Evans, J.S., Goldberg, C.S., Robinson, S., Spear, S.F., Dezzani, R., Delmelle, E., Vierling, L. & Waits, L.P. (2007) Putting the ‘landscape’ in landscape genetics. Heredity, 98, 128-142.

Sub Directorate of Statistics Indicator (2010) Statistical yearbook of Indonesia 2010. BPS - Statistics Indonesia,

Sugg, D.W., Chesser, R.K., Dobson, F.S. & Hoogland, J.L. (1996) Population genetics meets behavioral ecology. Trends in Ecology and Evolution, 11, 338-342.

Sugimoto, T., Nagata, J., Aramilev, V.V., Belozer, A., Higashi, S. & McCullough, D.R. (2006) Species and sex identification from faecal samples of sympatric carnivores, Amur leopard and Siberian tiger, in the Russian Far East. Conservation Genetics, 7, 799-802.

Sumner, J., Rousset, F., Estoup, A. & Moritz, C. (2001) ‘Neighbourhood’ size, dispersal and density estimates in the prickly forest skink (Gnypetoscincus queenslandiae) using individual genetic and demographic methods. Molecular Ecology, 10, 1917-1927.

Sunarto, S., Kelley, M.J., Parakkasi, K., Klenzendorf, S., Septayuda, E. & Kurniawan, H. (2012) Tigers need cover: multi-scale occupancy study of the big cat in Sumatran forest and plantation landscapes. PLoS One, 7, e30859.

Sundqvist, A.K., Ellegren, H. & Vila, C. (2008) Wolf or dog? Genetic identification of predators from saliva collected around bite wounds on prey. Conservation Genetics, 9, 1275-1279.

Sunquist, M., Karanth, K.U. & Sunquist, F. (1999) Ecology, behaviour and resilience of the tiger and its conservation needs. In Riding the Tiger: Tiger conservation in human- dominated landscapes., (Eds, Seidensticker, J., Christie, S. & Jackson, P.) Cambridge University Press, Cambridge, pp. 5-18.

Sunquist, M.E. (1981) The social organisation of tigers (Panthera tigris) in Royal Chitawan National Park, Nepal. Smithsonian Contributions to Zoology, 336,

Suyadi (2010) Tropical deforestation in Bukit Barisan Selatan National Park, Sumatra, Indonesia. Bogor Agricultural University,

- 203 - Szpiech, Z.A., Jakobsson, M. & Rosenberg, N.A. (2008) ADZE: a rarefaction approach for counting alleles private to combinations of populations. Bioinformatics, 24, 2498-2504.

Taberlet, P., Camarra, J.J., Griffin, S., Hanotte, O., Waits, L.P., Dubois-Paganon, C., Burke, T. & Bouvet, J. (1997) Noninvasive genetic tracking of the endangered Pyrenean brown bear population. Molecular Ecology, 6, 869-876.

Taberlet, P., Griffin, S., Goossens, B., Questiau, S., Manceau, V., Escaravage, N., Waits, L.P. & Bouvet, J. (1996) Reliable genotyping of samples with very low DNA quantities using PCR. Nucleic Acids Research, 24, 3189-3194.

Taberlet, P., Waits, L.P. & Luikart, G. (1999) Noninvasive genetic sampling: look before you leap. Trends in Ecology and Evolution, 14, 323-327.

Taberlet, P. & Luikart, G. (1999) Non-invasive genetic sampling and individual identification. Biological journal of the Linnean society, 68, 41-55.

Tallmon, D., Koyuk, A., Luikart, G. & Beaumont, M.A. (2008) ONeSAMP: a program to estimate effective population size using approximate Bayesian computation. Molecular Ecology Resources, 8, 299-301.

Tallmon, D.A., Luikart, G. & Beaumont, M.A. (2004) Comparative evaluation of a new effective population size estimator based on approximate Bayesian computation. Genetics, 167, 977-988.

Tigas, L.A., van Vuren, D.H. & Sauvajot, R.M. (2002) Behavioral responses of bobcats and coyotes to habitat fragmentation and corridors in an urban environment. Biological Conservation, 108, 299-306.

Tilman, D., May, R.M., Lehman, C.L. & Nowak, M.A. (1994) Habitat destruction and the extinction debt. Nature, 371, 65-66.

Tilson, R. & Brady, G. (1992) Summary report on Sumatran tiger regional captive breeding workshop. CBSG, AAZPA, 18.

Tilson, R.L. & Seal, U.S. (1987) Tigers of the world: The biology, biopolitics, management and conservation of an endangered species. Noyes Publications, New Jersey, USA.

Tilson, R.L., Foose, T.J., Princée, F. & Traylor-Holzer, K. (1993) Tiger Global Animal Survival Plan. Regional Captive Propagation Programs & IUCN/SSC Captive Breeding Specialist Group,

Tischendorf, L., Bender, D.J. & Fahrig, L. (2003) Evaluation of patch isolation metrics in mosaic landscapes for specialist vs. generalist dispersers. Landscape Ecology, 18, 41-50.

Tischendorf, L. & Fahrig, L. (2000) How should we measure landscape connectivity? Landscape Ecology, 15, 633-641.

Trizio, I., Crestanello, B., Galbusera, P., Wauters, L.A., Tosi, G., Matthysen, E. & Hauffe, H.C. (2005) Geographical distance and physical barriers shape the genetic structure of

- 204 - Eurasian red squirrels (Sciurus vulgaris) in the Italian Alps. Molecular Ecology, 2005, 469-481.

Turner, M.G. (1989) Landscape ecology: the effect of pattern on process. Annual Review of Ecology and Systematics, 20, 171-197.

Turner, M.G. (2005) Landscape ecology: What is the state of the science? Annual Review of Ecology, Evolution, and Systematics, 36, 319-344.

Turner, M.G., Gardner, R.H. & O’Neill, R.V. (2001) Landscape ecology in theory and practice: Pattern and process. Springer,

UN Population Division/DESA (2009) World population to exceed 9 billion by 2050 (Press Release).

Uphyrkina, O., Johnson, W., Quigley, H., Miquelle, D., Marker, L., Bush, M. & O’Brien, S.J. (2001) Phylogenetics, genome diversity and origin of modern leopard, Panthera pardus. Molecular Ecology, 10, 2617-2633.

Uphyrkina, O., Miquelle, D., Quigley, H., Driscoll, C. & O’Brien, S.J. (2002) Conservation genetics of the Far Eastern leopard (Panthera pardus orientalis). Journal of Heredity, 93, 303-311.

Uphyrkina, O. & O’Brien, S.J. (2003) Applying molecular genetic tools to the conservation and action plan for the critically endangered Far Eastern leopard (Panthera pardus orientalis). Comptes Rendus Biologies, 326, S93-S97.

Uryu, Y., Mott, C., Foead, N., Yulianto, K., Budiman, A., Setiabudi, Takakai, F., Nursamsu, Sunarto, Purastuti, E., Fadhli, N., Hutajulu, C.M.B., Jaenicke, J., Hatano, R., Siegert, F. & Stüwe, M. (2008) Deforestation, forest degradation, biodiversity loss and CO₂ emissions in Riau, Sumatra, Indonesia.

Valentini, A., Miquel, C., Nawaz, M.A., Bellemain, E., Coissac, E., Pompanon, F., Gielly, L., Cruaud, C., Nascetti, G., Wincker, P., Swenson, J.E. & Taberlet, P. (2009) New perspectives in diet analysis based on DNA barcoding and parallel pyrosequencing: the trnL approach. Molecular Ecology Resources, 9, 51-60.

Valière, N. (2002) GIMLET: a computer program for analysing genetic individual identification data. Molecular Ecology Notes, 2, 377-379.

Valière, N., Berthier, P., Mouchiroud, D. & Pontier, D. (2002) GEMINI: software for testing the effects of genotyping errors and multitubes approach for individual identification. Molecular Ecology Notes, 2, 83-86.

Valière, N. & Taberlet, P. (2000) Urine collected in the field as a source of DNA for species and individual identification. Molecular Ecology, 9, 2149-2154. van Loon, E.E., Cleary, D.F.R. & Fauvelot, C. (2007) ARES: software to compare allelic richness between uneven samples. Molecular Ecology Notes, 7, 579-582.

- 205 - van Oosterhout, C., Hutchinson, W.F., Wills, D.P.M. & Shipley, P. (2004) MICRO- CHECKER: software for identifying and correcting genotyping errors in microsatellite data. Molecular Ecology Notes, 4, 535-538. van Treuren, R. (1998) Estimating null allele frequencies at a microsatellite locus in the oystercatcher (Haematopus ostralegus). Molecular Ecology, 7, 1413-1417.

Vandermeer, J. & Carvajal, R. (2001) Metapopulation dynamics and the quality of the matrix. The American Naturalist, 158, 211-220.

Venturi, M., Hambly, R.J., Glinghammar, B., Rafter, J.J. & Rowland, I.R. (1997) Genotoxic activity in human faecal water and the role of bile acids: a study using the alkaline comet assay. Carcinogenesis, 18, 2353-2359.

Vignieri, S.N. (2005) Streams over mountains: influence of riparian connectivity on gene flow in Pacific jumping mouse (Zapus tinotatus). Molecular Ecology, 14, 1925-1937.

Waits, J.L. & Leberg, P.L. (2000) Biases associated with population estimation using molecular tagging. Animal Conservation, 3, 191-199.

Waits, L.P., Buckley, V.A., Johnson, W.E., Onorato, D. & McCarthy, T. (2007) A select panel of polymorphic microsatellite loi for individual identification of snow leopards (Panthera uncia). Molecular Ecology Notes, 7, 311-314.

Waits, L.P., Luikart, G. & Taberlet, P. (2001) Estimating the probability of identity among genotypes in natural populations: Cautions and guidelines. Molecular Ecology, 10, 249-256.

Waits, L.P. & Paetkau, D. (2005) Noninvasive genetic sampling tools for wildlife biologists: a review of applications and recommendations for accurate data collection. Journal of Wildlife Management, 69, 1419-1433.

Walker, R.S., Novaro, A.J. & Branch, L.C. (2003) Effects of patch attributes, barriers, and distance between patches on the distribution of a rock-dwelling rodent (Lagidium viscacia). Landscape Ecology, 18, 185-192.

Walston, J., Robinson, J.G., Bennett, E.L., Breitenmoser, U., da Fonseca, G.A.B., Goodrich, J., Gumal, M., Hunter, L., Johnson, A., Karanth, K.U., Leader-Williams, N., MacKinnon, K., Miquelle, D., Pattanavibool, A., Poole, C., Rabinowitz, A., Smith, J.L.D., Stokes, E.J., Stuart, S.N., Vongkhamheng, C. & Wibisono, H. (2010) Bringing the tiger back from the brink - the six percent solution. PLoS Biology, 8, e1000485.

Wang, J. (2004) Sibship reconstruction from genetic data with typing errors. Genetics, 166, 1963-1979.

Wang, J. (2012) Error method of Wang (2004) implemented in program ErrorGen.

Wang, J. (2009) A new method for estimating effective population sizes from a single sample of multilocus genotypes. Molecular Ecology, 18, 2148-2164.

Wang, J. (2005) Estimation of effective population sizes from data on genetic markers.

- 206 - Philosophical Transactions of the Royal Society of London Series B Biological Sciences, 360, 1395-1409.

Wang, J. & Caballero, A. (1999) Developments in predicting the effective size of subdivided populations. Heredity, 82, 212-226.

Waples, R.S. (2006) A bias correction for estimates of effective population size based on linkage disequilibrium at unlinked gene loci. Conservation Genetics, 7, 167-184.

Waples, R.S. & Do, C. (2008) LDNE: a program for estimating effective population size from data on linkage disequilibrium. Molecular Ecology Resources, 8, 753-756.

Waser, P.M. & Elliott, L.F. (1991) Dispersal and genetic structure in kangaroo rats. Evolution, 45, 935-943.

Wasser, S.K., Davenport, B., Ramage, E.R., Hunt, K.E., Parker, M., Clarke, C. & Stenhouse, G. (2004) Scat detection dogs in wildlife research and management: application to grizzly and black bears in the Yellowhead Ecosystem, Alberta, Canada. Canadian Journal of Zoology, 82, 475-492.

Wasser, S.K., Houston, C.S., Koehler, G.M., Cadd, G.G. & Fain, S.R. (1997) Techniques for application of faecal DNA methods to field studies of Ursids. Molecular Ecology, 6, 1091-1097.

Wasser, S.K., Mailand, C., Booth, R., Mutayoba, B., Kisamo, E., Clark, B. & Stephens, M. (2007) Using DNA to track the origin of the largest ivory seizure since the 1989 trade ban. Proceedings of the National Academy of Sciences of the United States of America, 104, 4228-4233.

Watts, P.C., Rouquette, J.R., Saccheri, I.J., Kemp, S.J. & Thompson, D.J. (2004) Molecular and ecological evidence for small-scale isolation by distance in an endangered damselfly, Coenagrion mercuriale. Molecular Ecology, 13, 2931-2945.

Watts, P.C., Rousset, F., Saccheri, I.J., Leblois, R., Kemp, S.J. & Thompson, D.J. (2007) Compatible genetic and ecological estimates of dispersal rates in insect (Coenagrion mercuriale: Odonata: Zygoptera) populations: analysis of ‘neighbourhood size’ using a more precise estimator. Molecular Ecology, 16, 737-751.

Weir, B.S. & Cockerham, C.C. (1984) Estimating F-Statistics for the analysis of population structure. Evolution, 38, 1358-1370.

Wetton, J.H., Tsang, C.S.F., Roney, C.A. & Spriggs, A.C. (2004) An extremely sensitive species-specific ARMS PCR test for the presence of tiger bone DNA. Forensic Science International, 126, 137-144.

Whitlock, M.C. (2011) G’st and D do not replace Fst. Molecular Ecology, 20,

Whitten, T., Damanik, S.J., Anwar, J. & Hisyam, N. (2000) The Ecology of Sumatra. Periplus, Singapore.

- 207 - Wibisono, H.T. & Pusparini, W. (2010) Sumatran tiger (Panthera tigris sumatrae): A review of conservation status. Integrative Zoology, 5, 313-323.

Wiegand, T., Moloney, K.A., Naves, J. & Knauer, F. (1999) Finding the missing link between landscape structure and population dynamics: a spatially explicit perspective. The American Naturalist, 154, 605-627.

Wikramanayake, E., Dinerstein, E. & Loucks, C. (2002) Terrestrial Ecoregions of the Indo- Paciﬁc: A Conservation Assessment. Island Press, USA.

Wikramanayake, E., Dinerstein, E., Seidensticker, J., Lumpkin, S., Pandav, B., Shrestha, M., Mishra, H., Ballou, J., Johnsingh, A.J.T., Chestin, I., Sunarto, S., Thinley, P., Thapa, K., Jiang, G., Elagupillay, S., Kafley, H., Pradhan, N.M.B., Jigme, K., Teak, S., Cutter, P., Aziz, M.A. & Than, U. (2011) A landscape-based conservation strategy to double the wild tiger population. Conservation Letters, 00, 1-9.

Wikramanayake, E., McKnight, M., Dinerstein, E., Joshi, A., Gurung, B. & Smith, D. (2004) Designing a conservation landscape for tigers in human-dominated environments. Conservation Biology, 18, 839-844.

Wikramanayake, E.D., Dinerstein, E., Robinson, J.G., Karanth, U.K., Rabinowitz, A., Olson, D., Mathew, T., Hedao, P., Conner, M., Hemley, G. & Bolze, D. (1998) An ecology-based method for defining priorities for large mammal conservation: The tiger as a case study. Conservation Biology, 12, 865-878.

Society, W.C. & Center for International Earth Science Information Network (2005) Last of the Wild Data v2: Global Human Footprint data set.

Wildt, D.E., Baas, E.J., Chakraborty, P.K., Wolfe, T.L. & Stewart, A.P. (1982) Influence of inbreeding on reproductive performance, ejaculate quality and testicular volume in the dog. Theriogenology, 17, 445-452.

Wildt, D.E., Bush, M., Goodrowe, K.L., Packer, C., Pusey, A.E., Brown, J.L., Joslin, P. & O’Brien, S.J. (1987) Reproductive and genetic consequences of founding isolated lion populations. Nature, 329, 328-331.

Williamson, J.E., Huebinger, R.M., Sommer, J.A., Louis Jr, E.E. & Barber, R.C. (2002) Development and cross-species amplification of 18 microsatellite markers in the Sumatran tiger (Panthera tigris sumatrae). Molecular Ecology Notes, 2, 110-112.

Wilson, G.A. & Rannala, B. (2003) Bayesian inference of recent migration rates using multilocus genotypes. Genetics, 163, 1177-1191.

With, K.A. (1997) The application of neutral landscape models in conservation biology. Conservation Biology, 11, 1069-1080.

With, K.A. (2004) Metapopulation dynamics: Perspectives from landscape ecology. In Ecology, Genetics, and Evolution of Metapopulations, Elsevier Academic Press, pp. 23-44.

With, K.A. & Crist, T.O. (1995) Critical thresholds in species’ responses to landscape

- 208 - structure. Ecology, 76, 2446-2459.

With, K.A. & King, A.W. (1999) Dispersal success on fractal landscapes: a consequence of lacunarity thresholds. Landscape Ecology, 14, 73-82.

With, K.A. & Pavuk, D.M. (2011) Habitat area trumps fragmentation effect on arthropods in an experimental landscape system. Landscape Ecology, 26, 1035-1048.

Woodroffe, R. & Ginsberg, J.R. (1998) Edge effects and the extinction of populations inside protected areas. Science, 280, 2126-2128.

Woods, J.G., Paetkau, D., Lewis, D., McLellan, B.N., Proctor, M. & Strobeck, C. (1999) Genetic tagging of free-ranging black and brown bears. Wildlife Society Bulletin, 27, 616-627.

Wright, S. (1940) Breeding structure of populations in relation to speciation. The American Naturalist, 74, 232-248.

Wright, S. (1943) Isolation by distance. Genetics, 28, 114-138.

Wright, S. (1946) Isolation by distance under diverse systems of mating. Genetics, 31, 39-59.

Wright, S. (1969) Population structure. In Volume 2: The Theory of Gene Frequencies., The University of Chicago Press, Chicago, pp. 290-344.

Wu, J., Lei, Y., Fang, S. & Wan, Q.-H. (2009) Twenty-one novel tri- and tetranucleotide microsatellite loci for the Amur tiger (Panthera tigris altaica). Conservation Genetics, 10, 567-570.

Xu, Y.C., Li, B., Li, W.S., Bai, S.Y., Li, X.P., Gu, M.B., Jing, S.Y. & Zhang, W. (2005) Individualization of tiger by using microsatellites. Forensic Science International, 151, 45-51.

Yan, L. & Zhang, D. (2004) Effects of sample size on various genetic diversity measures in population genetic study with microsatellite DNA markers. Acta Theriologica Sinica, 50, 279-290.

Zhan, X., Zheng, X., Bruford, M.W., Wei, F. & Tao, Y. (2010) A new method for quantifying genotyping errors for noninvasive genetic studies. Conservation Genetics, 11, 1567-1571.

Zhang, Z., Zhang, W., Yue, B., Shen, F., Zhang, L., Hou, R. & Zhu, M. (2006) Twelve polymorphic microsatellite loci for the South China tiger Panthera tigris amoyensis. Molecular Ecology Notes, 6, 24-26.

- 209 - Appendices

- 210 - Appendix 1 - Sample Collection & Genetic Analysis

A.1.1 Dog training protocol

Rudi, a 2 year old, male, Labrador Retriever was chosen for his high play drive and strong desire for a toy e.g. tennis ball. A good temperament was considered an important attribute for a dog working in a Muslim culture where many people are afraid of dogs and not used to working alongside them. An ideal detection dog must also want to work consistently for sustained periods of time, and should view the toy as a reward. Rudi’s training was built up from playing with a tennis ball to throw-and- retrieve games and blind searches during which he could not see where the ball was located and had to rely on scent. Two methods of reward association were used. In the first method fresh tiger scat was placed onto the ball so that the reward (the tennis ball) was associated with the target scent. In the second method, small fragments of tiger scat were placed in the search environment and the tennis ball produced as a reward when the scent was successfully located and indicated. A harness and long lead was used to signal when it was time to work.

A pilot survey was conducted in Way Kambas NP to acclimatise Rudi to field conditions. He then accompanied the field teams in the surveys of Kerinci Seblat NP and Batang Hari. It is recommended that surveys times are restricted to morning/ evening to ensure that the dog is not working during the hottest parts of the day. This has both a negative effect on the well-being of the dog and on his detection ability. Surveys were started at approximately 7.30 am and would finish early afternoon. The dog was worked in 20 minute blocks with 5-10 minute rest periods. A 30 minute rest period was also introduced after each hour to allow the teams to check the physical condition of the dog and to allow examination and/or treatment for mosquitoes, leeches, wounds etc.

- 211 - A.1.2 DNA Cleanup protocol

Each DNA eluate was subjected to a second extraction following the Qiagen ‘Cleanup of DNA’ protocol from the DNA Investigator Handbook. This protocol is designed to restore the suitability of DNA for PCR, and to increase the concentration of DNA.

1. Add up to 100µl of DNA (containing up to 10µg of DNA) to a 1.5ml microcentrifuge tube. If the volume of DNA is less than 100µl, add deionised water to a final volume of 100µl. 2. Add 10µl Buffer AW1. 3. Add 250µl Buffer AW2 and mix by pulse-vortexing for 10 seconds. 4. Transfer the entire sample from Step 3 to a QIAmp spin column (in a 2ml collection tube) without wetting the rim. Close the lid and centrifuge at 6000 x g (8000 rpm) for 1 min. Place the QIAmp spin column in a clean 2ml collection tube, discard the collection tube containing the flow-through. 5. Carefully open the QIAmp spin column and add 500µl Buffer AW2 without wetting the rim. Close the lid and centrifuge at 6000 x g (8000 rpm) for 1 min. Place the QIAmp spin column in a clean 2ml collection tube, and discard the collection tube containing the flow-through. 6. Centrifuge at full speed (20, 000 x g; 14, 000 rpm) for 3 min to dry the membrane completely. 7. Place the QIAmp spin column in a clean 1.5ml microcentrifuge tube and discard the collection tube containing the flow-through. Carefully open the lid of the QIAmp spin column and apply 20-100µl Buffer AE or distilled water to the centre of the membrane. 8. Close the lid and incubate at room temperature (15-25°C) for 1 min. Centrifuge at full speed (20, 000 x g; 14, 000 rpm) for 1 min.

- 212 - Table A.1.1 The allele size range and fluorescent tag for each microsatellite Fca locus used in this study. Marker Fluorescent label Allele size range (bp)

Fca 8 VIC 124 - 142

Fca 32 PET 192 - 202

Fca 44 6FAM 113 - 127

Fca 69 PET 103 - 111

Fca 77 VIC 137 - 157

Fca 90 6FAM 114 - 118

Fca 94 VIC 192 - 208

Fca 96 PET 208 - 218

Fca 105 NED 202 - 210

Fca 123 PET 148 - 154

Fca 129 VIC 171 - 191

Fca 139 6FAM 128 - 138

Fca 161 NED 168

Fca 176 6FAM 198 - 220

Fca 201 NED 119 - 135

Fca 211 6FAM 106 - 120

Fca 220 6FAM 200 - 206

Fca 229 VIC 154 - 172

Fca 290 VIC 208 - 216

Fca 293 VIC 190 - 192

Fca 304 PET 123 - 145

Fca 310 VIC 114 - 128

Fca 391 6FAM 202 - 218

Fca 441 PET 139 - 163

- 213 - Table A.1.2 Per-allele and per-genotype error rates for each Fca locus used in this study. Error rates were computed using a maximum likelihood method implemented in PEDANT v1.0. Per allele rate Per genotype rate Locus Allelic dropout False alleles Allelic dropout False alleles Fca 8 0.49 0.05 0.66 0.06 Fca 32 0.59 0.08 0.74 0.10 Fca 44 0.28 0.02 0.43 0.04 Fca 69 0.28 0.07 0.44 0.11 Fca 77 0.23 0.20 0.37 0.30 Fca 90 0.22 0.10 0.36 0.15 Fca 94 0.65 0.16 0.79 0.19 Fca 96 0.27 0.12 0.42 0.18 Fca 105 0.44 0.09 0.61 0.12 Fca 123 0.21 0.06 0.35 0.09 Fca 129 0.48 0.07 0.65 0.09 Fca 139 0.14 0.06 0.24 0.10 Fca 161 0.70 0.00 0.82 0.00 Fca 176 0.41 0.10 0.58 0.13 Fca 201 0.57 0.10 0.73 0.12 Fca 211 0.39 0.08 0.56 0.11 Fca 220 0.49 0.05 0.66 0.07 Fca 229 0.33 0.13 0.49 0.19 Fca 290 0.64 0.20 0.78 0.24 Fca 293 0.39 0.03 0.56 0.04 Fca 304 0.25 0.11 0.40 0.17 Fca 310 0.34 0.10 0.51 0.14 Fca 391 0.35 0.09 0.52 0.13 Fca 441 0.29 0.13 0.45 0.19 Mean 0.39 0.09 0.55 0.13

- 214 - Table A.1.3 Consensus genotypes for the 25 samples used in this study.

Loci: Fca8 Fca32 Fca44 Fca69 Fca77 Fca90 Fca94 Fca96 Fca105 Fca123 Fca129 Fca139 Fca161 Fca176 Fca201 Fca211 Fca220 Fca229 Fca290 Fca293 Fca304 Fca310 Fca391 Fca 441

TS1 , 134138 196198 121123 107107 145147 118118 200208 208208 204206 150150 175175 134138 168168 220220 129129 114114 200204 000000 000000 190190 131145 122122 202214 159163 TS2 , 134134 196196 123123 107111 145147 118118 206208 208210 210210 000000 000000 138138 168168 206220 129129 106114 204204 164172 216216 192192 139145 122128 202210 147163 TS3 , 128138 196198 115123 107107 145147 116118 194200 210218 204206 150154 175175 128138 168168 218220 119127 106114 200204 162172 000000 190192 145145 000000 202214 147163 TS4 , 134138 000000 115123 109109 145147 118118 194208 000000 206210 000000 175175 128138 000000 206220 000000 106114 204204 162172 000000 192192 131139 122128 202202 155163 TS5 , 138138 198198 121123 107109 145147 116116 194194 210212 210210 150152 187187 128138 168168 206206 129129 114114 200200 162172 210216 192192 131133 000000 202206 147155 TS6 , 128138 196198 115123 107107 000000 116118 198206 210216 204206 150154 175191 134138 000000 220220 119129 106114 200204 162172 212216 192192 131145 122122 202206 159163 TS7 , 000000 192196 121121 107111 147147 116118 194194 000000 000000 000000 000000 000000 000000 000000 000000 114114 000000 000000 000000 000000 131137 122122 202202 000000 TS8 , 000000 000000 113121 109111 000000 116118 200204 000000 204206 148148 175175 128138 168168 218218 119129 114116 204204 162172 212216 192192 131137 122122 202202 147155 TS10 , 140142 000000 113121 107107 145145 116118 200200 000000 000000 000000 175187 128138 000000 206218 000000 110114 200204 172172 000000 000000 131131 122128 202210 143147 ZSL1 , 138138 194196 115123 107109 147157 118118 000000 210212 204206 150150 189191 128138 000000 000000 000000 114114 200204 162172 000000 192192 131131 122122 000000 147155 R1 , 000000 192194 113121 000000 147157 000000 000000 000000 000000 000000 000000 000000 000000 000000 000000 110110 000000 000000 212216 000000 137137 000000 000000 147159 R9 , 136136 194194 115115 107111 143147 000000 192198 000000 206208 000000 000000 000000 000000 000000 000000 110114 000000 172172 000000 000000 000000 122128 000000 147159 KS15 , 136136 198200 121123 109109 000000 116118 192192 000000 206208 000000 000000 000000 168168 218218 000000 106106 202204 162172 000000 000000 131131 000000 000000 000000 KS24 , 134138 194196 000000 000000 147147 116118 198198 000000 000000 000000 000000 000000 000000 000000 000000 000000 202202 000000 000000 000000 000000 000000 000000 000000 BH66 , 138142 196200 121123 107109 145147 118118 198200 210216 204206 150152 189191 128138 168168 218220 129129 114114 200204 162172 216216 192192 131131 122128 202210 159163 BH74 , 138142 196198 123123 109109 145157 116118 198200 210212 206206 000000 000000 000000 168168 206218 127127 106110 200204 172172 000000 000000 123131 122128 202206 147147 UM3 , 136140 196196 121121 107109 000000 000000 202204 210212 206206 000000 175191 000000 168168 000000 000000 112114 000000 154164 000000 000000 141143 000000 206214 000000 UM4 , 136138 194196 113121 107109 000000 116118 198200 000000 000000 000000 175191 000000 168168 206218 129129 106114 200200 164172 000000 000000 131131 122128 210214 143155 UM5 , 136140 000000 121121 107109 000000 116118 200202 210212 204206 150150 000000 128138 168168 000000 000000 106114 200202 154164 000000 190192 141143 122122 206214 000000 WK2 , 000000 194194 000000 000000 145157 116118 000000 000000 208210 000000 000000 000000 000000 000000 000000 000000 000000 164172 000000 000000 135137 000000 000000 143143 WK3 , 138138 192194 121123 000000 143145 118118 000000 000000 210210 000000 000000 000000 000000 000000 000000 000000 200202 000000 000000 000000 123131 000000 000000 000000 WK4 , 138138 000000 000000 107109 145157 118118 198198 212214 206210 000000 187189 128138 000000 000000 127127 110112 200200 000000 000000 000000 131133 122122 202210 000000 WK5 , 000000 192192 121121 107109 145157 118118 000000 000000 208208 000000 000000 138138 000000 000000 129133 000000 200202 000000 210210 000000 000000 000000 210218 000000 WK6 , 000000 000000 121123 109109 145157 000000 000000 000000 000000 000000 000000 000000 000000 000000 133133 112114 200200 000000 000000 000000 000000 000000 210218 000000 BB1 , 000000 000000 121121 000000 143157 000000 000000 000000 204204 000000 000000 000000 000000 000000 000000 112116 206206 162172 000000 000000 000000 000000 000000 151151

- 215 - Appendix 2 - Sampling effects

- 216 - Figure A.2.1 Variance in mean number of alleles per locus, expected heterozygosity and pairwise FST with a reduction in sample size.

- 217 - Figure A.2.2 Variation in the modal value of !K with a reduction in sample size. The modal value of !K reflects the strength of STRUCTURE support for the estimated value of k.

- 218 - Appendix 3 - Inﬂuence of Genotyping Error

Table A.3.1 Values of the mean number of alleles per locus (NA), expected heterozygosity (He), pairwise FST and the number of clusters (k) with an increase in error rate (#) and a reduction in sample size (n).

Allelic dropout False alleles

n ! NA He FST ln P(X|K) "K n ! NA He FST ln P(X|K) "K 100 0 10.6 ± 0.06 0.83 ± 0.001 0.15 ± 0.002 3 3 100 0 10.6 ± 0.06 0.83 ± 0.001 0.15 ± 0.002 3 3 100 0.5 10.4 ± 0.06 0.83 ± 0.001 0.16 ± 0.002 3 3 100 0.04 16.4 ± 0.08 0.84 ± 0.001 0.14 ± 0.002 3 3 100 0.9 10.1 ± 0.06 0.82 ± 0.001 0.16 ± 0.002 3 3 100 0.10 23.6 ± 0.07 0.86 ± 0.001 0.13 ± 0.002 3 3 10 0 7.3 ± 0.13 0.83 ± 0.006 0.16 ± 0.009 3 3 10 0 7.3 ± 0.12 0.83 ± 0.005 0.15 ± 0.008 3 3 10 0.5 6.5 ± 0.12 0.81 ± 0.007 0.18 ± 0.009 3 3 10 0.04 7.8 ± 0.12 0.84 ± 0.005 0.14 ± 0.007 3 3 10 0.9 5.7 ± 0.10 0.79 ± 0.007 0.20 ± 0.009 > 3 3 10 0.10 8.7 ± 0.15 0.86 ± 0.005 0.12 ± 0.007 3 3 5 0 5.4 ± 0.15 0.83 ± 0.012 0.16 ± 0.016 NA NA 5 0 5.4 ± 0.16 0.83 ± 0.010 0.15 ± 0.015 NA NA 5 0.5 4.6 ± 0.14 0.78 ± 0.014 0.20 ± 0.019 NA NA 5 0.04 5.7 ± 0.17 0.84 ± 0.010 0.14 ± 0.014 NA NA 5 0.9 3.8 ± 0.12 0.75 ± 0.015 0.24 ± 0.021 NA NA 5 0.10 8.7 ± 0.15 0.86 ± 0.010 0.13 ± 0.013 NA NA

- 219 - Missing alleles Combined error

n ! NA He FST ln P(X|K) "K n ! NA He FST ln P(X|K) "K 100 0 10.7 ± 0.06 0.83 ± 0.001 0.16 ± 0.002 3 3 100 0 10.7 ± 0.06 0.83 ± 0.001 0.15 ± 0.002 3 3 100 0.5 10.3 ± 0.05 0.83 ± 0.001 0.15 ± 0.002 3 3 100 0.5 13.5 ± 0.08 0.84 ± 0.001 0.14 ± 0.002 3 3 100 0.9 7.1 ± 0.04 0.83 ± 0.002 0.12 ± 0.006 2 2 100 0.7 11.9 ± 0.06 0.84 ± 0.001 0.13 ± 0.002 >3 3 10 0 7.3 ± 0.14 0.83 ± 0.006 0.16 ± 0.009 3 3 10 0 7.3 ± 0.14 0.83 ± 0.006 0.15 ± 0.008 3 3 10 0.5 5.3 ± 0.14 0.83 ± 0.010 0.11 ± 0.020 1 2 10 0.5 4.9 ± 0.14 0.79 ± 0.011 0.14 ± 0.068 3 2 10 0.9 1.6 ± 0.11 0.53 ± 0.040 -0.34± 0.280 1 2 10 0.7 3.2 ± 0.11 0.68 ± 0.022 0.11 ± 0.049 1 2 5 0 5.5 ± 0.14 0.83 ± 0.011 0.15 ± 0.020 NA NA 5 0 5.5 ± 0.14 0.83 ± 0.010 0.15 ± 0.015 NA NA 5 0.5 3.4 ± 0.16 0.80 ± 0.028 0.06 ± 0.050 NA NA 5 0.5 3.0 ± 0.16 0.69 ± 0.036 0.12 ± 0.059 NA NA 5 0.9 0.83 ± 0.12 0.33 ± 0.055 -2.2x1014 NA NA 5 0.7 1.8 ± 0.14 0.45 ± 0.042 0.06 ± 0.160 NA NA ± 1.9x1015

- 220 - Appendix 4 - Population Genetics

Table A.4.1 Results of a locus-by-locus test for Hardy-Weinberg equilibrium performed in GENEPOP. Loci not in Hardy-Weinberg equilibrium are indicated with an asterisk (*).

Locus FIS estimate Fca8 0.1887 Fca32 0.1356 Fca44 0.0667 Fca69 -0.0022 Fca77 -0.1828 Fca90 -0.2294 Fca94 0.2184 Fca96 -0.2048 Fca105* 0.1429 Fca123 0.1765 Fca129 0.1630 Fca139 -0.5349 Fca161 monomorphic Fca176 0.1630 Fca201* 0.5056 Fca211 0.0328 Fca220 0.1788 Fca229* -0.2911 Fca290 0.0943 Fca293 0.4194 Fca304 0.1349 Fca310 -0.3000 Fca391 -0.1200 Fca441 0.0152 Global 0.0372

- 221 - Table A.4.2 Results of a Fisher's exact test for linkage disequilibrium performed in GENEPOP. Locus pairs in linkage disequilibrium are indicated by asterisks (**).

8 32 44 69 77 90 94 96 105 123 129 139 161 176 201 211 220 229 290 293 304 310 391 441 8 - ** 32 - ** 44 - ** 69 - ** 77 - ** ** 90 - 94 - 96 - 105 - ** 123 - 129 - 139 - 161 - 176 - (**) 201 - 211 - 220 - 229 - 290 - 293 - 304 - ** 310 - 391 - 441 - - 222 - Table A.4.3 Locus-by-locus values of observed and expected heterozygosity. Locus Sample size Observed Expected Unbiased heterozygosity heterozygosity expected heterozygosity Fca 8 18 0.61 0.73 0.75 Fca 32 18 0.67 0.75 0.77 Fca 44 22 0.64 0.67 0.68 Fca 69 20 0.60 0.58 0.60 Fca 77 19 0.84 0.70 0.72 Fca 90 20 0.55 0.44 0.45 Fca 94 18 0.67 0.82 0.85 Fca 96 11 0.91 0.73 0.76 Fca 105 19 0.63 0.72 0.73 Fca 123 8 0.50 0.56 0.60 Fca 129 12 0.58 0.66 0.69 Fca 139 13 0.85 0.54 0.56 Fca 161 11 0.00 0.00 0.00 Fca 176 12 0.58 0.66 0.69 Fca 201 12 0.33 0.63 0.66 Fca 211 21 0.67 0.67 0.69 Fca 220 20 0.55 0.65 0.67 Fca 229 17 0.82 0.63 0.64 Fca 290 7 0.57 0.58 0.63 Fca 293 10 0.20 0.32 0.34 Fca 304 20 0.65 0.73 0.75 Fca 310 14 0.50 0.38 0.39 Fca 391 17 0.82 0.72 0.74 Fca 441 16 0.81 0.80 0.83 Mean 15.6 0.61 ± 0.08 0.61 ± 0.07 0.63 ± 0.07

- 223 - Table A.4.4 Results of a test for population structure using AMOVA analysis performed in ARLEQUIN.

Grouping Among populations FST 2 % a % variation NWE-S 1.12 14.8 0.15 NS-EW 0.70 8.7 0.09 NW-E-S 0.64 9.2 0.09 NE-W-S 0.64 9.1 0.09 N-S-EW 0.62 9.2 0.09 N-E-W-S 0.57 8.7 0.09 NS-E-W 0.55 7.1 0.07 N-E-WS 0.42 5.6 0.06 NWS-E 0.39 5.1 0.05 NE-WS 0.36 4.6 0.05 WES-N 0.24 3.2 0.03 N-W-ES 0.19 2.6 0.03 NW-ES 0.16 2.1 0.02 ESN-W 0.10 1.3 0.01

- 224 - Appendix 5 - Population Structure

Table A.5.1 Summary settings for the Bayesian clustering programs used in this study.

Program General settings Spatial settings Inference of k Cluster membership assignment

STRUCTURE v2.3.3 Burn-in: 30,000 Locprior = TRUE Highest value of Average across 6 Iterations: 1x10 L(K) runs using CLUMPP Mixed ancestry Modal value of !K1 Correlated allele frequencies 5 runs at k = 1-10

TESS v2.3.1 Burn-in: 30,000 , = 0.6 Value of kmax with Average across 3 Sweeps: 100,000 the lowest DIC runs using CLUMPP CAR admixture model2 1 run at k = 2-10 to

determine kmax, then 100 runs at kmax

BAPS v5.4 Spatial clustering of Individual Highest log(ml) BAPS output individuals geographic value No admixture coordinates Highest posterior Vector of maximum probability for values of k (5 5 5 5 value of k4 5 10 10 10 10 10 15 15 15 15 15)

GENELAND Burn-in: 200 Spatial=TRUE Modal value of k 10 cycles at modal Thinning: 100 Spatial from plot of MCMC k; varnpop=FALSE, Iterations: 200,000 uncertainty: output npopinit=modal K freq.model= delta.coord=10 Then average ”Uncorrelated” across runs using npopmax=10 CLUMPP rate.max=25 nb.nuclei.max=75 1Earl 2009 2Durand et al. 2009 3François et al. 2006; Durand et al. 2009 4Latch et al. 2006

- 225 - Figure A.5.1 Resistance map using the human footprint index as a base layer. Resistance values are equivalent to the human footprint score.

- 226 - Figure A.5.2 Resistance map using FAO land cover data as a base layer. Resistance values are outlined in Table A.5.2

- 227 - Table A.5.2 The FAO land cover classification system (LCCS) for Sumatra. Resistance values were assigned based on a relative scale of tiger habitat preference from 1 (high preference, low resistance) to 5 (low preference, high resistance).

Description LCCS index Resistance value Post-flooding or irrigated croplands 11 3 Rainfed croplands 14 3 Mosaic cropland (50-70%) / vegetation (grassland/ 20 3 shrubland/forest) (20-50%) Mosaic vegetation (grassland/shrubland/forest) 30 2 (50-70%) / cropland (20-50%) Closed to open (>15%) broadleaved evergreen and/or 40 1 semi-deciduous forest (>5m) Closed (>40%) broadleaved deciduous forest (>5m) 50 1 Mosaic forest or shrubland (50-70%) and grassland 110 1 (20-50%) Closed to open (>15%) shrubland (<5m) 130 1 Closed (>40%) broadleaved forest regularly flooded, 160 3 fresh water Closed (>40%) broadleaved semi-deciduous and/or 170 3 evergreen forest regularly flooded, saline water Artificial surfaces and associated areas (Urban 190 5 areas >50%) Water bodies 210 3

- 228 - Table A.5.3 The settings used for CIRCUITSCAPE analysis Model Setting Source/ground modeling mode Pairwise: iterate across all pairs in focal node file Pairwise mode options Focal Regions: focal nodes may contain multiple cells Input habitat data Habitat data specify per-cell resistances Raster short-circuit region map Same as focal node file Cell connection scheme Connect eight neighbours Cell connection calculation Average resistance

- 229 -